blob: 99a81369e43ddb992322fbd7df73bc9dd3a250a0 [file] [log] [blame]
Chris Mason0b86a832008-03-24 15:01:56 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Mason8a4b83c2008-03-24 15:02:07 -040021#include <linux/buffer_head.h>
Chris Masonf2d8d742008-04-21 10:03:05 -040022#include <linux/blkdev.h>
Chris Masonb765ead2009-04-03 10:27:10 -040023#include <linux/iocontext.h>
Ben Hutchings6f88a442010-12-29 14:55:03 +000024#include <linux/capability.h>
Stefan Behrens442a4f62012-05-25 16:06:08 +020025#include <linux/ratelimit.h>
Ilya Dryomov59641012012-01-16 22:04:48 +020026#include <linux/kthread.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050027#include <linux/raid/pq.h>
Stefan Behrens803b2f52013-08-15 17:11:21 +020028#include <linux/semaphore.h>
Andy Shevchenko8da4b8c2016-05-20 17:01:00 -070029#include <linux/uuid.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050030#include <asm/div64.h>
Chris Mason0b86a832008-03-24 15:01:56 -040031#include "ctree.h"
32#include "extent_map.h"
33#include "disk-io.h"
34#include "transaction.h"
35#include "print-tree.h"
36#include "volumes.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050037#include "raid56.h"
Chris Mason8b712842008-06-11 16:50:36 -040038#include "async-thread.h"
Stefan Behrens21adbd52011-11-09 13:44:05 +010039#include "check-integrity.h"
Josef Bacik606686e2012-06-04 14:03:51 -040040#include "rcu-string.h"
Miao Xie3fed40c2012-09-13 04:51:36 -060041#include "math.h"
Stefan Behrens8dabb742012-11-06 13:15:27 +010042#include "dev-replace.h"
Anand Jain99994cd2014-06-03 11:36:00 +080043#include "sysfs.h"
Chris Mason0b86a832008-03-24 15:01:56 -040044
Zhao Leiaf902042015-09-15 21:08:06 +080045const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
47 .sub_stripes = 2,
48 .dev_stripes = 1,
49 .devs_max = 0, /* 0 == as many as possible */
50 .devs_min = 4,
Zhao Lei8789f4f2015-09-15 21:08:07 +080051 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080052 .devs_increment = 2,
53 .ncopies = 2,
54 },
55 [BTRFS_RAID_RAID1] = {
56 .sub_stripes = 1,
57 .dev_stripes = 1,
58 .devs_max = 2,
59 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080060 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080061 .devs_increment = 2,
62 .ncopies = 2,
63 },
64 [BTRFS_RAID_DUP] = {
65 .sub_stripes = 1,
66 .dev_stripes = 2,
67 .devs_max = 1,
68 .devs_min = 1,
Zhao Lei8789f4f2015-09-15 21:08:07 +080069 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080070 .devs_increment = 1,
71 .ncopies = 2,
72 },
73 [BTRFS_RAID_RAID0] = {
74 .sub_stripes = 1,
75 .dev_stripes = 1,
76 .devs_max = 0,
77 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080078 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080079 .devs_increment = 1,
80 .ncopies = 1,
81 },
82 [BTRFS_RAID_SINGLE] = {
83 .sub_stripes = 1,
84 .dev_stripes = 1,
85 .devs_max = 1,
86 .devs_min = 1,
Zhao Lei8789f4f2015-09-15 21:08:07 +080087 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080088 .devs_increment = 1,
89 .ncopies = 1,
90 },
91 [BTRFS_RAID_RAID5] = {
92 .sub_stripes = 1,
93 .dev_stripes = 1,
94 .devs_max = 0,
95 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080096 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080097 .devs_increment = 1,
98 .ncopies = 2,
99 },
100 [BTRFS_RAID_RAID6] = {
101 .sub_stripes = 1,
102 .dev_stripes = 1,
103 .devs_max = 0,
104 .devs_min = 3,
Zhao Lei8789f4f2015-09-15 21:08:07 +0800105 .tolerated_failures = 2,
Zhao Leiaf902042015-09-15 21:08:06 +0800106 .devs_increment = 1,
107 .ncopies = 3,
108 },
109};
110
Colin Ian Kingfb75d852016-01-19 00:05:28 +0000111const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
Zhao Leiaf902042015-09-15 21:08:06 +0800112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
119};
120
David Sterba621292b2016-02-15 16:28:03 +0100121/*
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
125 */
126const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134};
135
Yan Zheng2b820322008-11-17 21:11:30 -0500136static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200140static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
Eric Sandeen48a3b632013-04-25 20:41:01 +0000141static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200142static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
Yan Zheng2b820322008-11-17 21:11:30 -0500143
Miao Xie67a2c452014-09-03 21:35:43 +0800144DEFINE_MUTEX(uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400145static LIST_HEAD(fs_uuids);
Anand Jainc73eccf2015-03-10 06:38:30 +0800146struct list_head *btrfs_get_fs_uuids(void)
147{
148 return &fs_uuids;
149}
Chris Mason8a4b83c2008-03-24 15:02:07 -0400150
Ilya Dryomov2208a372013-08-12 14:33:03 +0300151static struct btrfs_fs_devices *__alloc_fs_devices(void)
152{
153 struct btrfs_fs_devices *fs_devs;
154
David Sterba78f2c9e2016-02-11 14:25:38 +0100155 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300156 if (!fs_devs)
157 return ERR_PTR(-ENOMEM);
158
159 mutex_init(&fs_devs->device_list_mutex);
160
161 INIT_LIST_HEAD(&fs_devs->devices);
Miao Xie935e5cc2014-09-03 21:35:33 +0800162 INIT_LIST_HEAD(&fs_devs->resized_devices);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300163 INIT_LIST_HEAD(&fs_devs->alloc_list);
164 INIT_LIST_HEAD(&fs_devs->list);
165
166 return fs_devs;
167}
168
169/**
170 * alloc_fs_devices - allocate struct btrfs_fs_devices
171 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
172 * generated.
173 *
174 * Return: a pointer to a new &struct btrfs_fs_devices on success;
175 * ERR_PTR() on error. Returned struct is not linked onto any lists and
176 * can be destroyed with kfree() right away.
177 */
178static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
179{
180 struct btrfs_fs_devices *fs_devs;
181
182 fs_devs = __alloc_fs_devices();
183 if (IS_ERR(fs_devs))
184 return fs_devs;
185
186 if (fsid)
187 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
188 else
189 generate_random_uuid(fs_devs->fsid);
190
191 return fs_devs;
192}
193
Yan Zhenge4404d62008-12-12 10:03:26 -0500194static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
195{
196 struct btrfs_device *device;
197 WARN_ON(fs_devices->opened);
198 while (!list_empty(&fs_devices->devices)) {
199 device = list_entry(fs_devices->devices.next,
200 struct btrfs_device, dev_list);
201 list_del(&device->dev_list);
Josef Bacik606686e2012-06-04 14:03:51 -0400202 rcu_string_free(device->name);
Yan Zhenge4404d62008-12-12 10:03:26 -0500203 kfree(device);
204 }
205 kfree(fs_devices);
206}
207
Lukas Czernerb8b8ff52012-12-06 19:25:48 +0000208static void btrfs_kobject_uevent(struct block_device *bdev,
209 enum kobject_action action)
210{
211 int ret;
212
213 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
214 if (ret)
Frank Holtonefe120a2013-12-20 11:37:06 -0500215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
Lukas Czernerb8b8ff52012-12-06 19:25:48 +0000216 action,
217 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218 &disk_to_dev(bdev->bd_disk)->kobj);
219}
220
Jeff Mahoney143bede2012-03-01 14:56:26 +0100221void btrfs_cleanup_fs_uuids(void)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400222{
223 struct btrfs_fs_devices *fs_devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400224
Yan Zheng2b820322008-11-17 21:11:30 -0500225 while (!list_empty(&fs_uuids)) {
226 fs_devices = list_entry(fs_uuids.next,
227 struct btrfs_fs_devices, list);
228 list_del(&fs_devices->list);
Yan Zhenge4404d62008-12-12 10:03:26 -0500229 free_fs_devices(fs_devices);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400230 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400231}
232
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300233static struct btrfs_device *__alloc_device(void)
234{
235 struct btrfs_device *dev;
236
David Sterba78f2c9e2016-02-11 14:25:38 +0100237 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300238 if (!dev)
239 return ERR_PTR(-ENOMEM);
240
241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
Miao Xie935e5cc2014-09-03 21:35:33 +0800243 INIT_LIST_HEAD(&dev->resized_list);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300244
245 spin_lock_init(&dev->io_lock);
246
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
Miao Xieaddc3fa2014-07-24 11:37:11 +0800249 atomic_set(&dev->dev_stats_ccnt, 0);
Sebastian Andrzej Siewior546bed62016-01-15 14:37:15 +0100250 btrfs_device_data_ordered_init(dev);
Mel Gormand0164ad2015-11-06 16:28:21 -0800251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300253
254 return dev;
255}
256
Chris Masona1b32a52008-09-05 16:09:51 -0400257static noinline struct btrfs_device *__find_device(struct list_head *head,
258 u64 devid, u8 *uuid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400259{
260 struct btrfs_device *dev;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400261
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500262 list_for_each_entry(dev, head, dev_list) {
Chris Masona4437552008-04-18 10:29:38 -0400263 if (dev->devid == devid &&
Chris Mason8f18cf12008-04-25 16:53:30 -0400264 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400265 return dev;
Chris Masona4437552008-04-18 10:29:38 -0400266 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400267 }
268 return NULL;
269}
270
Chris Masona1b32a52008-09-05 16:09:51 -0400271static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400272{
Chris Mason8a4b83c2008-03-24 15:02:07 -0400273 struct btrfs_fs_devices *fs_devices;
274
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500275 list_for_each_entry(fs_devices, &fs_uuids, list) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400276 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
277 return fs_devices;
278 }
279 return NULL;
280}
281
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100282static int
283btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284 int flush, struct block_device **bdev,
285 struct buffer_head **bh)
286{
287 int ret;
288
289 *bdev = blkdev_get_by_path(device_path, flags, holder);
290
291 if (IS_ERR(*bdev)) {
292 ret = PTR_ERR(*bdev);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100293 goto error;
294 }
295
296 if (flush)
297 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298 ret = set_blocksize(*bdev, 4096);
299 if (ret) {
300 blkdev_put(*bdev, flags);
301 goto error;
302 }
303 invalidate_bdev(*bdev);
304 *bh = btrfs_read_dev_super(*bdev);
Anand Jain92fc03f2015-08-14 18:32:51 +0800305 if (IS_ERR(*bh)) {
306 ret = PTR_ERR(*bh);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100307 blkdev_put(*bdev, flags);
308 goto error;
309 }
310
311 return 0;
312
313error:
314 *bdev = NULL;
315 *bh = NULL;
316 return ret;
317}
318
Chris Masonffbd5172009-04-20 15:50:09 -0400319static void requeue_list(struct btrfs_pending_bios *pending_bios,
320 struct bio *head, struct bio *tail)
321{
322
323 struct bio *old_head;
324
325 old_head = pending_bios->head;
326 pending_bios->head = head;
327 if (pending_bios->tail)
328 tail->bi_next = old_head;
329 else
330 pending_bios->tail = tail;
331}
332
Chris Mason8b712842008-06-11 16:50:36 -0400333/*
334 * we try to collect pending bios for a device so we don't get a large
335 * number of procs sending bios down to the same device. This greatly
336 * improves the schedulers ability to collect and merge the bios.
337 *
338 * But, it also turns into a long list of bios to process and that is sure
339 * to eventually make the worker thread block. The solution here is to
340 * make some progress and then put this work struct back at the end of
341 * the list if the block device is congested. This way, multiple devices
342 * can make progress from a single worker thread.
343 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100344static noinline void run_scheduled_bios(struct btrfs_device *device)
Chris Mason8b712842008-06-11 16:50:36 -0400345{
346 struct bio *pending;
347 struct backing_dev_info *bdi;
Chris Masonb64a2852008-08-20 13:39:41 -0400348 struct btrfs_fs_info *fs_info;
Chris Masonffbd5172009-04-20 15:50:09 -0400349 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -0400350 struct bio *tail;
351 struct bio *cur;
352 int again = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400353 unsigned long num_run;
Chris Masond644d8a2009-06-09 15:59:22 -0400354 unsigned long batch_run = 0;
Chris Masonb64a2852008-08-20 13:39:41 -0400355 unsigned long limit;
Chris Masonb765ead2009-04-03 10:27:10 -0400356 unsigned long last_waited = 0;
Chris Masond84275c2009-06-09 15:39:08 -0400357 int force_reg = 0;
Miao Xie0e588852011-08-05 09:32:37 +0000358 int sync_pending = 0;
Chris Mason211588a2011-04-19 20:12:40 -0400359 struct blk_plug plug;
360
361 /*
362 * this function runs all the bios we've collected for
363 * a particular device. We don't want to wander off to
364 * another device without first sending all of these down.
365 * So, setup a plug here and finish it off before we return
366 */
367 blk_start_plug(&plug);
Chris Mason8b712842008-06-11 16:50:36 -0400368
Chris Masonbedf7622009-04-03 10:32:58 -0400369 bdi = blk_get_backing_dev_info(device->bdev);
Chris Masonb64a2852008-08-20 13:39:41 -0400370 fs_info = device->dev_root->fs_info;
371 limit = btrfs_async_submit_limit(fs_info);
372 limit = limit * 2 / 3;
373
Chris Mason8b712842008-06-11 16:50:36 -0400374loop:
375 spin_lock(&device->io_lock);
376
Chris Masona6837052009-02-04 09:19:41 -0500377loop_lock:
Chris Masond84275c2009-06-09 15:39:08 -0400378 num_run = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400379
Chris Mason8b712842008-06-11 16:50:36 -0400380 /* take all the bios off the list at once and process them
381 * later on (without the lock held). But, remember the
382 * tail and other pointers so the bios can be properly reinserted
383 * into the list if we hit congestion
384 */
Chris Masond84275c2009-06-09 15:39:08 -0400385 if (!force_reg && device->pending_sync_bios.head) {
Chris Masonffbd5172009-04-20 15:50:09 -0400386 pending_bios = &device->pending_sync_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400387 force_reg = 1;
388 } else {
Chris Masonffbd5172009-04-20 15:50:09 -0400389 pending_bios = &device->pending_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400390 force_reg = 0;
391 }
Chris Masonffbd5172009-04-20 15:50:09 -0400392
393 pending = pending_bios->head;
394 tail = pending_bios->tail;
Chris Mason8b712842008-06-11 16:50:36 -0400395 WARN_ON(pending && !tail);
Chris Mason8b712842008-06-11 16:50:36 -0400396
397 /*
398 * if pending was null this time around, no bios need processing
399 * at all and we can stop. Otherwise it'll loop back up again
400 * and do an additional check so no bios are missed.
401 *
402 * device->running_pending is used to synchronize with the
403 * schedule_bio code.
404 */
Chris Masonffbd5172009-04-20 15:50:09 -0400405 if (device->pending_sync_bios.head == NULL &&
406 device->pending_bios.head == NULL) {
Chris Mason8b712842008-06-11 16:50:36 -0400407 again = 0;
408 device->running_pending = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400409 } else {
410 again = 1;
411 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400412 }
Chris Masonffbd5172009-04-20 15:50:09 -0400413
414 pending_bios->head = NULL;
415 pending_bios->tail = NULL;
416
Chris Mason8b712842008-06-11 16:50:36 -0400417 spin_unlock(&device->io_lock);
418
Chris Masond3977122009-01-05 21:25:51 -0500419 while (pending) {
Chris Masonffbd5172009-04-20 15:50:09 -0400420
421 rmb();
Chris Masond84275c2009-06-09 15:39:08 -0400422 /* we want to work on both lists, but do more bios on the
423 * sync list than the regular list
424 */
425 if ((num_run > 32 &&
426 pending_bios != &device->pending_sync_bios &&
427 device->pending_sync_bios.head) ||
428 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
429 device->pending_bios.head)) {
Chris Masonffbd5172009-04-20 15:50:09 -0400430 spin_lock(&device->io_lock);
431 requeue_list(pending_bios, pending, tail);
432 goto loop_lock;
433 }
434
Chris Mason8b712842008-06-11 16:50:36 -0400435 cur = pending;
436 pending = pending->bi_next;
437 cur->bi_next = NULL;
Chris Masonb64a2852008-08-20 13:39:41 -0400438
David Sterbaee863952015-02-16 19:41:40 +0100439 /*
440 * atomic_dec_return implies a barrier for waitqueue_active
441 */
Josef Bacik66657b32012-08-01 15:36:24 -0400442 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
Chris Masonb64a2852008-08-20 13:39:41 -0400443 waitqueue_active(&fs_info->async_submit_wait))
444 wake_up(&fs_info->async_submit_wait);
Chris Mason492bb6de2008-07-31 16:29:02 -0400445
Jens Axboedac56212015-04-17 16:23:59 -0600446 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
Chris Masond644d8a2009-06-09 15:59:22 -0400447
Chris Mason2ab1ba62011-08-04 14:28:36 -0400448 /*
449 * if we're doing the sync list, record that our
450 * plug has some sync requests on it
451 *
452 * If we're doing the regular list and there are
453 * sync requests sitting around, unplug before
454 * we add more
455 */
456 if (pending_bios == &device->pending_sync_bios) {
457 sync_pending = 1;
458 } else if (sync_pending) {
459 blk_finish_plug(&plug);
460 blk_start_plug(&plug);
461 sync_pending = 0;
462 }
463
Mike Christie4e49ea42016-06-05 14:31:41 -0500464 btrfsic_submit_bio(cur);
Chris Mason5ff7ba32010-03-15 10:21:30 -0400465 num_run++;
466 batch_run++;
David Sterba853d8ec2015-01-08 15:15:19 +0100467
468 cond_resched();
Chris Mason8b712842008-06-11 16:50:36 -0400469
470 /*
471 * we made progress, there is more work to do and the bdi
472 * is now congested. Back off and let other work structs
473 * run instead
474 */
Chris Mason57fd5a52009-08-07 09:59:15 -0400475 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
Chris Mason5f2cc082008-11-07 18:22:45 -0500476 fs_info->fs_devices->open_devices > 1) {
Chris Masonb765ead2009-04-03 10:27:10 -0400477 struct io_context *ioc;
Chris Mason8b712842008-06-11 16:50:36 -0400478
Chris Masonb765ead2009-04-03 10:27:10 -0400479 ioc = current->io_context;
480
481 /*
482 * the main goal here is that we don't want to
483 * block if we're going to be able to submit
484 * more requests without blocking.
485 *
486 * This code does two great things, it pokes into
487 * the elevator code from a filesystem _and_
488 * it makes assumptions about how batching works.
489 */
490 if (ioc && ioc->nr_batch_requests > 0 &&
491 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
492 (last_waited == 0 ||
493 ioc->last_waited == last_waited)) {
494 /*
495 * we want to go through our batch of
496 * requests and stop. So, we copy out
497 * the ioc->last_waited time and test
498 * against it before looping
499 */
500 last_waited = ioc->last_waited;
David Sterba853d8ec2015-01-08 15:15:19 +0100501 cond_resched();
Chris Masonb765ead2009-04-03 10:27:10 -0400502 continue;
503 }
Chris Mason8b712842008-06-11 16:50:36 -0400504 spin_lock(&device->io_lock);
Chris Masonffbd5172009-04-20 15:50:09 -0400505 requeue_list(pending_bios, pending, tail);
Chris Masona6837052009-02-04 09:19:41 -0500506 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400507
508 spin_unlock(&device->io_lock);
Qu Wenruoa8c93d4e2014-02-28 10:46:08 +0800509 btrfs_queue_work(fs_info->submit_workers,
510 &device->work);
Chris Mason8b712842008-06-11 16:50:36 -0400511 goto done;
512 }
Chris Masond85c8a6f2011-12-15 15:38:41 -0500513 /* unplug every 64 requests just for good measure */
514 if (batch_run % 64 == 0) {
515 blk_finish_plug(&plug);
516 blk_start_plug(&plug);
517 sync_pending = 0;
518 }
Chris Mason8b712842008-06-11 16:50:36 -0400519 }
Chris Masonffbd5172009-04-20 15:50:09 -0400520
Chris Mason51684082010-03-10 15:33:32 -0500521 cond_resched();
522 if (again)
523 goto loop;
524
525 spin_lock(&device->io_lock);
526 if (device->pending_bios.head || device->pending_sync_bios.head)
527 goto loop_lock;
528 spin_unlock(&device->io_lock);
529
Chris Mason8b712842008-06-11 16:50:36 -0400530done:
Chris Mason211588a2011-04-19 20:12:40 -0400531 blk_finish_plug(&plug);
Chris Mason8b712842008-06-11 16:50:36 -0400532}
533
Christoph Hellwigb2950862008-12-02 09:54:17 -0500534static void pending_bios_fn(struct btrfs_work *work)
Chris Mason8b712842008-06-11 16:50:36 -0400535{
536 struct btrfs_device *device;
537
538 device = container_of(work, struct btrfs_device, work);
539 run_scheduled_bios(device);
540}
541
Anand Jain4fde46f2015-06-17 21:10:48 +0800542
543void btrfs_free_stale_device(struct btrfs_device *cur_dev)
544{
545 struct btrfs_fs_devices *fs_devs;
546 struct btrfs_device *dev;
547
548 if (!cur_dev->name)
549 return;
550
551 list_for_each_entry(fs_devs, &fs_uuids, list) {
552 int del = 1;
553
554 if (fs_devs->opened)
555 continue;
556 if (fs_devs->seeding)
557 continue;
558
559 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
560
561 if (dev == cur_dev)
562 continue;
563 if (!dev->name)
564 continue;
565
566 /*
567 * Todo: This won't be enough. What if the same device
568 * comes back (with new uuid and) with its mapper path?
569 * But for now, this does help as mostly an admin will
570 * either use mapper or non mapper path throughout.
571 */
572 rcu_read_lock();
573 del = strcmp(rcu_str_deref(dev->name),
574 rcu_str_deref(cur_dev->name));
575 rcu_read_unlock();
576 if (!del)
577 break;
578 }
579
580 if (!del) {
581 /* delete the stale device */
582 if (fs_devs->num_devices == 1) {
583 btrfs_sysfs_remove_fsid(fs_devs);
584 list_del(&fs_devs->list);
585 free_fs_devices(fs_devs);
586 } else {
587 fs_devs->num_devices--;
588 list_del(&dev->dev_list);
589 rcu_string_free(dev->name);
590 kfree(dev);
591 }
592 break;
593 }
594 }
595}
596
David Sterba60999ca2014-03-26 18:26:36 +0100597/*
598 * Add new device to list of registered devices
599 *
600 * Returns:
601 * 1 - first time device is seen
602 * 0 - device already known
603 * < 0 - error
604 */
Chris Masona1b32a52008-09-05 16:09:51 -0400605static noinline int device_list_add(const char *path,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400606 struct btrfs_super_block *disk_super,
607 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
608{
609 struct btrfs_device *device;
610 struct btrfs_fs_devices *fs_devices;
Josef Bacik606686e2012-06-04 14:03:51 -0400611 struct rcu_string *name;
David Sterba60999ca2014-03-26 18:26:36 +0100612 int ret = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400613 u64 found_transid = btrfs_super_generation(disk_super);
614
615 fs_devices = find_fsid(disk_super->fsid);
616 if (!fs_devices) {
Ilya Dryomov2208a372013-08-12 14:33:03 +0300617 fs_devices = alloc_fs_devices(disk_super->fsid);
618 if (IS_ERR(fs_devices))
619 return PTR_ERR(fs_devices);
620
Chris Mason8a4b83c2008-03-24 15:02:07 -0400621 list_add(&fs_devices->list, &fs_uuids);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300622
Chris Mason8a4b83c2008-03-24 15:02:07 -0400623 device = NULL;
624 } else {
Chris Masona4437552008-04-18 10:29:38 -0400625 device = __find_device(&fs_devices->devices, devid,
626 disk_super->dev_item.uuid);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400627 }
Miao Xie443f24f2014-07-24 11:37:15 +0800628
Chris Mason8a4b83c2008-03-24 15:02:07 -0400629 if (!device) {
Yan Zheng2b820322008-11-17 21:11:30 -0500630 if (fs_devices->opened)
631 return -EBUSY;
632
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300633 device = btrfs_alloc_device(NULL, &devid,
634 disk_super->dev_item.uuid);
635 if (IS_ERR(device)) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400636 /* we can safely leave the fs_devices entry around */
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300637 return PTR_ERR(device);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400638 }
Josef Bacik606686e2012-06-04 14:03:51 -0400639
640 name = rcu_string_strdup(path, GFP_NOFS);
641 if (!name) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400642 kfree(device);
643 return -ENOMEM;
644 }
Josef Bacik606686e2012-06-04 14:03:51 -0400645 rcu_assign_pointer(device->name, name);
Arne Jansen90519d62011-05-23 14:30:00 +0200646
Chris Masone5e9a522009-06-10 15:17:02 -0400647 mutex_lock(&fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +0000648 list_add_rcu(&device->dev_list, &fs_devices->devices);
Filipe David Borba Mananaf7171752013-08-12 20:56:58 +0100649 fs_devices->num_devices++;
Chris Masone5e9a522009-06-10 15:17:02 -0400650 mutex_unlock(&fs_devices->device_list_mutex);
651
David Sterba60999ca2014-03-26 18:26:36 +0100652 ret = 1;
Yan Zheng2b820322008-11-17 21:11:30 -0500653 device->fs_devices = fs_devices;
Josef Bacik606686e2012-06-04 14:03:51 -0400654 } else if (!device->name || strcmp(device->name->str, path)) {
Anand Jainb96de002014-07-03 18:22:05 +0800655 /*
656 * When FS is already mounted.
657 * 1. If you are here and if the device->name is NULL that
658 * means this device was missing at time of FS mount.
659 * 2. If you are here and if the device->name is different
660 * from 'path' that means either
661 * a. The same device disappeared and reappeared with
662 * different name. or
663 * b. The missing-disk-which-was-replaced, has
664 * reappeared now.
665 *
666 * We must allow 1 and 2a above. But 2b would be a spurious
667 * and unintentional.
668 *
669 * Further in case of 1 and 2a above, the disk at 'path'
670 * would have missed some transaction when it was away and
671 * in case of 2a the stale bdev has to be updated as well.
672 * 2b must not be allowed at all time.
673 */
674
675 /*
Chris Mason0f23ae72014-09-18 07:49:05 -0700676 * For now, we do allow update to btrfs_fs_device through the
677 * btrfs dev scan cli after FS has been mounted. We're still
678 * tracking a problem where systems fail mount by subvolume id
679 * when we reject replacement on a mounted FS.
Anand Jainb96de002014-07-03 18:22:05 +0800680 */
Chris Mason0f23ae72014-09-18 07:49:05 -0700681 if (!fs_devices->opened && found_transid < device->generation) {
Anand Jain77bdae42014-07-03 18:22:06 +0800682 /*
683 * That is if the FS is _not_ mounted and if you
684 * are here, that means there is more than one
685 * disk with same uuid and devid.We keep the one
686 * with larger generation number or the last-in if
687 * generation are equal.
688 */
Chris Mason0f23ae72014-09-18 07:49:05 -0700689 return -EEXIST;
Anand Jain77bdae42014-07-03 18:22:06 +0800690 }
Anand Jainb96de002014-07-03 18:22:05 +0800691
Josef Bacik606686e2012-06-04 14:03:51 -0400692 name = rcu_string_strdup(path, GFP_NOFS);
TARUISI Hiroaki3a0524d2010-02-09 06:36:45 +0000693 if (!name)
694 return -ENOMEM;
Josef Bacik606686e2012-06-04 14:03:51 -0400695 rcu_string_free(device->name);
696 rcu_assign_pointer(device->name, name);
Chris Masoncd02dca2010-12-13 14:56:23 -0500697 if (device->missing) {
698 fs_devices->missing_devices--;
699 device->missing = 0;
700 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400701 }
702
Anand Jain77bdae42014-07-03 18:22:06 +0800703 /*
704 * Unmount does not free the btrfs_device struct but would zero
705 * generation along with most of the other members. So just update
706 * it back. We need it to pick the disk with largest generation
707 * (as above).
708 */
709 if (!fs_devices->opened)
710 device->generation = found_transid;
711
Anand Jain4fde46f2015-06-17 21:10:48 +0800712 /*
713 * if there is new btrfs on an already registered device,
714 * then remove the stale device entry.
715 */
Anand Jain02feae32016-02-13 10:01:40 +0800716 if (ret > 0)
717 btrfs_free_stale_device(device);
Anand Jain4fde46f2015-06-17 21:10:48 +0800718
Chris Mason8a4b83c2008-03-24 15:02:07 -0400719 *fs_devices_ret = fs_devices;
David Sterba60999ca2014-03-26 18:26:36 +0100720
721 return ret;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400722}
723
Yan Zhenge4404d62008-12-12 10:03:26 -0500724static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
725{
726 struct btrfs_fs_devices *fs_devices;
727 struct btrfs_device *device;
728 struct btrfs_device *orig_dev;
729
Ilya Dryomov2208a372013-08-12 14:33:03 +0300730 fs_devices = alloc_fs_devices(orig->fsid);
731 if (IS_ERR(fs_devices))
732 return fs_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -0500733
Miao Xieadbbb862014-09-03 21:35:42 +0800734 mutex_lock(&orig->device_list_mutex);
Josef Bacik02db0842012-06-21 16:03:58 -0400735 fs_devices->total_devices = orig->total_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -0500736
Xiao Guangrong46224702011-04-20 10:08:47 +0000737 /* We have held the volume lock, it is safe to get the devices. */
Yan Zhenge4404d62008-12-12 10:03:26 -0500738 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
Josef Bacik606686e2012-06-04 14:03:51 -0400739 struct rcu_string *name;
740
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300741 device = btrfs_alloc_device(NULL, &orig_dev->devid,
742 orig_dev->uuid);
743 if (IS_ERR(device))
Yan Zhenge4404d62008-12-12 10:03:26 -0500744 goto error;
745
Josef Bacik606686e2012-06-04 14:03:51 -0400746 /*
747 * This is ok to do without rcu read locked because we hold the
748 * uuid mutex so nothing we touch in here is going to disappear.
749 */
Anand Jaine755f782014-06-30 17:12:47 +0800750 if (orig_dev->name) {
David Sterba78f2c9e2016-02-11 14:25:38 +0100751 name = rcu_string_strdup(orig_dev->name->str,
752 GFP_KERNEL);
Anand Jaine755f782014-06-30 17:12:47 +0800753 if (!name) {
754 kfree(device);
755 goto error;
756 }
757 rcu_assign_pointer(device->name, name);
Julia Lawallfd2696f2009-09-29 13:51:04 -0400758 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500759
Yan Zhenge4404d62008-12-12 10:03:26 -0500760 list_add(&device->dev_list, &fs_devices->devices);
761 device->fs_devices = fs_devices;
762 fs_devices->num_devices++;
763 }
Miao Xieadbbb862014-09-03 21:35:42 +0800764 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500765 return fs_devices;
766error:
Miao Xieadbbb862014-09-03 21:35:42 +0800767 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500768 free_fs_devices(fs_devices);
769 return ERR_PTR(-ENOMEM);
770}
771
Eric Sandeen9eaed212014-08-01 18:12:35 -0500772void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
Chris Masondfe25022008-05-13 13:46:40 -0400773{
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500774 struct btrfs_device *device, *next;
Miao Xie443f24f2014-07-24 11:37:15 +0800775 struct btrfs_device *latest_dev = NULL;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500776
Chris Masondfe25022008-05-13 13:46:40 -0400777 mutex_lock(&uuid_mutex);
778again:
Xiao Guangrong46224702011-04-20 10:08:47 +0000779 /* This is the initialized path, it is safe to release the devices. */
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500780 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
Chris Masona6b0d5c2012-02-20 20:53:43 -0500781 if (device->in_fs_metadata) {
Stefan Behrens63a212a2012-11-05 18:29:28 +0100782 if (!device->is_tgtdev_for_dev_replace &&
Miao Xie443f24f2014-07-24 11:37:15 +0800783 (!latest_dev ||
784 device->generation > latest_dev->generation)) {
785 latest_dev = device;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500786 }
Yan Zheng2b820322008-11-17 21:11:30 -0500787 continue;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500788 }
Yan Zheng2b820322008-11-17 21:11:30 -0500789
Stefan Behrens8dabb742012-11-06 13:15:27 +0100790 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
791 /*
792 * In the first step, keep the device which has
793 * the correct fsid and the devid that is used
794 * for the dev_replace procedure.
795 * In the second step, the dev_replace state is
796 * read from the device tree and it is known
797 * whether the procedure is really active or
798 * not, which means whether this device is
799 * used or whether it should be removed.
800 */
801 if (step == 0 || device->is_tgtdev_for_dev_replace) {
802 continue;
803 }
804 }
Yan Zheng2b820322008-11-17 21:11:30 -0500805 if (device->bdev) {
Tejun Heod4d77622010-11-13 11:55:18 +0100806 blkdev_put(device->bdev, device->mode);
Yan Zheng2b820322008-11-17 21:11:30 -0500807 device->bdev = NULL;
808 fs_devices->open_devices--;
809 }
810 if (device->writeable) {
811 list_del_init(&device->dev_alloc_list);
812 device->writeable = 0;
Stefan Behrens8dabb742012-11-06 13:15:27 +0100813 if (!device->is_tgtdev_for_dev_replace)
814 fs_devices->rw_devices--;
Yan Zheng2b820322008-11-17 21:11:30 -0500815 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500816 list_del_init(&device->dev_list);
817 fs_devices->num_devices--;
Josef Bacik606686e2012-06-04 14:03:51 -0400818 rcu_string_free(device->name);
Yan Zhenge4404d62008-12-12 10:03:26 -0500819 kfree(device);
Chris Masondfe25022008-05-13 13:46:40 -0400820 }
Yan Zheng2b820322008-11-17 21:11:30 -0500821
822 if (fs_devices->seed) {
823 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -0500824 goto again;
825 }
826
Miao Xie443f24f2014-07-24 11:37:15 +0800827 fs_devices->latest_bdev = latest_dev->bdev;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500828
Chris Masondfe25022008-05-13 13:46:40 -0400829 mutex_unlock(&uuid_mutex);
Chris Masondfe25022008-05-13 13:46:40 -0400830}
Chris Masona0af4692008-05-13 16:03:06 -0400831
Xiao Guangrong1f781602011-04-20 10:09:16 +0000832static void __free_device(struct work_struct *work)
833{
834 struct btrfs_device *device;
835
836 device = container_of(work, struct btrfs_device, rcu_work);
Josef Bacik606686e2012-06-04 14:03:51 -0400837 rcu_string_free(device->name);
Xiao Guangrong1f781602011-04-20 10:09:16 +0000838 kfree(device);
839}
840
841static void free_device(struct rcu_head *head)
842{
843 struct btrfs_device *device;
844
845 device = container_of(head, struct btrfs_device, rcu);
846
847 INIT_WORK(&device->rcu_work, __free_device);
848 schedule_work(&device->rcu_work);
849}
850
Anand Jain14238812016-07-22 06:04:53 +0800851static void btrfs_close_bdev(struct btrfs_device *device)
852{
853 if (device->bdev && device->writeable) {
854 sync_blockdev(device->bdev);
855 invalidate_bdev(device->bdev);
856 }
857
858 if (device->bdev)
859 blkdev_put(device->bdev, device->mode);
860}
861
Anand Jain0ccd0522016-09-22 12:56:13 +0800862static void btrfs_prepare_close_one_device(struct btrfs_device *device)
Anand Jainf448341a2016-06-14 18:55:25 +0800863{
864 struct btrfs_fs_devices *fs_devices = device->fs_devices;
865 struct btrfs_device *new_device;
866 struct rcu_string *name;
867
868 if (device->bdev)
869 fs_devices->open_devices--;
870
871 if (device->writeable &&
872 device->devid != BTRFS_DEV_REPLACE_DEVID) {
873 list_del_init(&device->dev_alloc_list);
874 fs_devices->rw_devices--;
875 }
876
877 if (device->missing)
878 fs_devices->missing_devices--;
879
880 new_device = btrfs_alloc_device(NULL, &device->devid,
881 device->uuid);
882 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
883
884 /* Safe because we are under uuid_mutex */
885 if (device->name) {
886 name = rcu_string_strdup(device->name->str, GFP_NOFS);
887 BUG_ON(!name); /* -ENOMEM */
888 rcu_assign_pointer(new_device->name, name);
889 }
890
891 list_replace_rcu(&device->dev_list, &new_device->dev_list);
892 new_device->fs_devices = device->fs_devices;
Anand Jainf448341a2016-06-14 18:55:25 +0800893}
894
Yan Zheng2b820322008-11-17 21:11:30 -0500895static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400896{
Sasha Levin2037a092015-05-12 19:31:37 -0400897 struct btrfs_device *device, *tmp;
Anand Jain0ccd0522016-09-22 12:56:13 +0800898 struct list_head pending_put;
899
900 INIT_LIST_HEAD(&pending_put);
Yan Zhenge4404d62008-12-12 10:03:26 -0500901
Yan Zheng2b820322008-11-17 21:11:30 -0500902 if (--fs_devices->opened > 0)
903 return 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400904
Xiao Guangrongc9513ed2011-04-20 10:07:30 +0000905 mutex_lock(&fs_devices->device_list_mutex);
Sasha Levin2037a092015-05-12 19:31:37 -0400906 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
Anand Jain0ccd0522016-09-22 12:56:13 +0800907 btrfs_prepare_close_one_device(device);
908 list_add(&device->dev_list, &pending_put);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400909 }
Xiao Guangrongc9513ed2011-04-20 10:07:30 +0000910 mutex_unlock(&fs_devices->device_list_mutex);
911
Anand Jain0ccd0522016-09-22 12:56:13 +0800912 /*
913 * btrfs_show_devname() is using the device_list_mutex,
914 * sometimes call to blkdev_put() leads vfs calling
915 * into this func. So do put outside of device_list_mutex,
916 * as of now.
917 */
918 while (!list_empty(&pending_put)) {
919 device = list_first_entry(&pending_put,
920 struct btrfs_device, dev_list);
921 list_del(&device->dev_list);
922 btrfs_close_bdev(device);
923 call_rcu(&device->rcu, free_device);
924 }
925
Yan Zhenge4404d62008-12-12 10:03:26 -0500926 WARN_ON(fs_devices->open_devices);
927 WARN_ON(fs_devices->rw_devices);
Yan Zheng2b820322008-11-17 21:11:30 -0500928 fs_devices->opened = 0;
929 fs_devices->seeding = 0;
Yan Zheng2b820322008-11-17 21:11:30 -0500930
Chris Mason8a4b83c2008-03-24 15:02:07 -0400931 return 0;
932}
933
Yan Zheng2b820322008-11-17 21:11:30 -0500934int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
935{
Yan Zhenge4404d62008-12-12 10:03:26 -0500936 struct btrfs_fs_devices *seed_devices = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -0500937 int ret;
938
939 mutex_lock(&uuid_mutex);
940 ret = __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -0500941 if (!fs_devices->opened) {
942 seed_devices = fs_devices->seed;
943 fs_devices->seed = NULL;
944 }
Yan Zheng2b820322008-11-17 21:11:30 -0500945 mutex_unlock(&uuid_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500946
947 while (seed_devices) {
948 fs_devices = seed_devices;
949 seed_devices = fs_devices->seed;
950 __btrfs_close_devices(fs_devices);
951 free_fs_devices(fs_devices);
952 }
Eric Sandeenbc178622013-03-09 15:18:39 +0000953 /*
954 * Wait for rcu kworkers under __btrfs_close_devices
955 * to finish all blkdev_puts so device is really
956 * free when umount is done.
957 */
958 rcu_barrier();
Yan Zheng2b820322008-11-17 21:11:30 -0500959 return ret;
960}
961
Yan Zhenge4404d62008-12-12 10:03:26 -0500962static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
963 fmode_t flags, void *holder)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400964{
Josef Bacikd5e20032011-08-04 14:52:27 +0000965 struct request_queue *q;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400966 struct block_device *bdev;
967 struct list_head *head = &fs_devices->devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400968 struct btrfs_device *device;
Miao Xie443f24f2014-07-24 11:37:15 +0800969 struct btrfs_device *latest_dev = NULL;
Chris Masona0af4692008-05-13 16:03:06 -0400970 struct buffer_head *bh;
971 struct btrfs_super_block *disk_super;
Chris Masona0af4692008-05-13 16:03:06 -0400972 u64 devid;
Yan Zheng2b820322008-11-17 21:11:30 -0500973 int seeding = 1;
Chris Masona0af4692008-05-13 16:03:06 -0400974 int ret = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400975
Tejun Heod4d77622010-11-13 11:55:18 +0100976 flags |= FMODE_EXCL;
977
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500978 list_for_each_entry(device, head, dev_list) {
Chris Masonc1c4d912008-05-08 15:05:58 -0400979 if (device->bdev)
980 continue;
Chris Masondfe25022008-05-13 13:46:40 -0400981 if (!device->name)
982 continue;
983
Eric Sandeenf63e0cc2013-04-04 20:45:08 +0000984 /* Just open everything we can; ignore failures here */
985 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
986 &bdev, &bh))
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100987 continue;
Chris Masona0af4692008-05-13 16:03:06 -0400988
989 disk_super = (struct btrfs_super_block *)bh->b_data;
Xiao Guangronga3438322010-01-06 11:48:18 +0000990 devid = btrfs_stack_device_id(&disk_super->dev_item);
Chris Masona0af4692008-05-13 16:03:06 -0400991 if (devid != device->devid)
992 goto error_brelse;
993
Yan Zheng2b820322008-11-17 21:11:30 -0500994 if (memcmp(device->uuid, disk_super->dev_item.uuid,
995 BTRFS_UUID_SIZE))
996 goto error_brelse;
997
998 device->generation = btrfs_super_generation(disk_super);
Miao Xie443f24f2014-07-24 11:37:15 +0800999 if (!latest_dev ||
1000 device->generation > latest_dev->generation)
1001 latest_dev = device;
Chris Masona0af4692008-05-13 16:03:06 -04001002
Yan Zheng2b820322008-11-17 21:11:30 -05001003 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1004 device->writeable = 0;
1005 } else {
1006 device->writeable = !bdev_read_only(bdev);
1007 seeding = 0;
1008 }
1009
Josef Bacikd5e20032011-08-04 14:52:27 +00001010 q = bdev_get_queue(bdev);
Miao Xie90180da2014-09-03 21:35:30 +08001011 if (blk_queue_discard(q))
Josef Bacikd5e20032011-08-04 14:52:27 +00001012 device->can_discard = 1;
Josef Bacikd5e20032011-08-04 14:52:27 +00001013
Chris Mason8a4b83c2008-03-24 15:02:07 -04001014 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -04001015 device->in_fs_metadata = 0;
Chris Mason15916de2008-11-19 21:17:22 -05001016 device->mode = flags;
1017
Chris Masonc2898112009-06-10 09:51:32 -04001018 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1019 fs_devices->rotating = 1;
1020
Chris Masona0af4692008-05-13 16:03:06 -04001021 fs_devices->open_devices++;
Ilya Dryomov55e50e42013-09-01 18:56:44 +03001022 if (device->writeable &&
1023 device->devid != BTRFS_DEV_REPLACE_DEVID) {
Yan Zheng2b820322008-11-17 21:11:30 -05001024 fs_devices->rw_devices++;
1025 list_add(&device->dev_alloc_list,
1026 &fs_devices->alloc_list);
1027 }
Xiao Guangrong4f6c9322011-04-20 10:06:40 +00001028 brelse(bh);
Chris Masona0af4692008-05-13 16:03:06 -04001029 continue;
Chris Masona061fc82008-05-07 11:43:44 -04001030
Chris Masona0af4692008-05-13 16:03:06 -04001031error_brelse:
1032 brelse(bh);
Tejun Heod4d77622010-11-13 11:55:18 +01001033 blkdev_put(bdev, flags);
Chris Masona0af4692008-05-13 16:03:06 -04001034 continue;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001035 }
Chris Masona0af4692008-05-13 16:03:06 -04001036 if (fs_devices->open_devices == 0) {
Ilya Dryomov20bcd642011-10-20 00:06:20 +03001037 ret = -EINVAL;
Chris Masona0af4692008-05-13 16:03:06 -04001038 goto out;
1039 }
Yan Zheng2b820322008-11-17 21:11:30 -05001040 fs_devices->seeding = seeding;
1041 fs_devices->opened = 1;
Miao Xie443f24f2014-07-24 11:37:15 +08001042 fs_devices->latest_bdev = latest_dev->bdev;
Yan Zheng2b820322008-11-17 21:11:30 -05001043 fs_devices->total_rw_bytes = 0;
Chris Masona0af4692008-05-13 16:03:06 -04001044out:
Yan Zheng2b820322008-11-17 21:11:30 -05001045 return ret;
1046}
1047
1048int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
Christoph Hellwig97288f22008-12-02 06:36:09 -05001049 fmode_t flags, void *holder)
Yan Zheng2b820322008-11-17 21:11:30 -05001050{
1051 int ret;
1052
1053 mutex_lock(&uuid_mutex);
1054 if (fs_devices->opened) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001055 fs_devices->opened++;
1056 ret = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05001057 } else {
Chris Mason15916de2008-11-19 21:17:22 -05001058 ret = __btrfs_open_devices(fs_devices, flags, holder);
Yan Zheng2b820322008-11-17 21:11:30 -05001059 }
Chris Mason8a4b83c2008-03-24 15:02:07 -04001060 mutex_unlock(&uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001061 return ret;
1062}
1063
Anand Jain6cf86a002016-02-13 10:01:29 +08001064void btrfs_release_disk_super(struct page *page)
1065{
1066 kunmap(page);
1067 put_page(page);
1068}
1069
1070int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1071 struct page **page, struct btrfs_super_block **disk_super)
1072{
1073 void *p;
1074 pgoff_t index;
1075
1076 /* make sure our super fits in the device */
1077 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1078 return 1;
1079
1080 /* make sure our super fits in the page */
1081 if (sizeof(**disk_super) > PAGE_SIZE)
1082 return 1;
1083
1084 /* make sure our super doesn't straddle pages on disk */
1085 index = bytenr >> PAGE_SHIFT;
1086 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1087 return 1;
1088
1089 /* pull in the page with our super */
1090 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1091 index, GFP_KERNEL);
1092
1093 if (IS_ERR_OR_NULL(*page))
1094 return 1;
1095
1096 p = kmap(*page);
1097
1098 /* align our pointer to the offset of the super block */
1099 *disk_super = p + (bytenr & ~PAGE_MASK);
1100
1101 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1102 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1103 btrfs_release_disk_super(*page);
1104 return 1;
1105 }
1106
1107 if ((*disk_super)->label[0] &&
1108 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1109 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1110
1111 return 0;
1112}
1113
David Sterba6f60cbd2013-02-15 11:31:02 -07001114/*
1115 * Look for a btrfs signature on a device. This may be called out of the mount path
1116 * and we are not allowed to call set_blocksize during the scan. The superblock
1117 * is read via pagecache
1118 */
Christoph Hellwig97288f22008-12-02 06:36:09 -05001119int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
Chris Mason8a4b83c2008-03-24 15:02:07 -04001120 struct btrfs_fs_devices **fs_devices_ret)
1121{
1122 struct btrfs_super_block *disk_super;
1123 struct block_device *bdev;
David Sterba6f60cbd2013-02-15 11:31:02 -07001124 struct page *page;
David Sterba6f60cbd2013-02-15 11:31:02 -07001125 int ret = -EINVAL;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001126 u64 devid;
Chris Masonf2984462008-04-10 16:19:33 -04001127 u64 transid;
Josef Bacik02db0842012-06-21 16:03:58 -04001128 u64 total_devices;
David Sterba6f60cbd2013-02-15 11:31:02 -07001129 u64 bytenr;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001130
David Sterba6f60cbd2013-02-15 11:31:02 -07001131 /*
1132 * we would like to check all the supers, but that would make
1133 * a btrfs mount succeed after a mkfs from a different FS.
1134 * So, we need to add a special mount option to scan for
1135 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1136 */
1137 bytenr = btrfs_sb_offset(0);
Tejun Heod4d77622010-11-13 11:55:18 +01001138 flags |= FMODE_EXCL;
Al Viro10f63272011-11-17 15:05:22 -05001139 mutex_lock(&uuid_mutex);
David Sterba6f60cbd2013-02-15 11:31:02 -07001140
1141 bdev = blkdev_get_by_path(path, flags, holder);
David Sterba6f60cbd2013-02-15 11:31:02 -07001142 if (IS_ERR(bdev)) {
1143 ret = PTR_ERR(bdev);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +01001144 goto error;
David Sterba6f60cbd2013-02-15 11:31:02 -07001145 }
1146
Anand Jain6cf86a002016-02-13 10:01:29 +08001147 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
David Sterba6f60cbd2013-02-15 11:31:02 -07001148 goto error_bdev_put;
1149
Xiao Guangronga3438322010-01-06 11:48:18 +00001150 devid = btrfs_stack_device_id(&disk_super->dev_item);
Chris Masonf2984462008-04-10 16:19:33 -04001151 transid = btrfs_super_generation(disk_super);
Josef Bacik02db0842012-06-21 16:03:58 -04001152 total_devices = btrfs_super_num_devices(disk_super);
David Sterba6f60cbd2013-02-15 11:31:02 -07001153
Chris Mason8a4b83c2008-03-24 15:02:07 -04001154 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
David Sterba60999ca2014-03-26 18:26:36 +01001155 if (ret > 0) {
1156 if (disk_super->label[0]) {
Jeff Mahoney62e85572016-09-20 10:05:01 -04001157 pr_info("BTRFS: device label %s ", disk_super->label);
David Sterba60999ca2014-03-26 18:26:36 +01001158 } else {
Jeff Mahoney62e85572016-09-20 10:05:01 -04001159 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
David Sterba60999ca2014-03-26 18:26:36 +01001160 }
1161
Jeff Mahoney62e85572016-09-20 10:05:01 -04001162 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
David Sterba60999ca2014-03-26 18:26:36 +01001163 ret = 0;
1164 }
Josef Bacik02db0842012-06-21 16:03:58 -04001165 if (!ret && fs_devices_ret)
1166 (*fs_devices_ret)->total_devices = total_devices;
David Sterba6f60cbd2013-02-15 11:31:02 -07001167
Anand Jain6cf86a002016-02-13 10:01:29 +08001168 btrfs_release_disk_super(page);
David Sterba6f60cbd2013-02-15 11:31:02 -07001169
1170error_bdev_put:
Tejun Heod4d77622010-11-13 11:55:18 +01001171 blkdev_put(bdev, flags);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001172error:
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +01001173 mutex_unlock(&uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001174 return ret;
1175}
Chris Mason0b86a832008-03-24 15:01:56 -04001176
Miao Xie6d07bce2011-01-05 10:07:31 +00001177/* helper to account the used device space in the range */
1178int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1179 u64 end, u64 *length)
Chris Mason0b86a832008-03-24 15:01:56 -04001180{
1181 struct btrfs_key key;
1182 struct btrfs_root *root = device->dev_root;
Miao Xie6d07bce2011-01-05 10:07:31 +00001183 struct btrfs_dev_extent *dev_extent;
Yan Zheng2b820322008-11-17 21:11:30 -05001184 struct btrfs_path *path;
Miao Xie6d07bce2011-01-05 10:07:31 +00001185 u64 extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001186 int ret;
Miao Xie6d07bce2011-01-05 10:07:31 +00001187 int slot;
Chris Mason0b86a832008-03-24 15:01:56 -04001188 struct extent_buffer *l;
1189
Miao Xie6d07bce2011-01-05 10:07:31 +00001190 *length = 0;
1191
Stefan Behrens63a212a2012-11-05 18:29:28 +01001192 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
Miao Xie6d07bce2011-01-05 10:07:31 +00001193 return 0;
1194
Yan Zheng2b820322008-11-17 21:11:30 -05001195 path = btrfs_alloc_path();
1196 if (!path)
1197 return -ENOMEM;
David Sterbae4058b52015-11-27 16:31:35 +01001198 path->reada = READA_FORWARD;
Chris Mason8f18cf12008-04-25 16:53:30 -04001199
Chris Mason0b86a832008-03-24 15:01:56 -04001200 key.objectid = device->devid;
Miao Xie6d07bce2011-01-05 10:07:31 +00001201 key.offset = start;
Chris Mason0b86a832008-03-24 15:01:56 -04001202 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie6d07bce2011-01-05 10:07:31 +00001203
1204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001205 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001206 goto out;
Yan Zheng1fcbac52009-07-24 11:06:53 -04001207 if (ret > 0) {
1208 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1209 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001210 goto out;
Yan Zheng1fcbac52009-07-24 11:06:53 -04001211 }
Miao Xie6d07bce2011-01-05 10:07:31 +00001212
Chris Mason0b86a832008-03-24 15:01:56 -04001213 while (1) {
1214 l = path->nodes[0];
1215 slot = path->slots[0];
1216 if (slot >= btrfs_header_nritems(l)) {
1217 ret = btrfs_next_leaf(root, path);
1218 if (ret == 0)
1219 continue;
1220 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001221 goto out;
1222
1223 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001224 }
1225 btrfs_item_key_to_cpu(l, &key, slot);
1226
1227 if (key.objectid < device->devid)
1228 goto next;
1229
1230 if (key.objectid > device->devid)
Miao Xie6d07bce2011-01-05 10:07:31 +00001231 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001232
David Sterba962a2982014-06-04 18:41:45 +02001233 if (key.type != BTRFS_DEV_EXTENT_KEY)
Chris Mason0b86a832008-03-24 15:01:56 -04001234 goto next;
Chris Mason0b86a832008-03-24 15:01:56 -04001235
Chris Mason0b86a832008-03-24 15:01:56 -04001236 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
Miao Xie6d07bce2011-01-05 10:07:31 +00001237 extent_end = key.offset + btrfs_dev_extent_length(l,
1238 dev_extent);
1239 if (key.offset <= start && extent_end > end) {
1240 *length = end - start + 1;
1241 break;
1242 } else if (key.offset <= start && extent_end > start)
1243 *length += extent_end - start;
1244 else if (key.offset > start && extent_end <= end)
1245 *length += extent_end - key.offset;
1246 else if (key.offset > start && key.offset <= end) {
1247 *length += end - key.offset + 1;
1248 break;
1249 } else if (key.offset > end)
1250 break;
1251
1252next:
1253 path->slots[0]++;
1254 }
1255 ret = 0;
1256out:
1257 btrfs_free_path(path);
1258 return ret;
1259}
1260
Jeff Mahoney499f3772015-06-15 09:41:17 -04001261static int contains_pending_extent(struct btrfs_transaction *transaction,
Josef Bacik6df9a952013-06-27 13:22:46 -04001262 struct btrfs_device *device,
1263 u64 *start, u64 len)
1264{
Jeff Mahoney499f3772015-06-15 09:41:17 -04001265 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
Josef Bacik6df9a952013-06-27 13:22:46 -04001266 struct extent_map *em;
Jeff Mahoney499f3772015-06-15 09:41:17 -04001267 struct list_head *search_list = &fs_info->pinned_chunks;
Josef Bacik6df9a952013-06-27 13:22:46 -04001268 int ret = 0;
Forrest Liu1b984502015-02-09 17:30:47 +08001269 u64 physical_start = *start;
Josef Bacik6df9a952013-06-27 13:22:46 -04001270
Jeff Mahoney499f3772015-06-15 09:41:17 -04001271 if (transaction)
1272 search_list = &transaction->pending_chunks;
Filipe Manana04216822014-11-27 21:14:15 +00001273again:
1274 list_for_each_entry(em, search_list, list) {
Josef Bacik6df9a952013-06-27 13:22:46 -04001275 struct map_lookup *map;
1276 int i;
1277
Jeff Mahoney95617d62015-06-03 10:55:48 -04001278 map = em->map_lookup;
Josef Bacik6df9a952013-06-27 13:22:46 -04001279 for (i = 0; i < map->num_stripes; i++) {
Filipe Mananac152b632015-05-14 10:46:03 +01001280 u64 end;
1281
Josef Bacik6df9a952013-06-27 13:22:46 -04001282 if (map->stripes[i].dev != device)
1283 continue;
Forrest Liu1b984502015-02-09 17:30:47 +08001284 if (map->stripes[i].physical >= physical_start + len ||
Josef Bacik6df9a952013-06-27 13:22:46 -04001285 map->stripes[i].physical + em->orig_block_len <=
Forrest Liu1b984502015-02-09 17:30:47 +08001286 physical_start)
Josef Bacik6df9a952013-06-27 13:22:46 -04001287 continue;
Filipe Mananac152b632015-05-14 10:46:03 +01001288 /*
1289 * Make sure that while processing the pinned list we do
1290 * not override our *start with a lower value, because
1291 * we can have pinned chunks that fall within this
1292 * device hole and that have lower physical addresses
1293 * than the pending chunks we processed before. If we
1294 * do not take this special care we can end up getting
1295 * 2 pending chunks that start at the same physical
1296 * device offsets because the end offset of a pinned
1297 * chunk can be equal to the start offset of some
1298 * pending chunk.
1299 */
1300 end = map->stripes[i].physical + em->orig_block_len;
1301 if (end > *start) {
1302 *start = end;
1303 ret = 1;
1304 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001305 }
1306 }
Jeff Mahoney499f3772015-06-15 09:41:17 -04001307 if (search_list != &fs_info->pinned_chunks) {
1308 search_list = &fs_info->pinned_chunks;
Filipe Manana04216822014-11-27 21:14:15 +00001309 goto again;
1310 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001311
1312 return ret;
1313}
1314
1315
Chris Mason0b86a832008-03-24 15:01:56 -04001316/*
Jeff Mahoney499f3772015-06-15 09:41:17 -04001317 * find_free_dev_extent_start - find free space in the specified device
1318 * @device: the device which we search the free space in
1319 * @num_bytes: the size of the free space that we need
1320 * @search_start: the position from which to begin the search
1321 * @start: store the start of the free space.
1322 * @len: the size of the free space. that we find, or the size
1323 * of the max free space if we don't find suitable free space
Miao Xie7bfc8372011-01-05 10:07:26 +00001324 *
Chris Mason0b86a832008-03-24 15:01:56 -04001325 * this uses a pretty simple search, the expectation is that it is
1326 * called very infrequently and that a given device has a small number
1327 * of extents
Miao Xie7bfc8372011-01-05 10:07:26 +00001328 *
1329 * @start is used to store the start of the free space if we find. But if we
1330 * don't find suitable free space, it will be used to store the start position
1331 * of the max free space.
1332 *
1333 * @len is used to store the size of the free space that we find.
1334 * But if we don't find suitable free space, it is used to store the size of
1335 * the max free space.
Chris Mason0b86a832008-03-24 15:01:56 -04001336 */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001337int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1338 struct btrfs_device *device, u64 num_bytes,
1339 u64 search_start, u64 *start, u64 *len)
Chris Mason0b86a832008-03-24 15:01:56 -04001340{
1341 struct btrfs_key key;
1342 struct btrfs_root *root = device->dev_root;
Miao Xie7bfc8372011-01-05 10:07:26 +00001343 struct btrfs_dev_extent *dev_extent;
Chris Mason0b86a832008-03-24 15:01:56 -04001344 struct btrfs_path *path;
Miao Xie7bfc8372011-01-05 10:07:26 +00001345 u64 hole_size;
1346 u64 max_hole_start;
1347 u64 max_hole_size;
1348 u64 extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001349 u64 search_end = device->total_bytes;
1350 int ret;
Miao Xie7bfc8372011-01-05 10:07:26 +00001351 int slot;
Chris Mason0b86a832008-03-24 15:01:56 -04001352 struct extent_buffer *l;
Filipe Manana8cdc7c52016-01-06 22:42:35 +00001353 u64 min_search_start;
1354
1355 /*
1356 * We don't want to overwrite the superblock on the drive nor any area
1357 * used by the boot loader (grub for example), so we make sure to start
1358 * at an offset of at least 1MB.
1359 */
1360 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1361 search_start = max(search_start, min_search_start);
Chris Mason0b86a832008-03-24 15:01:56 -04001362
Josef Bacik6df9a952013-06-27 13:22:46 -04001363 path = btrfs_alloc_path();
1364 if (!path)
1365 return -ENOMEM;
Zhao Leif2ab7612015-02-16 18:52:17 +08001366
Miao Xie7bfc8372011-01-05 10:07:26 +00001367 max_hole_start = search_start;
1368 max_hole_size = 0;
1369
Zhao Leif2ab7612015-02-16 18:52:17 +08001370again:
Stefan Behrens63a212a2012-11-05 18:29:28 +01001371 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
Miao Xie7bfc8372011-01-05 10:07:26 +00001372 ret = -ENOSPC;
Josef Bacik6df9a952013-06-27 13:22:46 -04001373 goto out;
Miao Xie7bfc8372011-01-05 10:07:26 +00001374 }
1375
David Sterbae4058b52015-11-27 16:31:35 +01001376 path->reada = READA_FORWARD;
Josef Bacik6df9a952013-06-27 13:22:46 -04001377 path->search_commit_root = 1;
1378 path->skip_locking = 1;
Miao Xie7bfc8372011-01-05 10:07:26 +00001379
Chris Mason0b86a832008-03-24 15:01:56 -04001380 key.objectid = device->devid;
1381 key.offset = search_start;
1382 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie7bfc8372011-01-05 10:07:26 +00001383
Li Zefan125ccb02011-12-08 15:07:24 +08001384 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001385 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001386 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001387 if (ret > 0) {
1388 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1389 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001390 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001391 }
Miao Xie7bfc8372011-01-05 10:07:26 +00001392
Chris Mason0b86a832008-03-24 15:01:56 -04001393 while (1) {
1394 l = path->nodes[0];
1395 slot = path->slots[0];
1396 if (slot >= btrfs_header_nritems(l)) {
1397 ret = btrfs_next_leaf(root, path);
1398 if (ret == 0)
1399 continue;
1400 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001401 goto out;
1402
1403 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001404 }
1405 btrfs_item_key_to_cpu(l, &key, slot);
1406
1407 if (key.objectid < device->devid)
1408 goto next;
1409
1410 if (key.objectid > device->devid)
Miao Xie7bfc8372011-01-05 10:07:26 +00001411 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001412
David Sterba962a2982014-06-04 18:41:45 +02001413 if (key.type != BTRFS_DEV_EXTENT_KEY)
Chris Mason0b86a832008-03-24 15:01:56 -04001414 goto next;
1415
Miao Xie7bfc8372011-01-05 10:07:26 +00001416 if (key.offset > search_start) {
1417 hole_size = key.offset - search_start;
1418
Josef Bacik6df9a952013-06-27 13:22:46 -04001419 /*
1420 * Have to check before we set max_hole_start, otherwise
1421 * we could end up sending back this offset anyway.
1422 */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001423 if (contains_pending_extent(transaction, device,
Josef Bacik6df9a952013-06-27 13:22:46 -04001424 &search_start,
Forrest Liu1b984502015-02-09 17:30:47 +08001425 hole_size)) {
1426 if (key.offset >= search_start) {
1427 hole_size = key.offset - search_start;
1428 } else {
1429 WARN_ON_ONCE(1);
1430 hole_size = 0;
1431 }
1432 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001433
Miao Xie7bfc8372011-01-05 10:07:26 +00001434 if (hole_size > max_hole_size) {
1435 max_hole_start = search_start;
1436 max_hole_size = hole_size;
1437 }
1438
1439 /*
1440 * If this free space is greater than which we need,
1441 * it must be the max free space that we have found
1442 * until now, so max_hole_start must point to the start
1443 * of this free space and the length of this free space
1444 * is stored in max_hole_size. Thus, we return
1445 * max_hole_start and max_hole_size and go back to the
1446 * caller.
1447 */
1448 if (hole_size >= num_bytes) {
1449 ret = 0;
1450 goto out;
1451 }
1452 }
1453
Chris Mason0b86a832008-03-24 15:01:56 -04001454 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
Miao Xie7bfc8372011-01-05 10:07:26 +00001455 extent_end = key.offset + btrfs_dev_extent_length(l,
1456 dev_extent);
1457 if (extent_end > search_start)
1458 search_start = extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001459next:
1460 path->slots[0]++;
1461 cond_resched();
1462 }
Chris Mason0b86a832008-03-24 15:01:56 -04001463
liubo38c01b92011-08-02 02:39:03 +00001464 /*
1465 * At this point, search_start should be the end of
1466 * allocated dev extents, and when shrinking the device,
1467 * search_end may be smaller than search_start.
1468 */
Zhao Leif2ab7612015-02-16 18:52:17 +08001469 if (search_end > search_start) {
liubo38c01b92011-08-02 02:39:03 +00001470 hole_size = search_end - search_start;
1471
Jeff Mahoney499f3772015-06-15 09:41:17 -04001472 if (contains_pending_extent(transaction, device, &search_start,
Zhao Leif2ab7612015-02-16 18:52:17 +08001473 hole_size)) {
1474 btrfs_release_path(path);
1475 goto again;
1476 }
Chris Mason0b86a832008-03-24 15:01:56 -04001477
Zhao Leif2ab7612015-02-16 18:52:17 +08001478 if (hole_size > max_hole_size) {
1479 max_hole_start = search_start;
1480 max_hole_size = hole_size;
1481 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001482 }
1483
Miao Xie7bfc8372011-01-05 10:07:26 +00001484 /* See above. */
Zhao Leif2ab7612015-02-16 18:52:17 +08001485 if (max_hole_size < num_bytes)
Miao Xie7bfc8372011-01-05 10:07:26 +00001486 ret = -ENOSPC;
1487 else
1488 ret = 0;
1489
1490out:
Yan Zheng2b820322008-11-17 21:11:30 -05001491 btrfs_free_path(path);
Miao Xie7bfc8372011-01-05 10:07:26 +00001492 *start = max_hole_start;
Miao Xieb2117a32011-01-05 10:07:28 +00001493 if (len)
Miao Xie7bfc8372011-01-05 10:07:26 +00001494 *len = max_hole_size;
Chris Mason0b86a832008-03-24 15:01:56 -04001495 return ret;
1496}
1497
Jeff Mahoney499f3772015-06-15 09:41:17 -04001498int find_free_dev_extent(struct btrfs_trans_handle *trans,
1499 struct btrfs_device *device, u64 num_bytes,
1500 u64 *start, u64 *len)
1501{
Jeff Mahoney499f3772015-06-15 09:41:17 -04001502 /* FIXME use last free of some kind */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001503 return find_free_dev_extent_start(trans->transaction, device,
Filipe Manana8cdc7c52016-01-06 22:42:35 +00001504 num_bytes, 0, start, len);
Jeff Mahoney499f3772015-06-15 09:41:17 -04001505}
1506
Christoph Hellwigb2950862008-12-02 09:54:17 -05001507static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -04001508 struct btrfs_device *device,
Miao Xie2196d6e2014-09-03 21:35:41 +08001509 u64 start, u64 *dev_extent_len)
Chris Mason8f18cf12008-04-25 16:53:30 -04001510{
1511 int ret;
1512 struct btrfs_path *path;
1513 struct btrfs_root *root = device->dev_root;
1514 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -04001515 struct btrfs_key found_key;
1516 struct extent_buffer *leaf = NULL;
1517 struct btrfs_dev_extent *extent = NULL;
Chris Mason8f18cf12008-04-25 16:53:30 -04001518
1519 path = btrfs_alloc_path();
1520 if (!path)
1521 return -ENOMEM;
1522
1523 key.objectid = device->devid;
1524 key.offset = start;
1525 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie924cd8f2011-11-10 20:45:04 -05001526again:
Chris Mason8f18cf12008-04-25 16:53:30 -04001527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Chris Masona061fc82008-05-07 11:43:44 -04001528 if (ret > 0) {
1529 ret = btrfs_previous_item(root, path, key.objectid,
1530 BTRFS_DEV_EXTENT_KEY);
Tsutomu Itohb0b802d2011-05-19 07:03:42 +00001531 if (ret)
1532 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001533 leaf = path->nodes[0];
1534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 BUG_ON(found_key.offset > start || found_key.offset +
1538 btrfs_dev_extent_length(leaf, extent) < start);
Miao Xie924cd8f2011-11-10 20:45:04 -05001539 key = found_key;
1540 btrfs_release_path(path);
1541 goto again;
Chris Masona061fc82008-05-07 11:43:44 -04001542 } else if (ret == 0) {
1543 leaf = path->nodes[0];
1544 extent = btrfs_item_ptr(leaf, path->slots[0],
1545 struct btrfs_dev_extent);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001546 } else {
Anand Jain34d97002016-03-16 16:43:06 +08001547 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001548 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001549 }
Chris Mason8f18cf12008-04-25 16:53:30 -04001550
Miao Xie2196d6e2014-09-03 21:35:41 +08001551 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1552
Chris Mason8f18cf12008-04-25 16:53:30 -04001553 ret = btrfs_del_item(trans, root, path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001554 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08001555 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001556 "Failed to remove dev extent item");
Zhao Lei13212b52015-02-12 14:18:17 +08001557 } else {
Josef Bacik3204d332015-09-24 10:46:10 -04001558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001559 }
Tsutomu Itohb0b802d2011-05-19 07:03:42 +00001560out:
Chris Mason8f18cf12008-04-25 16:53:30 -04001561 btrfs_free_path(path);
1562 return ret;
1563}
1564
Eric Sandeen48a3b632013-04-25 20:41:01 +00001565static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566 struct btrfs_device *device,
1567 u64 chunk_tree, u64 chunk_objectid,
1568 u64 chunk_offset, u64 start, u64 num_bytes)
Chris Mason0b86a832008-03-24 15:01:56 -04001569{
1570 int ret;
1571 struct btrfs_path *path;
1572 struct btrfs_root *root = device->dev_root;
1573 struct btrfs_dev_extent *extent;
1574 struct extent_buffer *leaf;
1575 struct btrfs_key key;
1576
Chris Masondfe25022008-05-13 13:46:40 -04001577 WARN_ON(!device->in_fs_metadata);
Stefan Behrens63a212a2012-11-05 18:29:28 +01001578 WARN_ON(device->is_tgtdev_for_dev_replace);
Chris Mason0b86a832008-03-24 15:01:56 -04001579 path = btrfs_alloc_path();
1580 if (!path)
1581 return -ENOMEM;
1582
Chris Mason0b86a832008-03-24 15:01:56 -04001583 key.objectid = device->devid;
Yan Zheng2b820322008-11-17 21:11:30 -05001584 key.offset = start;
Chris Mason0b86a832008-03-24 15:01:56 -04001585 key.type = BTRFS_DEV_EXTENT_KEY;
1586 ret = btrfs_insert_empty_item(trans, root, path, &key,
1587 sizeof(*extent));
Mark Fasheh2cdcecb2011-09-08 17:14:32 -07001588 if (ret)
1589 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001590
1591 leaf = path->nodes[0];
1592 extent = btrfs_item_ptr(leaf, path->slots[0],
1593 struct btrfs_dev_extent);
Chris Masone17cade2008-04-15 15:41:47 -04001594 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1595 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1596 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1597
David Sterbad24ee972016-11-09 17:44:25 +01001598 write_extent_buffer_chunk_tree_uuid(leaf,
1599 root->fs_info->chunk_tree_uuid);
Chris Masone17cade2008-04-15 15:41:47 -04001600
Chris Mason0b86a832008-03-24 15:01:56 -04001601 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1602 btrfs_mark_buffer_dirty(leaf);
Mark Fasheh2cdcecb2011-09-08 17:14:32 -07001603out:
Chris Mason0b86a832008-03-24 15:01:56 -04001604 btrfs_free_path(path);
1605 return ret;
1606}
1607
Josef Bacik6df9a952013-06-27 13:22:46 -04001608static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
Chris Mason0b86a832008-03-24 15:01:56 -04001609{
Josef Bacik6df9a952013-06-27 13:22:46 -04001610 struct extent_map_tree *em_tree;
1611 struct extent_map *em;
1612 struct rb_node *n;
1613 u64 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04001614
Josef Bacik6df9a952013-06-27 13:22:46 -04001615 em_tree = &fs_info->mapping_tree.map_tree;
1616 read_lock(&em_tree->lock);
1617 n = rb_last(&em_tree->map);
1618 if (n) {
1619 em = rb_entry(n, struct extent_map, rb_node);
1620 ret = em->start + em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04001621 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001622 read_unlock(&em_tree->lock);
1623
Chris Mason0b86a832008-03-24 15:01:56 -04001624 return ret;
1625}
1626
Ilya Dryomov53f10652013-08-12 14:33:01 +03001627static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1628 u64 *devid_ret)
Chris Mason0b86a832008-03-24 15:01:56 -04001629{
1630 int ret;
1631 struct btrfs_key key;
1632 struct btrfs_key found_key;
Yan Zheng2b820322008-11-17 21:11:30 -05001633 struct btrfs_path *path;
1634
Yan Zheng2b820322008-11-17 21:11:30 -05001635 path = btrfs_alloc_path();
1636 if (!path)
1637 return -ENOMEM;
Chris Mason0b86a832008-03-24 15:01:56 -04001638
1639 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1640 key.type = BTRFS_DEV_ITEM_KEY;
1641 key.offset = (u64)-1;
1642
Ilya Dryomov53f10652013-08-12 14:33:01 +03001643 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001644 if (ret < 0)
1645 goto error;
1646
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001647 BUG_ON(ret == 0); /* Corruption */
Chris Mason0b86a832008-03-24 15:01:56 -04001648
Ilya Dryomov53f10652013-08-12 14:33:01 +03001649 ret = btrfs_previous_item(fs_info->chunk_root, path,
1650 BTRFS_DEV_ITEMS_OBJECTID,
Chris Mason0b86a832008-03-24 15:01:56 -04001651 BTRFS_DEV_ITEM_KEY);
1652 if (ret) {
Ilya Dryomov53f10652013-08-12 14:33:01 +03001653 *devid_ret = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04001654 } else {
1655 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656 path->slots[0]);
Ilya Dryomov53f10652013-08-12 14:33:01 +03001657 *devid_ret = found_key.offset + 1;
Chris Mason0b86a832008-03-24 15:01:56 -04001658 }
1659 ret = 0;
1660error:
Yan Zheng2b820322008-11-17 21:11:30 -05001661 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04001662 return ret;
1663}
1664
1665/*
1666 * the device information is stored in the chunk root
1667 * the btrfs_device struct should be fully filled in
1668 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00001669static int btrfs_add_device(struct btrfs_trans_handle *trans,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04001670 struct btrfs_fs_info *fs_info,
Eric Sandeen48a3b632013-04-25 20:41:01 +00001671 struct btrfs_device *device)
Chris Mason0b86a832008-03-24 15:01:56 -04001672{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04001673 struct btrfs_root *root = fs_info->chunk_root;
Chris Mason0b86a832008-03-24 15:01:56 -04001674 int ret;
1675 struct btrfs_path *path;
1676 struct btrfs_dev_item *dev_item;
1677 struct extent_buffer *leaf;
1678 struct btrfs_key key;
1679 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04001680
Chris Mason0b86a832008-03-24 15:01:56 -04001681 path = btrfs_alloc_path();
1682 if (!path)
1683 return -ENOMEM;
1684
Chris Mason0b86a832008-03-24 15:01:56 -04001685 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1686 key.type = BTRFS_DEV_ITEM_KEY;
Yan Zheng2b820322008-11-17 21:11:30 -05001687 key.offset = device->devid;
Chris Mason0b86a832008-03-24 15:01:56 -04001688
1689 ret = btrfs_insert_empty_item(trans, root, path, &key,
Chris Mason0d81ba52008-03-24 15:02:07 -04001690 sizeof(*dev_item));
Chris Mason0b86a832008-03-24 15:01:56 -04001691 if (ret)
1692 goto out;
1693
1694 leaf = path->nodes[0];
1695 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1696
1697 btrfs_set_device_id(leaf, dev_item, device->devid);
Yan Zheng2b820322008-11-17 21:11:30 -05001698 btrfs_set_device_generation(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001699 btrfs_set_device_type(leaf, dev_item, device->type);
1700 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1701 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1702 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Miao Xie7cc8e582014-09-03 21:35:38 +08001703 btrfs_set_device_total_bytes(leaf, dev_item,
1704 btrfs_device_get_disk_total_bytes(device));
1705 btrfs_set_device_bytes_used(leaf, dev_item,
1706 btrfs_device_get_bytes_used(device));
Chris Masone17cade2008-04-15 15:41:47 -04001707 btrfs_set_device_group(leaf, dev_item, 0);
1708 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1709 btrfs_set_device_bandwidth(leaf, dev_item, 0);
Chris Masonc3027eb2008-12-08 16:40:21 -05001710 btrfs_set_device_start_offset(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001711
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02001712 ptr = btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04001713 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02001714 ptr = btrfs_device_fsid(dev_item);
Yan Zheng2b820322008-11-17 21:11:30 -05001715 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04001716 btrfs_mark_buffer_dirty(leaf);
Chris Mason0b86a832008-03-24 15:01:56 -04001717
Yan Zheng2b820322008-11-17 21:11:30 -05001718 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04001719out:
1720 btrfs_free_path(path);
1721 return ret;
1722}
Chris Mason8f18cf12008-04-25 16:53:30 -04001723
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001724/*
1725 * Function to update ctime/mtime for a given device path.
1726 * Mainly used for ctime/mtime based probe like libblkid.
1727 */
1728static void update_dev_time(char *path_name)
1729{
1730 struct file *filp;
1731
1732 filp = filp_open(path_name, O_RDWR, 0);
Al Viro98af5922014-12-14 02:59:17 -05001733 if (IS_ERR(filp))
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001734 return;
1735 file_update_time(filp);
1736 filp_close(filp, NULL);
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001737}
1738
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04001739static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
Chris Masona061fc82008-05-07 11:43:44 -04001740 struct btrfs_device *device)
1741{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04001742 struct btrfs_root *root = fs_info->chunk_root;
Chris Masona061fc82008-05-07 11:43:44 -04001743 int ret;
1744 struct btrfs_path *path;
Chris Masona061fc82008-05-07 11:43:44 -04001745 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -04001746 struct btrfs_trans_handle *trans;
1747
Chris Masona061fc82008-05-07 11:43:44 -04001748 path = btrfs_alloc_path();
1749 if (!path)
1750 return -ENOMEM;
1751
Yan, Zhenga22285a2010-05-16 10:48:46 -04001752 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00001753 if (IS_ERR(trans)) {
1754 btrfs_free_path(path);
1755 return PTR_ERR(trans);
1756 }
Chris Masona061fc82008-05-07 11:43:44 -04001757 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1758 key.type = BTRFS_DEV_ITEM_KEY;
1759 key.offset = device->devid;
1760
1761 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762 if (ret < 0)
1763 goto out;
1764
1765 if (ret > 0) {
1766 ret = -ENOENT;
1767 goto out;
1768 }
1769
1770 ret = btrfs_del_item(trans, root, path);
1771 if (ret)
1772 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001773out:
1774 btrfs_free_path(path);
1775 btrfs_commit_transaction(trans, root);
1776 return ret;
1777}
1778
David Sterba3cc31a02016-02-15 16:00:26 +01001779/*
1780 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781 * filesystem. It's up to the caller to adjust that number regarding eg. device
1782 * replace.
1783 */
1784static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1785 u64 num_devices)
Chris Masona061fc82008-05-07 11:43:44 -04001786{
Chris Masona061fc82008-05-07 11:43:44 -04001787 u64 all_avail;
Miao Xiede98ced2013-01-29 10:13:12 +00001788 unsigned seq;
David Sterba418775a2016-02-15 16:28:14 +01001789 int i;
Chris Masona061fc82008-05-07 11:43:44 -04001790
Miao Xiede98ced2013-01-29 10:13:12 +00001791 do {
Anand Jainbd45ffb2016-02-13 10:01:34 +08001792 seq = read_seqbegin(&fs_info->profiles_lock);
Miao Xiede98ced2013-01-29 10:13:12 +00001793
Anand Jainbd45ffb2016-02-13 10:01:34 +08001794 all_avail = fs_info->avail_data_alloc_bits |
1795 fs_info->avail_system_alloc_bits |
1796 fs_info->avail_metadata_alloc_bits;
1797 } while (read_seqretry(&fs_info->profiles_lock, seq));
Anand Jainf1fa7f22016-02-13 10:01:33 +08001798
David Sterba418775a2016-02-15 16:28:14 +01001799 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1800 if (!(all_avail & btrfs_raid_group[i]))
1801 continue;
Chris Masona061fc82008-05-07 11:43:44 -04001802
David Sterba418775a2016-02-15 16:28:14 +01001803 if (num_devices < btrfs_raid_array[i].devs_min) {
1804 int ret = btrfs_raid_mindev_error[i];
Chris Masona061fc82008-05-07 11:43:44 -04001805
David Sterba418775a2016-02-15 16:28:14 +01001806 if (ret)
1807 return ret;
1808 }
Anand Jainbd45ffb2016-02-13 10:01:34 +08001809 }
1810
1811 return 0;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001812}
1813
Anand Jain88acff62016-05-03 17:44:43 +08001814struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1815 struct btrfs_device *device)
1816{
1817 struct btrfs_device *next_device;
1818
1819 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1820 if (next_device != device &&
1821 !next_device->missing && next_device->bdev)
1822 return next_device;
1823 }
1824
1825 return NULL;
1826}
1827
1828/*
1829 * Helper function to check if the given device is part of s_bdev / latest_bdev
1830 * and replace it with the provided or the next active device, in the context
1831 * where this function called, there should be always be another device (or
1832 * this_dev) which is active.
1833 */
1834void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1835 struct btrfs_device *device, struct btrfs_device *this_dev)
1836{
1837 struct btrfs_device *next_device;
1838
1839 if (this_dev)
1840 next_device = this_dev;
1841 else
1842 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1843 device);
1844 ASSERT(next_device);
1845
1846 if (fs_info->sb->s_bdev &&
1847 (fs_info->sb->s_bdev == device->bdev))
1848 fs_info->sb->s_bdev = next_device->bdev;
1849
1850 if (fs_info->fs_devices->latest_bdev == device->bdev)
1851 fs_info->fs_devices->latest_bdev = next_device->bdev;
1852}
1853
Anand Jain6b526ed2016-02-13 10:01:39 +08001854int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
Anand Jainf1fa7f22016-02-13 10:01:33 +08001855{
1856 struct btrfs_device *device;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001857 struct btrfs_fs_devices *cur_devices;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001858 u64 num_devices;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001859 int ret = 0;
1860 bool clear_super = false;
1861
1862 mutex_lock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001863
1864 num_devices = root->fs_info->fs_devices->num_devices;
1865 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1866 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1867 WARN_ON(num_devices < 1);
1868 num_devices--;
1869 }
1870 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1871
David Sterba3cc31a02016-02-15 16:00:26 +01001872 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
Anand Jainf1fa7f22016-02-13 10:01:33 +08001873 if (ret)
Chris Masona061fc82008-05-07 11:43:44 -04001874 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001875
David Sterba5c5c0df2016-02-15 16:39:55 +01001876 ret = btrfs_find_device_by_devspec(root, devid, device_path,
Anand Jain24fc5722016-02-13 10:01:36 +08001877 &device);
1878 if (ret)
Chris Masona061fc82008-05-07 11:43:44 -04001879 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05001880
Stefan Behrens63a212a2012-11-05 18:29:28 +01001881 if (device->is_tgtdev_for_dev_replace) {
Anand Jain183860f2013-05-17 10:52:45 +00001882 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
Anand Jain24fc5722016-02-13 10:01:36 +08001883 goto out;
Stefan Behrens63a212a2012-11-05 18:29:28 +01001884 }
1885
Yan Zheng2b820322008-11-17 21:11:30 -05001886 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
Anand Jain183860f2013-05-17 10:52:45 +00001887 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
Anand Jain24fc5722016-02-13 10:01:36 +08001888 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05001889 }
1890
1891 if (device->writeable) {
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001892 lock_chunks(root);
Yan Zheng2b820322008-11-17 21:11:30 -05001893 list_del_init(&device->dev_alloc_list);
Miao Xiec3929c32014-09-03 21:35:47 +08001894 device->fs_devices->rw_devices--;
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001895 unlock_chunks(root);
Xiao Guangrong1f781602011-04-20 10:09:16 +00001896 clear_super = true;
Yan Zheng2b820322008-11-17 21:11:30 -05001897 }
Chris Masona061fc82008-05-07 11:43:44 -04001898
Carey Underwoodd7901552013-03-04 16:37:06 -06001899 mutex_unlock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001900 ret = btrfs_shrink_device(device, 0);
Carey Underwoodd7901552013-03-04 16:37:06 -06001901 mutex_lock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001902 if (ret)
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001903 goto error_undo;
Chris Masona061fc82008-05-07 11:43:44 -04001904
Stefan Behrens63a212a2012-11-05 18:29:28 +01001905 /*
1906 * TODO: the superblock still includes this device in its num_devices
1907 * counter although write_all_supers() is not locked out. This
1908 * could give a filesystem state which requires a degraded mount.
1909 */
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04001910 ret = btrfs_rm_dev_item(root->fs_info, device);
Chris Masona061fc82008-05-07 11:43:44 -04001911 if (ret)
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001912 goto error_undo;
Chris Masona061fc82008-05-07 11:43:44 -04001913
Yan Zheng2b820322008-11-17 21:11:30 -05001914 device->in_fs_metadata = 0;
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01001915 btrfs_scrub_cancel_dev(root->fs_info, device);
Chris Masone5e9a522009-06-10 15:17:02 -04001916
1917 /*
1918 * the device list mutex makes sure that we don't change
1919 * the device list while someone else is writing out all
Filipe David Borba Mananad7306802013-08-09 15:41:36 +01001920 * the device supers. Whoever is writing all supers, should
1921 * lock the device list mutex before getting the number of
1922 * devices in the super block (super_copy). Conversely,
1923 * whoever updates the number of devices in the super block
1924 * (super_copy) should hold the device list mutex.
Chris Masone5e9a522009-06-10 15:17:02 -04001925 */
Xiao Guangrong1f781602011-04-20 10:09:16 +00001926
1927 cur_devices = device->fs_devices;
Chris Masone5e9a522009-06-10 15:17:02 -04001928 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +00001929 list_del_rcu(&device->dev_list);
Chris Masone5e9a522009-06-10 15:17:02 -04001930
Yan Zhenge4404d62008-12-12 10:03:26 -05001931 device->fs_devices->num_devices--;
Josef Bacik02db0842012-06-21 16:03:58 -04001932 device->fs_devices->total_devices--;
Yan Zheng2b820322008-11-17 21:11:30 -05001933
Chris Masoncd02dca2010-12-13 14:56:23 -05001934 if (device->missing)
Miao Xie3a7d55c2014-07-16 18:38:01 +08001935 device->fs_devices->missing_devices--;
Chris Masoncd02dca2010-12-13 14:56:23 -05001936
Anand Jain88acff62016-05-03 17:44:43 +08001937 btrfs_assign_next_active_device(root->fs_info, device, NULL);
Yan Zheng2b820322008-11-17 21:11:30 -05001938
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001939 if (device->bdev) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001940 device->fs_devices->open_devices--;
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001941 /* remove sysfs entry */
Anand Jain32576042015-08-14 18:32:49 +08001942 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001943 }
Anand Jain99994cd2014-06-03 11:36:00 +08001944
David Sterba6c417612011-04-13 15:41:04 +02001945 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1946 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
Filipe David Borba Mananad7306802013-08-09 15:41:36 +01001947 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05001948
Jeff Mahoneycea67ab2016-09-20 08:50:21 -04001949 /*
1950 * at this point, the device is zero sized and detached from
1951 * the devices list. All that's left is to zero out the old
1952 * supers and free the device.
1953 */
1954 if (device->writeable)
1955 btrfs_scratch_superblocks(device->bdev, device->name->str);
1956
1957 btrfs_close_bdev(device);
1958 call_rcu(&device->rcu, free_device);
1959
Xiao Guangrong1f781602011-04-20 10:09:16 +00001960 if (cur_devices->open_devices == 0) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001961 struct btrfs_fs_devices *fs_devices;
1962 fs_devices = root->fs_info->fs_devices;
1963 while (fs_devices) {
Rickard Strandqvist8321cf22014-05-22 22:43:43 +02001964 if (fs_devices->seed == cur_devices) {
1965 fs_devices->seed = cur_devices->seed;
Yan Zhenge4404d62008-12-12 10:03:26 -05001966 break;
Rickard Strandqvist8321cf22014-05-22 22:43:43 +02001967 }
Yan Zhenge4404d62008-12-12 10:03:26 -05001968 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -05001969 }
Xiao Guangrong1f781602011-04-20 10:09:16 +00001970 cur_devices->seed = NULL;
Xiao Guangrong1f781602011-04-20 10:09:16 +00001971 __btrfs_close_devices(cur_devices);
Xiao Guangrong1f781602011-04-20 10:09:16 +00001972 free_fs_devices(cur_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05001973 }
1974
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02001975 root->fs_info->num_tolerated_disk_barrier_failures =
1976 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1977
Chris Masona061fc82008-05-07 11:43:44 -04001978out:
1979 mutex_unlock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001980 return ret;
Anand Jain24fc5722016-02-13 10:01:36 +08001981
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001982error_undo:
1983 if (device->writeable) {
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001984 lock_chunks(root);
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001985 list_add(&device->dev_alloc_list,
1986 &root->fs_info->fs_devices->alloc_list);
Miao Xiec3929c32014-09-03 21:35:47 +08001987 device->fs_devices->rw_devices++;
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001988 unlock_chunks(root);
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001989 }
Anand Jain24fc5722016-02-13 10:01:36 +08001990 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001991}
1992
Qu Wenruo084b6e7c2014-10-30 16:52:31 +08001993void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1994 struct btrfs_device *srcdev)
Stefan Behrense93c89c2012-11-05 17:33:06 +01001995{
Anand Jaind51908c2014-08-13 14:24:19 +08001996 struct btrfs_fs_devices *fs_devices;
1997
Stefan Behrense93c89c2012-11-05 17:33:06 +01001998 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
Ilya Dryomov13572722013-10-02 20:41:01 +03001999
Anand Jain25e8e912014-08-20 10:56:56 +08002000 /*
2001 * in case of fs with no seed, srcdev->fs_devices will point
2002 * to fs_devices of fs_info. However when the dev being replaced is
2003 * a seed dev it will point to the seed's local fs_devices. In short
2004 * srcdev will have its correct fs_devices in both the cases.
2005 */
2006 fs_devices = srcdev->fs_devices;
Anand Jaind51908c2014-08-13 14:24:19 +08002007
Stefan Behrense93c89c2012-11-05 17:33:06 +01002008 list_del_rcu(&srcdev->dev_list);
2009 list_del_rcu(&srcdev->dev_alloc_list);
Anand Jaind51908c2014-08-13 14:24:19 +08002010 fs_devices->num_devices--;
Miao Xie82372bc2014-09-03 21:35:44 +08002011 if (srcdev->missing)
Anand Jaind51908c2014-08-13 14:24:19 +08002012 fs_devices->missing_devices--;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002013
Anand Jain48b3b9d2016-04-12 21:36:16 +08002014 if (srcdev->writeable)
Miao Xie82372bc2014-09-03 21:35:44 +08002015 fs_devices->rw_devices--;
Ilya Dryomov13572722013-10-02 20:41:01 +03002016
Miao Xie82372bc2014-09-03 21:35:44 +08002017 if (srcdev->bdev)
Anand Jaind51908c2014-08-13 14:24:19 +08002018 fs_devices->open_devices--;
Qu Wenruo084b6e7c2014-10-30 16:52:31 +08002019}
2020
2021void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2022 struct btrfs_device *srcdev)
2023{
2024 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002025
Anand Jain48b3b9d2016-04-12 21:36:16 +08002026 if (srcdev->writeable) {
2027 /* zero out the old super if it is writable */
2028 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2029 }
Anand Jain14238812016-07-22 06:04:53 +08002030
2031 btrfs_close_bdev(srcdev);
2032
Stefan Behrense93c89c2012-11-05 17:33:06 +01002033 call_rcu(&srcdev->rcu, free_device);
Anand Jain94d5f0c2014-08-13 14:24:22 +08002034
2035 /*
2036 * unless fs_devices is seed fs, num_devices shouldn't go
2037 * zero
2038 */
2039 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2040
2041 /* if this is no devs we rather delete the fs_devices */
2042 if (!fs_devices->num_devices) {
2043 struct btrfs_fs_devices *tmp_fs_devices;
2044
2045 tmp_fs_devices = fs_info->fs_devices;
2046 while (tmp_fs_devices) {
2047 if (tmp_fs_devices->seed == fs_devices) {
2048 tmp_fs_devices->seed = fs_devices->seed;
2049 break;
2050 }
2051 tmp_fs_devices = tmp_fs_devices->seed;
2052 }
2053 fs_devices->seed = NULL;
Anand Jain8bef8402014-08-13 14:24:23 +08002054 __btrfs_close_devices(fs_devices);
2055 free_fs_devices(fs_devices);
Anand Jain94d5f0c2014-08-13 14:24:22 +08002056 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002057}
2058
2059void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2060 struct btrfs_device *tgtdev)
2061{
Miao Xie67a2c452014-09-03 21:35:43 +08002062 mutex_lock(&uuid_mutex);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002063 WARN_ON(!tgtdev);
2064 mutex_lock(&fs_info->fs_devices->device_list_mutex);
Anand Jaind2ff1b22015-03-10 06:38:42 +08002065
Anand Jain32576042015-08-14 18:32:49 +08002066 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
Anand Jaind2ff1b22015-03-10 06:38:42 +08002067
Anand Jain779bf3fe2016-04-18 16:51:23 +08002068 if (tgtdev->bdev)
Stefan Behrense93c89c2012-11-05 17:33:06 +01002069 fs_info->fs_devices->open_devices--;
Anand Jain779bf3fe2016-04-18 16:51:23 +08002070
Stefan Behrense93c89c2012-11-05 17:33:06 +01002071 fs_info->fs_devices->num_devices--;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002072
Anand Jain88acff62016-05-03 17:44:43 +08002073 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002074
Stefan Behrense93c89c2012-11-05 17:33:06 +01002075 list_del_rcu(&tgtdev->dev_list);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002076
2077 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Miao Xie67a2c452014-09-03 21:35:43 +08002078 mutex_unlock(&uuid_mutex);
Anand Jain779bf3fe2016-04-18 16:51:23 +08002079
2080 /*
2081 * The update_dev_time() with in btrfs_scratch_superblocks()
2082 * may lead to a call to btrfs_show_devname() which will try
2083 * to hold device_list_mutex. And here this device
2084 * is already out of device list, so we don't have to hold
2085 * the device_list_mutex lock.
2086 */
2087 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
Anand Jain14238812016-07-22 06:04:53 +08002088
2089 btrfs_close_bdev(tgtdev);
Anand Jain779bf3fe2016-04-18 16:51:23 +08002090 call_rcu(&tgtdev->rcu, free_device);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002091}
2092
Eric Sandeen48a3b632013-04-25 20:41:01 +00002093static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2094 struct btrfs_device **device)
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002095{
2096 int ret = 0;
2097 struct btrfs_super_block *disk_super;
2098 u64 devid;
2099 u8 *dev_uuid;
2100 struct block_device *bdev;
2101 struct buffer_head *bh;
2102
2103 *device = NULL;
2104 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2105 root->fs_info->bdev_holder, 0, &bdev, &bh);
2106 if (ret)
2107 return ret;
2108 disk_super = (struct btrfs_super_block *)bh->b_data;
2109 devid = btrfs_stack_device_id(&disk_super->dev_item);
2110 dev_uuid = disk_super->dev_item.uuid;
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002111 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002112 disk_super->fsid);
2113 brelse(bh);
2114 if (!*device)
2115 ret = -ENOENT;
2116 blkdev_put(bdev, FMODE_READ);
2117 return ret;
2118}
2119
2120int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2121 char *device_path,
2122 struct btrfs_device **device)
2123{
2124 *device = NULL;
2125 if (strcmp(device_path, "missing") == 0) {
2126 struct list_head *devices;
2127 struct btrfs_device *tmp;
2128
2129 devices = &root->fs_info->fs_devices->devices;
2130 /*
2131 * It is safe to read the devices since the volume_mutex
2132 * is held by the caller.
2133 */
2134 list_for_each_entry(tmp, devices, dev_list) {
2135 if (tmp->in_fs_metadata && !tmp->bdev) {
2136 *device = tmp;
2137 break;
2138 }
2139 }
2140
Anand Jaind74a6252015-08-14 18:32:56 +08002141 if (!*device)
2142 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002143
2144 return 0;
2145 } else {
2146 return btrfs_find_device_by_path(root, device_path, device);
2147 }
2148}
2149
Yan Zheng2b820322008-11-17 21:11:30 -05002150/*
David Sterba5c5c0df2016-02-15 16:39:55 +01002151 * Lookup a device given by device id, or the path if the id is 0.
2152 */
2153int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2154 char *devpath,
Anand Jain24e04742016-02-13 10:01:35 +08002155 struct btrfs_device **device)
2156{
2157 int ret;
2158
David Sterba5c5c0df2016-02-15 16:39:55 +01002159 if (devid) {
Anand Jain24e04742016-02-13 10:01:35 +08002160 ret = 0;
David Sterba5c5c0df2016-02-15 16:39:55 +01002161 *device = btrfs_find_device(root->fs_info, devid, NULL,
Anand Jain24e04742016-02-13 10:01:35 +08002162 NULL);
2163 if (!*device)
2164 ret = -ENOENT;
2165 } else {
David Sterba5c5c0df2016-02-15 16:39:55 +01002166 if (!devpath || !devpath[0])
Anand Jainb3d1b152016-02-13 10:01:37 +08002167 return -EINVAL;
2168
David Sterba5c5c0df2016-02-15 16:39:55 +01002169 ret = btrfs_find_device_missing_or_by_path(root, devpath,
Anand Jain24e04742016-02-13 10:01:35 +08002170 device);
2171 }
2172 return ret;
2173}
2174
Yan Zheng2b820322008-11-17 21:11:30 -05002175/*
2176 * does all the dirty work required for changing file system's UUID.
2177 */
Li Zefan125ccb02011-12-08 15:07:24 +08002178static int btrfs_prepare_sprout(struct btrfs_root *root)
Yan Zheng2b820322008-11-17 21:11:30 -05002179{
2180 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2181 struct btrfs_fs_devices *old_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -05002182 struct btrfs_fs_devices *seed_devices;
David Sterba6c417612011-04-13 15:41:04 +02002183 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
Yan Zheng2b820322008-11-17 21:11:30 -05002184 struct btrfs_device *device;
2185 u64 super_flags;
2186
2187 BUG_ON(!mutex_is_locked(&uuid_mutex));
Yan Zhenge4404d62008-12-12 10:03:26 -05002188 if (!fs_devices->seeding)
Yan Zheng2b820322008-11-17 21:11:30 -05002189 return -EINVAL;
2190
Ilya Dryomov2208a372013-08-12 14:33:03 +03002191 seed_devices = __alloc_fs_devices();
2192 if (IS_ERR(seed_devices))
2193 return PTR_ERR(seed_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002194
Yan Zhenge4404d62008-12-12 10:03:26 -05002195 old_devices = clone_fs_devices(fs_devices);
2196 if (IS_ERR(old_devices)) {
2197 kfree(seed_devices);
2198 return PTR_ERR(old_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002199 }
Yan Zhenge4404d62008-12-12 10:03:26 -05002200
Yan Zheng2b820322008-11-17 21:11:30 -05002201 list_add(&old_devices->list, &fs_uuids);
2202
Yan Zhenge4404d62008-12-12 10:03:26 -05002203 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2204 seed_devices->opened = 1;
2205 INIT_LIST_HEAD(&seed_devices->devices);
2206 INIT_LIST_HEAD(&seed_devices->alloc_list);
Chris Masone5e9a522009-06-10 15:17:02 -04002207 mutex_init(&seed_devices->device_list_mutex);
Xiao Guangrongc9513ed2011-04-20 10:07:30 +00002208
2209 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +00002210 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2211 synchronize_rcu);
Miao Xie2196d6e2014-09-03 21:35:41 +08002212 list_for_each_entry(device, &seed_devices->devices, dev_list)
Yan Zhenge4404d62008-12-12 10:03:26 -05002213 device->fs_devices = seed_devices;
Miao Xie2196d6e2014-09-03 21:35:41 +08002214
2215 lock_chunks(root);
2216 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2217 unlock_chunks(root);
Yan Zhenge4404d62008-12-12 10:03:26 -05002218
Yan Zheng2b820322008-11-17 21:11:30 -05002219 fs_devices->seeding = 0;
2220 fs_devices->num_devices = 0;
2221 fs_devices->open_devices = 0;
Miao Xie69611ac2014-07-03 18:22:12 +08002222 fs_devices->missing_devices = 0;
Miao Xie69611ac2014-07-03 18:22:12 +08002223 fs_devices->rotating = 0;
Yan Zhenge4404d62008-12-12 10:03:26 -05002224 fs_devices->seed = seed_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05002225
2226 generate_random_uuid(fs_devices->fsid);
2227 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2228 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
Filipe David Borba Mananaf7171752013-08-12 20:56:58 +01002229 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2230
Yan Zheng2b820322008-11-17 21:11:30 -05002231 super_flags = btrfs_super_flags(disk_super) &
2232 ~BTRFS_SUPER_FLAG_SEEDING;
2233 btrfs_set_super_flags(disk_super, super_flags);
2234
2235 return 0;
2236}
2237
2238/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04002239 * Store the expected generation for seed devices in device items.
Yan Zheng2b820322008-11-17 21:11:30 -05002240 */
2241static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002242 struct btrfs_fs_info *fs_info)
Yan Zheng2b820322008-11-17 21:11:30 -05002243{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002244 struct btrfs_root *root = fs_info->chunk_root;
Yan Zheng2b820322008-11-17 21:11:30 -05002245 struct btrfs_path *path;
2246 struct extent_buffer *leaf;
2247 struct btrfs_dev_item *dev_item;
2248 struct btrfs_device *device;
2249 struct btrfs_key key;
2250 u8 fs_uuid[BTRFS_UUID_SIZE];
2251 u8 dev_uuid[BTRFS_UUID_SIZE];
2252 u64 devid;
2253 int ret;
2254
2255 path = btrfs_alloc_path();
2256 if (!path)
2257 return -ENOMEM;
2258
Yan Zheng2b820322008-11-17 21:11:30 -05002259 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2260 key.offset = 0;
2261 key.type = BTRFS_DEV_ITEM_KEY;
2262
2263 while (1) {
2264 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2265 if (ret < 0)
2266 goto error;
2267
2268 leaf = path->nodes[0];
2269next_slot:
2270 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2271 ret = btrfs_next_leaf(root, path);
2272 if (ret > 0)
2273 break;
2274 if (ret < 0)
2275 goto error;
2276 leaf = path->nodes[0];
2277 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002278 btrfs_release_path(path);
Yan Zheng2b820322008-11-17 21:11:30 -05002279 continue;
2280 }
2281
2282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2284 key.type != BTRFS_DEV_ITEM_KEY)
2285 break;
2286
2287 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2288 struct btrfs_dev_item);
2289 devid = btrfs_device_id(leaf, dev_item);
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02002290 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05002291 BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02002292 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05002293 BTRFS_UUID_SIZE);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002294 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2295 fs_uuid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002296 BUG_ON(!device); /* Logic error */
Yan Zheng2b820322008-11-17 21:11:30 -05002297
2298 if (device->fs_devices->seeding) {
2299 btrfs_set_device_generation(leaf, dev_item,
2300 device->generation);
2301 btrfs_mark_buffer_dirty(leaf);
2302 }
2303
2304 path->slots[0]++;
2305 goto next_slot;
2306 }
2307 ret = 0;
2308error:
2309 btrfs_free_path(path);
2310 return ret;
2311}
2312
Chris Mason788f20e2008-04-28 15:29:42 -04002313int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2314{
Josef Bacikd5e20032011-08-04 14:52:27 +00002315 struct request_queue *q;
Chris Mason788f20e2008-04-28 15:29:42 -04002316 struct btrfs_trans_handle *trans;
2317 struct btrfs_device *device;
2318 struct block_device *bdev;
Chris Mason788f20e2008-04-28 15:29:42 -04002319 struct list_head *devices;
Yan Zheng2b820322008-11-17 21:11:30 -05002320 struct super_block *sb = root->fs_info->sb;
Josef Bacik606686e2012-06-04 14:03:51 -04002321 struct rcu_string *name;
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002322 u64 tmp;
Yan Zheng2b820322008-11-17 21:11:30 -05002323 int seeding_dev = 0;
Chris Mason788f20e2008-04-28 15:29:42 -04002324 int ret = 0;
2325
Yan Zheng2b820322008-11-17 21:11:30 -05002326 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
Liu Bof8c5d0b2012-05-10 18:10:38 +08002327 return -EROFS;
Chris Mason788f20e2008-04-28 15:29:42 -04002328
Li Zefana5d163332011-12-07 20:08:40 -05002329 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
Tejun Heod4d77622010-11-13 11:55:18 +01002330 root->fs_info->bdev_holder);
Josef Bacik7f592032010-01-27 02:09:00 +00002331 if (IS_ERR(bdev))
2332 return PTR_ERR(bdev);
Chris Masona2135012008-06-25 16:01:30 -04002333
Yan Zheng2b820322008-11-17 21:11:30 -05002334 if (root->fs_info->fs_devices->seeding) {
2335 seeding_dev = 1;
2336 down_write(&sb->s_umount);
2337 mutex_lock(&uuid_mutex);
2338 }
2339
Chris Mason8c8bee12008-09-29 11:19:10 -04002340 filemap_write_and_wait(bdev->bd_inode->i_mapping);
Chris Masona2135012008-06-25 16:01:30 -04002341
Chris Mason788f20e2008-04-28 15:29:42 -04002342 devices = &root->fs_info->fs_devices->devices;
Liu Bod25628b2012-11-14 14:35:30 +00002343
2344 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05002345 list_for_each_entry(device, devices, dev_list) {
Chris Mason788f20e2008-04-28 15:29:42 -04002346 if (device->bdev == bdev) {
2347 ret = -EEXIST;
Liu Bod25628b2012-11-14 14:35:30 +00002348 mutex_unlock(
2349 &root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002350 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002351 }
2352 }
Liu Bod25628b2012-11-14 14:35:30 +00002353 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04002354
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002355 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2356 if (IS_ERR(device)) {
Chris Mason788f20e2008-04-28 15:29:42 -04002357 /* we can safely leave the fs_devices entry around */
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002358 ret = PTR_ERR(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002359 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002360 }
2361
David Sterba78f2c9e2016-02-11 14:25:38 +01002362 name = rcu_string_strdup(device_path, GFP_KERNEL);
Josef Bacik606686e2012-06-04 14:03:51 -04002363 if (!name) {
Chris Mason788f20e2008-04-28 15:29:42 -04002364 kfree(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002365 ret = -ENOMEM;
2366 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002367 }
Josef Bacik606686e2012-06-04 14:03:51 -04002368 rcu_assign_pointer(device->name, name);
Yan Zheng2b820322008-11-17 21:11:30 -05002369
Yan, Zhenga22285a2010-05-16 10:48:46 -04002370 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002371 if (IS_ERR(trans)) {
Josef Bacik606686e2012-06-04 14:03:51 -04002372 rcu_string_free(device->name);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002373 kfree(device);
2374 ret = PTR_ERR(trans);
2375 goto error;
2376 }
2377
Josef Bacikd5e20032011-08-04 14:52:27 +00002378 q = bdev_get_queue(bdev);
2379 if (blk_queue_discard(q))
2380 device->can_discard = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05002381 device->writeable = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05002382 device->generation = trans->transid;
Chris Mason788f20e2008-04-28 15:29:42 -04002383 device->io_width = root->sectorsize;
2384 device->io_align = root->sectorsize;
2385 device->sector_size = root->sectorsize;
2386 device->total_bytes = i_size_read(bdev->bd_inode);
Yan Zheng2cc3c552009-06-04 09:23:50 -04002387 device->disk_total_bytes = device->total_bytes;
Miao Xie935e5cc2014-09-03 21:35:33 +08002388 device->commit_total_bytes = device->total_bytes;
Chris Mason788f20e2008-04-28 15:29:42 -04002389 device->dev_root = root->fs_info->dev_root;
2390 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -04002391 device->in_fs_metadata = 1;
Stefan Behrens63a212a2012-11-05 18:29:28 +01002392 device->is_tgtdev_for_dev_replace = 0;
Ilya Dryomovfb01aa82011-02-15 18:12:57 +00002393 device->mode = FMODE_EXCL;
Stefan Behrens27087f32013-10-11 15:20:42 +02002394 device->dev_stats_valid = 1;
Zheng Yan325cd4b2008-09-05 16:43:54 -04002395 set_blocksize(device->bdev, 4096);
2396
Yan Zheng2b820322008-11-17 21:11:30 -05002397 if (seeding_dev) {
2398 sb->s_flags &= ~MS_RDONLY;
Li Zefan125ccb02011-12-08 15:07:24 +08002399 ret = btrfs_prepare_sprout(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002400 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2b820322008-11-17 21:11:30 -05002401 }
2402
2403 device->fs_devices = root->fs_info->fs_devices;
Chris Masone5e9a522009-06-10 15:17:02 -04002404
Chris Masone5e9a522009-06-10 15:17:02 -04002405 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Miao Xie2196d6e2014-09-03 21:35:41 +08002406 lock_chunks(root);
Xiao Guangrong1f781602011-04-20 10:09:16 +00002407 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002408 list_add(&device->dev_alloc_list,
2409 &root->fs_info->fs_devices->alloc_list);
2410 root->fs_info->fs_devices->num_devices++;
2411 root->fs_info->fs_devices->open_devices++;
2412 root->fs_info->fs_devices->rw_devices++;
Josef Bacik02db0842012-06-21 16:03:58 -04002413 root->fs_info->fs_devices->total_devices++;
Yan Zheng2b820322008-11-17 21:11:30 -05002414 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2415
Josef Bacik2bf64752011-09-26 17:12:22 -04002416 spin_lock(&root->fs_info->free_chunk_lock);
2417 root->fs_info->free_chunk_space += device->total_bytes;
2418 spin_unlock(&root->fs_info->free_chunk_lock);
2419
Chris Masonc2898112009-06-10 09:51:32 -04002420 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2421 root->fs_info->fs_devices->rotating = 1;
2422
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002423 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
David Sterba6c417612011-04-13 15:41:04 +02002424 btrfs_set_super_total_bytes(root->fs_info->super_copy,
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002425 tmp + device->total_bytes);
Chris Mason788f20e2008-04-28 15:29:42 -04002426
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002427 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
David Sterba6c417612011-04-13 15:41:04 +02002428 btrfs_set_super_num_devices(root->fs_info->super_copy,
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002429 tmp + 1);
Anand Jain0d393762014-06-03 11:36:01 +08002430
2431 /* add sysfs device entry */
Anand Jaine3bd6972015-08-14 18:32:48 +08002432 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
Anand Jain0d393762014-06-03 11:36:01 +08002433
Miao Xie2196d6e2014-09-03 21:35:41 +08002434 /*
2435 * we've got more storage, clear any full flags on the space
2436 * infos
2437 */
2438 btrfs_clear_space_info_full(root->fs_info);
2439
2440 unlock_chunks(root);
Chris Masone5e9a522009-06-10 15:17:02 -04002441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04002442
Yan Zheng2b820322008-11-17 21:11:30 -05002443 if (seeding_dev) {
Miao Xie2196d6e2014-09-03 21:35:41 +08002444 lock_chunks(root);
Yan Zheng2b820322008-11-17 21:11:30 -05002445 ret = init_first_rw_device(trans, root, device);
Miao Xie2196d6e2014-09-03 21:35:41 +08002446 unlock_chunks(root);
David Sterba005d6422012-09-18 07:52:32 -06002447 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002448 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002449 goto error_trans;
David Sterba005d6422012-09-18 07:52:32 -06002450 }
Miao Xie2196d6e2014-09-03 21:35:41 +08002451 }
2452
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002453 ret = btrfs_add_device(trans, root->fs_info, device);
Miao Xie2196d6e2014-09-03 21:35:41 +08002454 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002455 btrfs_abort_transaction(trans, ret);
Miao Xie2196d6e2014-09-03 21:35:41 +08002456 goto error_trans;
2457 }
2458
2459 if (seeding_dev) {
2460 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2461
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002462 ret = btrfs_finish_sprout(trans, root->fs_info);
David Sterba005d6422012-09-18 07:52:32 -06002463 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002464 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002465 goto error_trans;
David Sterba005d6422012-09-18 07:52:32 -06002466 }
Anand Jainb2373f22014-06-03 11:36:03 +08002467
2468 /* Sprouting would change fsid of the mounted root,
2469 * so rename the fsid on the sysfs
2470 */
2471 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2472 root->fs_info->fsid);
Anand Jainc1b7e472015-08-14 18:32:50 +08002473 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
Anand Jain2e7910d2015-03-10 06:38:29 +08002474 fsid_buf))
David Sterbaf14d1042015-10-08 11:37:06 +02002475 btrfs_warn(root->fs_info,
2476 "sysfs: failed to create fsid for sprout");
Yan Zheng2b820322008-11-17 21:11:30 -05002477 }
2478
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002479 root->fs_info->num_tolerated_disk_barrier_failures =
2480 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002481 ret = btrfs_commit_transaction(trans, root);
Yan Zheng2b820322008-11-17 21:11:30 -05002482
2483 if (seeding_dev) {
2484 mutex_unlock(&uuid_mutex);
2485 up_write(&sb->s_umount);
2486
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002487 if (ret) /* transaction commit */
2488 return ret;
2489
Yan Zheng2b820322008-11-17 21:11:30 -05002490 ret = btrfs_relocate_sys_chunks(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002491 if (ret < 0)
Anand Jain34d97002016-03-16 16:43:06 +08002492 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002493 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
Miao Xie671415b2012-10-16 11:26:46 +00002494 trans = btrfs_attach_transaction(root);
2495 if (IS_ERR(trans)) {
2496 if (PTR_ERR(trans) == -ENOENT)
2497 return 0;
2498 return PTR_ERR(trans);
2499 }
2500 ret = btrfs_commit_transaction(trans, root);
Yan Zheng2b820322008-11-17 21:11:30 -05002501 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02002502
Qu Wenruo5a1972b2014-04-16 17:02:32 +08002503 /* Update ctime/mtime for libblkid */
2504 update_dev_time(device_path);
Chris Mason788f20e2008-04-28 15:29:42 -04002505 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002506
2507error_trans:
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002508 btrfs_end_transaction(trans, root);
Josef Bacik606686e2012-06-04 14:03:51 -04002509 rcu_string_free(device->name);
Anand Jain32576042015-08-14 18:32:49 +08002510 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002511 kfree(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002512error:
Tejun Heoe525fd82010-11-13 11:55:17 +01002513 blkdev_put(bdev, FMODE_EXCL);
Yan Zheng2b820322008-11-17 21:11:30 -05002514 if (seeding_dev) {
2515 mutex_unlock(&uuid_mutex);
2516 up_write(&sb->s_umount);
2517 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02002518 return ret;
Chris Mason788f20e2008-04-28 15:29:42 -04002519}
2520
Stefan Behrense93c89c2012-11-05 17:33:06 +01002521int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
Miao Xie1c433662014-09-03 21:35:32 +08002522 struct btrfs_device *srcdev,
Stefan Behrense93c89c2012-11-05 17:33:06 +01002523 struct btrfs_device **device_out)
2524{
2525 struct request_queue *q;
2526 struct btrfs_device *device;
2527 struct block_device *bdev;
2528 struct btrfs_fs_info *fs_info = root->fs_info;
2529 struct list_head *devices;
2530 struct rcu_string *name;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002531 u64 devid = BTRFS_DEV_REPLACE_DEVID;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002532 int ret = 0;
2533
2534 *device_out = NULL;
Miao Xie1c433662014-09-03 21:35:32 +08002535 if (fs_info->fs_devices->seeding) {
2536 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
Stefan Behrense93c89c2012-11-05 17:33:06 +01002537 return -EINVAL;
Miao Xie1c433662014-09-03 21:35:32 +08002538 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002539
2540 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2541 fs_info->bdev_holder);
Miao Xie1c433662014-09-03 21:35:32 +08002542 if (IS_ERR(bdev)) {
2543 btrfs_err(fs_info, "target device %s is invalid!", device_path);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002544 return PTR_ERR(bdev);
Miao Xie1c433662014-09-03 21:35:32 +08002545 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002546
2547 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2548
2549 devices = &fs_info->fs_devices->devices;
2550 list_for_each_entry(device, devices, dev_list) {
2551 if (device->bdev == bdev) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002552 btrfs_err(fs_info,
2553 "target device is in the filesystem!");
Stefan Behrense93c89c2012-11-05 17:33:06 +01002554 ret = -EEXIST;
2555 goto error;
2556 }
2557 }
2558
Miao Xie1c433662014-09-03 21:35:32 +08002559
Miao Xie7cc8e582014-09-03 21:35:38 +08002560 if (i_size_read(bdev->bd_inode) <
2561 btrfs_device_get_total_bytes(srcdev)) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002562 btrfs_err(fs_info,
2563 "target device is smaller than source device!");
Miao Xie1c433662014-09-03 21:35:32 +08002564 ret = -EINVAL;
2565 goto error;
2566 }
2567
2568
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002569 device = btrfs_alloc_device(NULL, &devid, NULL);
2570 if (IS_ERR(device)) {
2571 ret = PTR_ERR(device);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002572 goto error;
2573 }
2574
2575 name = rcu_string_strdup(device_path, GFP_NOFS);
2576 if (!name) {
2577 kfree(device);
2578 ret = -ENOMEM;
2579 goto error;
2580 }
2581 rcu_assign_pointer(device->name, name);
2582
2583 q = bdev_get_queue(bdev);
2584 if (blk_queue_discard(q))
2585 device->can_discard = 1;
2586 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2587 device->writeable = 1;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002588 device->generation = 0;
2589 device->io_width = root->sectorsize;
2590 device->io_align = root->sectorsize;
2591 device->sector_size = root->sectorsize;
Miao Xie7cc8e582014-09-03 21:35:38 +08002592 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2593 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2594 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
Miao Xie935e5cc2014-09-03 21:35:33 +08002595 ASSERT(list_empty(&srcdev->resized_list));
2596 device->commit_total_bytes = srcdev->commit_total_bytes;
Miao Xiece7213c2014-09-03 21:35:34 +08002597 device->commit_bytes_used = device->bytes_used;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002598 device->dev_root = fs_info->dev_root;
2599 device->bdev = bdev;
2600 device->in_fs_metadata = 1;
2601 device->is_tgtdev_for_dev_replace = 1;
2602 device->mode = FMODE_EXCL;
Stefan Behrens27087f32013-10-11 15:20:42 +02002603 device->dev_stats_valid = 1;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002604 set_blocksize(device->bdev, 4096);
2605 device->fs_devices = fs_info->fs_devices;
2606 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2607 fs_info->fs_devices->num_devices++;
2608 fs_info->fs_devices->open_devices++;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002609 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2610
2611 *device_out = device;
2612 return ret;
2613
2614error:
2615 blkdev_put(bdev, FMODE_EXCL);
2616 return ret;
2617}
2618
2619void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2620 struct btrfs_device *tgtdev)
2621{
2622 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2623 tgtdev->io_width = fs_info->dev_root->sectorsize;
2624 tgtdev->io_align = fs_info->dev_root->sectorsize;
2625 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2626 tgtdev->dev_root = fs_info->dev_root;
2627 tgtdev->in_fs_metadata = 1;
2628}
2629
Chris Masond3977122009-01-05 21:25:51 -05002630static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2631 struct btrfs_device *device)
Chris Mason0b86a832008-03-24 15:01:56 -04002632{
2633 int ret;
2634 struct btrfs_path *path;
2635 struct btrfs_root *root;
2636 struct btrfs_dev_item *dev_item;
2637 struct extent_buffer *leaf;
2638 struct btrfs_key key;
2639
2640 root = device->dev_root->fs_info->chunk_root;
2641
2642 path = btrfs_alloc_path();
2643 if (!path)
2644 return -ENOMEM;
2645
2646 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2647 key.type = BTRFS_DEV_ITEM_KEY;
2648 key.offset = device->devid;
2649
2650 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2651 if (ret < 0)
2652 goto out;
2653
2654 if (ret > 0) {
2655 ret = -ENOENT;
2656 goto out;
2657 }
2658
2659 leaf = path->nodes[0];
2660 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2661
2662 btrfs_set_device_id(leaf, dev_item, device->devid);
2663 btrfs_set_device_type(leaf, dev_item, device->type);
2664 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2665 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2666 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Miao Xie7cc8e582014-09-03 21:35:38 +08002667 btrfs_set_device_total_bytes(leaf, dev_item,
2668 btrfs_device_get_disk_total_bytes(device));
2669 btrfs_set_device_bytes_used(leaf, dev_item,
2670 btrfs_device_get_bytes_used(device));
Chris Mason0b86a832008-03-24 15:01:56 -04002671 btrfs_mark_buffer_dirty(leaf);
2672
2673out:
2674 btrfs_free_path(path);
2675 return ret;
2676}
2677
Miao Xie2196d6e2014-09-03 21:35:41 +08002678int btrfs_grow_device(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -04002679 struct btrfs_device *device, u64 new_size)
2680{
2681 struct btrfs_super_block *super_copy =
David Sterba6c417612011-04-13 15:41:04 +02002682 device->dev_root->fs_info->super_copy;
Miao Xie935e5cc2014-09-03 21:35:33 +08002683 struct btrfs_fs_devices *fs_devices;
Miao Xie2196d6e2014-09-03 21:35:41 +08002684 u64 old_total;
2685 u64 diff;
Chris Mason8f18cf12008-04-25 16:53:30 -04002686
Yan Zheng2b820322008-11-17 21:11:30 -05002687 if (!device->writeable)
2688 return -EACCES;
Miao Xie2196d6e2014-09-03 21:35:41 +08002689
2690 lock_chunks(device->dev_root);
2691 old_total = btrfs_super_total_bytes(super_copy);
2692 diff = new_size - device->total_bytes;
2693
Stefan Behrens63a212a2012-11-05 18:29:28 +01002694 if (new_size <= device->total_bytes ||
Miao Xie2196d6e2014-09-03 21:35:41 +08002695 device->is_tgtdev_for_dev_replace) {
2696 unlock_chunks(device->dev_root);
Yan Zheng2b820322008-11-17 21:11:30 -05002697 return -EINVAL;
Miao Xie2196d6e2014-09-03 21:35:41 +08002698 }
Yan Zheng2b820322008-11-17 21:11:30 -05002699
Miao Xie935e5cc2014-09-03 21:35:33 +08002700 fs_devices = device->dev_root->fs_info->fs_devices;
Chris Mason8f18cf12008-04-25 16:53:30 -04002701
2702 btrfs_set_super_total_bytes(super_copy, old_total + diff);
Yan Zheng2b820322008-11-17 21:11:30 -05002703 device->fs_devices->total_rw_bytes += diff;
2704
Miao Xie7cc8e582014-09-03 21:35:38 +08002705 btrfs_device_set_total_bytes(device, new_size);
2706 btrfs_device_set_disk_total_bytes(device, new_size);
Chris Mason4184ea72009-03-10 12:39:20 -04002707 btrfs_clear_space_info_full(device->dev_root->fs_info);
Miao Xie935e5cc2014-09-03 21:35:33 +08002708 if (list_empty(&device->resized_list))
2709 list_add_tail(&device->resized_list,
2710 &fs_devices->resized_devices);
Miao Xie2196d6e2014-09-03 21:35:41 +08002711 unlock_chunks(device->dev_root);
Chris Mason4184ea72009-03-10 12:39:20 -04002712
Chris Mason8f18cf12008-04-25 16:53:30 -04002713 return btrfs_update_device(trans, device);
2714}
2715
2716static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002717 struct btrfs_fs_info *fs_info, u64 chunk_objectid,
Chris Mason8f18cf12008-04-25 16:53:30 -04002718 u64 chunk_offset)
2719{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002720 struct btrfs_root *root = fs_info->chunk_root;
Chris Mason8f18cf12008-04-25 16:53:30 -04002721 int ret;
2722 struct btrfs_path *path;
2723 struct btrfs_key key;
2724
Chris Mason8f18cf12008-04-25 16:53:30 -04002725 path = btrfs_alloc_path();
2726 if (!path)
2727 return -ENOMEM;
2728
2729 key.objectid = chunk_objectid;
2730 key.offset = chunk_offset;
2731 key.type = BTRFS_CHUNK_ITEM_KEY;
2732
2733 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002734 if (ret < 0)
2735 goto out;
2736 else if (ret > 0) { /* Logic error or corruption */
Anand Jain34d97002016-03-16 16:43:06 +08002737 btrfs_handle_fs_error(root->fs_info, -ENOENT,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002738 "Failed lookup while freeing chunk.");
2739 ret = -ENOENT;
2740 goto out;
2741 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002742
2743 ret = btrfs_del_item(trans, root, path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002744 if (ret < 0)
Anand Jain34d97002016-03-16 16:43:06 +08002745 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002746 "Failed to delete chunk item.");
2747out:
Chris Mason8f18cf12008-04-25 16:53:30 -04002748 btrfs_free_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00002749 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002750}
2751
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002752static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2753 u64 chunk_objectid, u64 chunk_offset)
Chris Mason8f18cf12008-04-25 16:53:30 -04002754{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002755 struct btrfs_root *root = fs_info->chunk_root;
David Sterba6c417612011-04-13 15:41:04 +02002756 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason8f18cf12008-04-25 16:53:30 -04002757 struct btrfs_disk_key *disk_key;
2758 struct btrfs_chunk *chunk;
2759 u8 *ptr;
2760 int ret = 0;
2761 u32 num_stripes;
2762 u32 array_size;
2763 u32 len = 0;
2764 u32 cur;
2765 struct btrfs_key key;
2766
Miao Xie2196d6e2014-09-03 21:35:41 +08002767 lock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002768 array_size = btrfs_super_sys_array_size(super_copy);
2769
2770 ptr = super_copy->sys_chunk_array;
2771 cur = 0;
2772
2773 while (cur < array_size) {
2774 disk_key = (struct btrfs_disk_key *)ptr;
2775 btrfs_disk_key_to_cpu(&key, disk_key);
2776
2777 len = sizeof(*disk_key);
2778
2779 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2780 chunk = (struct btrfs_chunk *)(ptr + len);
2781 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2782 len += btrfs_chunk_item_size(num_stripes);
2783 } else {
2784 ret = -EIO;
2785 break;
2786 }
2787 if (key.objectid == chunk_objectid &&
2788 key.offset == chunk_offset) {
2789 memmove(ptr, ptr + len, array_size - (cur + len));
2790 array_size -= len;
2791 btrfs_set_super_sys_array_size(super_copy, array_size);
2792 } else {
2793 ptr += len;
2794 cur += len;
2795 }
2796 }
Miao Xie2196d6e2014-09-03 21:35:41 +08002797 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002798 return ret;
2799}
2800
Josef Bacik47ab2a62014-09-18 11:20:02 -04002801int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002802 struct btrfs_fs_info *fs_info, u64 chunk_offset)
Josef Bacik47ab2a62014-09-18 11:20:02 -04002803{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002804 struct btrfs_root *root = fs_info->chunk_root;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002805 struct extent_map_tree *em_tree;
2806 struct extent_map *em;
2807 struct btrfs_root *extent_root = root->fs_info->extent_root;
2808 struct map_lookup *map;
2809 u64 dev_extent_len = 0;
2810 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002811 int i, ret = 0;
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002812 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002813
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002814 em_tree = &fs_info->mapping_tree.map_tree;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002815
2816 read_lock(&em_tree->lock);
2817 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2818 read_unlock(&em_tree->lock);
2819
2820 if (!em || em->start > chunk_offset ||
2821 em->start + em->len < chunk_offset) {
2822 /*
2823 * This is a logic error, but we don't want to just rely on the
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08002824 * user having built with ASSERT enabled, so if ASSERT doesn't
Josef Bacik47ab2a62014-09-18 11:20:02 -04002825 * do anything we still error out.
2826 */
2827 ASSERT(0);
2828 if (em)
2829 free_extent_map(em);
2830 return -EINVAL;
2831 }
Jeff Mahoney95617d62015-06-03 10:55:48 -04002832 map = em->map_lookup;
Filipe Manana39c2d7f2015-05-20 14:01:55 +01002833 lock_chunks(root->fs_info->chunk_root);
Filipe Manana4617ea32015-06-09 17:48:21 +01002834 check_system_chunk(trans, extent_root, map->type);
Filipe Manana39c2d7f2015-05-20 14:01:55 +01002835 unlock_chunks(root->fs_info->chunk_root);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002836
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002837 /*
2838 * Take the device list mutex to prevent races with the final phase of
2839 * a device replace operation that replaces the device object associated
2840 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2841 */
2842 mutex_lock(&fs_devices->device_list_mutex);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002843 for (i = 0; i < map->num_stripes; i++) {
2844 struct btrfs_device *device = map->stripes[i].dev;
2845 ret = btrfs_free_dev_extent(trans, device,
2846 map->stripes[i].physical,
2847 &dev_extent_len);
2848 if (ret) {
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002849 mutex_unlock(&fs_devices->device_list_mutex);
Jeff Mahoney66642832016-06-10 18:19:25 -04002850 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002851 goto out;
2852 }
2853
2854 if (device->bytes_used > 0) {
2855 lock_chunks(root);
2856 btrfs_device_set_bytes_used(device,
2857 device->bytes_used - dev_extent_len);
2858 spin_lock(&root->fs_info->free_chunk_lock);
2859 root->fs_info->free_chunk_space += dev_extent_len;
2860 spin_unlock(&root->fs_info->free_chunk_lock);
2861 btrfs_clear_space_info_full(root->fs_info);
2862 unlock_chunks(root);
2863 }
2864
2865 if (map->stripes[i].dev) {
2866 ret = btrfs_update_device(trans, map->stripes[i].dev);
2867 if (ret) {
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002868 mutex_unlock(&fs_devices->device_list_mutex);
Jeff Mahoney66642832016-06-10 18:19:25 -04002869 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002870 goto out;
2871 }
2872 }
2873 }
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002874 mutex_unlock(&fs_devices->device_list_mutex);
2875
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002876 ret = btrfs_free_chunk(trans, root->fs_info, chunk_objectid,
2877 chunk_offset);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002878 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002879 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002880 goto out;
2881 }
2882
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002883 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002884
2885 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002886 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2887 chunk_offset);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002888 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002889 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002890 goto out;
2891 }
2892 }
2893
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002894 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002895 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002896 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002897 goto out;
2898 }
2899
Josef Bacik47ab2a62014-09-18 11:20:02 -04002900out:
2901 /* once for us */
2902 free_extent_map(em);
Chris Mason8f18cf12008-04-25 16:53:30 -04002903 return ret;
2904}
2905
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002906static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
Chris Mason8f18cf12008-04-25 16:53:30 -04002907{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002908 struct btrfs_root *root = fs_info->chunk_root;
2909 struct btrfs_root *extent_root = fs_info->extent_root;
Chris Mason19c4d2f2016-10-10 13:43:31 -07002910 struct btrfs_trans_handle *trans;
Chris Mason8f18cf12008-04-25 16:53:30 -04002911 int ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002912
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002913 /*
2914 * Prevent races with automatic removal of unused block groups.
2915 * After we relocate and before we remove the chunk with offset
2916 * chunk_offset, automatic removal of the block group can kick in,
2917 * resulting in a failure when calling btrfs_remove_chunk() below.
2918 *
2919 * Make sure to acquire this mutex before doing a tree search (dev
2920 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2921 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2922 * we release the path used to search the chunk/dev tree and before
2923 * the current task acquires this mutex and calls us.
2924 */
2925 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2926
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002927 ret = btrfs_can_relocate(root->fs_info, chunk_offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04002928 if (ret)
2929 return -ENOSPC;
2930
Chris Mason8f18cf12008-04-25 16:53:30 -04002931 /* step one, relocate all the extents inside this chunk */
Zhaolei55e3a602015-08-05 16:43:30 +08002932 btrfs_scrub_pause(root);
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04002933 ret = btrfs_relocate_block_group(root->fs_info, chunk_offset);
Zhaolei55e3a602015-08-05 16:43:30 +08002934 btrfs_scrub_continue(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002935 if (ret)
2936 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002937
Chris Mason19c4d2f2016-10-10 13:43:31 -07002938 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2939 chunk_offset);
2940 if (IS_ERR(trans)) {
2941 ret = PTR_ERR(trans);
2942 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2943 return ret;
2944 }
Naohiro Aota5d8eb6f2016-09-02 16:46:32 +09002945
Chris Mason19c4d2f2016-10-10 13:43:31 -07002946 /*
2947 * step two, delete the device extents and the
2948 * chunk tree entries
2949 */
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04002950 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
Chris Mason19c4d2f2016-10-10 13:43:31 -07002951 btrfs_end_transaction(trans, extent_root);
2952 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002953}
2954
Yan Zheng2b820322008-11-17 21:11:30 -05002955static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2956{
2957 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2958 struct btrfs_path *path;
2959 struct extent_buffer *leaf;
2960 struct btrfs_chunk *chunk;
2961 struct btrfs_key key;
2962 struct btrfs_key found_key;
Yan Zheng2b820322008-11-17 21:11:30 -05002963 u64 chunk_type;
Josef Bacikba1bf482009-09-11 16:11:19 -04002964 bool retried = false;
2965 int failed = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05002966 int ret;
2967
2968 path = btrfs_alloc_path();
2969 if (!path)
2970 return -ENOMEM;
2971
Josef Bacikba1bf482009-09-11 16:11:19 -04002972again:
Yan Zheng2b820322008-11-17 21:11:30 -05002973 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2974 key.offset = (u64)-1;
2975 key.type = BTRFS_CHUNK_ITEM_KEY;
2976
2977 while (1) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002978 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002979 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002980 if (ret < 0) {
2981 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002982 goto error;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002983 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002984 BUG_ON(ret == 0); /* Corruption */
Yan Zheng2b820322008-11-17 21:11:30 -05002985
2986 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2987 key.type);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002988 if (ret)
2989 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002990 if (ret < 0)
2991 goto error;
2992 if (ret > 0)
2993 break;
2994
2995 leaf = path->nodes[0];
2996 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2997
2998 chunk = btrfs_item_ptr(leaf, path->slots[0],
2999 struct btrfs_chunk);
3000 chunk_type = btrfs_chunk_type(leaf, chunk);
David Sterbab3b4aa72011-04-21 01:20:15 +02003001 btrfs_release_path(path);
Yan Zheng2b820322008-11-17 21:11:30 -05003002
3003 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04003004 ret = btrfs_relocate_chunk(root->fs_info,
Yan Zheng2b820322008-11-17 21:11:30 -05003005 found_key.offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04003006 if (ret == -ENOSPC)
3007 failed++;
HIMANGI SARAOGI14586652014-07-09 03:51:41 +05303008 else
3009 BUG_ON(ret);
Yan Zheng2b820322008-11-17 21:11:30 -05003010 }
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003011 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05003012
3013 if (found_key.offset == 0)
3014 break;
3015 key.offset = found_key.offset - 1;
3016 }
3017 ret = 0;
Josef Bacikba1bf482009-09-11 16:11:19 -04003018 if (failed && !retried) {
3019 failed = 0;
3020 retried = true;
3021 goto again;
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303022 } else if (WARN_ON(failed && retried)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04003023 ret = -ENOSPC;
3024 }
Yan Zheng2b820322008-11-17 21:11:30 -05003025error:
3026 btrfs_free_path(path);
3027 return ret;
3028}
3029
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003030static int insert_balance_item(struct btrfs_fs_info *fs_info,
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003031 struct btrfs_balance_control *bctl)
3032{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003033 struct btrfs_root *root = fs_info->tree_root;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003034 struct btrfs_trans_handle *trans;
3035 struct btrfs_balance_item *item;
3036 struct btrfs_disk_balance_args disk_bargs;
3037 struct btrfs_path *path;
3038 struct extent_buffer *leaf;
3039 struct btrfs_key key;
3040 int ret, err;
3041
3042 path = btrfs_alloc_path();
3043 if (!path)
3044 return -ENOMEM;
3045
3046 trans = btrfs_start_transaction(root, 0);
3047 if (IS_ERR(trans)) {
3048 btrfs_free_path(path);
3049 return PTR_ERR(trans);
3050 }
3051
3052 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003053 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003054 key.offset = 0;
3055
3056 ret = btrfs_insert_empty_item(trans, root, path, &key,
3057 sizeof(*item));
3058 if (ret)
3059 goto out;
3060
3061 leaf = path->nodes[0];
3062 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3063
David Sterbab159fa22016-11-08 18:09:03 +01003064 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003065
3066 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3067 btrfs_set_balance_data(leaf, item, &disk_bargs);
3068 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3069 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3070 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3071 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3072
3073 btrfs_set_balance_flags(leaf, item, bctl->flags);
3074
3075 btrfs_mark_buffer_dirty(leaf);
3076out:
3077 btrfs_free_path(path);
3078 err = btrfs_commit_transaction(trans, root);
3079 if (err && !ret)
3080 ret = err;
3081 return ret;
3082}
3083
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003084static int del_balance_item(struct btrfs_fs_info *fs_info)
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003085{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003086 struct btrfs_root *root = fs_info->tree_root;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003087 struct btrfs_trans_handle *trans;
3088 struct btrfs_path *path;
3089 struct btrfs_key key;
3090 int ret, err;
3091
3092 path = btrfs_alloc_path();
3093 if (!path)
3094 return -ENOMEM;
3095
3096 trans = btrfs_start_transaction(root, 0);
3097 if (IS_ERR(trans)) {
3098 btrfs_free_path(path);
3099 return PTR_ERR(trans);
3100 }
3101
3102 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003103 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003104 key.offset = 0;
3105
3106 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3107 if (ret < 0)
3108 goto out;
3109 if (ret > 0) {
3110 ret = -ENOENT;
3111 goto out;
3112 }
3113
3114 ret = btrfs_del_item(trans, root, path);
3115out:
3116 btrfs_free_path(path);
3117 err = btrfs_commit_transaction(trans, root);
3118 if (err && !ret)
3119 ret = err;
3120 return ret;
3121}
3122
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003123/*
Ilya Dryomov59641012012-01-16 22:04:48 +02003124 * This is a heuristic used to reduce the number of chunks balanced on
3125 * resume after balance was interrupted.
3126 */
3127static void update_balance_args(struct btrfs_balance_control *bctl)
3128{
3129 /*
3130 * Turn on soft mode for chunk types that were being converted.
3131 */
3132 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3133 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3134 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138
3139 /*
3140 * Turn on usage filter if is not already used. The idea is
3141 * that chunks that we have already balanced should be
3142 * reasonably full. Don't do it for chunks that are being
3143 * converted - that will keep us from relocating unconverted
3144 * (albeit full) chunks.
3145 */
3146 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003147 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003148 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3149 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3150 bctl->data.usage = 90;
3151 }
3152 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003153 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003154 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3155 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3156 bctl->sys.usage = 90;
3157 }
3158 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003159 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003160 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162 bctl->meta.usage = 90;
3163 }
3164}
3165
3166/*
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003167 * Should be called with both balance and volume mutexes held to
3168 * serialize other volume operations (add_dev/rm_dev/resize) with
3169 * restriper. Same goes for unset_balance_control.
3170 */
3171static void set_balance_control(struct btrfs_balance_control *bctl)
3172{
3173 struct btrfs_fs_info *fs_info = bctl->fs_info;
3174
3175 BUG_ON(fs_info->balance_ctl);
3176
3177 spin_lock(&fs_info->balance_lock);
3178 fs_info->balance_ctl = bctl;
3179 spin_unlock(&fs_info->balance_lock);
3180}
3181
3182static void unset_balance_control(struct btrfs_fs_info *fs_info)
3183{
3184 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3185
3186 BUG_ON(!fs_info->balance_ctl);
3187
3188 spin_lock(&fs_info->balance_lock);
3189 fs_info->balance_ctl = NULL;
3190 spin_unlock(&fs_info->balance_lock);
3191
3192 kfree(bctl);
3193}
3194
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003195/*
3196 * Balance filters. Return 1 if chunk should be filtered out
3197 * (should not be balanced).
3198 */
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003199static int chunk_profiles_filter(u64 chunk_type,
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003200 struct btrfs_balance_args *bargs)
3201{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003202 chunk_type = chunk_to_extended(chunk_type) &
3203 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003204
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003205 if (bargs->profiles & chunk_type)
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003206 return 0;
3207
3208 return 1;
3209}
3210
Holger Hoffstättedba72cb2015-11-17 12:29:32 +01003211static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003212 struct btrfs_balance_args *bargs)
3213{
3214 struct btrfs_block_group_cache *cache;
David Sterbabc309462015-10-20 18:22:13 +02003215 u64 chunk_used;
3216 u64 user_thresh_min;
3217 u64 user_thresh_max;
3218 int ret = 1;
3219
3220 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3221 chunk_used = btrfs_block_group_used(&cache->item);
3222
3223 if (bargs->usage_min == 0)
3224 user_thresh_min = 0;
3225 else
3226 user_thresh_min = div_factor_fine(cache->key.offset,
3227 bargs->usage_min);
3228
3229 if (bargs->usage_max == 0)
3230 user_thresh_max = 1;
3231 else if (bargs->usage_max > 100)
3232 user_thresh_max = cache->key.offset;
3233 else
3234 user_thresh_max = div_factor_fine(cache->key.offset,
3235 bargs->usage_max);
3236
3237 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3238 ret = 0;
3239
3240 btrfs_put_block_group(cache);
3241 return ret;
3242}
3243
Holger Hoffstättedba72cb2015-11-17 12:29:32 +01003244static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
David Sterbabc309462015-10-20 18:22:13 +02003245 u64 chunk_offset, struct btrfs_balance_args *bargs)
3246{
3247 struct btrfs_block_group_cache *cache;
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003248 u64 chunk_used, user_thresh;
3249 int ret = 1;
3250
3251 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3252 chunk_used = btrfs_block_group_used(&cache->item);
3253
David Sterbabc309462015-10-20 18:22:13 +02003254 if (bargs->usage_min == 0)
Ilya Dryomov3e39cea2013-02-12 16:28:59 +00003255 user_thresh = 1;
Ilya Dryomova105bb82013-01-21 15:15:56 +02003256 else if (bargs->usage > 100)
3257 user_thresh = cache->key.offset;
3258 else
3259 user_thresh = div_factor_fine(cache->key.offset,
3260 bargs->usage);
3261
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003262 if (chunk_used < user_thresh)
3263 ret = 0;
3264
3265 btrfs_put_block_group(cache);
3266 return ret;
3267}
3268
Ilya Dryomov409d4042012-01-16 22:04:47 +02003269static int chunk_devid_filter(struct extent_buffer *leaf,
3270 struct btrfs_chunk *chunk,
3271 struct btrfs_balance_args *bargs)
3272{
3273 struct btrfs_stripe *stripe;
3274 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3275 int i;
3276
3277 for (i = 0; i < num_stripes; i++) {
3278 stripe = btrfs_stripe_nr(chunk, i);
3279 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3280 return 0;
3281 }
3282
3283 return 1;
3284}
3285
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003286/* [pstart, pend) */
3287static int chunk_drange_filter(struct extent_buffer *leaf,
3288 struct btrfs_chunk *chunk,
3289 u64 chunk_offset,
3290 struct btrfs_balance_args *bargs)
3291{
3292 struct btrfs_stripe *stripe;
3293 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3294 u64 stripe_offset;
3295 u64 stripe_length;
3296 int factor;
3297 int i;
3298
3299 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3300 return 0;
3301
3302 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
David Woodhouse53b381b2013-01-29 18:40:14 -05003303 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3304 factor = num_stripes / 2;
3305 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3306 factor = num_stripes - 1;
3307 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3308 factor = num_stripes - 2;
3309 } else {
3310 factor = num_stripes;
3311 }
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003312
3313 for (i = 0; i < num_stripes; i++) {
3314 stripe = btrfs_stripe_nr(chunk, i);
3315 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3316 continue;
3317
3318 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3319 stripe_length = btrfs_chunk_length(leaf, chunk);
David Sterbab8b93ad2015-01-16 17:26:13 +01003320 stripe_length = div_u64(stripe_length, factor);
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003321
3322 if (stripe_offset < bargs->pend &&
3323 stripe_offset + stripe_length > bargs->pstart)
3324 return 0;
3325 }
3326
3327 return 1;
3328}
3329
Ilya Dryomovea671762012-01-16 22:04:48 +02003330/* [vstart, vend) */
3331static int chunk_vrange_filter(struct extent_buffer *leaf,
3332 struct btrfs_chunk *chunk,
3333 u64 chunk_offset,
3334 struct btrfs_balance_args *bargs)
3335{
3336 if (chunk_offset < bargs->vend &&
3337 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3338 /* at least part of the chunk is inside this vrange */
3339 return 0;
3340
3341 return 1;
3342}
3343
Gabríel Arthúr Péturssondee32d02015-09-28 22:32:41 +00003344static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3345 struct btrfs_chunk *chunk,
3346 struct btrfs_balance_args *bargs)
3347{
3348 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3349
3350 if (bargs->stripes_min <= num_stripes
3351 && num_stripes <= bargs->stripes_max)
3352 return 0;
3353
3354 return 1;
3355}
3356
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003357static int chunk_soft_convert_filter(u64 chunk_type,
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003358 struct btrfs_balance_args *bargs)
3359{
3360 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3361 return 0;
3362
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003363 chunk_type = chunk_to_extended(chunk_type) &
3364 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003365
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003366 if (bargs->target == chunk_type)
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003367 return 1;
3368
3369 return 0;
3370}
3371
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003372static int should_balance_chunk(struct btrfs_root *root,
3373 struct extent_buffer *leaf,
3374 struct btrfs_chunk *chunk, u64 chunk_offset)
3375{
3376 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3377 struct btrfs_balance_args *bargs = NULL;
3378 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3379
3380 /* type filter */
3381 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3382 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3383 return 0;
3384 }
3385
3386 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3387 bargs = &bctl->data;
3388 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3389 bargs = &bctl->sys;
3390 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3391 bargs = &bctl->meta;
3392
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003393 /* profiles filter */
3394 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3395 chunk_profiles_filter(chunk_type, bargs)) {
3396 return 0;
3397 }
3398
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003399 /* usage filter */
3400 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3401 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3402 return 0;
David Sterbabc309462015-10-20 18:22:13 +02003403 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3404 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3405 return 0;
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003406 }
3407
Ilya Dryomov409d4042012-01-16 22:04:47 +02003408 /* devid filter */
3409 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3410 chunk_devid_filter(leaf, chunk, bargs)) {
3411 return 0;
3412 }
3413
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003414 /* drange filter, makes sense only with devid filter */
3415 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3416 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3417 return 0;
3418 }
3419
Ilya Dryomovea671762012-01-16 22:04:48 +02003420 /* vrange filter */
3421 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3422 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3423 return 0;
3424 }
3425
Gabríel Arthúr Péturssondee32d02015-09-28 22:32:41 +00003426 /* stripes filter */
3427 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3428 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3429 return 0;
3430 }
3431
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003432 /* soft profile changing mode */
3433 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3434 chunk_soft_convert_filter(chunk_type, bargs)) {
3435 return 0;
3436 }
3437
David Sterba7d824b62014-05-07 17:37:51 +02003438 /*
3439 * limited by count, must be the last filter
3440 */
3441 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3442 if (bargs->limit == 0)
3443 return 0;
3444 else
3445 bargs->limit--;
David Sterba12907fc2015-10-10 17:16:50 +02003446 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3447 /*
3448 * Same logic as the 'limit' filter; the minimum cannot be
Nicholas D Steeves01327612016-05-19 21:18:45 -04003449 * determined here because we do not have the global information
David Sterba12907fc2015-10-10 17:16:50 +02003450 * about the count of all chunks that satisfy the filters.
3451 */
3452 if (bargs->limit_max == 0)
3453 return 0;
3454 else
3455 bargs->limit_max--;
David Sterba7d824b62014-05-07 17:37:51 +02003456 }
3457
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003458 return 1;
3459}
3460
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003461static int __btrfs_balance(struct btrfs_fs_info *fs_info)
Chris Masonec44a352008-04-28 15:29:52 -04003462{
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003463 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003464 struct btrfs_root *chunk_root = fs_info->chunk_root;
3465 struct btrfs_root *dev_root = fs_info->dev_root;
3466 struct list_head *devices;
Chris Masonec44a352008-04-28 15:29:52 -04003467 struct btrfs_device *device;
3468 u64 old_size;
3469 u64 size_to_free;
David Sterba12907fc2015-10-10 17:16:50 +02003470 u64 chunk_type;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003471 struct btrfs_chunk *chunk;
Liu Bo5a488b92016-07-12 11:24:21 -07003472 struct btrfs_path *path = NULL;
Chris Masonec44a352008-04-28 15:29:52 -04003473 struct btrfs_key key;
Chris Masonec44a352008-04-28 15:29:52 -04003474 struct btrfs_key found_key;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003475 struct btrfs_trans_handle *trans;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003476 struct extent_buffer *leaf;
3477 int slot;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003478 int ret;
3479 int enospc_errors = 0;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003480 bool counting = true;
David Sterba12907fc2015-10-10 17:16:50 +02003481 /* The single value limit and min/max limits use the same bytes in the */
David Sterba7d824b62014-05-07 17:37:51 +02003482 u64 limit_data = bctl->data.limit;
3483 u64 limit_meta = bctl->meta.limit;
3484 u64 limit_sys = bctl->sys.limit;
David Sterba12907fc2015-10-10 17:16:50 +02003485 u32 count_data = 0;
3486 u32 count_meta = 0;
3487 u32 count_sys = 0;
Zhao Lei2c9fe832015-11-09 11:51:32 +08003488 int chunk_reserved = 0;
Liu Bocf25ce52015-12-14 18:29:32 -08003489 u64 bytes_used = 0;
Chris Masonec44a352008-04-28 15:29:52 -04003490
Chris Masonec44a352008-04-28 15:29:52 -04003491 /* step one make some room on all the devices */
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003492 devices = &fs_info->fs_devices->devices;
Qinghuang Fengc6e30872009-01-21 10:59:08 -05003493 list_for_each_entry(device, devices, dev_list) {
Miao Xie7cc8e582014-09-03 21:35:38 +08003494 old_size = btrfs_device_get_total_bytes(device);
Chris Masonec44a352008-04-28 15:29:52 -04003495 size_to_free = div_factor(old_size, 1);
Byongho Leeee221842015-12-15 01:42:10 +09003496 size_to_free = min_t(u64, size_to_free, SZ_1M);
Yan Zheng2b820322008-11-17 21:11:30 -05003497 if (!device->writeable ||
Miao Xie7cc8e582014-09-03 21:35:38 +08003498 btrfs_device_get_total_bytes(device) -
3499 btrfs_device_get_bytes_used(device) > size_to_free ||
Stefan Behrens63a212a2012-11-05 18:29:28 +01003500 device->is_tgtdev_for_dev_replace)
Chris Masonec44a352008-04-28 15:29:52 -04003501 continue;
3502
3503 ret = btrfs_shrink_device(device, old_size - size_to_free);
Josef Bacikba1bf482009-09-11 16:11:19 -04003504 if (ret == -ENOSPC)
3505 break;
Liu Bo5a488b92016-07-12 11:24:21 -07003506 if (ret) {
3507 /* btrfs_shrink_device never returns ret > 0 */
3508 WARN_ON(ret > 0);
3509 goto error;
3510 }
Chris Masonec44a352008-04-28 15:29:52 -04003511
Yan, Zhenga22285a2010-05-16 10:48:46 -04003512 trans = btrfs_start_transaction(dev_root, 0);
Liu Bo5a488b92016-07-12 11:24:21 -07003513 if (IS_ERR(trans)) {
3514 ret = PTR_ERR(trans);
3515 btrfs_info_in_rcu(fs_info,
3516 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3517 rcu_str_deref(device->name), ret,
3518 old_size, old_size - size_to_free);
3519 goto error;
3520 }
Chris Masonec44a352008-04-28 15:29:52 -04003521
3522 ret = btrfs_grow_device(trans, device, old_size);
Liu Bo5a488b92016-07-12 11:24:21 -07003523 if (ret) {
3524 btrfs_end_transaction(trans, dev_root);
3525 /* btrfs_grow_device never returns ret > 0 */
3526 WARN_ON(ret > 0);
3527 btrfs_info_in_rcu(fs_info,
3528 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device->name), ret,
3530 old_size, old_size - size_to_free);
3531 goto error;
3532 }
Chris Masonec44a352008-04-28 15:29:52 -04003533
3534 btrfs_end_transaction(trans, dev_root);
3535 }
3536
3537 /* step two, relocate all the chunks */
3538 path = btrfs_alloc_path();
Mark Fasheh17e9f792011-07-12 11:10:23 -07003539 if (!path) {
3540 ret = -ENOMEM;
3541 goto error;
3542 }
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003543
3544 /* zero out stat counters */
3545 spin_lock(&fs_info->balance_lock);
3546 memset(&bctl->stat, 0, sizeof(bctl->stat));
3547 spin_unlock(&fs_info->balance_lock);
3548again:
David Sterba7d824b62014-05-07 17:37:51 +02003549 if (!counting) {
David Sterba12907fc2015-10-10 17:16:50 +02003550 /*
3551 * The single value limit and min/max limits use the same bytes
3552 * in the
3553 */
David Sterba7d824b62014-05-07 17:37:51 +02003554 bctl->data.limit = limit_data;
3555 bctl->meta.limit = limit_meta;
3556 bctl->sys.limit = limit_sys;
3557 }
Chris Masonec44a352008-04-28 15:29:52 -04003558 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3559 key.offset = (u64)-1;
3560 key.type = BTRFS_CHUNK_ITEM_KEY;
3561
Chris Masond3977122009-01-05 21:25:51 -05003562 while (1) {
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003563 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003564 atomic_read(&fs_info->balance_cancel_req)) {
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003565 ret = -ECANCELED;
3566 goto error;
3567 }
3568
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003569 mutex_lock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003570 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003571 if (ret < 0) {
3572 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003573 goto error;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003574 }
Chris Masonec44a352008-04-28 15:29:52 -04003575
3576 /*
3577 * this shouldn't happen, it means the last relocate
3578 * failed
3579 */
3580 if (ret == 0)
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003581 BUG(); /* FIXME break ? */
Chris Masonec44a352008-04-28 15:29:52 -04003582
3583 ret = btrfs_previous_item(chunk_root, path, 0,
3584 BTRFS_CHUNK_ITEM_KEY);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003585 if (ret) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003586 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003587 ret = 0;
Chris Masonec44a352008-04-28 15:29:52 -04003588 break;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003589 }
Chris Mason7d9eb122008-07-08 14:19:17 -04003590
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003591 leaf = path->nodes[0];
3592 slot = path->slots[0];
3593 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3594
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003595 if (found_key.objectid != key.objectid) {
3596 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003597 break;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003598 }
Chris Mason7d9eb122008-07-08 14:19:17 -04003599
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003600 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
David Sterba12907fc2015-10-10 17:16:50 +02003601 chunk_type = btrfs_chunk_type(leaf, chunk);
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003602
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003603 if (!counting) {
3604 spin_lock(&fs_info->balance_lock);
3605 bctl->stat.considered++;
3606 spin_unlock(&fs_info->balance_lock);
3607 }
3608
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003609 ret = should_balance_chunk(chunk_root, leaf, chunk,
3610 found_key.offset);
Zhao Lei2c9fe832015-11-09 11:51:32 +08003611
David Sterbab3b4aa72011-04-21 01:20:15 +02003612 btrfs_release_path(path);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003613 if (!ret) {
3614 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003615 goto loop;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003616 }
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003617
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003618 if (counting) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003619 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003620 spin_lock(&fs_info->balance_lock);
3621 bctl->stat.expected++;
3622 spin_unlock(&fs_info->balance_lock);
David Sterba12907fc2015-10-10 17:16:50 +02003623
3624 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3625 count_data++;
3626 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3627 count_sys++;
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3629 count_meta++;
3630
3631 goto loop;
3632 }
3633
3634 /*
3635 * Apply limit_min filter, no need to check if the LIMITS
3636 * filter is used, limit_min is 0 by default
3637 */
3638 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3639 count_data < bctl->data.limit_min)
3640 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3641 count_meta < bctl->meta.limit_min)
3642 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3643 count_sys < bctl->sys.limit_min)) {
3644 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003645 goto loop;
3646 }
3647
Liu Bocf25ce52015-12-14 18:29:32 -08003648 ASSERT(fs_info->data_sinfo);
3649 spin_lock(&fs_info->data_sinfo->lock);
3650 bytes_used = fs_info->data_sinfo->bytes_used;
3651 spin_unlock(&fs_info->data_sinfo->lock);
3652
3653 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3654 !chunk_reserved && !bytes_used) {
Zhao Lei2c9fe832015-11-09 11:51:32 +08003655 trans = btrfs_start_transaction(chunk_root, 0);
3656 if (IS_ERR(trans)) {
3657 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3658 ret = PTR_ERR(trans);
3659 goto error;
3660 }
3661
3662 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3663 BTRFS_BLOCK_GROUP_DATA);
Filipe Manana8a7d6562015-12-10 11:08:59 +00003664 btrfs_end_transaction(trans, chunk_root);
Zhao Lei2c9fe832015-11-09 11:51:32 +08003665 if (ret < 0) {
3666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3667 goto error;
3668 }
Zhao Lei2c9fe832015-11-09 11:51:32 +08003669 chunk_reserved = 1;
3670 }
3671
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04003672 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003673 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Josef Bacik508794e2011-07-02 21:24:41 +00003674 if (ret && ret != -ENOSPC)
3675 goto error;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003676 if (ret == -ENOSPC) {
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003677 enospc_errors++;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003678 } else {
3679 spin_lock(&fs_info->balance_lock);
3680 bctl->stat.completed++;
3681 spin_unlock(&fs_info->balance_lock);
3682 }
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003683loop:
Ilya Dryomov795a3322013-08-27 13:50:44 +03003684 if (found_key.offset == 0)
3685 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04003686 key.offset = found_key.offset - 1;
Chris Masonec44a352008-04-28 15:29:52 -04003687 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003688
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003689 if (counting) {
3690 btrfs_release_path(path);
3691 counting = false;
3692 goto again;
3693 }
Chris Masonec44a352008-04-28 15:29:52 -04003694error:
3695 btrfs_free_path(path);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003696 if (enospc_errors) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003697 btrfs_info(fs_info, "%d enospc errors during balance",
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003698 enospc_errors);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003699 if (!ret)
3700 ret = -ENOSPC;
3701 }
3702
Chris Masonec44a352008-04-28 15:29:52 -04003703 return ret;
3704}
3705
Ilya Dryomov0c460c02012-03-27 17:09:17 +03003706/**
3707 * alloc_profile_is_valid - see if a given profile is valid and reduced
3708 * @flags: profile to validate
3709 * @extended: if true @flags is treated as an extended profile
3710 */
3711static int alloc_profile_is_valid(u64 flags, int extended)
3712{
3713 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3714 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3715
3716 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3717
3718 /* 1) check that all other bits are zeroed */
3719 if (flags & ~mask)
3720 return 0;
3721
3722 /* 2) see if profile is reduced */
3723 if (flags == 0)
3724 return !extended; /* "0" is valid for usual profiles */
3725
3726 /* true if exactly one bit set */
3727 return (flags & (flags - 1)) == 0;
3728}
3729
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003730static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3731{
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003732 /* cancel requested || normal exit path */
3733 return atomic_read(&fs_info->balance_cancel_req) ||
3734 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3735 atomic_read(&fs_info->balance_cancel_req) == 0);
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003736}
3737
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003738static void __cancel_balance(struct btrfs_fs_info *fs_info)
3739{
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003740 int ret;
3741
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003742 unset_balance_control(fs_info);
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003743 ret = del_balance_item(fs_info);
Liu Bo0f788c52013-03-04 16:25:40 +00003744 if (ret)
Anand Jain34d97002016-03-16 16:43:06 +08003745 btrfs_handle_fs_error(fs_info, ret, NULL);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003746
3747 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003748}
3749
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003750/* Non-zero return value signifies invalidity */
3751static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3752 u64 allowed)
3753{
3754 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3755 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3756 (bctl_arg->target & ~allowed)));
3757}
3758
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003759/*
3760 * Should be called with both balance and volume mutexes held
3761 */
3762int btrfs_balance(struct btrfs_balance_control *bctl,
3763 struct btrfs_ioctl_balance_args *bargs)
3764{
3765 struct btrfs_fs_info *fs_info = bctl->fs_info;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003766 u64 allowed;
Ilya Dryomove4837f82012-03-27 17:09:17 +03003767 int mixed = 0;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003768 int ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01003769 u64 num_devices;
Miao Xiede98ced2013-01-29 10:13:12 +00003770 unsigned seq;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003771
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003772 if (btrfs_fs_closing(fs_info) ||
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003773 atomic_read(&fs_info->balance_pause_req) ||
3774 atomic_read(&fs_info->balance_cancel_req)) {
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003775 ret = -EINVAL;
3776 goto out;
3777 }
3778
Ilya Dryomove4837f82012-03-27 17:09:17 +03003779 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3780 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3781 mixed = 1;
3782
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003783 /*
3784 * In case of mixed groups both data and meta should be picked,
3785 * and identical options should be given for both of them.
3786 */
Ilya Dryomove4837f82012-03-27 17:09:17 +03003787 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3788 if (mixed && (bctl->flags & allowed)) {
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003789 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3790 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3791 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003792 btrfs_err(fs_info,
3793 "with mixed groups data and metadata balance options must be the same");
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003794 ret = -EINVAL;
3795 goto out;
3796 }
3797 }
3798
Stefan Behrens8dabb742012-11-06 13:15:27 +01003799 num_devices = fs_info->fs_devices->num_devices;
Liu Bo73beece2015-07-17 16:49:19 +08003800 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
Stefan Behrens8dabb742012-11-06 13:15:27 +01003801 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3802 BUG_ON(num_devices < 1);
3803 num_devices--;
3804 }
Liu Bo73beece2015-07-17 16:49:19 +08003805 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
Austin S. Hemmelgarn88be1592016-03-23 14:22:59 -04003806 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3807 if (num_devices > 1)
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003808 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
Andreas Philipp8250dab2013-05-11 11:13:03 +00003809 if (num_devices > 2)
3810 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3811 if (num_devices > 3)
3812 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3813 BTRFS_BLOCK_GROUP_RAID6);
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003814 if (validate_convert_profile(&bctl->data, allowed)) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003815 btrfs_err(fs_info,
3816 "unable to start balance with target data profile %llu",
3817 bctl->data.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003818 ret = -EINVAL;
3819 goto out;
3820 }
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003821 if (validate_convert_profile(&bctl->meta, allowed)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003822 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003823 "unable to start balance with target metadata profile %llu",
3824 bctl->meta.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003825 ret = -EINVAL;
3826 goto out;
3827 }
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003828 if (validate_convert_profile(&bctl->sys, allowed)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003829 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003830 "unable to start balance with target system profile %llu",
3831 bctl->sys.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003832 ret = -EINVAL;
3833 goto out;
3834 }
3835
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003836 /* allow to reduce meta or sys integrity only if force set */
3837 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05003838 BTRFS_BLOCK_GROUP_RAID10 |
3839 BTRFS_BLOCK_GROUP_RAID5 |
3840 BTRFS_BLOCK_GROUP_RAID6;
Miao Xiede98ced2013-01-29 10:13:12 +00003841 do {
3842 seq = read_seqbegin(&fs_info->profiles_lock);
3843
3844 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3845 (fs_info->avail_system_alloc_bits & allowed) &&
3846 !(bctl->sys.target & allowed)) ||
3847 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3848 (fs_info->avail_metadata_alloc_bits & allowed) &&
3849 !(bctl->meta.target & allowed))) {
3850 if (bctl->flags & BTRFS_BALANCE_FORCE) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003851 btrfs_info(fs_info,
3852 "force reducing metadata integrity");
Miao Xiede98ced2013-01-29 10:13:12 +00003853 } else {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003854 btrfs_err(fs_info,
3855 "balance will reduce metadata integrity, use force if you want this");
Miao Xiede98ced2013-01-29 10:13:12 +00003856 ret = -EINVAL;
3857 goto out;
3858 }
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003859 }
Miao Xiede98ced2013-01-29 10:13:12 +00003860 } while (read_seqretry(&fs_info->profiles_lock, seq));
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003861
Sam Tygieree592d02016-01-06 08:46:12 +00003862 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3863 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3864 btrfs_warn(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003865 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3866 bctl->meta.target, bctl->data.target);
Sam Tygieree592d02016-01-06 08:46:12 +00003867 }
3868
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02003869 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
Zhao Lei943c6e92015-08-19 15:54:15 +08003870 fs_info->num_tolerated_disk_barrier_failures = min(
3871 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3872 btrfs_get_num_tolerated_disk_barrier_failures(
3873 bctl->sys.target));
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02003874 }
3875
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04003876 ret = insert_balance_item(fs_info, bctl);
Ilya Dryomov59641012012-01-16 22:04:48 +02003877 if (ret && ret != -EEXIST)
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003878 goto out;
3879
Ilya Dryomov59641012012-01-16 22:04:48 +02003880 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3881 BUG_ON(ret == -EEXIST);
3882 set_balance_control(bctl);
3883 } else {
3884 BUG_ON(ret != -EEXIST);
3885 spin_lock(&fs_info->balance_lock);
3886 update_balance_args(bctl);
3887 spin_unlock(&fs_info->balance_lock);
3888 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003889
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003890 atomic_inc(&fs_info->balance_running);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003891 mutex_unlock(&fs_info->balance_mutex);
3892
3893 ret = __btrfs_balance(fs_info);
3894
3895 mutex_lock(&fs_info->balance_mutex);
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003896 atomic_dec(&fs_info->balance_running);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003897
Ilya Dryomovbf023ec2013-02-12 16:27:46 +00003898 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3899 fs_info->num_tolerated_disk_barrier_failures =
3900 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3901 }
3902
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003903 if (bargs) {
3904 memset(bargs, 0, sizeof(*bargs));
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003905 update_ioctl_balance_args(fs_info, 0, bargs);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003906 }
3907
Ilya Dryomov3a01aa72013-03-06 01:57:55 -07003908 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3909 balance_need_close(fs_info)) {
3910 __cancel_balance(fs_info);
3911 }
3912
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003913 wake_up(&fs_info->balance_wait_q);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003914
3915 return ret;
3916out:
Ilya Dryomov59641012012-01-16 22:04:48 +02003917 if (bctl->flags & BTRFS_BALANCE_RESUME)
3918 __cancel_balance(fs_info);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003919 else {
Ilya Dryomov59641012012-01-16 22:04:48 +02003920 kfree(bctl);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003921 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3922 }
Ilya Dryomov59641012012-01-16 22:04:48 +02003923 return ret;
3924}
3925
3926static int balance_kthread(void *data)
3927{
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003928 struct btrfs_fs_info *fs_info = data;
Ilya Dryomov9555c6c2012-01-16 22:04:48 +02003929 int ret = 0;
Ilya Dryomov59641012012-01-16 22:04:48 +02003930
3931 mutex_lock(&fs_info->volume_mutex);
3932 mutex_lock(&fs_info->balance_mutex);
3933
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003934 if (fs_info->balance_ctl) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003935 btrfs_info(fs_info, "continuing balance");
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003936 ret = btrfs_balance(fs_info->balance_ctl, NULL);
Ilya Dryomov9555c6c2012-01-16 22:04:48 +02003937 }
Ilya Dryomov59641012012-01-16 22:04:48 +02003938
3939 mutex_unlock(&fs_info->balance_mutex);
3940 mutex_unlock(&fs_info->volume_mutex);
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003941
Ilya Dryomov59641012012-01-16 22:04:48 +02003942 return ret;
3943}
3944
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003945int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3946{
3947 struct task_struct *tsk;
3948
3949 spin_lock(&fs_info->balance_lock);
3950 if (!fs_info->balance_ctl) {
3951 spin_unlock(&fs_info->balance_lock);
3952 return 0;
3953 }
3954 spin_unlock(&fs_info->balance_lock);
3955
Jeff Mahoney3cdde222016-06-09 21:38:35 -04003956 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003957 btrfs_info(fs_info, "force skipping balance");
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003958 return 0;
3959 }
3960
3961 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
Sachin Kamatcd633972013-07-15 16:52:18 +05303962 return PTR_ERR_OR_ZERO(tsk);
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003963}
3964
Ilya Dryomov68310a52012-06-22 12:24:12 -06003965int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
Ilya Dryomov59641012012-01-16 22:04:48 +02003966{
Ilya Dryomov59641012012-01-16 22:04:48 +02003967 struct btrfs_balance_control *bctl;
3968 struct btrfs_balance_item *item;
3969 struct btrfs_disk_balance_args disk_bargs;
3970 struct btrfs_path *path;
3971 struct extent_buffer *leaf;
3972 struct btrfs_key key;
3973 int ret;
3974
3975 path = btrfs_alloc_path();
3976 if (!path)
3977 return -ENOMEM;
3978
Ilya Dryomov68310a52012-06-22 12:24:12 -06003979 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003980 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov68310a52012-06-22 12:24:12 -06003981 key.offset = 0;
3982
3983 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3984 if (ret < 0)
3985 goto out;
3986 if (ret > 0) { /* ret = -ENOENT; */
3987 ret = 0;
3988 goto out;
3989 }
3990
Ilya Dryomov59641012012-01-16 22:04:48 +02003991 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3992 if (!bctl) {
3993 ret = -ENOMEM;
3994 goto out;
3995 }
3996
Ilya Dryomov59641012012-01-16 22:04:48 +02003997 leaf = path->nodes[0];
3998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3999
Ilya Dryomov68310a52012-06-22 12:24:12 -06004000 bctl->fs_info = fs_info;
4001 bctl->flags = btrfs_balance_flags(leaf, item);
4002 bctl->flags |= BTRFS_BALANCE_RESUME;
Ilya Dryomov59641012012-01-16 22:04:48 +02004003
4004 btrfs_balance_data(leaf, item, &disk_bargs);
4005 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4006 btrfs_balance_meta(leaf, item, &disk_bargs);
4007 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4008 btrfs_balance_sys(leaf, item, &disk_bargs);
4009 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4010
Ilya Dryomoved0fb782013-01-20 15:57:57 +02004011 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4012
Ilya Dryomov68310a52012-06-22 12:24:12 -06004013 mutex_lock(&fs_info->volume_mutex);
4014 mutex_lock(&fs_info->balance_mutex);
Ilya Dryomov59641012012-01-16 22:04:48 +02004015
Ilya Dryomov68310a52012-06-22 12:24:12 -06004016 set_balance_control(bctl);
4017
4018 mutex_unlock(&fs_info->balance_mutex);
4019 mutex_unlock(&fs_info->volume_mutex);
Ilya Dryomov59641012012-01-16 22:04:48 +02004020out:
4021 btrfs_free_path(path);
Chris Mason8f18cf12008-04-25 16:53:30 -04004022 return ret;
4023}
4024
Ilya Dryomov837d5b62012-01-16 22:04:49 +02004025int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4026{
4027 int ret = 0;
4028
4029 mutex_lock(&fs_info->balance_mutex);
4030 if (!fs_info->balance_ctl) {
4031 mutex_unlock(&fs_info->balance_mutex);
4032 return -ENOTCONN;
4033 }
4034
4035 if (atomic_read(&fs_info->balance_running)) {
4036 atomic_inc(&fs_info->balance_pause_req);
4037 mutex_unlock(&fs_info->balance_mutex);
4038
4039 wait_event(fs_info->balance_wait_q,
4040 atomic_read(&fs_info->balance_running) == 0);
4041
4042 mutex_lock(&fs_info->balance_mutex);
4043 /* we are good with balance_ctl ripped off from under us */
4044 BUG_ON(atomic_read(&fs_info->balance_running));
4045 atomic_dec(&fs_info->balance_pause_req);
4046 } else {
4047 ret = -ENOTCONN;
4048 }
4049
4050 mutex_unlock(&fs_info->balance_mutex);
4051 return ret;
4052}
4053
Ilya Dryomova7e99c62012-01-16 22:04:49 +02004054int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4055{
Ilya Dryomove649e582013-10-10 20:40:21 +03004056 if (fs_info->sb->s_flags & MS_RDONLY)
4057 return -EROFS;
4058
Ilya Dryomova7e99c62012-01-16 22:04:49 +02004059 mutex_lock(&fs_info->balance_mutex);
4060 if (!fs_info->balance_ctl) {
4061 mutex_unlock(&fs_info->balance_mutex);
4062 return -ENOTCONN;
4063 }
4064
4065 atomic_inc(&fs_info->balance_cancel_req);
4066 /*
4067 * if we are running just wait and return, balance item is
4068 * deleted in btrfs_balance in this case
4069 */
4070 if (atomic_read(&fs_info->balance_running)) {
4071 mutex_unlock(&fs_info->balance_mutex);
4072 wait_event(fs_info->balance_wait_q,
4073 atomic_read(&fs_info->balance_running) == 0);
4074 mutex_lock(&fs_info->balance_mutex);
4075 } else {
4076 /* __cancel_balance needs volume_mutex */
4077 mutex_unlock(&fs_info->balance_mutex);
4078 mutex_lock(&fs_info->volume_mutex);
4079 mutex_lock(&fs_info->balance_mutex);
4080
4081 if (fs_info->balance_ctl)
4082 __cancel_balance(fs_info);
4083
4084 mutex_unlock(&fs_info->volume_mutex);
4085 }
4086
4087 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4088 atomic_dec(&fs_info->balance_cancel_req);
4089 mutex_unlock(&fs_info->balance_mutex);
4090 return 0;
4091}
4092
Stefan Behrens803b2f52013-08-15 17:11:21 +02004093static int btrfs_uuid_scan_kthread(void *data)
4094{
4095 struct btrfs_fs_info *fs_info = data;
4096 struct btrfs_root *root = fs_info->tree_root;
4097 struct btrfs_key key;
4098 struct btrfs_key max_key;
4099 struct btrfs_path *path = NULL;
4100 int ret = 0;
4101 struct extent_buffer *eb;
4102 int slot;
4103 struct btrfs_root_item root_item;
4104 u32 item_size;
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004105 struct btrfs_trans_handle *trans = NULL;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004106
4107 path = btrfs_alloc_path();
4108 if (!path) {
4109 ret = -ENOMEM;
4110 goto out;
4111 }
4112
4113 key.objectid = 0;
4114 key.type = BTRFS_ROOT_ITEM_KEY;
4115 key.offset = 0;
4116
4117 max_key.objectid = (u64)-1;
4118 max_key.type = BTRFS_ROOT_ITEM_KEY;
4119 max_key.offset = (u64)-1;
4120
Stefan Behrens803b2f52013-08-15 17:11:21 +02004121 while (1) {
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004122 ret = btrfs_search_forward(root, &key, path, 0);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004123 if (ret) {
4124 if (ret > 0)
4125 ret = 0;
4126 break;
4127 }
4128
4129 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4130 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4131 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4132 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4133 goto skip;
4134
4135 eb = path->nodes[0];
4136 slot = path->slots[0];
4137 item_size = btrfs_item_size_nr(eb, slot);
4138 if (item_size < sizeof(root_item))
4139 goto skip;
4140
Stefan Behrens803b2f52013-08-15 17:11:21 +02004141 read_extent_buffer(eb, &root_item,
4142 btrfs_item_ptr_offset(eb, slot),
4143 (int)sizeof(root_item));
4144 if (btrfs_root_refs(&root_item) == 0)
4145 goto skip;
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004146
4147 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4148 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4149 if (trans)
4150 goto update_tree;
4151
4152 btrfs_release_path(path);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004153 /*
4154 * 1 - subvol uuid item
4155 * 1 - received_subvol uuid item
4156 */
4157 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4158 if (IS_ERR(trans)) {
4159 ret = PTR_ERR(trans);
4160 break;
4161 }
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004162 continue;
4163 } else {
4164 goto skip;
4165 }
4166update_tree:
4167 if (!btrfs_is_empty_uuid(root_item.uuid)) {
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04004168 ret = btrfs_uuid_tree_add(trans, fs_info,
Stefan Behrens803b2f52013-08-15 17:11:21 +02004169 root_item.uuid,
4170 BTRFS_UUID_KEY_SUBVOL,
4171 key.objectid);
4172 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004173 btrfs_warn(fs_info, "uuid_tree_add failed %d",
Stefan Behrens803b2f52013-08-15 17:11:21 +02004174 ret);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004175 break;
4176 }
4177 }
4178
4179 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04004180 ret = btrfs_uuid_tree_add(trans, fs_info,
Stefan Behrens803b2f52013-08-15 17:11:21 +02004181 root_item.received_uuid,
4182 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4183 key.objectid);
4184 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004185 btrfs_warn(fs_info, "uuid_tree_add failed %d",
Stefan Behrens803b2f52013-08-15 17:11:21 +02004186 ret);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004187 break;
4188 }
4189 }
4190
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004191skip:
Stefan Behrens803b2f52013-08-15 17:11:21 +02004192 if (trans) {
4193 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004194 trans = NULL;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004195 if (ret)
4196 break;
4197 }
4198
Stefan Behrens803b2f52013-08-15 17:11:21 +02004199 btrfs_release_path(path);
4200 if (key.offset < (u64)-1) {
4201 key.offset++;
4202 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4203 key.offset = 0;
4204 key.type = BTRFS_ROOT_ITEM_KEY;
4205 } else if (key.objectid < (u64)-1) {
4206 key.offset = 0;
4207 key.type = BTRFS_ROOT_ITEM_KEY;
4208 key.objectid++;
4209 } else {
4210 break;
4211 }
4212 cond_resched();
4213 }
4214
4215out:
4216 btrfs_free_path(path);
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004217 if (trans && !IS_ERR(trans))
4218 btrfs_end_transaction(trans, fs_info->uuid_root);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004219 if (ret)
Frank Holtonefe120a2013-12-20 11:37:06 -05004220 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
Stefan Behrens70f80172013-08-15 17:11:23 +02004221 else
Josef Bacikafcdd122016-09-02 15:40:02 -04004222 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004223 up(&fs_info->uuid_tree_rescan_sem);
4224 return 0;
4225}
4226
Stefan Behrens70f80172013-08-15 17:11:23 +02004227/*
4228 * Callback for btrfs_uuid_tree_iterate().
4229 * returns:
4230 * 0 check succeeded, the entry is not outdated.
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08004231 * < 0 if an error occurred.
Stefan Behrens70f80172013-08-15 17:11:23 +02004232 * > 0 if the check failed, which means the caller shall remove the entry.
4233 */
4234static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4235 u8 *uuid, u8 type, u64 subid)
4236{
4237 struct btrfs_key key;
4238 int ret = 0;
4239 struct btrfs_root *subvol_root;
4240
4241 if (type != BTRFS_UUID_KEY_SUBVOL &&
4242 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4243 goto out;
4244
4245 key.objectid = subid;
4246 key.type = BTRFS_ROOT_ITEM_KEY;
4247 key.offset = (u64)-1;
4248 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4249 if (IS_ERR(subvol_root)) {
4250 ret = PTR_ERR(subvol_root);
4251 if (ret == -ENOENT)
4252 ret = 1;
4253 goto out;
4254 }
4255
4256 switch (type) {
4257 case BTRFS_UUID_KEY_SUBVOL:
4258 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4259 ret = 1;
4260 break;
4261 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4262 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4263 BTRFS_UUID_SIZE))
4264 ret = 1;
4265 break;
4266 }
4267
4268out:
4269 return ret;
4270}
4271
4272static int btrfs_uuid_rescan_kthread(void *data)
4273{
4274 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4275 int ret;
4276
4277 /*
4278 * 1st step is to iterate through the existing UUID tree and
4279 * to delete all entries that contain outdated data.
4280 * 2nd step is to add all missing entries to the UUID tree.
4281 */
4282 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4283 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004284 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
Stefan Behrens70f80172013-08-15 17:11:23 +02004285 up(&fs_info->uuid_tree_rescan_sem);
4286 return ret;
4287 }
4288 return btrfs_uuid_scan_kthread(data);
4289}
4290
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004291int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4292{
4293 struct btrfs_trans_handle *trans;
4294 struct btrfs_root *tree_root = fs_info->tree_root;
4295 struct btrfs_root *uuid_root;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004296 struct task_struct *task;
4297 int ret;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004298
4299 /*
4300 * 1 - root node
4301 * 1 - root item
4302 */
4303 trans = btrfs_start_transaction(tree_root, 2);
4304 if (IS_ERR(trans))
4305 return PTR_ERR(trans);
4306
4307 uuid_root = btrfs_create_tree(trans, fs_info,
4308 BTRFS_UUID_TREE_OBJECTID);
4309 if (IS_ERR(uuid_root)) {
David Sterba6d13f542015-04-24 19:12:01 +02004310 ret = PTR_ERR(uuid_root);
Jeff Mahoney66642832016-06-10 18:19:25 -04004311 btrfs_abort_transaction(trans, ret);
Josef Bacik65d4f4c2015-09-23 15:00:37 -04004312 btrfs_end_transaction(trans, tree_root);
David Sterba6d13f542015-04-24 19:12:01 +02004313 return ret;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004314 }
4315
4316 fs_info->uuid_root = uuid_root;
4317
Stefan Behrens803b2f52013-08-15 17:11:21 +02004318 ret = btrfs_commit_transaction(trans, tree_root);
4319 if (ret)
4320 return ret;
4321
4322 down(&fs_info->uuid_tree_rescan_sem);
4323 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4324 if (IS_ERR(task)) {
Stefan Behrens70f80172013-08-15 17:11:23 +02004325 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
Frank Holtonefe120a2013-12-20 11:37:06 -05004326 btrfs_warn(fs_info, "failed to start uuid_scan task");
Stefan Behrens803b2f52013-08-15 17:11:21 +02004327 up(&fs_info->uuid_tree_rescan_sem);
4328 return PTR_ERR(task);
4329 }
4330
4331 return 0;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004332}
Stefan Behrens803b2f52013-08-15 17:11:21 +02004333
Stefan Behrens70f80172013-08-15 17:11:23 +02004334int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4335{
4336 struct task_struct *task;
4337
4338 down(&fs_info->uuid_tree_rescan_sem);
4339 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4340 if (IS_ERR(task)) {
4341 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
Frank Holtonefe120a2013-12-20 11:37:06 -05004342 btrfs_warn(fs_info, "failed to start uuid_rescan task");
Stefan Behrens70f80172013-08-15 17:11:23 +02004343 up(&fs_info->uuid_tree_rescan_sem);
4344 return PTR_ERR(task);
4345 }
4346
4347 return 0;
4348}
4349
Chris Mason8f18cf12008-04-25 16:53:30 -04004350/*
4351 * shrinking a device means finding all of the device extents past
4352 * the new size, and then following the back refs to the chunks.
4353 * The chunk relocation code actually frees the device extent
4354 */
4355int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4356{
4357 struct btrfs_trans_handle *trans;
4358 struct btrfs_root *root = device->dev_root;
4359 struct btrfs_dev_extent *dev_extent = NULL;
4360 struct btrfs_path *path;
4361 u64 length;
Chris Mason8f18cf12008-04-25 16:53:30 -04004362 u64 chunk_offset;
4363 int ret;
4364 int slot;
Josef Bacikba1bf482009-09-11 16:11:19 -04004365 int failed = 0;
4366 bool retried = false;
Filipe Manana53e489b2015-06-02 14:43:21 +01004367 bool checked_pending_chunks = false;
Chris Mason8f18cf12008-04-25 16:53:30 -04004368 struct extent_buffer *l;
4369 struct btrfs_key key;
David Sterba6c417612011-04-13 15:41:04 +02004370 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason8f18cf12008-04-25 16:53:30 -04004371 u64 old_total = btrfs_super_total_bytes(super_copy);
Miao Xie7cc8e582014-09-03 21:35:38 +08004372 u64 old_size = btrfs_device_get_total_bytes(device);
4373 u64 diff = old_size - new_size;
Chris Mason8f18cf12008-04-25 16:53:30 -04004374
Stefan Behrens63a212a2012-11-05 18:29:28 +01004375 if (device->is_tgtdev_for_dev_replace)
4376 return -EINVAL;
4377
Chris Mason8f18cf12008-04-25 16:53:30 -04004378 path = btrfs_alloc_path();
4379 if (!path)
4380 return -ENOMEM;
4381
David Sterbae4058b52015-11-27 16:31:35 +01004382 path->reada = READA_FORWARD;
Chris Mason8f18cf12008-04-25 16:53:30 -04004383
Chris Mason7d9eb122008-07-08 14:19:17 -04004384 lock_chunks(root);
4385
Miao Xie7cc8e582014-09-03 21:35:38 +08004386 btrfs_device_set_total_bytes(device, new_size);
Josef Bacik2bf64752011-09-26 17:12:22 -04004387 if (device->writeable) {
Yan Zheng2b820322008-11-17 21:11:30 -05004388 device->fs_devices->total_rw_bytes -= diff;
Josef Bacik2bf64752011-09-26 17:12:22 -04004389 spin_lock(&root->fs_info->free_chunk_lock);
4390 root->fs_info->free_chunk_space -= diff;
4391 spin_unlock(&root->fs_info->free_chunk_lock);
4392 }
Chris Mason7d9eb122008-07-08 14:19:17 -04004393 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04004394
Josef Bacikba1bf482009-09-11 16:11:19 -04004395again:
Chris Mason8f18cf12008-04-25 16:53:30 -04004396 key.objectid = device->devid;
4397 key.offset = (u64)-1;
4398 key.type = BTRFS_DEV_EXTENT_KEY;
4399
Ilya Dryomov213e64d2012-03-27 17:09:18 +03004400 do {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004401 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004402 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004403 if (ret < 0) {
4404 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004405 goto done;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004406 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004407
4408 ret = btrfs_previous_item(root, path, 0, key.type);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004409 if (ret)
4410 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004411 if (ret < 0)
4412 goto done;
4413 if (ret) {
4414 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02004415 btrfs_release_path(path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04004416 break;
Chris Mason8f18cf12008-04-25 16:53:30 -04004417 }
4418
4419 l = path->nodes[0];
4420 slot = path->slots[0];
4421 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4422
Josef Bacikba1bf482009-09-11 16:11:19 -04004423 if (key.objectid != device->devid) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004424 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
David Sterbab3b4aa72011-04-21 01:20:15 +02004425 btrfs_release_path(path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04004426 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04004427 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004428
4429 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4430 length = btrfs_dev_extent_length(l, dev_extent);
4431
Josef Bacikba1bf482009-09-11 16:11:19 -04004432 if (key.offset + length <= new_size) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004433 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
David Sterbab3b4aa72011-04-21 01:20:15 +02004434 btrfs_release_path(path);
Chris Balld6397ba2009-04-27 07:29:03 -04004435 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04004436 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004437
Chris Mason8f18cf12008-04-25 16:53:30 -04004438 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
David Sterbab3b4aa72011-04-21 01:20:15 +02004439 btrfs_release_path(path);
Chris Mason8f18cf12008-04-25 16:53:30 -04004440
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04004441 ret = btrfs_relocate_chunk(root->fs_info, chunk_offset);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004442 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Josef Bacikba1bf482009-09-11 16:11:19 -04004443 if (ret && ret != -ENOSPC)
Chris Mason8f18cf12008-04-25 16:53:30 -04004444 goto done;
Josef Bacikba1bf482009-09-11 16:11:19 -04004445 if (ret == -ENOSPC)
4446 failed++;
Ilya Dryomov213e64d2012-03-27 17:09:18 +03004447 } while (key.offset-- > 0);
Josef Bacikba1bf482009-09-11 16:11:19 -04004448
4449 if (failed && !retried) {
4450 failed = 0;
4451 retried = true;
4452 goto again;
4453 } else if (failed && retried) {
4454 ret = -ENOSPC;
Josef Bacikba1bf482009-09-11 16:11:19 -04004455 goto done;
Chris Mason8f18cf12008-04-25 16:53:30 -04004456 }
4457
Chris Balld6397ba2009-04-27 07:29:03 -04004458 /* Shrinking succeeded, else we would be at "done". */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004459 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00004460 if (IS_ERR(trans)) {
4461 ret = PTR_ERR(trans);
4462 goto done;
4463 }
4464
Chris Balld6397ba2009-04-27 07:29:03 -04004465 lock_chunks(root);
Filipe Manana53e489b2015-06-02 14:43:21 +01004466
4467 /*
4468 * We checked in the above loop all device extents that were already in
4469 * the device tree. However before we have updated the device's
4470 * total_bytes to the new size, we might have had chunk allocations that
4471 * have not complete yet (new block groups attached to transaction
4472 * handles), and therefore their device extents were not yet in the
4473 * device tree and we missed them in the loop above. So if we have any
4474 * pending chunk using a device extent that overlaps the device range
4475 * that we can not use anymore, commit the current transaction and
4476 * repeat the search on the device tree - this way we guarantee we will
4477 * not have chunks using device extents that end beyond 'new_size'.
4478 */
4479 if (!checked_pending_chunks) {
4480 u64 start = new_size;
4481 u64 len = old_size - new_size;
4482
Jeff Mahoney499f3772015-06-15 09:41:17 -04004483 if (contains_pending_extent(trans->transaction, device,
4484 &start, len)) {
Filipe Manana53e489b2015-06-02 14:43:21 +01004485 unlock_chunks(root);
4486 checked_pending_chunks = true;
4487 failed = 0;
4488 retried = false;
4489 ret = btrfs_commit_transaction(trans, root);
4490 if (ret)
4491 goto done;
4492 goto again;
4493 }
4494 }
4495
Miao Xie7cc8e582014-09-03 21:35:38 +08004496 btrfs_device_set_disk_total_bytes(device, new_size);
Miao Xie935e5cc2014-09-03 21:35:33 +08004497 if (list_empty(&device->resized_list))
4498 list_add_tail(&device->resized_list,
4499 &root->fs_info->fs_devices->resized_devices);
Chris Balld6397ba2009-04-27 07:29:03 -04004500
Chris Balld6397ba2009-04-27 07:29:03 -04004501 WARN_ON(diff > old_total);
4502 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4503 unlock_chunks(root);
Miao Xie2196d6e2014-09-03 21:35:41 +08004504
4505 /* Now btrfs_update_device() will change the on-disk size. */
4506 ret = btrfs_update_device(trans, device);
Chris Balld6397ba2009-04-27 07:29:03 -04004507 btrfs_end_transaction(trans, root);
Chris Mason8f18cf12008-04-25 16:53:30 -04004508done:
4509 btrfs_free_path(path);
Filipe Manana53e489b2015-06-02 14:43:21 +01004510 if (ret) {
4511 lock_chunks(root);
4512 btrfs_device_set_total_bytes(device, old_size);
4513 if (device->writeable)
4514 device->fs_devices->total_rw_bytes += diff;
4515 spin_lock(&root->fs_info->free_chunk_lock);
4516 root->fs_info->free_chunk_space += diff;
4517 spin_unlock(&root->fs_info->free_chunk_lock);
4518 unlock_chunks(root);
4519 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004520 return ret;
4521}
4522
Li Zefan125ccb02011-12-08 15:07:24 +08004523static int btrfs_add_system_chunk(struct btrfs_root *root,
Chris Mason0b86a832008-03-24 15:01:56 -04004524 struct btrfs_key *key,
4525 struct btrfs_chunk *chunk, int item_size)
4526{
David Sterba6c417612011-04-13 15:41:04 +02004527 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason0b86a832008-03-24 15:01:56 -04004528 struct btrfs_disk_key disk_key;
4529 u32 array_size;
4530 u8 *ptr;
4531
Miao Xiefe48a5c2014-09-03 21:35:39 +08004532 lock_chunks(root);
Chris Mason0b86a832008-03-24 15:01:56 -04004533 array_size = btrfs_super_sys_array_size(super_copy);
Gui Hecheng5f43f862014-04-21 20:13:11 +08004534 if (array_size + item_size + sizeof(disk_key)
Miao Xiefe48a5c2014-09-03 21:35:39 +08004535 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4536 unlock_chunks(root);
Chris Mason0b86a832008-03-24 15:01:56 -04004537 return -EFBIG;
Miao Xiefe48a5c2014-09-03 21:35:39 +08004538 }
Chris Mason0b86a832008-03-24 15:01:56 -04004539
4540 ptr = super_copy->sys_chunk_array + array_size;
4541 btrfs_cpu_key_to_disk(&disk_key, key);
4542 memcpy(ptr, &disk_key, sizeof(disk_key));
4543 ptr += sizeof(disk_key);
4544 memcpy(ptr, chunk, item_size);
4545 item_size += sizeof(disk_key);
4546 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
Miao Xiefe48a5c2014-09-03 21:35:39 +08004547 unlock_chunks(root);
4548
Chris Mason0b86a832008-03-24 15:01:56 -04004549 return 0;
4550}
4551
Miao Xieb2117a32011-01-05 10:07:28 +00004552/*
Arne Jansen73c5de02011-04-12 12:07:57 +02004553 * sort the devices in descending order by max_avail, total_avail
Miao Xieb2117a32011-01-05 10:07:28 +00004554 */
Arne Jansen73c5de02011-04-12 12:07:57 +02004555static int btrfs_cmp_device_info(const void *a, const void *b)
Miao Xieb2117a32011-01-05 10:07:28 +00004556{
Arne Jansen73c5de02011-04-12 12:07:57 +02004557 const struct btrfs_device_info *di_a = a;
4558 const struct btrfs_device_info *di_b = b;
Miao Xieb2117a32011-01-05 10:07:28 +00004559
Arne Jansen73c5de02011-04-12 12:07:57 +02004560 if (di_a->max_avail > di_b->max_avail)
4561 return -1;
4562 if (di_a->max_avail < di_b->max_avail)
4563 return 1;
4564 if (di_a->total_avail > di_b->total_avail)
4565 return -1;
4566 if (di_a->total_avail < di_b->total_avail)
4567 return 1;
Miao Xieb2117a32011-01-05 10:07:28 +00004568 return 0;
4569}
4570
David Woodhouse53b381b2013-01-29 18:40:14 -05004571static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4572{
4573 /* TODO allow them to set a preferred stripe size */
Byongho Leeee221842015-12-15 01:42:10 +09004574 return SZ_64K;
David Woodhouse53b381b2013-01-29 18:40:14 -05004575}
4576
4577static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4578{
Zhao Leiffe2d202015-01-20 15:11:44 +08004579 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
David Woodhouse53b381b2013-01-29 18:40:14 -05004580 return;
4581
Miao Xieceda0862013-04-11 10:30:16 +00004582 btrfs_set_fs_incompat(info, RAID56);
David Woodhouse53b381b2013-01-29 18:40:14 -05004583}
4584
Jeff Mahoney14a1e062016-06-15 10:25:38 -04004585#define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004586 - sizeof(struct btrfs_chunk)) \
4587 / sizeof(struct btrfs_stripe) + 1)
4588
4589#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4590 - 2 * sizeof(struct btrfs_disk_key) \
4591 - 2 * sizeof(struct btrfs_chunk)) \
4592 / sizeof(struct btrfs_stripe) + 1)
4593
Miao Xieb2117a32011-01-05 10:07:28 +00004594static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
Josef Bacik6df9a952013-06-27 13:22:46 -04004595 struct btrfs_root *extent_root, u64 start,
4596 u64 type)
Miao Xieb2117a32011-01-05 10:07:28 +00004597{
4598 struct btrfs_fs_info *info = extent_root->fs_info;
Miao Xieb2117a32011-01-05 10:07:28 +00004599 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4600 struct list_head *cur;
Arne Jansen73c5de02011-04-12 12:07:57 +02004601 struct map_lookup *map = NULL;
Miao Xieb2117a32011-01-05 10:07:28 +00004602 struct extent_map_tree *em_tree;
4603 struct extent_map *em;
Arne Jansen73c5de02011-04-12 12:07:57 +02004604 struct btrfs_device_info *devices_info = NULL;
4605 u64 total_avail;
4606 int num_stripes; /* total number of stripes to allocate */
David Woodhouse53b381b2013-01-29 18:40:14 -05004607 int data_stripes; /* number of stripes that count for
4608 block group size */
Arne Jansen73c5de02011-04-12 12:07:57 +02004609 int sub_stripes; /* sub_stripes info for map */
4610 int dev_stripes; /* stripes per dev */
4611 int devs_max; /* max devs to use */
4612 int devs_min; /* min devs needed */
4613 int devs_increment; /* ndevs has to be a multiple of this */
4614 int ncopies; /* how many copies to data has */
Miao Xieb2117a32011-01-05 10:07:28 +00004615 int ret;
Arne Jansen73c5de02011-04-12 12:07:57 +02004616 u64 max_stripe_size;
4617 u64 max_chunk_size;
4618 u64 stripe_size;
4619 u64 num_bytes;
David Woodhouse53b381b2013-01-29 18:40:14 -05004620 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
Arne Jansen73c5de02011-04-12 12:07:57 +02004621 int ndevs;
4622 int i;
4623 int j;
Liu Bo31e50222012-11-21 14:18:10 +00004624 int index;
Miao Xieb2117a32011-01-05 10:07:28 +00004625
Ilya Dryomov0c460c02012-03-27 17:09:17 +03004626 BUG_ON(!alloc_profile_is_valid(type, 0));
Arne Jansen73c5de02011-04-12 12:07:57 +02004627
Miao Xieb2117a32011-01-05 10:07:28 +00004628 if (list_empty(&fs_devices->alloc_list))
4629 return -ENOSPC;
4630
Liu Bo31e50222012-11-21 14:18:10 +00004631 index = __get_raid_index(type);
Arne Jansen73c5de02011-04-12 12:07:57 +02004632
Liu Bo31e50222012-11-21 14:18:10 +00004633 sub_stripes = btrfs_raid_array[index].sub_stripes;
4634 dev_stripes = btrfs_raid_array[index].dev_stripes;
4635 devs_max = btrfs_raid_array[index].devs_max;
4636 devs_min = btrfs_raid_array[index].devs_min;
4637 devs_increment = btrfs_raid_array[index].devs_increment;
4638 ncopies = btrfs_raid_array[index].ncopies;
Arne Jansen73c5de02011-04-12 12:07:57 +02004639
4640 if (type & BTRFS_BLOCK_GROUP_DATA) {
Byongho Leeee221842015-12-15 01:42:10 +09004641 max_stripe_size = SZ_1G;
Arne Jansen73c5de02011-04-12 12:07:57 +02004642 max_chunk_size = 10 * max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004643 if (!devs_max)
4644 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
Arne Jansen73c5de02011-04-12 12:07:57 +02004645 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
Chris Mason11003732012-01-06 15:47:38 -05004646 /* for larger filesystems, use larger metadata chunks */
Byongho Leeee221842015-12-15 01:42:10 +09004647 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4648 max_stripe_size = SZ_1G;
Chris Mason11003732012-01-06 15:47:38 -05004649 else
Byongho Leeee221842015-12-15 01:42:10 +09004650 max_stripe_size = SZ_256M;
Arne Jansen73c5de02011-04-12 12:07:57 +02004651 max_chunk_size = max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004652 if (!devs_max)
4653 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
Arne Jansen73c5de02011-04-12 12:07:57 +02004654 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
Byongho Leeee221842015-12-15 01:42:10 +09004655 max_stripe_size = SZ_32M;
Arne Jansen73c5de02011-04-12 12:07:57 +02004656 max_chunk_size = 2 * max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004657 if (!devs_max)
4658 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
Arne Jansen73c5de02011-04-12 12:07:57 +02004659 } else {
David Sterba351fd352014-05-15 16:48:20 +02004660 btrfs_err(info, "invalid chunk type 0x%llx requested",
Arne Jansen73c5de02011-04-12 12:07:57 +02004661 type);
4662 BUG_ON(1);
4663 }
4664
4665 /* we don't want a chunk larger than 10% of writeable space */
4666 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4667 max_chunk_size);
Miao Xieb2117a32011-01-05 10:07:28 +00004668
David Sterba31e818f2015-02-20 18:00:26 +01004669 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
Miao Xieb2117a32011-01-05 10:07:28 +00004670 GFP_NOFS);
4671 if (!devices_info)
4672 return -ENOMEM;
4673
Arne Jansen73c5de02011-04-12 12:07:57 +02004674 cur = fs_devices->alloc_list.next;
4675
4676 /*
4677 * in the first pass through the devices list, we gather information
4678 * about the available holes on each device.
4679 */
4680 ndevs = 0;
4681 while (cur != &fs_devices->alloc_list) {
4682 struct btrfs_device *device;
4683 u64 max_avail;
4684 u64 dev_offset;
4685
4686 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4687
4688 cur = cur->next;
4689
4690 if (!device->writeable) {
Julia Lawall31b1a2b2012-11-03 10:58:34 +00004691 WARN(1, KERN_ERR
Frank Holtonefe120a2013-12-20 11:37:06 -05004692 "BTRFS: read-only device in alloc_list\n");
Arne Jansen73c5de02011-04-12 12:07:57 +02004693 continue;
4694 }
4695
Stefan Behrens63a212a2012-11-05 18:29:28 +01004696 if (!device->in_fs_metadata ||
4697 device->is_tgtdev_for_dev_replace)
Arne Jansen73c5de02011-04-12 12:07:57 +02004698 continue;
4699
4700 if (device->total_bytes > device->bytes_used)
4701 total_avail = device->total_bytes - device->bytes_used;
4702 else
4703 total_avail = 0;
liubo38c01b92011-08-02 02:39:03 +00004704
4705 /* If there is no space on this device, skip it. */
4706 if (total_avail == 0)
4707 continue;
Arne Jansen73c5de02011-04-12 12:07:57 +02004708
Josef Bacik6df9a952013-06-27 13:22:46 -04004709 ret = find_free_dev_extent(trans, device,
Arne Jansen73c5de02011-04-12 12:07:57 +02004710 max_stripe_size * dev_stripes,
4711 &dev_offset, &max_avail);
4712 if (ret && ret != -ENOSPC)
4713 goto error;
4714
4715 if (ret == 0)
4716 max_avail = max_stripe_size * dev_stripes;
4717
4718 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4719 continue;
4720
Eric Sandeen063d0062013-01-31 00:55:01 +00004721 if (ndevs == fs_devices->rw_devices) {
4722 WARN(1, "%s: found more than %llu devices\n",
4723 __func__, fs_devices->rw_devices);
4724 break;
4725 }
Arne Jansen73c5de02011-04-12 12:07:57 +02004726 devices_info[ndevs].dev_offset = dev_offset;
4727 devices_info[ndevs].max_avail = max_avail;
4728 devices_info[ndevs].total_avail = total_avail;
4729 devices_info[ndevs].dev = device;
4730 ++ndevs;
4731 }
4732
4733 /*
4734 * now sort the devices by hole size / available space
4735 */
4736 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4737 btrfs_cmp_device_info, NULL);
4738
4739 /* round down to number of usable stripes */
4740 ndevs -= ndevs % devs_increment;
4741
4742 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4743 ret = -ENOSPC;
4744 goto error;
4745 }
4746
4747 if (devs_max && ndevs > devs_max)
4748 ndevs = devs_max;
4749 /*
4750 * the primary goal is to maximize the number of stripes, so use as many
4751 * devices as possible, even if the stripes are not maximum sized.
4752 */
4753 stripe_size = devices_info[ndevs-1].max_avail;
4754 num_stripes = ndevs * dev_stripes;
4755
David Woodhouse53b381b2013-01-29 18:40:14 -05004756 /*
4757 * this will have to be fixed for RAID1 and RAID10 over
4758 * more drives
4759 */
4760 data_stripes = num_stripes / ncopies;
4761
David Woodhouse53b381b2013-01-29 18:40:14 -05004762 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4763 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
Chandan Rajendrab7f67052016-06-23 15:16:44 +05304764 extent_root->stripesize);
David Woodhouse53b381b2013-01-29 18:40:14 -05004765 data_stripes = num_stripes - 1;
4766 }
4767 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4768 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
Chandan Rajendrab7f67052016-06-23 15:16:44 +05304769 extent_root->stripesize);
David Woodhouse53b381b2013-01-29 18:40:14 -05004770 data_stripes = num_stripes - 2;
4771 }
Chris Mason86db2572013-02-20 16:23:40 -05004772
4773 /*
4774 * Use the number of data stripes to figure out how big this chunk
4775 * is really going to be in terms of logical address space,
4776 * and compare that answer with the max chunk size
4777 */
4778 if (stripe_size * data_stripes > max_chunk_size) {
4779 u64 mask = (1ULL << 24) - 1;
David Sterbab8b93ad2015-01-16 17:26:13 +01004780
4781 stripe_size = div_u64(max_chunk_size, data_stripes);
Chris Mason86db2572013-02-20 16:23:40 -05004782
4783 /* bump the answer up to a 16MB boundary */
4784 stripe_size = (stripe_size + mask) & ~mask;
4785
4786 /* but don't go higher than the limits we found
4787 * while searching for free extents
4788 */
4789 if (stripe_size > devices_info[ndevs-1].max_avail)
4790 stripe_size = devices_info[ndevs-1].max_avail;
4791 }
4792
David Sterbab8b93ad2015-01-16 17:26:13 +01004793 stripe_size = div_u64(stripe_size, dev_stripes);
Ilya Dryomov37db63a2012-04-13 17:05:08 +03004794
4795 /* align to BTRFS_STRIPE_LEN */
David Sterbab8b93ad2015-01-16 17:26:13 +01004796 stripe_size = div_u64(stripe_size, raid_stripe_len);
David Woodhouse53b381b2013-01-29 18:40:14 -05004797 stripe_size *= raid_stripe_len;
Arne Jansen73c5de02011-04-12 12:07:57 +02004798
Miao Xieb2117a32011-01-05 10:07:28 +00004799 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4800 if (!map) {
4801 ret = -ENOMEM;
4802 goto error;
4803 }
4804 map->num_stripes = num_stripes;
Chris Mason9b3f68b2008-04-18 10:29:51 -04004805
Arne Jansen73c5de02011-04-12 12:07:57 +02004806 for (i = 0; i < ndevs; ++i) {
4807 for (j = 0; j < dev_stripes; ++j) {
4808 int s = i * dev_stripes + j;
4809 map->stripes[s].dev = devices_info[i].dev;
4810 map->stripes[s].physical = devices_info[i].dev_offset +
4811 j * stripe_size;
Chris Masona40a90a2008-04-18 11:55:51 -04004812 }
Chris Mason6324fbf2008-03-24 15:01:59 -04004813 }
Chris Mason593060d2008-03-25 16:50:33 -04004814 map->sector_size = extent_root->sectorsize;
David Woodhouse53b381b2013-01-29 18:40:14 -05004815 map->stripe_len = raid_stripe_len;
4816 map->io_align = raid_stripe_len;
4817 map->io_width = raid_stripe_len;
Chris Mason593060d2008-03-25 16:50:33 -04004818 map->type = type;
Chris Mason321aecc2008-04-16 10:49:51 -04004819 map->sub_stripes = sub_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04004820
David Woodhouse53b381b2013-01-29 18:40:14 -05004821 num_bytes = stripe_size * data_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04004822
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04004823 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
liubo1abe9b82011-03-24 11:18:59 +00004824
David Sterba172ddd62011-04-21 00:48:27 +02004825 em = alloc_extent_map();
Yan Zheng2b820322008-11-17 21:11:30 -05004826 if (!em) {
Wang Shilong298a8f92014-06-19 10:42:52 +08004827 kfree(map);
Miao Xieb2117a32011-01-05 10:07:28 +00004828 ret = -ENOMEM;
4829 goto error;
Yan Zheng2b820322008-11-17 21:11:30 -05004830 }
Wang Shilong298a8f92014-06-19 10:42:52 +08004831 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
Jeff Mahoney95617d62015-06-03 10:55:48 -04004832 em->map_lookup = map;
Yan Zheng2b820322008-11-17 21:11:30 -05004833 em->start = start;
Arne Jansen73c5de02011-04-12 12:07:57 +02004834 em->len = num_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04004835 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04004836 em->block_len = em->len;
Josef Bacik6df9a952013-06-27 13:22:46 -04004837 em->orig_block_len = stripe_size;
Chris Mason0b86a832008-03-24 15:01:56 -04004838
Chris Mason0b86a832008-03-24 15:01:56 -04004839 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
Chris Mason890871b2009-09-02 16:24:52 -04004840 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004841 ret = add_extent_mapping(em_tree, em, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04004842 if (!ret) {
4843 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4844 atomic_inc(&em->refs);
4845 }
Chris Mason890871b2009-09-02 16:24:52 -04004846 write_unlock(&em_tree->lock);
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004847 if (ret) {
4848 free_extent_map(em);
Mark Fasheh1dd46022011-09-08 17:29:00 -07004849 goto error;
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004850 }
Yan Zheng2b820322008-11-17 21:11:30 -05004851
Josef Bacik04487482013-01-29 15:03:37 -05004852 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4853 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4854 start, num_bytes);
Josef Bacik6df9a952013-06-27 13:22:46 -04004855 if (ret)
4856 goto error_del_extent;
Yan Zheng2b820322008-11-17 21:11:30 -05004857
Miao Xie7cc8e582014-09-03 21:35:38 +08004858 for (i = 0; i < map->num_stripes; i++) {
4859 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4860 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4861 }
Miao Xie43530c42014-09-03 21:35:36 +08004862
Miao Xie1c116182014-09-03 21:35:37 +08004863 spin_lock(&extent_root->fs_info->free_chunk_lock);
4864 extent_root->fs_info->free_chunk_space -= (stripe_size *
4865 map->num_stripes);
4866 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4867
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004868 free_extent_map(em);
David Woodhouse53b381b2013-01-29 18:40:14 -05004869 check_raid56_incompat_flag(extent_root->fs_info, type);
4870
Miao Xieb2117a32011-01-05 10:07:28 +00004871 kfree(devices_info);
Yan Zheng2b820322008-11-17 21:11:30 -05004872 return 0;
Miao Xieb2117a32011-01-05 10:07:28 +00004873
Josef Bacik6df9a952013-06-27 13:22:46 -04004874error_del_extent:
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004875 write_lock(&em_tree->lock);
4876 remove_extent_mapping(em_tree, em);
4877 write_unlock(&em_tree->lock);
4878
4879 /* One for our allocation */
4880 free_extent_map(em);
4881 /* One for the tree reference */
4882 free_extent_map(em);
Filipe Manana495e64f2014-12-02 18:07:30 +00004883 /* One for the pending_chunks list reference */
4884 free_extent_map(em);
Miao Xieb2117a32011-01-05 10:07:28 +00004885error:
Miao Xieb2117a32011-01-05 10:07:28 +00004886 kfree(devices_info);
4887 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05004888}
4889
Josef Bacik6df9a952013-06-27 13:22:46 -04004890int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04004891 struct btrfs_fs_info *fs_info,
Josef Bacik6df9a952013-06-27 13:22:46 -04004892 u64 chunk_offset, u64 chunk_size)
Yan Zheng2b820322008-11-17 21:11:30 -05004893{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04004894 struct btrfs_root *extent_root = fs_info->extent_root;
4895 struct btrfs_root *chunk_root = fs_info->chunk_root;
Yan Zheng2b820322008-11-17 21:11:30 -05004896 struct btrfs_key key;
Yan Zheng2b820322008-11-17 21:11:30 -05004897 struct btrfs_device *device;
4898 struct btrfs_chunk *chunk;
4899 struct btrfs_stripe *stripe;
Josef Bacik6df9a952013-06-27 13:22:46 -04004900 struct extent_map_tree *em_tree;
4901 struct extent_map *em;
4902 struct map_lookup *map;
4903 size_t item_size;
4904 u64 dev_offset;
4905 u64 stripe_size;
4906 int i = 0;
Chris Mason140e6392015-12-23 13:30:51 -08004907 int ret = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05004908
Josef Bacik6df9a952013-06-27 13:22:46 -04004909 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4910 read_lock(&em_tree->lock);
4911 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4912 read_unlock(&em_tree->lock);
Yan Zheng2b820322008-11-17 21:11:30 -05004913
Josef Bacik6df9a952013-06-27 13:22:46 -04004914 if (!em) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04004915 btrfs_crit(extent_root->fs_info,
4916 "unable to find logical %Lu len %Lu",
4917 chunk_offset, chunk_size);
Josef Bacik6df9a952013-06-27 13:22:46 -04004918 return -EINVAL;
4919 }
4920
4921 if (em->start != chunk_offset || em->len != chunk_size) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04004922 btrfs_crit(extent_root->fs_info,
4923 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4924 chunk_offset, chunk_size, em->start, em->len);
Josef Bacik6df9a952013-06-27 13:22:46 -04004925 free_extent_map(em);
4926 return -EINVAL;
4927 }
4928
Jeff Mahoney95617d62015-06-03 10:55:48 -04004929 map = em->map_lookup;
Josef Bacik6df9a952013-06-27 13:22:46 -04004930 item_size = btrfs_chunk_item_size(map->num_stripes);
4931 stripe_size = em->orig_block_len;
4932
4933 chunk = kzalloc(item_size, GFP_NOFS);
4934 if (!chunk) {
4935 ret = -ENOMEM;
4936 goto out;
4937 }
4938
Filipe Manana50460e32015-11-20 10:42:47 +00004939 /*
4940 * Take the device list mutex to prevent races with the final phase of
4941 * a device replace operation that replaces the device object associated
4942 * with the map's stripes, because the device object's id can change
4943 * at any time during that final phase of the device replace operation
4944 * (dev-replace.c:btrfs_dev_replace_finishing()).
4945 */
4946 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
Josef Bacik6df9a952013-06-27 13:22:46 -04004947 for (i = 0; i < map->num_stripes; i++) {
4948 device = map->stripes[i].dev;
4949 dev_offset = map->stripes[i].physical;
4950
Yan Zheng2b820322008-11-17 21:11:30 -05004951 ret = btrfs_update_device(trans, device);
Mark Fasheh3acd3952011-09-08 17:40:01 -07004952 if (ret)
Filipe Manana50460e32015-11-20 10:42:47 +00004953 break;
Josef Bacik6df9a952013-06-27 13:22:46 -04004954 ret = btrfs_alloc_dev_extent(trans, device,
4955 chunk_root->root_key.objectid,
4956 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4957 chunk_offset, dev_offset,
4958 stripe_size);
4959 if (ret)
Filipe Manana50460e32015-11-20 10:42:47 +00004960 break;
4961 }
4962 if (ret) {
4963 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4964 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05004965 }
4966
Yan Zheng2b820322008-11-17 21:11:30 -05004967 stripe = &chunk->stripe;
Josef Bacik6df9a952013-06-27 13:22:46 -04004968 for (i = 0; i < map->num_stripes; i++) {
4969 device = map->stripes[i].dev;
4970 dev_offset = map->stripes[i].physical;
Yan Zheng2b820322008-11-17 21:11:30 -05004971
4972 btrfs_set_stack_stripe_devid(stripe, device->devid);
4973 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4974 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4975 stripe++;
Yan Zheng2b820322008-11-17 21:11:30 -05004976 }
Filipe Manana50460e32015-11-20 10:42:47 +00004977 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05004978
4979 btrfs_set_stack_chunk_length(chunk, chunk_size);
4980 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4981 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4982 btrfs_set_stack_chunk_type(chunk, map->type);
4983 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4984 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4985 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4986 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4987 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4988
4989 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4990 key.type = BTRFS_CHUNK_ITEM_KEY;
4991 key.offset = chunk_offset;
4992
4993 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
Mark Fasheh4ed1d162011-08-10 12:32:10 -07004994 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4995 /*
4996 * TODO: Cleanup of inserted chunk root in case of
4997 * failure.
4998 */
Li Zefan125ccb02011-12-08 15:07:24 +08004999 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
Yan Zheng2b820322008-11-17 21:11:30 -05005000 item_size);
Yan Zheng2b820322008-11-17 21:11:30 -05005001 }
liubo1abe9b82011-03-24 11:18:59 +00005002
Josef Bacik6df9a952013-06-27 13:22:46 -04005003out:
Yan Zheng2b820322008-11-17 21:11:30 -05005004 kfree(chunk);
Josef Bacik6df9a952013-06-27 13:22:46 -04005005 free_extent_map(em);
Mark Fasheh4ed1d162011-08-10 12:32:10 -07005006 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005007}
5008
5009/*
5010 * Chunk allocation falls into two parts. The first part does works
5011 * that make the new allocated chunk useable, but not do any operation
5012 * that modifies the chunk tree. The second part does the works that
5013 * require modifying the chunk tree. This division is important for the
5014 * bootstrap process of adding storage to a seed btrfs.
5015 */
5016int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *extent_root, u64 type)
5018{
5019 u64 chunk_offset;
Yan Zheng2b820322008-11-17 21:11:30 -05005020
Filipe Mananaa9629592015-05-18 19:11:40 +01005021 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
Josef Bacik6df9a952013-06-27 13:22:46 -04005022 chunk_offset = find_next_chunk(extent_root->fs_info);
5023 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
Yan Zheng2b820322008-11-17 21:11:30 -05005024}
5025
Chris Masond3977122009-01-05 21:25:51 -05005026static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
Yan Zheng2b820322008-11-17 21:11:30 -05005027 struct btrfs_root *root,
5028 struct btrfs_device *device)
5029{
5030 u64 chunk_offset;
5031 u64 sys_chunk_offset;
Yan Zheng2b820322008-11-17 21:11:30 -05005032 u64 alloc_profile;
Yan Zheng2b820322008-11-17 21:11:30 -05005033 struct btrfs_fs_info *fs_info = root->fs_info;
5034 struct btrfs_root *extent_root = fs_info->extent_root;
5035 int ret;
5036
Josef Bacik6df9a952013-06-27 13:22:46 -04005037 chunk_offset = find_next_chunk(fs_info);
Miao Xiede98ced2013-01-29 10:13:12 +00005038 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04005039 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5040 alloc_profile);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005041 if (ret)
5042 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005043
Josef Bacik6df9a952013-06-27 13:22:46 -04005044 sys_chunk_offset = find_next_chunk(root->fs_info);
Miao Xiede98ced2013-01-29 10:13:12 +00005045 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04005046 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5047 alloc_profile);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005048 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005049}
5050
Miao Xied20983b2014-07-03 18:22:13 +08005051static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5052{
5053 int max_errors;
5054
5055 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056 BTRFS_BLOCK_GROUP_RAID10 |
5057 BTRFS_BLOCK_GROUP_RAID5 |
5058 BTRFS_BLOCK_GROUP_DUP)) {
5059 max_errors = 1;
5060 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5061 max_errors = 2;
5062 } else {
5063 max_errors = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05005064 }
5065
Miao Xied20983b2014-07-03 18:22:13 +08005066 return max_errors;
Yan Zheng2b820322008-11-17 21:11:30 -05005067}
5068
5069int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5070{
5071 struct extent_map *em;
5072 struct map_lookup *map;
5073 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5074 int readonly = 0;
Miao Xied20983b2014-07-03 18:22:13 +08005075 int miss_ndevs = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05005076 int i;
5077
Chris Mason890871b2009-09-02 16:24:52 -04005078 read_lock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05005079 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
Chris Mason890871b2009-09-02 16:24:52 -04005080 read_unlock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05005081 if (!em)
5082 return 1;
5083
Jeff Mahoney95617d62015-06-03 10:55:48 -04005084 map = em->map_lookup;
Yan Zheng2b820322008-11-17 21:11:30 -05005085 for (i = 0; i < map->num_stripes; i++) {
Miao Xied20983b2014-07-03 18:22:13 +08005086 if (map->stripes[i].dev->missing) {
5087 miss_ndevs++;
5088 continue;
5089 }
5090
Yan Zheng2b820322008-11-17 21:11:30 -05005091 if (!map->stripes[i].dev->writeable) {
5092 readonly = 1;
Miao Xied20983b2014-07-03 18:22:13 +08005093 goto end;
Yan Zheng2b820322008-11-17 21:11:30 -05005094 }
5095 }
Miao Xied20983b2014-07-03 18:22:13 +08005096
5097 /*
5098 * If the number of missing devices is larger than max errors,
5099 * we can not write the data into that chunk successfully, so
5100 * set it readonly.
5101 */
5102 if (miss_ndevs > btrfs_chunk_max_errors(map))
5103 readonly = 1;
5104end:
Yan Zheng2b820322008-11-17 21:11:30 -05005105 free_extent_map(em);
5106 return readonly;
Chris Mason0b86a832008-03-24 15:01:56 -04005107}
5108
5109void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5110{
David Sterbaa8067e02011-04-21 00:34:43 +02005111 extent_map_tree_init(&tree->map_tree);
Chris Mason0b86a832008-03-24 15:01:56 -04005112}
5113
5114void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5115{
5116 struct extent_map *em;
5117
Chris Masond3977122009-01-05 21:25:51 -05005118 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04005119 write_lock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005120 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5121 if (em)
5122 remove_extent_mapping(&tree->map_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04005123 write_unlock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005124 if (!em)
5125 break;
Chris Mason0b86a832008-03-24 15:01:56 -04005126 /* once for us */
5127 free_extent_map(em);
5128 /* once for the tree */
5129 free_extent_map(em);
5130 }
5131}
5132
Stefan Behrens5d964052012-11-05 14:59:07 +01005133int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
Chris Masonf1885912008-04-09 16:28:12 -04005134{
Stefan Behrens5d964052012-11-05 14:59:07 +01005135 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
Chris Masonf1885912008-04-09 16:28:12 -04005136 struct extent_map *em;
5137 struct map_lookup *map;
5138 struct extent_map_tree *em_tree = &map_tree->map_tree;
5139 int ret;
5140
Chris Mason890871b2009-09-02 16:24:52 -04005141 read_lock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04005142 em = lookup_extent_mapping(em_tree, logical, len);
Chris Mason890871b2009-09-02 16:24:52 -04005143 read_unlock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04005144
Josef Bacikfb7669b2013-04-23 10:53:18 -04005145 /*
5146 * We could return errors for these cases, but that could get ugly and
5147 * we'd probably do the same thing which is just not do anything else
5148 * and exit, so return 1 so the callers don't try to use other copies.
5149 */
5150 if (!em) {
David Sterba351fd352014-05-15 16:48:20 +02005151 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
Josef Bacikfb7669b2013-04-23 10:53:18 -04005152 logical+len);
5153 return 1;
5154 }
5155
5156 if (em->start > logical || em->start + em->len < logical) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005157 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5158 logical, logical+len, em->start,
5159 em->start + em->len);
Liu Bo7d3d1742013-09-29 10:33:16 +08005160 free_extent_map(em);
Josef Bacikfb7669b2013-04-23 10:53:18 -04005161 return 1;
5162 }
5163
Jeff Mahoney95617d62015-06-03 10:55:48 -04005164 map = em->map_lookup;
Chris Masonf1885912008-04-09 16:28:12 -04005165 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5166 ret = map->num_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04005167 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5168 ret = map->sub_stripes;
David Woodhouse53b381b2013-01-29 18:40:14 -05005169 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5170 ret = 2;
5171 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5172 ret = 3;
Chris Masonf1885912008-04-09 16:28:12 -04005173 else
5174 ret = 1;
5175 free_extent_map(em);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005176
Liu Bo73beece2015-07-17 16:49:19 +08005177 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005178 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5179 ret++;
Liu Bo73beece2015-07-17 16:49:19 +08005180 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005181
Chris Masonf1885912008-04-09 16:28:12 -04005182 return ret;
5183}
5184
David Woodhouse53b381b2013-01-29 18:40:14 -05005185unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5186 struct btrfs_mapping_tree *map_tree,
5187 u64 logical)
5188{
5189 struct extent_map *em;
5190 struct map_lookup *map;
5191 struct extent_map_tree *em_tree = &map_tree->map_tree;
5192 unsigned long len = root->sectorsize;
5193
5194 read_lock(&em_tree->lock);
5195 em = lookup_extent_mapping(em_tree, logical, len);
5196 read_unlock(&em_tree->lock);
5197 BUG_ON(!em);
5198
5199 BUG_ON(em->start > logical || em->start + em->len < logical);
Jeff Mahoney95617d62015-06-03 10:55:48 -04005200 map = em->map_lookup;
Zhao Leiffe2d202015-01-20 15:11:44 +08005201 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05005202 len = map->stripe_len * nr_data_stripes(map);
David Woodhouse53b381b2013-01-29 18:40:14 -05005203 free_extent_map(em);
5204 return len;
5205}
5206
5207int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5208 u64 logical, u64 len, int mirror_num)
5209{
5210 struct extent_map *em;
5211 struct map_lookup *map;
5212 struct extent_map_tree *em_tree = &map_tree->map_tree;
5213 int ret = 0;
5214
5215 read_lock(&em_tree->lock);
5216 em = lookup_extent_mapping(em_tree, logical, len);
5217 read_unlock(&em_tree->lock);
5218 BUG_ON(!em);
5219
5220 BUG_ON(em->start > logical || em->start + em->len < logical);
Jeff Mahoney95617d62015-06-03 10:55:48 -04005221 map = em->map_lookup;
Zhao Leiffe2d202015-01-20 15:11:44 +08005222 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05005223 ret = 1;
5224 free_extent_map(em);
5225 return ret;
5226}
5227
Stefan Behrens30d98612012-11-06 14:52:18 +01005228static int find_live_mirror(struct btrfs_fs_info *fs_info,
5229 struct map_lookup *map, int first, int num,
5230 int optimal, int dev_replace_is_ongoing)
Chris Masondfe25022008-05-13 13:46:40 -04005231{
5232 int i;
Stefan Behrens30d98612012-11-06 14:52:18 +01005233 int tolerance;
5234 struct btrfs_device *srcdev;
5235
5236 if (dev_replace_is_ongoing &&
5237 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5238 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5239 srcdev = fs_info->dev_replace.srcdev;
5240 else
5241 srcdev = NULL;
5242
5243 /*
5244 * try to avoid the drive that is the source drive for a
5245 * dev-replace procedure, only choose it if no other non-missing
5246 * mirror is available
5247 */
5248 for (tolerance = 0; tolerance < 2; tolerance++) {
5249 if (map->stripes[optimal].dev->bdev &&
5250 (tolerance || map->stripes[optimal].dev != srcdev))
5251 return optimal;
5252 for (i = first; i < first + num; i++) {
5253 if (map->stripes[i].dev->bdev &&
5254 (tolerance || map->stripes[i].dev != srcdev))
5255 return i;
5256 }
Chris Masondfe25022008-05-13 13:46:40 -04005257 }
Stefan Behrens30d98612012-11-06 14:52:18 +01005258
Chris Masondfe25022008-05-13 13:46:40 -04005259 /* we couldn't find one that doesn't fail. Just return something
5260 * and the io error handling code will clean up eventually
5261 */
5262 return optimal;
5263}
5264
David Woodhouse53b381b2013-01-29 18:40:14 -05005265static inline int parity_smaller(u64 a, u64 b)
5266{
5267 return a > b;
5268}
5269
5270/* Bubble-sort the stripe set to put the parity/syndrome stripes last */
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005271static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
David Woodhouse53b381b2013-01-29 18:40:14 -05005272{
5273 struct btrfs_bio_stripe s;
5274 int i;
5275 u64 l;
5276 int again = 1;
5277
5278 while (again) {
5279 again = 0;
Zhao Leicc7539e2015-01-20 15:11:32 +08005280 for (i = 0; i < num_stripes - 1; i++) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005281 if (parity_smaller(bbio->raid_map[i],
5282 bbio->raid_map[i+1])) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005283 s = bbio->stripes[i];
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005284 l = bbio->raid_map[i];
David Woodhouse53b381b2013-01-29 18:40:14 -05005285 bbio->stripes[i] = bbio->stripes[i+1];
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005286 bbio->raid_map[i] = bbio->raid_map[i+1];
David Woodhouse53b381b2013-01-29 18:40:14 -05005287 bbio->stripes[i+1] = s;
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005288 bbio->raid_map[i+1] = l;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005289
David Woodhouse53b381b2013-01-29 18:40:14 -05005290 again = 1;
5291 }
5292 }
5293 }
5294}
5295
Zhao Lei6e9606d2015-01-20 15:11:34 +08005296static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5297{
5298 struct btrfs_bio *bbio = kzalloc(
Chris Masone57cf212015-02-19 17:51:39 -08005299 /* the size of the btrfs_bio */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005300 sizeof(struct btrfs_bio) +
Chris Masone57cf212015-02-19 17:51:39 -08005301 /* plus the variable array for the stripes */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005302 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
Chris Masone57cf212015-02-19 17:51:39 -08005303 /* plus the variable array for the tgt dev */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005304 sizeof(int) * (real_stripes) +
Chris Masone57cf212015-02-19 17:51:39 -08005305 /*
5306 * plus the raid_map, which includes both the tgt dev
5307 * and the stripes
5308 */
5309 sizeof(u64) * (total_stripes),
Michal Hocko277fb5f2015-08-19 14:17:41 +02005310 GFP_NOFS|__GFP_NOFAIL);
Zhao Lei6e9606d2015-01-20 15:11:34 +08005311
5312 atomic_set(&bbio->error, 0);
5313 atomic_set(&bbio->refs, 1);
5314
5315 return bbio;
5316}
5317
5318void btrfs_get_bbio(struct btrfs_bio *bbio)
5319{
5320 WARN_ON(!atomic_read(&bbio->refs));
5321 atomic_inc(&bbio->refs);
5322}
5323
5324void btrfs_put_bbio(struct btrfs_bio *bbio)
5325{
5326 if (!bbio)
5327 return;
5328 if (atomic_dec_and_test(&bbio->refs))
5329 kfree(bbio);
5330}
5331
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005332static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5333 enum btrfs_map_op op,
Chris Masonf2d8d742008-04-21 10:03:05 -04005334 u64 logical, u64 *length,
Jan Schmidta1d3c472011-08-04 17:15:33 +02005335 struct btrfs_bio **bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005336 int mirror_num, int need_raid_map)
Chris Mason0b86a832008-03-24 15:01:56 -04005337{
5338 struct extent_map *em;
5339 struct map_lookup *map;
Stefan Behrens3ec706c2012-11-05 15:46:42 +01005340 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
Chris Mason0b86a832008-03-24 15:01:56 -04005341 struct extent_map_tree *em_tree = &map_tree->map_tree;
5342 u64 offset;
Chris Mason593060d2008-03-25 16:50:33 -04005343 u64 stripe_offset;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005344 u64 stripe_end_offset;
Chris Mason593060d2008-03-25 16:50:33 -04005345 u64 stripe_nr;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005346 u64 stripe_nr_orig;
5347 u64 stripe_nr_end;
David Woodhouse53b381b2013-01-29 18:40:14 -05005348 u64 stripe_len;
David Sterba9d644a62015-02-20 18:42:11 +01005349 u32 stripe_index;
Chris Masoncea9e442008-04-09 16:28:12 -04005350 int i;
Li Zefande11cc12011-12-01 12:55:47 +08005351 int ret = 0;
Chris Masonf2d8d742008-04-21 10:03:05 -04005352 int num_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04005353 int max_errors = 0;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005354 int tgtdev_indexes = 0;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005355 struct btrfs_bio *bbio = NULL;
Stefan Behrens472262f2012-11-06 14:43:46 +01005356 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5357 int dev_replace_is_ongoing = 0;
5358 int num_alloc_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005359 int patch_the_first_stripe_for_dev_replace = 0;
5360 u64 physical_to_patch_in_first_stripe = 0;
David Woodhouse53b381b2013-01-29 18:40:14 -05005361 u64 raid56_full_stripe_start = (u64)-1;
Chris Mason0b86a832008-03-24 15:01:56 -04005362
Chris Mason890871b2009-09-02 16:24:52 -04005363 read_lock(&em_tree->lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005364 em = lookup_extent_mapping(em_tree, logical, *length);
Chris Mason890871b2009-09-02 16:24:52 -04005365 read_unlock(&em_tree->lock);
Chris Masonf2d8d742008-04-21 10:03:05 -04005366
Chris Mason3b951512008-04-17 11:29:12 -04005367 if (!em) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005368 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02005369 logical, *length);
Josef Bacik9bb91872013-04-19 23:45:33 -04005370 return -EINVAL;
Chris Mason3b951512008-04-17 11:29:12 -04005371 }
Chris Mason0b86a832008-03-24 15:01:56 -04005372
Josef Bacik9bb91872013-04-19 23:45:33 -04005373 if (em->start > logical || em->start + em->len < logical) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005374 btrfs_crit(fs_info,
5375 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5376 logical, em->start, em->start + em->len);
Liu Bo7d3d1742013-09-29 10:33:16 +08005377 free_extent_map(em);
Josef Bacik9bb91872013-04-19 23:45:33 -04005378 return -EINVAL;
5379 }
5380
Jeff Mahoney95617d62015-06-03 10:55:48 -04005381 map = em->map_lookup;
Chris Mason0b86a832008-03-24 15:01:56 -04005382 offset = logical - em->start;
Chris Mason593060d2008-03-25 16:50:33 -04005383
David Woodhouse53b381b2013-01-29 18:40:14 -05005384 stripe_len = map->stripe_len;
Chris Mason593060d2008-03-25 16:50:33 -04005385 stripe_nr = offset;
5386 /*
5387 * stripe_nr counts the total number of stripes we have to stride
5388 * to get to this block
5389 */
David Sterba47c57132015-02-20 18:43:47 +01005390 stripe_nr = div64_u64(stripe_nr, stripe_len);
Chris Mason593060d2008-03-25 16:50:33 -04005391
David Woodhouse53b381b2013-01-29 18:40:14 -05005392 stripe_offset = stripe_nr * stripe_len;
Josef Bacike042d1e2016-04-12 12:54:40 -04005393 if (offset < stripe_offset) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005394 btrfs_crit(fs_info,
5395 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
Josef Bacike042d1e2016-04-12 12:54:40 -04005396 stripe_offset, offset, em->start, logical,
5397 stripe_len);
5398 free_extent_map(em);
5399 return -EINVAL;
5400 }
Chris Mason593060d2008-03-25 16:50:33 -04005401
5402 /* stripe_offset is the offset of this block in its stripe*/
5403 stripe_offset = offset - stripe_offset;
5404
David Woodhouse53b381b2013-01-29 18:40:14 -05005405 /* if we're here for raid56, we need to know the stripe aligned start */
Zhao Leiffe2d202015-01-20 15:11:44 +08005406 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005407 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5408 raid56_full_stripe_start = offset;
5409
5410 /* allow a write of a full stripe, but make sure we don't
5411 * allow straddling of stripes
5412 */
David Sterba47c57132015-02-20 18:43:47 +01005413 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5414 full_stripe_len);
David Woodhouse53b381b2013-01-29 18:40:14 -05005415 raid56_full_stripe_start *= full_stripe_len;
5416 }
5417
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005418 if (op == BTRFS_MAP_DISCARD) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005419 /* we don't discard raid56 yet */
Zhao Leiffe2d202015-01-20 15:11:44 +08005420 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005421 ret = -EOPNOTSUPP;
5422 goto out;
5423 }
Li Dongyangfce3bb92011-03-24 10:24:26 +00005424 *length = min_t(u64, em->len - offset, *length);
David Woodhouse53b381b2013-01-29 18:40:14 -05005425 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5426 u64 max_len;
5427 /* For writes to RAID[56], allow a full stripeset across all disks.
5428 For other RAID types and for RAID[56] reads, just allow a single
5429 stripe (on a single disk). */
Zhao Leiffe2d202015-01-20 15:11:44 +08005430 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005431 (op == BTRFS_MAP_WRITE)) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005432 max_len = stripe_len * nr_data_stripes(map) -
5433 (offset - raid56_full_stripe_start);
5434 } else {
5435 /* we limit the length of each bio to what fits in a stripe */
5436 max_len = stripe_len - stripe_offset;
5437 }
5438 *length = min_t(u64, em->len - offset, max_len);
Chris Masoncea9e442008-04-09 16:28:12 -04005439 } else {
5440 *length = em->len - offset;
5441 }
Chris Masonf2d8d742008-04-21 10:03:05 -04005442
David Woodhouse53b381b2013-01-29 18:40:14 -05005443 /* This is for when we're called from btrfs_merge_bio_hook() and all
5444 it cares about is the length */
Jan Schmidta1d3c472011-08-04 17:15:33 +02005445 if (!bbio_ret)
Chris Masoncea9e442008-04-09 16:28:12 -04005446 goto out;
5447
Liu Bo73beece2015-07-17 16:49:19 +08005448 btrfs_dev_replace_lock(dev_replace, 0);
Stefan Behrens472262f2012-11-06 14:43:46 +01005449 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5450 if (!dev_replace_is_ongoing)
Liu Bo73beece2015-07-17 16:49:19 +08005451 btrfs_dev_replace_unlock(dev_replace, 0);
5452 else
5453 btrfs_dev_replace_set_lock_blocking(dev_replace);
Stefan Behrens472262f2012-11-06 14:43:46 +01005454
Stefan Behrensad6d6202012-11-06 15:06:47 +01005455 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005456 op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5457 op != BTRFS_MAP_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
Stefan Behrensad6d6202012-11-06 15:06:47 +01005458 /*
5459 * in dev-replace case, for repair case (that's the only
5460 * case where the mirror is selected explicitly when
5461 * calling btrfs_map_block), blocks left of the left cursor
5462 * can also be read from the target drive.
5463 * For REQ_GET_READ_MIRRORS, the target drive is added as
5464 * the last one to the array of stripes. For READ, it also
5465 * needs to be supported using the same mirror number.
5466 * If the requested block is not left of the left cursor,
5467 * EIO is returned. This can happen because btrfs_num_copies()
5468 * returns one more in the dev-replace case.
5469 */
5470 u64 tmp_length = *length;
5471 struct btrfs_bio *tmp_bbio = NULL;
5472 int tmp_num_stripes;
5473 u64 srcdev_devid = dev_replace->srcdev->devid;
5474 int index_srcdev = 0;
5475 int found = 0;
5476 u64 physical_of_found = 0;
5477
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005478 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005479 logical, &tmp_length, &tmp_bbio, 0, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005480 if (ret) {
5481 WARN_ON(tmp_bbio != NULL);
5482 goto out;
5483 }
5484
5485 tmp_num_stripes = tmp_bbio->num_stripes;
5486 if (mirror_num > tmp_num_stripes) {
5487 /*
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005488 * BTRFS_MAP_GET_READ_MIRRORS does not contain this
Stefan Behrensad6d6202012-11-06 15:06:47 +01005489 * mirror, that means that the requested area
5490 * is not left of the left cursor
5491 */
5492 ret = -EIO;
Zhao Lei6e9606d2015-01-20 15:11:34 +08005493 btrfs_put_bbio(tmp_bbio);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005494 goto out;
5495 }
5496
5497 /*
5498 * process the rest of the function using the mirror_num
5499 * of the source drive. Therefore look it up first.
5500 * At the end, patch the device pointer to the one of the
5501 * target drive.
5502 */
5503 for (i = 0; i < tmp_num_stripes; i++) {
Zhao Lei94a97df2015-12-09 21:03:49 +08005504 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5505 continue;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005506
Zhao Lei94a97df2015-12-09 21:03:49 +08005507 /*
5508 * In case of DUP, in order to keep it simple, only add
5509 * the mirror with the lowest physical address
5510 */
5511 if (found &&
5512 physical_of_found <= tmp_bbio->stripes[i].physical)
5513 continue;
5514
5515 index_srcdev = i;
5516 found = 1;
5517 physical_of_found = tmp_bbio->stripes[i].physical;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005518 }
5519
Zhao Lei6e9606d2015-01-20 15:11:34 +08005520 btrfs_put_bbio(tmp_bbio);
Zhao Lei94a97df2015-12-09 21:03:49 +08005521
5522 if (!found) {
5523 WARN_ON(1);
5524 ret = -EIO;
5525 goto out;
5526 }
5527
5528 mirror_num = index_srcdev + 1;
5529 patch_the_first_stripe_for_dev_replace = 1;
5530 physical_to_patch_in_first_stripe = physical_of_found;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005531 } else if (mirror_num > map->num_stripes) {
5532 mirror_num = 0;
5533 }
5534
Chris Masonf2d8d742008-04-21 10:03:05 -04005535 num_stripes = 1;
Chris Masoncea9e442008-04-09 16:28:12 -04005536 stripe_index = 0;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005537 stripe_nr_orig = stripe_nr;
Qu Wenruofda28322013-02-26 08:10:22 +00005538 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
David Sterbab8b93ad2015-01-16 17:26:13 +01005539 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
Li Dongyangfce3bb92011-03-24 10:24:26 +00005540 stripe_end_offset = stripe_nr_end * map->stripe_len -
5541 (offset + *length);
David Woodhouse53b381b2013-01-29 18:40:14 -05005542
Li Dongyangfce3bb92011-03-24 10:24:26 +00005543 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005544 if (op == BTRFS_MAP_DISCARD)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005545 num_stripes = min_t(u64, map->num_stripes,
5546 stripe_nr_end - stripe_nr_orig);
David Sterba47c57132015-02-20 18:43:47 +01005547 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5548 &stripe_index);
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005549 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5550 op != BTRFS_MAP_GET_READ_MIRRORS)
Miao Xie28e1cc72014-09-12 18:44:02 +08005551 mirror_num = 1;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005552 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005553 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5554 op == BTRFS_MAP_GET_READ_MIRRORS)
Chris Masonf2d8d742008-04-21 10:03:05 -04005555 num_stripes = map->num_stripes;
Chris Mason2fff7342008-04-29 14:12:09 -04005556 else if (mirror_num)
Chris Masonf1885912008-04-09 16:28:12 -04005557 stripe_index = mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04005558 else {
Stefan Behrens30d98612012-11-06 14:52:18 +01005559 stripe_index = find_live_mirror(fs_info, map, 0,
Chris Masondfe25022008-05-13 13:46:40 -04005560 map->num_stripes,
Stefan Behrens30d98612012-11-06 14:52:18 +01005561 current->pid % map->num_stripes,
5562 dev_replace_is_ongoing);
Jan Schmidta1d3c472011-08-04 17:15:33 +02005563 mirror_num = stripe_index + 1;
Chris Masondfe25022008-05-13 13:46:40 -04005564 }
Chris Mason2fff7342008-04-29 14:12:09 -04005565
Chris Mason611f0e02008-04-03 16:29:03 -04005566 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005567 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5568 op == BTRFS_MAP_GET_READ_MIRRORS) {
Chris Masonf2d8d742008-04-21 10:03:05 -04005569 num_stripes = map->num_stripes;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005570 } else if (mirror_num) {
Chris Masonf1885912008-04-09 16:28:12 -04005571 stripe_index = mirror_num - 1;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005572 } else {
5573 mirror_num = 1;
5574 }
Chris Mason2fff7342008-04-29 14:12:09 -04005575
Chris Mason321aecc2008-04-16 10:49:51 -04005576 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
David Sterba9d644a62015-02-20 18:42:11 +01005577 u32 factor = map->num_stripes / map->sub_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04005578
David Sterba47c57132015-02-20 18:43:47 +01005579 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
Chris Mason321aecc2008-04-16 10:49:51 -04005580 stripe_index *= map->sub_stripes;
5581
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005582 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
Chris Masonf2d8d742008-04-21 10:03:05 -04005583 num_stripes = map->sub_stripes;
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005584 else if (op == BTRFS_MAP_DISCARD)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005585 num_stripes = min_t(u64, map->sub_stripes *
5586 (stripe_nr_end - stripe_nr_orig),
5587 map->num_stripes);
Chris Mason321aecc2008-04-16 10:49:51 -04005588 else if (mirror_num)
5589 stripe_index += mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04005590 else {
Jan Schmidt3e743172012-04-27 12:41:45 -04005591 int old_stripe_index = stripe_index;
Stefan Behrens30d98612012-11-06 14:52:18 +01005592 stripe_index = find_live_mirror(fs_info, map,
5593 stripe_index,
Chris Masondfe25022008-05-13 13:46:40 -04005594 map->sub_stripes, stripe_index +
Stefan Behrens30d98612012-11-06 14:52:18 +01005595 current->pid % map->sub_stripes,
5596 dev_replace_is_ongoing);
Jan Schmidt3e743172012-04-27 12:41:45 -04005597 mirror_num = stripe_index - old_stripe_index + 1;
Chris Masondfe25022008-05-13 13:46:40 -04005598 }
David Woodhouse53b381b2013-01-29 18:40:14 -05005599
Zhao Leiffe2d202015-01-20 15:11:44 +08005600 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005601 if (need_raid_map &&
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005602 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
Miao Xieaf8e2d12014-10-23 14:42:50 +08005603 mirror_num > 1)) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005604 /* push stripe_nr back to the start of the full stripe */
David Sterbab8b93ad2015-01-16 17:26:13 +01005605 stripe_nr = div_u64(raid56_full_stripe_start,
5606 stripe_len * nr_data_stripes(map));
David Woodhouse53b381b2013-01-29 18:40:14 -05005607
5608 /* RAID[56] write or recovery. Return all stripes */
5609 num_stripes = map->num_stripes;
5610 max_errors = nr_parity_stripes(map);
5611
David Woodhouse53b381b2013-01-29 18:40:14 -05005612 *length = map->stripe_len;
5613 stripe_index = 0;
5614 stripe_offset = 0;
5615 } else {
5616 /*
5617 * Mirror #0 or #1 means the original data block.
5618 * Mirror #2 is RAID5 parity block.
5619 * Mirror #3 is RAID6 Q block.
5620 */
David Sterba47c57132015-02-20 18:43:47 +01005621 stripe_nr = div_u64_rem(stripe_nr,
5622 nr_data_stripes(map), &stripe_index);
David Woodhouse53b381b2013-01-29 18:40:14 -05005623 if (mirror_num > 1)
5624 stripe_index = nr_data_stripes(map) +
5625 mirror_num - 2;
5626
5627 /* We distribute the parity blocks across stripes */
David Sterba47c57132015-02-20 18:43:47 +01005628 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5629 &stripe_index);
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005630 if ((op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5631 op != BTRFS_MAP_GET_READ_MIRRORS) && mirror_num <= 1)
Miao Xie28e1cc72014-09-12 18:44:02 +08005632 mirror_num = 1;
David Woodhouse53b381b2013-01-29 18:40:14 -05005633 }
Chris Mason8790d502008-04-03 16:29:03 -04005634 } else {
5635 /*
David Sterba47c57132015-02-20 18:43:47 +01005636 * after this, stripe_nr is the number of stripes on this
5637 * device we have to walk to find the data, and stripe_index is
5638 * the number of our device in the stripe array
Chris Mason8790d502008-04-03 16:29:03 -04005639 */
David Sterba47c57132015-02-20 18:43:47 +01005640 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5641 &stripe_index);
Jan Schmidta1d3c472011-08-04 17:15:33 +02005642 mirror_num = stripe_index + 1;
Chris Mason8790d502008-04-03 16:29:03 -04005643 }
Josef Bacike042d1e2016-04-12 12:54:40 -04005644 if (stripe_index >= map->num_stripes) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005645 btrfs_crit(fs_info,
5646 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
Josef Bacike042d1e2016-04-12 12:54:40 -04005647 stripe_index, map->num_stripes);
5648 ret = -EINVAL;
5649 goto out;
5650 }
Chris Mason593060d2008-03-25 16:50:33 -04005651
Stefan Behrens472262f2012-11-06 14:43:46 +01005652 num_alloc_stripes = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005653 if (dev_replace_is_ongoing) {
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005654 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD)
Stefan Behrensad6d6202012-11-06 15:06:47 +01005655 num_alloc_stripes <<= 1;
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005656 if (op == BTRFS_MAP_GET_READ_MIRRORS)
Stefan Behrensad6d6202012-11-06 15:06:47 +01005657 num_alloc_stripes++;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005658 tgtdev_indexes = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005659 }
Miao Xie2c8cdd62014-11-14 16:06:25 +08005660
Zhao Lei6e9606d2015-01-20 15:11:34 +08005661 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
Li Zefande11cc12011-12-01 12:55:47 +08005662 if (!bbio) {
5663 ret = -ENOMEM;
5664 goto out;
5665 }
Miao Xie2c8cdd62014-11-14 16:06:25 +08005666 if (dev_replace_is_ongoing)
5667 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
Li Zefande11cc12011-12-01 12:55:47 +08005668
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005669 /* build raid_map */
Zhao Leiffe2d202015-01-20 15:11:44 +08005670 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
Mike Christieb3d3fa52016-06-05 14:31:53 -05005671 need_raid_map &&
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005672 ((op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) ||
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005673 mirror_num > 1)) {
5674 u64 tmp;
David Sterba9d644a62015-02-20 18:42:11 +01005675 unsigned rot;
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005676
5677 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5678 sizeof(struct btrfs_bio_stripe) *
5679 num_alloc_stripes +
5680 sizeof(int) * tgtdev_indexes);
5681
5682 /* Work out the disk rotation on this stripe-set */
David Sterba47c57132015-02-20 18:43:47 +01005683 div_u64_rem(stripe_nr, num_stripes, &rot);
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005684
5685 /* Fill in the logical address of each stripe */
5686 tmp = stripe_nr * nr_data_stripes(map);
5687 for (i = 0; i < nr_data_stripes(map); i++)
5688 bbio->raid_map[(i+rot) % num_stripes] =
5689 em->start + (tmp + i) * map->stripe_len;
5690
5691 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5692 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5693 bbio->raid_map[(i+rot+1) % num_stripes] =
5694 RAID6_Q_STRIPE;
5695 }
5696
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005697 if (op == BTRFS_MAP_DISCARD) {
David Sterba9d644a62015-02-20 18:42:11 +01005698 u32 factor = 0;
5699 u32 sub_stripes = 0;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005700 u64 stripes_per_dev = 0;
5701 u32 remaining_stripes = 0;
Liu Bob89203f2012-04-12 16:03:56 -04005702 u32 last_stripe = 0;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005703
5704 if (map->type &
5705 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5706 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5707 sub_stripes = 1;
5708 else
5709 sub_stripes = map->sub_stripes;
5710
5711 factor = map->num_stripes / sub_stripes;
5712 stripes_per_dev = div_u64_rem(stripe_nr_end -
5713 stripe_nr_orig,
5714 factor,
5715 &remaining_stripes);
Liu Bob89203f2012-04-12 16:03:56 -04005716 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5717 last_stripe *= sub_stripes;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005718 }
5719
Li Dongyangfce3bb92011-03-24 10:24:26 +00005720 for (i = 0; i < num_stripes; i++) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02005721 bbio->stripes[i].physical =
Chris Masonf2d8d742008-04-21 10:03:05 -04005722 map->stripes[stripe_index].physical +
5723 stripe_offset + stripe_nr * map->stripe_len;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005724 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005725
Li Zefanec9ef7a2011-12-01 14:06:42 +08005726 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5727 BTRFS_BLOCK_GROUP_RAID10)) {
5728 bbio->stripes[i].length = stripes_per_dev *
5729 map->stripe_len;
Liu Bob89203f2012-04-12 16:03:56 -04005730
Li Zefanec9ef7a2011-12-01 14:06:42 +08005731 if (i / sub_stripes < remaining_stripes)
5732 bbio->stripes[i].length +=
5733 map->stripe_len;
Liu Bob89203f2012-04-12 16:03:56 -04005734
5735 /*
5736 * Special for the first stripe and
5737 * the last stripe:
5738 *
5739 * |-------|...|-------|
5740 * |----------|
5741 * off end_off
5742 */
Li Zefanec9ef7a2011-12-01 14:06:42 +08005743 if (i < sub_stripes)
Jan Schmidta1d3c472011-08-04 17:15:33 +02005744 bbio->stripes[i].length -=
Li Dongyangfce3bb92011-03-24 10:24:26 +00005745 stripe_offset;
Liu Bob89203f2012-04-12 16:03:56 -04005746
5747 if (stripe_index >= last_stripe &&
5748 stripe_index <= (last_stripe +
5749 sub_stripes - 1))
Li Zefanec9ef7a2011-12-01 14:06:42 +08005750 bbio->stripes[i].length -=
5751 stripe_end_offset;
Liu Bob89203f2012-04-12 16:03:56 -04005752
Li Zefanec9ef7a2011-12-01 14:06:42 +08005753 if (i == sub_stripes - 1)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005754 stripe_offset = 0;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005755 } else
Jan Schmidta1d3c472011-08-04 17:15:33 +02005756 bbio->stripes[i].length = *length;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005757
5758 stripe_index++;
5759 if (stripe_index == map->num_stripes) {
5760 /* This could only happen for RAID0/10 */
5761 stripe_index = 0;
5762 stripe_nr++;
5763 }
Chris Masonf2d8d742008-04-21 10:03:05 -04005764 }
Li Dongyangfce3bb92011-03-24 10:24:26 +00005765 } else {
5766 for (i = 0; i < num_stripes; i++) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02005767 bbio->stripes[i].physical =
Linus Torvalds212a17a2011-03-28 15:31:05 -07005768 map->stripes[stripe_index].physical +
5769 stripe_offset +
5770 stripe_nr * map->stripe_len;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005771 bbio->stripes[i].dev =
Linus Torvalds212a17a2011-03-28 15:31:05 -07005772 map->stripes[stripe_index].dev;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005773 stripe_index++;
5774 }
Chris Mason593060d2008-03-25 16:50:33 -04005775 }
Li Zefande11cc12011-12-01 12:55:47 +08005776
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005777 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
Miao Xied20983b2014-07-03 18:22:13 +08005778 max_errors = btrfs_chunk_max_errors(map);
Li Zefande11cc12011-12-01 12:55:47 +08005779
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005780 if (bbio->raid_map)
5781 sort_parity_stripes(bbio, num_stripes);
Zhao Leicc7539e2015-01-20 15:11:32 +08005782
Miao Xie2c8cdd62014-11-14 16:06:25 +08005783 tgtdev_indexes = 0;
Mike Christieb3d3fa52016-06-05 14:31:53 -05005784 if (dev_replace_is_ongoing &&
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005785 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD) &&
Stefan Behrens472262f2012-11-06 14:43:46 +01005786 dev_replace->tgtdev != NULL) {
5787 int index_where_to_add;
5788 u64 srcdev_devid = dev_replace->srcdev->devid;
5789
5790 /*
5791 * duplicate the write operations while the dev replace
5792 * procedure is running. Since the copying of the old disk
5793 * to the new disk takes place at run time while the
5794 * filesystem is mounted writable, the regular write
5795 * operations to the old disk have to be duplicated to go
5796 * to the new disk as well.
5797 * Note that device->missing is handled by the caller, and
5798 * that the write to the old disk is already set up in the
5799 * stripes array.
5800 */
5801 index_where_to_add = num_stripes;
5802 for (i = 0; i < num_stripes; i++) {
5803 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5804 /* write to new disk, too */
5805 struct btrfs_bio_stripe *new =
5806 bbio->stripes + index_where_to_add;
5807 struct btrfs_bio_stripe *old =
5808 bbio->stripes + i;
5809
5810 new->physical = old->physical;
5811 new->length = old->length;
5812 new->dev = dev_replace->tgtdev;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005813 bbio->tgtdev_map[i] = index_where_to_add;
Stefan Behrens472262f2012-11-06 14:43:46 +01005814 index_where_to_add++;
5815 max_errors++;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005816 tgtdev_indexes++;
Stefan Behrens472262f2012-11-06 14:43:46 +01005817 }
5818 }
5819 num_stripes = index_where_to_add;
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005820 } else if (dev_replace_is_ongoing &&
5821 op == BTRFS_MAP_GET_READ_MIRRORS &&
Stefan Behrensad6d6202012-11-06 15:06:47 +01005822 dev_replace->tgtdev != NULL) {
5823 u64 srcdev_devid = dev_replace->srcdev->devid;
5824 int index_srcdev = 0;
5825 int found = 0;
5826 u64 physical_of_found = 0;
5827
5828 /*
5829 * During the dev-replace procedure, the target drive can
5830 * also be used to read data in case it is needed to repair
5831 * a corrupt block elsewhere. This is possible if the
5832 * requested area is left of the left cursor. In this area,
5833 * the target drive is a full copy of the source drive.
5834 */
5835 for (i = 0; i < num_stripes; i++) {
5836 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5837 /*
5838 * In case of DUP, in order to keep it
5839 * simple, only add the mirror with the
5840 * lowest physical address
5841 */
5842 if (found &&
5843 physical_of_found <=
5844 bbio->stripes[i].physical)
5845 continue;
5846 index_srcdev = i;
5847 found = 1;
5848 physical_of_found = bbio->stripes[i].physical;
5849 }
5850 }
5851 if (found) {
Filipe Manana22ab04e2016-05-18 20:29:44 +01005852 struct btrfs_bio_stripe *tgtdev_stripe =
5853 bbio->stripes + num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005854
Filipe Manana22ab04e2016-05-18 20:29:44 +01005855 tgtdev_stripe->physical = physical_of_found;
5856 tgtdev_stripe->length =
5857 bbio->stripes[index_srcdev].length;
5858 tgtdev_stripe->dev = dev_replace->tgtdev;
5859 bbio->tgtdev_map[index_srcdev] = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005860
Filipe Manana22ab04e2016-05-18 20:29:44 +01005861 tgtdev_indexes++;
5862 num_stripes++;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005863 }
Stefan Behrens472262f2012-11-06 14:43:46 +01005864 }
5865
Li Zefande11cc12011-12-01 12:55:47 +08005866 *bbio_ret = bbio;
Zhao Lei10f11902015-01-20 15:11:43 +08005867 bbio->map_type = map->type;
Li Zefande11cc12011-12-01 12:55:47 +08005868 bbio->num_stripes = num_stripes;
5869 bbio->max_errors = max_errors;
5870 bbio->mirror_num = mirror_num;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005871 bbio->num_tgtdevs = tgtdev_indexes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005872
5873 /*
5874 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5875 * mirror_num == num_stripes + 1 && dev_replace target drive is
5876 * available as a mirror
5877 */
5878 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5879 WARN_ON(num_stripes > 1);
5880 bbio->stripes[0].dev = dev_replace->tgtdev;
5881 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5882 bbio->mirror_num = map->num_stripes + 1;
5883 }
Chris Masoncea9e442008-04-09 16:28:12 -04005884out:
Liu Bo73beece2015-07-17 16:49:19 +08005885 if (dev_replace_is_ongoing) {
5886 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5887 btrfs_dev_replace_unlock(dev_replace, 0);
5888 }
Chris Mason0b86a832008-03-24 15:01:56 -04005889 free_extent_map(em);
Li Zefande11cc12011-12-01 12:55:47 +08005890 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -04005891}
5892
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005893int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
Chris Masonf2d8d742008-04-21 10:03:05 -04005894 u64 logical, u64 *length,
Jan Schmidta1d3c472011-08-04 17:15:33 +02005895 struct btrfs_bio **bbio_ret, int mirror_num)
Chris Masonf2d8d742008-04-21 10:03:05 -04005896{
Mike Christieb3d3fa52016-06-05 14:31:53 -05005897 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005898 mirror_num, 0);
Chris Masonf2d8d742008-04-21 10:03:05 -04005899}
5900
Miao Xieaf8e2d12014-10-23 14:42:50 +08005901/* For Scrub/replace */
Christoph Hellwigcf8cddd2016-10-27 09:27:36 +02005902int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
Miao Xieaf8e2d12014-10-23 14:42:50 +08005903 u64 logical, u64 *length,
5904 struct btrfs_bio **bbio_ret, int mirror_num,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005905 int need_raid_map)
Miao Xieaf8e2d12014-10-23 14:42:50 +08005906{
Mike Christieb3d3fa52016-06-05 14:31:53 -05005907 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005908 mirror_num, need_raid_map);
Miao Xieaf8e2d12014-10-23 14:42:50 +08005909}
5910
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005911int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
Yan Zhenga512bbf2008-12-08 16:46:26 -05005912 u64 chunk_start, u64 physical, u64 devid,
5913 u64 **logical, int *naddrs, int *stripe_len)
5914{
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005915 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005916 struct extent_map_tree *em_tree = &map_tree->map_tree;
5917 struct extent_map *em;
5918 struct map_lookup *map;
5919 u64 *buf;
5920 u64 bytenr;
5921 u64 length;
5922 u64 stripe_nr;
David Woodhouse53b381b2013-01-29 18:40:14 -05005923 u64 rmap_len;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005924 int i, j, nr = 0;
5925
Chris Mason890871b2009-09-02 16:24:52 -04005926 read_lock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005927 em = lookup_extent_mapping(em_tree, chunk_start, 1);
Chris Mason890871b2009-09-02 16:24:52 -04005928 read_unlock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005929
Josef Bacik835d9742013-03-19 12:13:25 -04005930 if (!em) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005931 btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5932 chunk_start);
Josef Bacik835d9742013-03-19 12:13:25 -04005933 return -EIO;
5934 }
5935
5936 if (em->start != chunk_start) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005937 btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
Josef Bacik835d9742013-03-19 12:13:25 -04005938 em->start, chunk_start);
5939 free_extent_map(em);
5940 return -EIO;
5941 }
Jeff Mahoney95617d62015-06-03 10:55:48 -04005942 map = em->map_lookup;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005943
5944 length = em->len;
David Woodhouse53b381b2013-01-29 18:40:14 -05005945 rmap_len = map->stripe_len;
5946
Yan Zhenga512bbf2008-12-08 16:46:26 -05005947 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
David Sterbab8b93ad2015-01-16 17:26:13 +01005948 length = div_u64(length, map->num_stripes / map->sub_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005949 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
David Sterbab8b93ad2015-01-16 17:26:13 +01005950 length = div_u64(length, map->num_stripes);
Zhao Leiffe2d202015-01-20 15:11:44 +08005951 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Sterbab8b93ad2015-01-16 17:26:13 +01005952 length = div_u64(length, nr_data_stripes(map));
David Woodhouse53b381b2013-01-29 18:40:14 -05005953 rmap_len = map->stripe_len * nr_data_stripes(map);
5954 }
Yan Zhenga512bbf2008-12-08 16:46:26 -05005955
David Sterba31e818f2015-02-20 18:00:26 +01005956 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005957 BUG_ON(!buf); /* -ENOMEM */
Yan Zhenga512bbf2008-12-08 16:46:26 -05005958
5959 for (i = 0; i < map->num_stripes; i++) {
5960 if (devid && map->stripes[i].dev->devid != devid)
5961 continue;
5962 if (map->stripes[i].physical > physical ||
5963 map->stripes[i].physical + length <= physical)
5964 continue;
5965
5966 stripe_nr = physical - map->stripes[i].physical;
David Sterbab8b93ad2015-01-16 17:26:13 +01005967 stripe_nr = div_u64(stripe_nr, map->stripe_len);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005968
5969 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5970 stripe_nr = stripe_nr * map->num_stripes + i;
David Sterbab8b93ad2015-01-16 17:26:13 +01005971 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005972 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5973 stripe_nr = stripe_nr * map->num_stripes + i;
David Woodhouse53b381b2013-01-29 18:40:14 -05005974 } /* else if RAID[56], multiply by nr_data_stripes().
5975 * Alternatively, just use rmap_len below instead of
5976 * map->stripe_len */
5977
5978 bytenr = chunk_start + stripe_nr * rmap_len;
Chris Mason934d3752008-12-08 16:43:10 -05005979 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005980 for (j = 0; j < nr; j++) {
5981 if (buf[j] == bytenr)
5982 break;
5983 }
Chris Mason934d3752008-12-08 16:43:10 -05005984 if (j == nr) {
5985 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005986 buf[nr++] = bytenr;
Chris Mason934d3752008-12-08 16:43:10 -05005987 }
Yan Zhenga512bbf2008-12-08 16:46:26 -05005988 }
5989
Yan Zhenga512bbf2008-12-08 16:46:26 -05005990 *logical = buf;
5991 *naddrs = nr;
David Woodhouse53b381b2013-01-29 18:40:14 -05005992 *stripe_len = rmap_len;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005993
5994 free_extent_map(em);
5995 return 0;
5996}
5997
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005998static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
Miao Xie8408c712014-06-19 10:42:55 +08005999{
Mike Snitzer326e1db2015-05-22 09:14:03 -04006000 bio->bi_private = bbio->private;
6001 bio->bi_end_io = bbio->end_io;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006002 bio_endio(bio);
Mike Snitzer326e1db2015-05-22 09:14:03 -04006003
Zhao Lei6e9606d2015-01-20 15:11:34 +08006004 btrfs_put_bbio(bbio);
Miao Xie8408c712014-06-19 10:42:55 +08006005}
6006
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006007static void btrfs_end_bio(struct bio *bio)
Chris Mason8790d502008-04-03 16:29:03 -04006008{
Chris Mason9be33952013-05-17 18:30:14 -04006009 struct btrfs_bio *bbio = bio->bi_private;
Chris Mason7d2b4da2008-08-05 10:13:57 -04006010 int is_orig_bio = 0;
Chris Mason8790d502008-04-03 16:29:03 -04006011
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006012 if (bio->bi_error) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02006013 atomic_inc(&bbio->error);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006014 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
Stefan Behrens442a4f62012-05-25 16:06:08 +02006015 unsigned int stripe_index =
Chris Mason9be33952013-05-17 18:30:14 -04006016 btrfs_io_bio(bio)->stripe_index;
Zhao Lei65f53332015-06-08 20:05:50 +08006017 struct btrfs_device *dev;
Stefan Behrens442a4f62012-05-25 16:06:08 +02006018
6019 BUG_ON(stripe_index >= bbio->num_stripes);
6020 dev = bbio->stripes[stripe_index].dev;
Stefan Behrens597a60f2012-06-14 16:42:31 +02006021 if (dev->bdev) {
Mike Christie37226b22016-06-05 14:31:52 -05006022 if (bio_op(bio) == REQ_OP_WRITE)
Stefan Behrens597a60f2012-06-14 16:42:31 +02006023 btrfs_dev_stat_inc(dev,
6024 BTRFS_DEV_STAT_WRITE_ERRS);
6025 else
6026 btrfs_dev_stat_inc(dev,
6027 BTRFS_DEV_STAT_READ_ERRS);
Jens Axboe1eff9d32016-08-05 15:35:16 -06006028 if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH)
Stefan Behrens597a60f2012-06-14 16:42:31 +02006029 btrfs_dev_stat_inc(dev,
6030 BTRFS_DEV_STAT_FLUSH_ERRS);
6031 btrfs_dev_stat_print_on_error(dev);
6032 }
Stefan Behrens442a4f62012-05-25 16:06:08 +02006033 }
6034 }
Chris Mason8790d502008-04-03 16:29:03 -04006035
Jan Schmidta1d3c472011-08-04 17:15:33 +02006036 if (bio == bbio->orig_bio)
Chris Mason7d2b4da2008-08-05 10:13:57 -04006037 is_orig_bio = 1;
6038
Miao Xiec404e0d2014-01-30 16:46:55 +08006039 btrfs_bio_counter_dec(bbio->fs_info);
6040
Jan Schmidta1d3c472011-08-04 17:15:33 +02006041 if (atomic_dec_and_test(&bbio->stripes_pending)) {
Chris Mason7d2b4da2008-08-05 10:13:57 -04006042 if (!is_orig_bio) {
6043 bio_put(bio);
Jan Schmidta1d3c472011-08-04 17:15:33 +02006044 bio = bbio->orig_bio;
Chris Mason7d2b4da2008-08-05 10:13:57 -04006045 }
Muthu Kumarc7b22bb2014-01-08 14:19:52 -07006046
Chris Mason9be33952013-05-17 18:30:14 -04006047 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
Chris Masona236aed2008-04-29 09:38:00 -04006048 /* only send an error to the higher layers if it is
David Woodhouse53b381b2013-01-29 18:40:14 -05006049 * beyond the tolerance of the btrfs bio
Chris Masona236aed2008-04-29 09:38:00 -04006050 */
Jan Schmidta1d3c472011-08-04 17:15:33 +02006051 if (atomic_read(&bbio->error) > bbio->max_errors) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006052 bio->bi_error = -EIO;
Chris Mason5dbc8fc2011-12-09 11:07:37 -05006053 } else {
Chris Mason1259ab72008-05-12 13:39:03 -04006054 /*
6055 * this bio is actually up to date, we didn't
6056 * go over the max number of errors
6057 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006058 bio->bi_error = 0;
Chris Mason1259ab72008-05-12 13:39:03 -04006059 }
Miao Xiec55f1392014-06-19 10:42:54 +08006060
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006061 btrfs_end_bbio(bbio, bio);
Chris Mason7d2b4da2008-08-05 10:13:57 -04006062 } else if (!is_orig_bio) {
Chris Mason8790d502008-04-03 16:29:03 -04006063 bio_put(bio);
6064 }
Chris Mason8790d502008-04-03 16:29:03 -04006065}
6066
Chris Mason8b712842008-06-11 16:50:36 -04006067/*
6068 * see run_scheduled_bios for a description of why bios are collected for
6069 * async submit.
6070 *
6071 * This will add one bio to the pending list for a device and make sure
6072 * the work struct is scheduled.
6073 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00006074static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6075 struct btrfs_device *device,
Mike Christie4e49ea42016-06-05 14:31:41 -05006076 struct bio *bio)
Chris Mason8b712842008-06-11 16:50:36 -04006077{
6078 int should_queue = 1;
Chris Masonffbd5172009-04-20 15:50:09 -04006079 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04006080
David Woodhouse53b381b2013-01-29 18:40:14 -05006081 if (device->missing || !device->bdev) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006082 bio_io_error(bio);
David Woodhouse53b381b2013-01-29 18:40:14 -05006083 return;
6084 }
6085
Chris Mason8b712842008-06-11 16:50:36 -04006086 /* don't bother with additional async steps for reads, right now */
Mike Christie37226b22016-06-05 14:31:52 -05006087 if (bio_op(bio) == REQ_OP_READ) {
Chris Mason492bb6de2008-07-31 16:29:02 -04006088 bio_get(bio);
Mike Christie4e49ea42016-06-05 14:31:41 -05006089 btrfsic_submit_bio(bio);
Chris Mason492bb6de2008-07-31 16:29:02 -04006090 bio_put(bio);
Jeff Mahoney143bede2012-03-01 14:56:26 +01006091 return;
Chris Mason8b712842008-06-11 16:50:36 -04006092 }
6093
6094 /*
Chris Mason0986fe9e2008-08-15 15:34:15 -04006095 * nr_async_bios allows us to reliably return congestion to the
Chris Mason8b712842008-06-11 16:50:36 -04006096 * higher layers. Otherwise, the async bio makes it appear we have
6097 * made progress against dirty pages when we've really just put it
6098 * on a queue for later
6099 */
Chris Mason0986fe9e2008-08-15 15:34:15 -04006100 atomic_inc(&root->fs_info->nr_async_bios);
Chris Mason492bb6de2008-07-31 16:29:02 -04006101 WARN_ON(bio->bi_next);
Chris Mason8b712842008-06-11 16:50:36 -04006102 bio->bi_next = NULL;
Chris Mason8b712842008-06-11 16:50:36 -04006103
6104 spin_lock(&device->io_lock);
Jens Axboe1eff9d32016-08-05 15:35:16 -06006105 if (bio->bi_opf & REQ_SYNC)
Chris Masonffbd5172009-04-20 15:50:09 -04006106 pending_bios = &device->pending_sync_bios;
6107 else
6108 pending_bios = &device->pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04006109
Chris Masonffbd5172009-04-20 15:50:09 -04006110 if (pending_bios->tail)
6111 pending_bios->tail->bi_next = bio;
Chris Mason8b712842008-06-11 16:50:36 -04006112
Chris Masonffbd5172009-04-20 15:50:09 -04006113 pending_bios->tail = bio;
6114 if (!pending_bios->head)
6115 pending_bios->head = bio;
Chris Mason8b712842008-06-11 16:50:36 -04006116 if (device->running_pending)
6117 should_queue = 0;
6118
6119 spin_unlock(&device->io_lock);
6120
6121 if (should_queue)
Qu Wenruoa8c93d4e2014-02-28 10:46:08 +08006122 btrfs_queue_work(root->fs_info->submit_workers,
6123 &device->work);
Chris Mason8b712842008-06-11 16:50:36 -04006124}
6125
Josef Bacikde1ee922012-10-19 16:50:56 -04006126static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6127 struct bio *bio, u64 physical, int dev_nr,
Mike Christie81a75f672016-06-05 14:31:54 -05006128 int async)
Josef Bacikde1ee922012-10-19 16:50:56 -04006129{
6130 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6131
6132 bio->bi_private = bbio;
Chris Mason9be33952013-05-17 18:30:14 -04006133 btrfs_io_bio(bio)->stripe_index = dev_nr;
Josef Bacikde1ee922012-10-19 16:50:56 -04006134 bio->bi_end_io = btrfs_end_bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006135 bio->bi_iter.bi_sector = physical >> 9;
Josef Bacikde1ee922012-10-19 16:50:56 -04006136#ifdef DEBUG
6137 {
6138 struct rcu_string *name;
6139
6140 rcu_read_lock();
6141 name = rcu_dereference(dev->name);
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006142 btrfs_debug(fs_info,
6143 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6144 bio_op(bio), bio->bi_opf,
6145 (u64)bio->bi_iter.bi_sector,
6146 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6147 bio->bi_iter.bi_size);
Josef Bacikde1ee922012-10-19 16:50:56 -04006148 rcu_read_unlock();
6149 }
6150#endif
6151 bio->bi_bdev = dev->bdev;
Miao Xiec404e0d2014-01-30 16:46:55 +08006152
6153 btrfs_bio_counter_inc_noblocked(root->fs_info);
6154
Josef Bacikde1ee922012-10-19 16:50:56 -04006155 if (async)
Mike Christie4e49ea42016-06-05 14:31:41 -05006156 btrfs_schedule_bio(root, dev, bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006157 else
Mike Christie4e49ea42016-06-05 14:31:41 -05006158 btrfsic_submit_bio(bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006159}
6160
Josef Bacikde1ee922012-10-19 16:50:56 -04006161static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6162{
6163 atomic_inc(&bbio->error);
6164 if (atomic_dec_and_test(&bbio->stripes_pending)) {
Nicholas D Steeves01327612016-05-19 21:18:45 -04006165 /* Should be the original bio. */
Miao Xie8408c712014-06-19 10:42:55 +08006166 WARN_ON(bio != bbio->orig_bio);
6167
Chris Mason9be33952013-05-17 18:30:14 -04006168 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006169 bio->bi_iter.bi_sector = logical >> 9;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006170 bio->bi_error = -EIO;
6171 btrfs_end_bbio(bbio, bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006172 }
6173}
6174
Mike Christie81a75f672016-06-05 14:31:54 -05006175int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
Chris Mason8b712842008-06-11 16:50:36 -04006176 int mirror_num, int async_submit)
Chris Mason0b86a832008-03-24 15:01:56 -04006177{
Chris Mason0b86a832008-03-24 15:01:56 -04006178 struct btrfs_device *dev;
Chris Mason8790d502008-04-03 16:29:03 -04006179 struct bio *first_bio = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006180 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
Chris Mason0b86a832008-03-24 15:01:56 -04006181 u64 length = 0;
6182 u64 map_length;
Chris Mason0b86a832008-03-24 15:01:56 -04006183 int ret;
Zhao Lei08da7572015-02-12 15:42:16 +08006184 int dev_nr;
6185 int total_devs;
Jan Schmidta1d3c472011-08-04 17:15:33 +02006186 struct btrfs_bio *bbio = NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04006187
Kent Overstreet4f024f32013-10-11 15:44:27 -07006188 length = bio->bi_iter.bi_size;
Chris Mason0b86a832008-03-24 15:01:56 -04006189 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04006190
Miao Xiec404e0d2014-01-30 16:46:55 +08006191 btrfs_bio_counter_inc_blocked(root->fs_info);
Mike Christie37226b22016-06-05 14:31:52 -05006192 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6193 &map_length, &bbio, mirror_num, 1);
Miao Xiec404e0d2014-01-30 16:46:55 +08006194 if (ret) {
6195 btrfs_bio_counter_dec(root->fs_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006196 return ret;
Miao Xiec404e0d2014-01-30 16:46:55 +08006197 }
Chris Masoncea9e442008-04-09 16:28:12 -04006198
Jan Schmidta1d3c472011-08-04 17:15:33 +02006199 total_devs = bbio->num_stripes;
David Woodhouse53b381b2013-01-29 18:40:14 -05006200 bbio->orig_bio = first_bio;
6201 bbio->private = first_bio->bi_private;
6202 bbio->end_io = first_bio->bi_end_io;
Miao Xiec404e0d2014-01-30 16:46:55 +08006203 bbio->fs_info = root->fs_info;
David Woodhouse53b381b2013-01-29 18:40:14 -05006204 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6205
Zhao Leiad1ba2a2015-12-15 18:18:09 +08006206 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
Mike Christie37226b22016-06-05 14:31:52 -05006207 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
David Woodhouse53b381b2013-01-29 18:40:14 -05006208 /* In this case, map_length has been set to the length of
6209 a single stripe; not the whole write */
Mike Christie37226b22016-06-05 14:31:52 -05006210 if (bio_op(bio) == REQ_OP_WRITE) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08006211 ret = raid56_parity_write(root, bio, bbio, map_length);
David Woodhouse53b381b2013-01-29 18:40:14 -05006212 } else {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08006213 ret = raid56_parity_recover(root, bio, bbio, map_length,
Miao Xie42452152014-11-25 16:39:28 +08006214 mirror_num, 1);
David Woodhouse53b381b2013-01-29 18:40:14 -05006215 }
Miao Xie42452152014-11-25 16:39:28 +08006216
Miao Xiec404e0d2014-01-30 16:46:55 +08006217 btrfs_bio_counter_dec(root->fs_info);
6218 return ret;
David Woodhouse53b381b2013-01-29 18:40:14 -05006219 }
6220
Chris Masoncea9e442008-04-09 16:28:12 -04006221 if (map_length < length) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04006222 btrfs_crit(root->fs_info,
6223 "mapping failed logical %llu bio len %llu len %llu",
6224 logical, length, map_length);
Chris Masoncea9e442008-04-09 16:28:12 -04006225 BUG();
6226 }
Jan Schmidta1d3c472011-08-04 17:15:33 +02006227
Zhao Lei08da7572015-02-12 15:42:16 +08006228 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
Josef Bacikde1ee922012-10-19 16:50:56 -04006229 dev = bbio->stripes[dev_nr].dev;
Mike Christie37226b22016-06-05 14:31:52 -05006230 if (!dev || !dev->bdev ||
6231 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
Josef Bacikde1ee922012-10-19 16:50:56 -04006232 bbio_error(bbio, first_bio, logical);
Josef Bacikde1ee922012-10-19 16:50:56 -04006233 continue;
6234 }
6235
Jan Schmidta1d3c472011-08-04 17:15:33 +02006236 if (dev_nr < total_devs - 1) {
Chris Mason9be33952013-05-17 18:30:14 -04006237 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006238 BUG_ON(!bio); /* -ENOMEM */
Mike Snitzer326e1db2015-05-22 09:14:03 -04006239 } else
Jan Schmidta1d3c472011-08-04 17:15:33 +02006240 bio = first_bio;
Josef Bacik606686e2012-06-04 14:03:51 -04006241
Josef Bacikde1ee922012-10-19 16:50:56 -04006242 submit_stripe_bio(root, bbio, bio,
Mike Christie81a75f672016-06-05 14:31:54 -05006243 bbio->stripes[dev_nr].physical, dev_nr,
Josef Bacikde1ee922012-10-19 16:50:56 -04006244 async_submit);
Chris Mason239b14b2008-03-24 15:02:07 -04006245 }
Miao Xiec404e0d2014-01-30 16:46:55 +08006246 btrfs_bio_counter_dec(root->fs_info);
Chris Mason0b86a832008-03-24 15:01:56 -04006247 return 0;
6248}
6249
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006250struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
Yan Zheng2b820322008-11-17 21:11:30 -05006251 u8 *uuid, u8 *fsid)
Chris Mason0b86a832008-03-24 15:01:56 -04006252{
Yan Zheng2b820322008-11-17 21:11:30 -05006253 struct btrfs_device *device;
6254 struct btrfs_fs_devices *cur_devices;
Chris Mason0b86a832008-03-24 15:01:56 -04006255
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006256 cur_devices = fs_info->fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006257 while (cur_devices) {
6258 if (!fsid ||
6259 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6260 device = __find_device(&cur_devices->devices,
6261 devid, uuid);
6262 if (device)
6263 return device;
6264 }
6265 cur_devices = cur_devices->seed;
6266 }
6267 return NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04006268}
6269
Chris Masondfe25022008-05-13 13:46:40 -04006270static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
Miao Xie5f375832014-09-03 21:35:46 +08006271 struct btrfs_fs_devices *fs_devices,
Chris Masondfe25022008-05-13 13:46:40 -04006272 u64 devid, u8 *dev_uuid)
6273{
6274 struct btrfs_device *device;
Chris Masondfe25022008-05-13 13:46:40 -04006275
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006276 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6277 if (IS_ERR(device))
yanhai zhu7cbd8a82008-11-12 14:38:54 -05006278 return NULL;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006279
6280 list_add(&device->dev_list, &fs_devices->devices);
Yan Zhenge4404d62008-12-12 10:03:26 -05006281 device->fs_devices = fs_devices;
Chris Masondfe25022008-05-13 13:46:40 -04006282 fs_devices->num_devices++;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006283
6284 device->missing = 1;
Chris Masoncd02dca2010-12-13 14:56:23 -05006285 fs_devices->missing_devices++;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006286
Chris Masondfe25022008-05-13 13:46:40 -04006287 return device;
6288}
6289
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006290/**
6291 * btrfs_alloc_device - allocate struct btrfs_device
6292 * @fs_info: used only for generating a new devid, can be NULL if
6293 * devid is provided (i.e. @devid != NULL).
6294 * @devid: a pointer to devid for this device. If NULL a new devid
6295 * is generated.
6296 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6297 * is generated.
6298 *
6299 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6300 * on error. Returned struct is not linked onto any lists and can be
6301 * destroyed with kfree() right away.
6302 */
6303struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6304 const u64 *devid,
6305 const u8 *uuid)
6306{
6307 struct btrfs_device *dev;
6308 u64 tmp;
6309
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05306310 if (WARN_ON(!devid && !fs_info))
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006311 return ERR_PTR(-EINVAL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006312
6313 dev = __alloc_device();
6314 if (IS_ERR(dev))
6315 return dev;
6316
6317 if (devid)
6318 tmp = *devid;
6319 else {
6320 int ret;
6321
6322 ret = find_next_devid(fs_info, &tmp);
6323 if (ret) {
6324 kfree(dev);
6325 return ERR_PTR(ret);
6326 }
6327 }
6328 dev->devid = tmp;
6329
6330 if (uuid)
6331 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6332 else
6333 generate_random_uuid(dev->uuid);
6334
Liu Bo9e0af232014-08-15 23:36:53 +08006335 btrfs_init_work(&dev->work, btrfs_submit_helper,
6336 pending_bios_fn, NULL, NULL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006337
6338 return dev;
6339}
6340
Liu Boe06cd3d2016-06-03 12:05:15 -07006341/* Return -EIO if any error, otherwise return 0. */
6342static int btrfs_check_chunk_valid(struct btrfs_root *root,
6343 struct extent_buffer *leaf,
6344 struct btrfs_chunk *chunk, u64 logical)
6345{
6346 u64 length;
6347 u64 stripe_len;
6348 u16 num_stripes;
6349 u16 sub_stripes;
6350 u64 type;
6351
6352 length = btrfs_chunk_length(leaf, chunk);
6353 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6354 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6355 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6356 type = btrfs_chunk_type(leaf, chunk);
6357
6358 if (!num_stripes) {
6359 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6360 num_stripes);
6361 return -EIO;
6362 }
6363 if (!IS_ALIGNED(logical, root->sectorsize)) {
6364 btrfs_err(root->fs_info,
6365 "invalid chunk logical %llu", logical);
6366 return -EIO;
6367 }
6368 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6369 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6370 btrfs_chunk_sector_size(leaf, chunk));
6371 return -EIO;
6372 }
6373 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6374 btrfs_err(root->fs_info,
6375 "invalid chunk length %llu", length);
6376 return -EIO;
6377 }
6378 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6379 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6380 stripe_len);
6381 return -EIO;
6382 }
6383 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6384 type) {
6385 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6386 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6387 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6388 btrfs_chunk_type(leaf, chunk));
6389 return -EIO;
6390 }
6391 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6392 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6393 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6394 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6395 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6396 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6397 num_stripes != 1)) {
6398 btrfs_err(root->fs_info,
6399 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6400 num_stripes, sub_stripes,
6401 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6402 return -EIO;
6403 }
6404
6405 return 0;
6406}
6407
Chris Mason0b86a832008-03-24 15:01:56 -04006408static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6409 struct extent_buffer *leaf,
6410 struct btrfs_chunk *chunk)
6411{
6412 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6413 struct map_lookup *map;
6414 struct extent_map *em;
6415 u64 logical;
6416 u64 length;
Qu Wenruof04b7722015-12-15 09:14:37 +08006417 u64 stripe_len;
Chris Mason0b86a832008-03-24 15:01:56 -04006418 u64 devid;
Chris Masona4437552008-04-18 10:29:38 -04006419 u8 uuid[BTRFS_UUID_SIZE];
Chris Mason593060d2008-03-25 16:50:33 -04006420 int num_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04006421 int ret;
Chris Mason593060d2008-03-25 16:50:33 -04006422 int i;
Chris Mason0b86a832008-03-24 15:01:56 -04006423
Chris Masone17cade2008-04-15 15:41:47 -04006424 logical = key->offset;
6425 length = btrfs_chunk_length(leaf, chunk);
Qu Wenruof04b7722015-12-15 09:14:37 +08006426 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6427 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
Liu Boe06cd3d2016-06-03 12:05:15 -07006428
6429 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6430 if (ret)
6431 return ret;
Chris Masona061fc82008-05-07 11:43:44 -04006432
Chris Mason890871b2009-09-02 16:24:52 -04006433 read_lock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04006434 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
Chris Mason890871b2009-09-02 16:24:52 -04006435 read_unlock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04006436
6437 /* already mapped? */
6438 if (em && em->start <= logical && em->start + em->len > logical) {
6439 free_extent_map(em);
Chris Mason0b86a832008-03-24 15:01:56 -04006440 return 0;
6441 } else if (em) {
6442 free_extent_map(em);
6443 }
Chris Mason0b86a832008-03-24 15:01:56 -04006444
David Sterba172ddd62011-04-21 00:48:27 +02006445 em = alloc_extent_map();
Chris Mason0b86a832008-03-24 15:01:56 -04006446 if (!em)
6447 return -ENOMEM;
Chris Mason593060d2008-03-25 16:50:33 -04006448 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
Chris Mason0b86a832008-03-24 15:01:56 -04006449 if (!map) {
6450 free_extent_map(em);
6451 return -ENOMEM;
6452 }
6453
Wang Shilong298a8f92014-06-19 10:42:52 +08006454 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
Jeff Mahoney95617d62015-06-03 10:55:48 -04006455 em->map_lookup = map;
Chris Mason0b86a832008-03-24 15:01:56 -04006456 em->start = logical;
6457 em->len = length;
Josef Bacik70c8a912012-10-11 16:54:30 -04006458 em->orig_start = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006459 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04006460 em->block_len = em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04006461
Chris Mason593060d2008-03-25 16:50:33 -04006462 map->num_stripes = num_stripes;
6463 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6464 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6465 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6466 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6467 map->type = btrfs_chunk_type(leaf, chunk);
Chris Mason321aecc2008-04-16 10:49:51 -04006468 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
Chris Mason593060d2008-03-25 16:50:33 -04006469 for (i = 0; i < num_stripes; i++) {
6470 map->stripes[i].physical =
6471 btrfs_stripe_offset_nr(leaf, chunk, i);
6472 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
Chris Masona4437552008-04-18 10:29:38 -04006473 read_extent_buffer(leaf, uuid, (unsigned long)
6474 btrfs_stripe_dev_uuid_nr(chunk, i),
6475 BTRFS_UUID_SIZE);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006476 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6477 uuid, NULL);
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006478 if (!map->stripes[i].dev &&
6479 !btrfs_test_opt(root->fs_info, DEGRADED)) {
Chris Mason593060d2008-03-25 16:50:33 -04006480 free_extent_map(em);
6481 return -EIO;
6482 }
Chris Masondfe25022008-05-13 13:46:40 -04006483 if (!map->stripes[i].dev) {
6484 map->stripes[i].dev =
Miao Xie5f375832014-09-03 21:35:46 +08006485 add_missing_dev(root, root->fs_info->fs_devices,
6486 devid, uuid);
Chris Masondfe25022008-05-13 13:46:40 -04006487 if (!map->stripes[i].dev) {
Chris Masondfe25022008-05-13 13:46:40 -04006488 free_extent_map(em);
6489 return -EIO;
6490 }
Jeff Mahoney5d163e02016-09-20 10:05:00 -04006491 btrfs_warn(root->fs_info,
6492 "devid %llu uuid %pU is missing",
6493 devid, uuid);
Chris Masondfe25022008-05-13 13:46:40 -04006494 }
6495 map->stripes[i].dev->in_fs_metadata = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04006496 }
6497
Chris Mason890871b2009-09-02 16:24:52 -04006498 write_lock(&map_tree->map_tree.lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006499 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
Chris Mason890871b2009-09-02 16:24:52 -04006500 write_unlock(&map_tree->map_tree.lock);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006501 BUG_ON(ret); /* Tree corruption */
Chris Mason0b86a832008-03-24 15:01:56 -04006502 free_extent_map(em);
6503
6504 return 0;
6505}
6506
Jeff Mahoney143bede2012-03-01 14:56:26 +01006507static void fill_device_from_item(struct extent_buffer *leaf,
Chris Mason0b86a832008-03-24 15:01:56 -04006508 struct btrfs_dev_item *dev_item,
6509 struct btrfs_device *device)
6510{
6511 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04006512
6513 device->devid = btrfs_device_id(leaf, dev_item);
Chris Balld6397ba2009-04-27 07:29:03 -04006514 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6515 device->total_bytes = device->disk_total_bytes;
Miao Xie935e5cc2014-09-03 21:35:33 +08006516 device->commit_total_bytes = device->disk_total_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04006517 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
Miao Xiece7213c2014-09-03 21:35:34 +08006518 device->commit_bytes_used = device->bytes_used;
Chris Mason0b86a832008-03-24 15:01:56 -04006519 device->type = btrfs_device_type(leaf, dev_item);
6520 device->io_align = btrfs_device_io_align(leaf, dev_item);
6521 device->io_width = btrfs_device_io_width(leaf, dev_item);
6522 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
Stefan Behrens8dabb742012-11-06 13:15:27 +01006523 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
Stefan Behrens63a212a2012-11-05 18:29:28 +01006524 device->is_tgtdev_for_dev_replace = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006525
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02006526 ptr = btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04006527 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04006528}
6529
Miao Xie5f375832014-09-03 21:35:46 +08006530static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6531 u8 *fsid)
Yan Zheng2b820322008-11-17 21:11:30 -05006532{
6533 struct btrfs_fs_devices *fs_devices;
6534 int ret;
6535
Li Zefanb367e472011-12-07 11:38:24 +08006536 BUG_ON(!mutex_is_locked(&uuid_mutex));
Yan Zheng2b820322008-11-17 21:11:30 -05006537
6538 fs_devices = root->fs_info->fs_devices->seed;
6539 while (fs_devices) {
Miao Xie5f375832014-09-03 21:35:46 +08006540 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6541 return fs_devices;
6542
Yan Zheng2b820322008-11-17 21:11:30 -05006543 fs_devices = fs_devices->seed;
6544 }
6545
6546 fs_devices = find_fsid(fsid);
6547 if (!fs_devices) {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006548 if (!btrfs_test_opt(root->fs_info, DEGRADED))
Miao Xie5f375832014-09-03 21:35:46 +08006549 return ERR_PTR(-ENOENT);
6550
6551 fs_devices = alloc_fs_devices(fsid);
6552 if (IS_ERR(fs_devices))
6553 return fs_devices;
6554
6555 fs_devices->seeding = 1;
6556 fs_devices->opened = 1;
6557 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006558 }
Yan Zhenge4404d62008-12-12 10:03:26 -05006559
6560 fs_devices = clone_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006561 if (IS_ERR(fs_devices))
6562 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006563
Christoph Hellwig97288f22008-12-02 06:36:09 -05006564 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
Chris Mason15916de2008-11-19 21:17:22 -05006565 root->fs_info->bdev_holder);
Julia Lawall48d28232012-04-14 11:24:33 +02006566 if (ret) {
6567 free_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006568 fs_devices = ERR_PTR(ret);
Yan Zheng2b820322008-11-17 21:11:30 -05006569 goto out;
Julia Lawall48d28232012-04-14 11:24:33 +02006570 }
Yan Zheng2b820322008-11-17 21:11:30 -05006571
6572 if (!fs_devices->seeding) {
6573 __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -05006574 free_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006575 fs_devices = ERR_PTR(-EINVAL);
Yan Zheng2b820322008-11-17 21:11:30 -05006576 goto out;
6577 }
6578
6579 fs_devices->seed = root->fs_info->fs_devices->seed;
6580 root->fs_info->fs_devices->seed = fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006581out:
Miao Xie5f375832014-09-03 21:35:46 +08006582 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006583}
6584
Chris Mason0d81ba52008-03-24 15:02:07 -04006585static int read_one_dev(struct btrfs_root *root,
Chris Mason0b86a832008-03-24 15:01:56 -04006586 struct extent_buffer *leaf,
6587 struct btrfs_dev_item *dev_item)
6588{
Miao Xie5f375832014-09-03 21:35:46 +08006589 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
Chris Mason0b86a832008-03-24 15:01:56 -04006590 struct btrfs_device *device;
6591 u64 devid;
6592 int ret;
Yan Zheng2b820322008-11-17 21:11:30 -05006593 u8 fs_uuid[BTRFS_UUID_SIZE];
Chris Masona4437552008-04-18 10:29:38 -04006594 u8 dev_uuid[BTRFS_UUID_SIZE];
6595
Chris Mason0b86a832008-03-24 15:01:56 -04006596 devid = btrfs_device_id(leaf, dev_item);
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02006597 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
Chris Masona4437552008-04-18 10:29:38 -04006598 BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02006599 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05006600 BTRFS_UUID_SIZE);
6601
6602 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
Miao Xie5f375832014-09-03 21:35:46 +08006603 fs_devices = open_seed_devices(root, fs_uuid);
6604 if (IS_ERR(fs_devices))
6605 return PTR_ERR(fs_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05006606 }
6607
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006608 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
Miao Xie5f375832014-09-03 21:35:46 +08006609 if (!device) {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006610 if (!btrfs_test_opt(root->fs_info, DEGRADED))
Yan Zheng2b820322008-11-17 21:11:30 -05006611 return -EIO;
6612
Miao Xie5f375832014-09-03 21:35:46 +08006613 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6614 if (!device)
6615 return -ENOMEM;
Anand Jain33b97e42015-05-08 04:34:35 +08006616 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6617 devid, dev_uuid);
Miao Xie5f375832014-09-03 21:35:46 +08006618 } else {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006619 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
Miao Xie5f375832014-09-03 21:35:46 +08006620 return -EIO;
6621
6622 if(!device->bdev && !device->missing) {
Chris Masoncd02dca2010-12-13 14:56:23 -05006623 /*
6624 * this happens when a device that was properly setup
6625 * in the device info lists suddenly goes bad.
6626 * device->bdev is NULL, and so we have to set
6627 * device->missing to one here
6628 */
Miao Xie5f375832014-09-03 21:35:46 +08006629 device->fs_devices->missing_devices++;
Chris Masoncd02dca2010-12-13 14:56:23 -05006630 device->missing = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05006631 }
Miao Xie5f375832014-09-03 21:35:46 +08006632
6633 /* Move the device to its own fs_devices */
6634 if (device->fs_devices != fs_devices) {
6635 ASSERT(device->missing);
6636
6637 list_move(&device->dev_list, &fs_devices->devices);
6638 device->fs_devices->num_devices--;
6639 fs_devices->num_devices++;
6640
6641 device->fs_devices->missing_devices--;
6642 fs_devices->missing_devices++;
6643
6644 device->fs_devices = fs_devices;
6645 }
Yan Zheng2b820322008-11-17 21:11:30 -05006646 }
6647
6648 if (device->fs_devices != root->fs_info->fs_devices) {
6649 BUG_ON(device->writeable);
6650 if (device->generation !=
6651 btrfs_device_generation(leaf, dev_item))
6652 return -EINVAL;
Chris Mason6324fbf2008-03-24 15:01:59 -04006653 }
Chris Mason0b86a832008-03-24 15:01:56 -04006654
6655 fill_device_from_item(leaf, dev_item, device);
Chris Masondfe25022008-05-13 13:46:40 -04006656 device->in_fs_metadata = 1;
Stefan Behrens63a212a2012-11-05 18:29:28 +01006657 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
Yan Zheng2b820322008-11-17 21:11:30 -05006658 device->fs_devices->total_rw_bytes += device->total_bytes;
Josef Bacik2bf64752011-09-26 17:12:22 -04006659 spin_lock(&root->fs_info->free_chunk_lock);
6660 root->fs_info->free_chunk_space += device->total_bytes -
6661 device->bytes_used;
6662 spin_unlock(&root->fs_info->free_chunk_lock);
6663 }
Chris Mason0b86a832008-03-24 15:01:56 -04006664 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006665 return ret;
6666}
6667
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04006668int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
Chris Mason0b86a832008-03-24 15:01:56 -04006669{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04006670 struct btrfs_root *root = fs_info->tree_root;
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006671 struct btrfs_super_block *super_copy = fs_info->super_copy;
Chris Masona061fc82008-05-07 11:43:44 -04006672 struct extent_buffer *sb;
Chris Mason0b86a832008-03-24 15:01:56 -04006673 struct btrfs_disk_key *disk_key;
Chris Mason0b86a832008-03-24 15:01:56 -04006674 struct btrfs_chunk *chunk;
David Sterba1ffb22c2014-10-31 19:02:42 +01006675 u8 *array_ptr;
6676 unsigned long sb_array_offset;
Chris Mason84eed902008-04-25 09:04:37 -04006677 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006678 u32 num_stripes;
6679 u32 array_size;
6680 u32 len = 0;
David Sterba1ffb22c2014-10-31 19:02:42 +01006681 u32 cur_offset;
Liu Boe06cd3d2016-06-03 12:05:15 -07006682 u64 type;
Chris Mason84eed902008-04-25 09:04:37 -04006683 struct btrfs_key key;
Chris Mason0b86a832008-03-24 15:01:56 -04006684
David Sterbaa83fffb2014-06-15 02:39:54 +02006685 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6686 /*
6687 * This will create extent buffer of nodesize, superblock size is
6688 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6689 * overallocate but we can keep it as-is, only the first page is used.
6690 */
6691 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
Liu Boc871b0f2016-06-06 12:01:23 -07006692 if (IS_ERR(sb))
6693 return PTR_ERR(sb);
David Sterba4db8c522015-12-03 13:06:46 +01006694 set_extent_buffer_uptodate(sb);
Chris Mason85d4e462011-07-26 16:11:19 -04006695 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
David Sterba8a334422011-10-07 18:06:13 +02006696 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04006697 * The sb extent buffer is artificial and just used to read the system array.
David Sterba4db8c522015-12-03 13:06:46 +01006698 * set_extent_buffer_uptodate() call does not properly mark all it's
David Sterba8a334422011-10-07 18:06:13 +02006699 * pages up-to-date when the page is larger: extent does not cover the
6700 * whole page and consequently check_page_uptodate does not find all
6701 * the page's extents up-to-date (the hole beyond sb),
6702 * write_extent_buffer then triggers a WARN_ON.
6703 *
6704 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6705 * but sb spans only this function. Add an explicit SetPageUptodate call
6706 * to silence the warning eg. on PowerPC 64.
6707 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006708 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
Chris Mason727011e2010-08-06 13:21:20 -04006709 SetPageUptodate(sb->pages[0]);
Chris Mason4008c042009-02-12 14:09:45 -05006710
Chris Masona061fc82008-05-07 11:43:44 -04006711 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04006712 array_size = btrfs_super_sys_array_size(super_copy);
6713
David Sterba1ffb22c2014-10-31 19:02:42 +01006714 array_ptr = super_copy->sys_chunk_array;
6715 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6716 cur_offset = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006717
David Sterba1ffb22c2014-10-31 19:02:42 +01006718 while (cur_offset < array_size) {
6719 disk_key = (struct btrfs_disk_key *)array_ptr;
David Sterbae3540ea2014-11-05 15:24:51 +01006720 len = sizeof(*disk_key);
6721 if (cur_offset + len > array_size)
6722 goto out_short_read;
6723
Chris Mason0b86a832008-03-24 15:01:56 -04006724 btrfs_disk_key_to_cpu(&key, disk_key);
6725
David Sterba1ffb22c2014-10-31 19:02:42 +01006726 array_ptr += len;
6727 sb_array_offset += len;
6728 cur_offset += len;
Chris Mason0b86a832008-03-24 15:01:56 -04006729
Chris Mason0d81ba52008-03-24 15:02:07 -04006730 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
David Sterba1ffb22c2014-10-31 19:02:42 +01006731 chunk = (struct btrfs_chunk *)sb_array_offset;
David Sterbae3540ea2014-11-05 15:24:51 +01006732 /*
6733 * At least one btrfs_chunk with one stripe must be
6734 * present, exact stripe count check comes afterwards
6735 */
6736 len = btrfs_chunk_item_size(1);
6737 if (cur_offset + len > array_size)
6738 goto out_short_read;
6739
6740 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
David Sterbaf5cdedd2015-11-30 17:27:06 +01006741 if (!num_stripes) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006742 btrfs_err(fs_info,
6743 "invalid number of stripes %u in sys_array at offset %u",
David Sterbaf5cdedd2015-11-30 17:27:06 +01006744 num_stripes, cur_offset);
6745 ret = -EIO;
6746 break;
6747 }
6748
Liu Boe06cd3d2016-06-03 12:05:15 -07006749 type = btrfs_chunk_type(sb, chunk);
6750 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006751 btrfs_err(fs_info,
Liu Boe06cd3d2016-06-03 12:05:15 -07006752 "invalid chunk type %llu in sys_array at offset %u",
6753 type, cur_offset);
6754 ret = -EIO;
6755 break;
6756 }
6757
David Sterbae3540ea2014-11-05 15:24:51 +01006758 len = btrfs_chunk_item_size(num_stripes);
6759 if (cur_offset + len > array_size)
6760 goto out_short_read;
6761
Chris Mason0d81ba52008-03-24 15:02:07 -04006762 ret = read_one_chunk(root, &key, sb, chunk);
Chris Mason84eed902008-04-25 09:04:37 -04006763 if (ret)
6764 break;
Chris Mason0b86a832008-03-24 15:01:56 -04006765 } else {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006766 btrfs_err(fs_info,
6767 "unexpected item type %u in sys_array at offset %u",
6768 (u32)key.type, cur_offset);
Chris Mason84eed902008-04-25 09:04:37 -04006769 ret = -EIO;
6770 break;
Chris Mason0b86a832008-03-24 15:01:56 -04006771 }
David Sterba1ffb22c2014-10-31 19:02:42 +01006772 array_ptr += len;
6773 sb_array_offset += len;
6774 cur_offset += len;
Chris Mason0b86a832008-03-24 15:01:56 -04006775 }
Liu Bod8651772016-06-03 17:41:42 -07006776 clear_extent_buffer_uptodate(sb);
Liu Bo1c8b5b62016-05-13 17:06:59 -07006777 free_extent_buffer_stale(sb);
Chris Mason84eed902008-04-25 09:04:37 -04006778 return ret;
David Sterbae3540ea2014-11-05 15:24:51 +01006779
6780out_short_read:
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04006781 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
David Sterbae3540ea2014-11-05 15:24:51 +01006782 len, cur_offset);
Liu Bod8651772016-06-03 17:41:42 -07006783 clear_extent_buffer_uptodate(sb);
Liu Bo1c8b5b62016-05-13 17:06:59 -07006784 free_extent_buffer_stale(sb);
David Sterbae3540ea2014-11-05 15:24:51 +01006785 return -EIO;
Chris Mason0b86a832008-03-24 15:01:56 -04006786}
6787
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04006788int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
Chris Mason0b86a832008-03-24 15:01:56 -04006789{
Jeff Mahoney5b4aace2016-06-21 10:40:19 -04006790 struct btrfs_root *root = fs_info->chunk_root;
Chris Mason0b86a832008-03-24 15:01:56 -04006791 struct btrfs_path *path;
6792 struct extent_buffer *leaf;
6793 struct btrfs_key key;
6794 struct btrfs_key found_key;
6795 int ret;
6796 int slot;
Liu Bo99e3ecf2016-06-03 12:05:14 -07006797 u64 total_dev = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006798
Chris Mason0b86a832008-03-24 15:01:56 -04006799 path = btrfs_alloc_path();
6800 if (!path)
6801 return -ENOMEM;
6802
Li Zefanb367e472011-12-07 11:38:24 +08006803 mutex_lock(&uuid_mutex);
6804 lock_chunks(root);
6805
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006806 /*
6807 * Read all device items, and then all the chunk items. All
6808 * device items are found before any chunk item (their object id
6809 * is smaller than the lowest possible object id for a chunk
6810 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
Chris Mason0b86a832008-03-24 15:01:56 -04006811 */
6812 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6813 key.offset = 0;
6814 key.type = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006815 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Zhao Leiab593812010-03-25 12:34:49 +00006816 if (ret < 0)
6817 goto error;
Chris Masond3977122009-01-05 21:25:51 -05006818 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04006819 leaf = path->nodes[0];
6820 slot = path->slots[0];
6821 if (slot >= btrfs_header_nritems(leaf)) {
6822 ret = btrfs_next_leaf(root, path);
6823 if (ret == 0)
6824 continue;
6825 if (ret < 0)
6826 goto error;
6827 break;
6828 }
6829 btrfs_item_key_to_cpu(leaf, &found_key, slot);
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006830 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6831 struct btrfs_dev_item *dev_item;
6832 dev_item = btrfs_item_ptr(leaf, slot,
Chris Mason0b86a832008-03-24 15:01:56 -04006833 struct btrfs_dev_item);
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006834 ret = read_one_dev(root, leaf, dev_item);
6835 if (ret)
6836 goto error;
Liu Bo99e3ecf2016-06-03 12:05:14 -07006837 total_dev++;
Chris Mason0b86a832008-03-24 15:01:56 -04006838 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6839 struct btrfs_chunk *chunk;
6840 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6841 ret = read_one_chunk(root, &found_key, leaf, chunk);
Yan Zheng2b820322008-11-17 21:11:30 -05006842 if (ret)
6843 goto error;
Chris Mason0b86a832008-03-24 15:01:56 -04006844 }
6845 path->slots[0]++;
6846 }
Liu Bo99e3ecf2016-06-03 12:05:14 -07006847
6848 /*
6849 * After loading chunk tree, we've got all device information,
6850 * do another round of validation checks.
6851 */
6852 if (total_dev != root->fs_info->fs_devices->total_devices) {
6853 btrfs_err(root->fs_info,
6854 "super_num_devices %llu mismatch with num_devices %llu found here",
6855 btrfs_super_num_devices(root->fs_info->super_copy),
6856 total_dev);
6857 ret = -EINVAL;
6858 goto error;
6859 }
6860 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6861 root->fs_info->fs_devices->total_rw_bytes) {
6862 btrfs_err(root->fs_info,
6863 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6864 btrfs_super_total_bytes(root->fs_info->super_copy),
6865 root->fs_info->fs_devices->total_rw_bytes);
6866 ret = -EINVAL;
6867 goto error;
6868 }
Chris Mason0b86a832008-03-24 15:01:56 -04006869 ret = 0;
6870error:
Li Zefanb367e472011-12-07 11:38:24 +08006871 unlock_chunks(root);
6872 mutex_unlock(&uuid_mutex);
6873
Yan Zheng2b820322008-11-17 21:11:30 -05006874 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04006875 return ret;
6876}
Stefan Behrens442a4f62012-05-25 16:06:08 +02006877
Miao Xiecb517ea2013-05-15 07:48:19 +00006878void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6879{
6880 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6881 struct btrfs_device *device;
6882
Liu Bo29cc83f2014-05-11 23:14:59 +08006883 while (fs_devices) {
6884 mutex_lock(&fs_devices->device_list_mutex);
6885 list_for_each_entry(device, &fs_devices->devices, dev_list)
6886 device->dev_root = fs_info->dev_root;
6887 mutex_unlock(&fs_devices->device_list_mutex);
6888
6889 fs_devices = fs_devices->seed;
6890 }
Miao Xiecb517ea2013-05-15 07:48:19 +00006891}
6892
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006893static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6894{
6895 int i;
6896
6897 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6898 btrfs_dev_stat_reset(dev, i);
6899}
6900
6901int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6902{
6903 struct btrfs_key key;
6904 struct btrfs_key found_key;
6905 struct btrfs_root *dev_root = fs_info->dev_root;
6906 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6907 struct extent_buffer *eb;
6908 int slot;
6909 int ret = 0;
6910 struct btrfs_device *device;
6911 struct btrfs_path *path = NULL;
6912 int i;
6913
6914 path = btrfs_alloc_path();
6915 if (!path) {
6916 ret = -ENOMEM;
6917 goto out;
6918 }
6919
6920 mutex_lock(&fs_devices->device_list_mutex);
6921 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6922 int item_size;
6923 struct btrfs_dev_stats_item *ptr;
6924
David Sterba242e2952016-01-25 17:51:31 +01006925 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6926 key.type = BTRFS_PERSISTENT_ITEM_KEY;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006927 key.offset = device->devid;
6928 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6929 if (ret) {
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006930 __btrfs_reset_dev_stats(device);
6931 device->dev_stats_valid = 1;
6932 btrfs_release_path(path);
6933 continue;
6934 }
6935 slot = path->slots[0];
6936 eb = path->nodes[0];
6937 btrfs_item_key_to_cpu(eb, &found_key, slot);
6938 item_size = btrfs_item_size_nr(eb, slot);
6939
6940 ptr = btrfs_item_ptr(eb, slot,
6941 struct btrfs_dev_stats_item);
6942
6943 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6944 if (item_size >= (1 + i) * sizeof(__le64))
6945 btrfs_dev_stat_set(device, i,
6946 btrfs_dev_stats_value(eb, ptr, i));
6947 else
6948 btrfs_dev_stat_reset(device, i);
6949 }
6950
6951 device->dev_stats_valid = 1;
6952 btrfs_dev_stat_print_on_load(device);
6953 btrfs_release_path(path);
6954 }
6955 mutex_unlock(&fs_devices->device_list_mutex);
6956
6957out:
6958 btrfs_free_path(path);
6959 return ret < 0 ? ret : 0;
6960}
6961
6962static int update_dev_stat_item(struct btrfs_trans_handle *trans,
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04006963 struct btrfs_fs_info *fs_info,
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006964 struct btrfs_device *device)
6965{
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04006966 struct btrfs_root *dev_root = fs_info->dev_root;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006967 struct btrfs_path *path;
6968 struct btrfs_key key;
6969 struct extent_buffer *eb;
6970 struct btrfs_dev_stats_item *ptr;
6971 int ret;
6972 int i;
6973
David Sterba242e2952016-01-25 17:51:31 +01006974 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6975 key.type = BTRFS_PERSISTENT_ITEM_KEY;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006976 key.offset = device->devid;
6977
6978 path = btrfs_alloc_path();
6979 BUG_ON(!path);
6980 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6981 if (ret < 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02006982 btrfs_warn_in_rcu(dev_root->fs_info,
6983 "error %d while searching for dev_stats item for device %s",
Josef Bacik606686e2012-06-04 14:03:51 -04006984 ret, rcu_str_deref(device->name));
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006985 goto out;
6986 }
6987
6988 if (ret == 0 &&
6989 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6990 /* need to delete old one and insert a new one */
6991 ret = btrfs_del_item(trans, dev_root, path);
6992 if (ret != 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02006993 btrfs_warn_in_rcu(dev_root->fs_info,
6994 "delete too small dev_stats item for device %s failed %d",
Josef Bacik606686e2012-06-04 14:03:51 -04006995 rcu_str_deref(device->name), ret);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006996 goto out;
6997 }
6998 ret = 1;
6999 }
7000
7001 if (ret == 1) {
7002 /* need to insert a new item */
7003 btrfs_release_path(path);
7004 ret = btrfs_insert_empty_item(trans, dev_root, path,
7005 &key, sizeof(*ptr));
7006 if (ret < 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02007007 btrfs_warn_in_rcu(dev_root->fs_info,
7008 "insert dev_stats item for device %s failed %d",
7009 rcu_str_deref(device->name), ret);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007010 goto out;
7011 }
7012 }
7013
7014 eb = path->nodes[0];
7015 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7016 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7017 btrfs_set_dev_stats_value(eb, ptr, i,
7018 btrfs_dev_stat_read(device, i));
7019 btrfs_mark_buffer_dirty(eb);
7020
7021out:
7022 btrfs_free_path(path);
7023 return ret;
7024}
7025
7026/*
7027 * called from commit_transaction. Writes all changed device stats to disk.
7028 */
7029int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7030 struct btrfs_fs_info *fs_info)
7031{
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007032 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7033 struct btrfs_device *device;
Miao Xieaddc3fa2014-07-24 11:37:11 +08007034 int stats_cnt;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007035 int ret = 0;
7036
7037 mutex_lock(&fs_devices->device_list_mutex);
7038 list_for_each_entry(device, &fs_devices->devices, dev_list) {
Miao Xieaddc3fa2014-07-24 11:37:11 +08007039 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007040 continue;
7041
Miao Xieaddc3fa2014-07-24 11:37:11 +08007042 stats_cnt = atomic_read(&device->dev_stats_ccnt);
Jeff Mahoney6bccf3a2016-06-21 21:16:51 -04007043 ret = update_dev_stat_item(trans, fs_info, device);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007044 if (!ret)
Miao Xieaddc3fa2014-07-24 11:37:11 +08007045 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007046 }
7047 mutex_unlock(&fs_devices->device_list_mutex);
7048
7049 return ret;
7050}
7051
Stefan Behrens442a4f62012-05-25 16:06:08 +02007052void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7053{
7054 btrfs_dev_stat_inc(dev, index);
7055 btrfs_dev_stat_print_on_error(dev);
7056}
7057
Eric Sandeen48a3b632013-04-25 20:41:01 +00007058static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
Stefan Behrens442a4f62012-05-25 16:06:08 +02007059{
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007060 if (!dev->dev_stats_valid)
7061 return;
David Sterbab14af3b2015-10-08 10:43:10 +02007062 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7063 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
Josef Bacik606686e2012-06-04 14:03:51 -04007064 rcu_str_deref(dev->name),
Stefan Behrens442a4f62012-05-25 16:06:08 +02007065 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7066 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7067 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
Frank Holtonefe120a2013-12-20 11:37:06 -05007068 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7069 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
Stefan Behrens442a4f62012-05-25 16:06:08 +02007070}
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007071
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007072static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7073{
Stefan Behrensa98cdb82012-07-17 09:02:11 -06007074 int i;
7075
7076 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7077 if (btrfs_dev_stat_read(dev, i) != 0)
7078 break;
7079 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7080 return; /* all values == 0, suppress message */
7081
David Sterbaecaeb142015-10-08 09:01:03 +02007082 btrfs_info_in_rcu(dev->dev_root->fs_info,
7083 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
Josef Bacik606686e2012-06-04 14:03:51 -04007084 rcu_str_deref(dev->name),
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007085 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7086 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7087 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7088 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7089 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7090}
7091
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007092int btrfs_get_dev_stats(struct btrfs_root *root,
David Sterbab27f7c02012-06-22 06:30:39 -06007093 struct btrfs_ioctl_get_dev_stats *stats)
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007094{
7095 struct btrfs_device *dev;
7096 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7097 int i;
7098
7099 mutex_lock(&fs_devices->device_list_mutex);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01007100 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007101 mutex_unlock(&fs_devices->device_list_mutex);
7102
7103 if (!dev) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04007104 btrfs_warn(root->fs_info,
7105 "get dev_stats failed, device not found");
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007106 return -ENODEV;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007107 } else if (!dev->dev_stats_valid) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04007108 btrfs_warn(root->fs_info,
7109 "get dev_stats failed, not yet valid");
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007110 return -ENODEV;
David Sterbab27f7c02012-06-22 06:30:39 -06007111 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007112 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7113 if (stats->nr_items > i)
7114 stats->values[i] =
7115 btrfs_dev_stat_read_and_reset(dev, i);
7116 else
7117 btrfs_dev_stat_reset(dev, i);
7118 }
7119 } else {
7120 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7121 if (stats->nr_items > i)
7122 stats->values[i] = btrfs_dev_stat_read(dev, i);
7123 }
7124 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7125 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7126 return 0;
7127}
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007128
Anand Jain12b1c262015-08-14 18:32:59 +08007129void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007130{
7131 struct buffer_head *bh;
7132 struct btrfs_super_block *disk_super;
Anand Jain12b1c262015-08-14 18:32:59 +08007133 int copy_num;
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007134
Anand Jain12b1c262015-08-14 18:32:59 +08007135 if (!bdev)
7136 return;
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007137
Anand Jain12b1c262015-08-14 18:32:59 +08007138 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7139 copy_num++) {
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007140
Anand Jain12b1c262015-08-14 18:32:59 +08007141 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7142 continue;
7143
7144 disk_super = (struct btrfs_super_block *)bh->b_data;
7145
7146 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7147 set_buffer_dirty(bh);
7148 sync_dirty_buffer(bh);
7149 brelse(bh);
7150 }
7151
7152 /* Notify udev that device has changed */
7153 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7154
7155 /* Update ctime/mtime for device path for libblkid */
7156 update_dev_time(device_path);
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007157}
Miao Xie935e5cc2014-09-03 21:35:33 +08007158
7159/*
7160 * Update the size of all devices, which is used for writing out the
7161 * super blocks.
7162 */
7163void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7164{
7165 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7166 struct btrfs_device *curr, *next;
7167
7168 if (list_empty(&fs_devices->resized_devices))
7169 return;
7170
7171 mutex_lock(&fs_devices->device_list_mutex);
7172 lock_chunks(fs_info->dev_root);
7173 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7174 resized_list) {
7175 list_del_init(&curr->resized_list);
7176 curr->commit_total_bytes = curr->disk_total_bytes;
7177 }
7178 unlock_chunks(fs_info->dev_root);
7179 mutex_unlock(&fs_devices->device_list_mutex);
7180}
Miao Xiece7213c2014-09-03 21:35:34 +08007181
7182/* Must be invoked during the transaction commit */
7183void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7184 struct btrfs_transaction *transaction)
7185{
7186 struct extent_map *em;
7187 struct map_lookup *map;
7188 struct btrfs_device *dev;
7189 int i;
7190
7191 if (list_empty(&transaction->pending_chunks))
7192 return;
7193
7194 /* In order to kick the device replace finish process */
7195 lock_chunks(root);
7196 list_for_each_entry(em, &transaction->pending_chunks, list) {
Jeff Mahoney95617d62015-06-03 10:55:48 -04007197 map = em->map_lookup;
Miao Xiece7213c2014-09-03 21:35:34 +08007198
7199 for (i = 0; i < map->num_stripes; i++) {
7200 dev = map->stripes[i].dev;
7201 dev->commit_bytes_used = dev->bytes_used;
7202 }
7203 }
7204 unlock_chunks(root);
7205}
Anand Jain5a13f432015-03-10 06:38:31 +08007206
7207void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7208{
7209 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7210 while (fs_devices) {
7211 fs_devices->fs_info = fs_info;
7212 fs_devices = fs_devices->seed;
7213 }
7214}
7215
7216void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7217{
7218 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7219 while (fs_devices) {
7220 fs_devices->fs_info = NULL;
7221 fs_devices = fs_devices->seed;
7222 }
7223}