blob: 8b8dd82da3c4486f434606904a07a6e157963130 [file] [log] [blame]
Philipp Reisnerb411b362009-09-25 16:07:19 -07001/*
2 drbd.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
12
13 drbd is free software; you can redistribute it and/or modify
14 it under the terms of the GNU General Public License as published by
15 the Free Software Foundation; either version 2, or (at your option)
16 any later version.
17
18 drbd is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
22
23 You should have received a copy of the GNU General Public License
24 along with drbd; see the file COPYING. If not, write to
25 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26
27 */
28
Lars Ellenbergf88c5d92014-03-27 14:10:55 +010029#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
Philipp Reisnerb411b362009-09-25 16:07:19 -070031#include <linux/module.h>
Lars Ellenberge5f891b2013-11-22 12:32:01 +010032#include <linux/jiffies.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070033#include <linux/drbd.h>
Fabian Frederick7e5fec32016-06-14 00:26:35 +020034#include <linux/uaccess.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070035#include <asm/types.h>
36#include <net/sock.h>
37#include <linux/ctype.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020038#include <linux/mutex.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070039#include <linux/fs.h>
40#include <linux/file.h>
41#include <linux/proc_fs.h>
42#include <linux/init.h>
43#include <linux/mm.h>
44#include <linux/memcontrol.h>
45#include <linux/mm_inline.h>
46#include <linux/slab.h>
47#include <linux/random.h>
48#include <linux/reboot.h>
49#include <linux/notifier.h>
50#include <linux/kthread.h>
Lars Ellenberg113fef92013-03-22 18:14:40 -060051#include <linux/workqueue.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070052#define __KERNEL_SYSCALLS__
53#include <linux/unistd.h>
54#include <linux/vmalloc.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010055#include <linux/sched/signal.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070056
57#include <linux/drbd_limits.h>
58#include "drbd_int.h"
Andreas Gruenbachera3603a62011-05-30 11:47:37 +020059#include "drbd_protocol.h"
Philipp Reisnerb411b362009-09-25 16:07:19 -070060#include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
Philipp Reisnerb411b362009-09-25 16:07:19 -070061#include "drbd_vli.h"
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +020062#include "drbd_debugfs.h"
Philipp Reisnerb411b362009-09-25 16:07:19 -070063
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020064static DEFINE_MUTEX(drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -070065static int drbd_open(struct block_device *bdev, fmode_t mode);
Al Virodb2a1442013-05-05 21:52:57 -040066static void drbd_release(struct gendisk *gd, fmode_t mode);
Philipp Reisnerb411b362009-09-25 16:07:19 -070067static void md_sync_timer_fn(unsigned long data);
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +010068static int w_bitmap_io(struct drbd_work *w, int unused);
Philipp Reisnerb411b362009-09-25 16:07:19 -070069
Philipp Reisnerb411b362009-09-25 16:07:19 -070070MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
71 "Lars Ellenberg <lars@linbit.com>");
72MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
73MODULE_VERSION(REL_VERSION);
74MODULE_LICENSE("GPL");
Philipp Reisner81a5d602011-02-22 19:53:16 -050075MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
Philipp Reisner2b8a90b2011-01-10 11:15:17 +010076 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
Philipp Reisnerb411b362009-09-25 16:07:19 -070077MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
78
79#include <linux/moduleparam.h>
80/* allow_open_on_secondary */
81MODULE_PARM_DESC(allow_oos, "DONT USE!");
82/* thanks to these macros, if compiled into the kernel (not-module),
83 * this becomes the boot parameter drbd.minor_count */
84module_param(minor_count, uint, 0444);
85module_param(disable_sendpage, bool, 0644);
86module_param(allow_oos, bool, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -070087module_param(proc_details, int, 0644);
88
89#ifdef CONFIG_DRBD_FAULT_INJECTION
90int enable_faults;
91int fault_rate;
92static int fault_count;
93int fault_devs;
94/* bitmap of enabled faults */
95module_param(enable_faults, int, 0664);
96/* fault rate % value - applies to all enabled faults */
97module_param(fault_rate, int, 0664);
98/* count of faults inserted */
99module_param(fault_count, int, 0664);
100/* bitmap of devices to insert faults on */
101module_param(fault_devs, int, 0644);
102#endif
103
104/* module parameter, defined */
Philipp Reisner2b8a90b2011-01-10 11:15:17 +0100105unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
Rusty Russell90ab5ee2012-01-13 09:32:20 +1030106bool disable_sendpage;
107bool allow_oos;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700108int proc_details; /* Detail level in proc drbd*/
109
110/* Module parameter for setting the user mode helper program
111 * to run. Default is /sbin/drbdadm */
112char usermode_helper[80] = "/sbin/drbdadm";
113
114module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
115
116/* in 2.6.x, our device mapping and config info contains our virtual gendisks
117 * as member "struct gendisk *vdisk;"
118 */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +0200119struct idr drbd_devices;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +0200120struct list_head drbd_resources;
Andreas Gruenbacher28bc3b82014-08-14 18:33:30 +0200121struct mutex resources_mutex;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700122
123struct kmem_cache *drbd_request_cache;
Andreas Gruenbacher6c852be2011-02-04 15:38:52 +0100124struct kmem_cache *drbd_ee_cache; /* peer requests */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700125struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
126struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
127mempool_t *drbd_request_mempool;
128mempool_t *drbd_ee_mempool;
Lars Ellenberg42818082011-02-23 12:39:46 +0100129mempool_t *drbd_md_io_page_pool;
Lars Ellenberg9476f392011-02-23 17:02:01 +0100130struct bio_set *drbd_md_io_bio_set;
NeilBrown8cb0def2017-06-18 14:38:58 +1000131struct bio_set *drbd_io_bio_set;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700132
133/* I do not use a standard mempool, because:
134 1) I want to hand out the pre-allocated objects first.
135 2) I want to be able to interrupt sleeping allocation with a signal.
136 Note: This is a single linked list, the next pointer is the private
137 member of struct page.
138 */
139struct page *drbd_pp_pool;
140spinlock_t drbd_pp_lock;
141int drbd_pp_vacant;
142wait_queue_head_t drbd_pp_wait;
143
144DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
145
Emese Revfy7d4e9d02009-12-14 00:59:30 +0100146static const struct block_device_operations drbd_ops = {
Philipp Reisnerb411b362009-09-25 16:07:19 -0700147 .owner = THIS_MODULE,
148 .open = drbd_open,
149 .release = drbd_release,
150};
151
Lars Ellenberg9476f392011-02-23 17:02:01 +0100152struct bio *bio_alloc_drbd(gfp_t gfp_mask)
153{
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100154 struct bio *bio;
155
Lars Ellenberg9476f392011-02-23 17:02:01 +0100156 if (!drbd_md_io_bio_set)
157 return bio_alloc(gfp_mask, 1);
158
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100159 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
160 if (!bio)
161 return NULL;
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100162 return bio;
Lars Ellenberg9476f392011-02-23 17:02:01 +0100163}
164
Philipp Reisnerb411b362009-09-25 16:07:19 -0700165#ifdef __CHECKER__
166/* When checking with sparse, and this is an inline function, sparse will
167 give tons of false positives. When this is a real functions sparse works.
168 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200169int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700170{
171 int io_allowed;
172
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200173 atomic_inc(&device->local_cnt);
174 io_allowed = (device->state.disk >= mins);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700175 if (!io_allowed) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200176 if (atomic_dec_and_test(&device->local_cnt))
177 wake_up(&device->misc_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700178 }
179 return io_allowed;
180}
181
182#endif
183
184/**
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100185 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200186 * @connection: DRBD connection.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700187 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
188 * @set_size: Expected number of requests before that barrier.
189 *
190 * In case the passed barrier_nr or set_size does not match the oldest
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100191 * epoch of not yet barrier-acked requests, this function will cause a
192 * termination of the connection.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700193 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200194void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
Philipp Reisner2f5cdd02011-02-21 14:29:27 +0100195 unsigned int set_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700196{
Philipp Reisnerb411b362009-09-25 16:07:19 -0700197 struct drbd_request *r;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100198 struct drbd_request *req = NULL;
199 int expect_epoch = 0;
200 int expect_size = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700201
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200202 spin_lock_irq(&connection->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700203
Philipp Reisner98683652012-11-09 14:18:43 +0100204 /* find oldest not yet barrier-acked write request,
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100205 * count writes in its epoch. */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200206 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
Lars Ellenberga0d856d2012-01-24 17:19:42 +0100207 const unsigned s = r->rq_state;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100208 if (!req) {
209 if (!(s & RQ_WRITE))
210 continue;
211 if (!(s & RQ_NET_MASK))
212 continue;
213 if (s & RQ_NET_DONE)
214 continue;
215 req = r;
216 expect_epoch = req->epoch;
217 expect_size ++;
218 } else {
219 if (r->epoch != expect_epoch)
220 break;
221 if (!(s & RQ_WRITE))
222 continue;
223 /* if (s & RQ_DONE): not expected */
224 /* if (!(s & RQ_NET_MASK)): not expected */
225 expect_size++;
226 }
227 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700228
229 /* first some paranoia code */
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100230 if (req == NULL) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200231 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
Philipp Reisner2f5cdd02011-02-21 14:29:27 +0100232 barrier_nr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700233 goto bail;
234 }
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100235 if (expect_epoch != barrier_nr) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200236 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100237 barrier_nr, expect_epoch);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700238 goto bail;
239 }
240
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100241 if (expect_size != set_size) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200242 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100243 barrier_nr, set_size, expect_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700244 goto bail;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700245 }
246
Philipp Reisner98683652012-11-09 14:18:43 +0100247 /* Clean up list of requests processed during current epoch. */
248 /* this extra list walk restart is paranoia,
249 * to catch requests being barrier-acked "unexpectedly".
250 * It usually should find the same req again, or some READ preceding it. */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200251 list_for_each_entry(req, &connection->transfer_log, tl_requests)
Philipp Reisner98683652012-11-09 14:18:43 +0100252 if (req->epoch == expect_epoch)
253 break;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200254 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100255 if (req->epoch != expect_epoch)
256 break;
257 _req_mod(req, BARRIER_ACKED);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700258 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200259 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700260
261 return;
262
263bail:
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200264 spin_unlock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200265 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700266}
267
Philipp Reisner617049a2010-12-22 12:48:31 +0100268
Philipp Reisner11b58e72010-05-12 17:08:26 +0200269/**
270 * _tl_restart() - Walks the transfer log, and applies an action to all requests
Lars Ellenberge5f891b2013-11-22 12:32:01 +0100271 * @connection: DRBD connection to operate on.
Philipp Reisner11b58e72010-05-12 17:08:26 +0200272 * @what: The action/event to perform with all request objects
273 *
Andreas Gruenbacher8554df12011-01-25 15:37:43 +0100274 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
275 * RESTART_FROZEN_DISK_IO.
Philipp Reisner11b58e72010-05-12 17:08:26 +0200276 */
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100277/* must hold resource->req_lock */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200278void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
Philipp Reisner11b58e72010-05-12 17:08:26 +0200279{
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100280 struct drbd_request *req, *r;
Philipp Reisner11b58e72010-05-12 17:08:26 +0200281
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200282 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
Philipp Reisner509fc012012-07-31 11:22:58 +0200283 _req_mod(req, what);
Philipp Reisner11b58e72010-05-12 17:08:26 +0200284}
285
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200286void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100287{
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200288 spin_lock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200289 _tl_restart(connection, what);
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200290 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnercdfda632011-07-05 15:38:59 +0200291}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700292
293/**
294 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200295 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700296 *
297 * This is called after the connection to the peer was lost. The storage covered
298 * by the requests on the transfer gets marked as our of sync. Called from the
299 * receiver thread and the worker thread.
300 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200301void tl_clear(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700302{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200303 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700304}
305
306/**
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200307 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
308 * @device: DRBD device.
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200309 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200310void tl_abort_disk_io(struct drbd_device *device)
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200311{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +0200312 struct drbd_connection *connection = first_peer_device(device)->connection;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100313 struct drbd_request *req, *r;
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200314
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200315 spin_lock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200316 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200317 if (!(req->rq_state & RQ_LOCAL_PENDING))
318 continue;
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +0200319 if (req->device != device)
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100320 continue;
321 _req_mod(req, ABORT_DISK_IO);
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200322 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200323 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200324}
325
Philipp Reisnerb411b362009-09-25 16:07:19 -0700326static int drbd_thread_setup(void *arg)
327{
328 struct drbd_thread *thi = (struct drbd_thread *) arg;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200329 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700330 unsigned long flags;
331 int retval;
332
Philipp Reisnerf1b3a6e2011-02-08 15:35:58 +0100333 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +0200334 thi->name[0],
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200335 resource->name);
Philipp Reisnerf1b3a6e2011-02-08 15:35:58 +0100336
Philipp Reisnerb411b362009-09-25 16:07:19 -0700337restart:
338 retval = thi->function(thi);
339
340 spin_lock_irqsave(&thi->t_lock, flags);
341
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100342 /* if the receiver has been "EXITING", the last thing it did
Philipp Reisnerb411b362009-09-25 16:07:19 -0700343 * was set the conn state to "StandAlone",
344 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
345 * and receiver thread will be "started".
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100346 * drbd_thread_start needs to set "RESTARTING" in that case.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700347 * t_state check and assignment needs to be within the same spinlock,
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100348 * so either thread_start sees EXITING, and can remap to RESTARTING,
349 * or thread_start see NONE, and can proceed as normal.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700350 */
351
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100352 if (thi->t_state == RESTARTING) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200353 drbd_info(resource, "Restarting %s thread\n", thi->name);
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100354 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700355 spin_unlock_irqrestore(&thi->t_lock, flags);
356 goto restart;
357 }
358
359 thi->task = NULL;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100360 thi->t_state = NONE;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700361 smp_mb();
Lars Ellenberg992d6e92011-05-02 11:47:18 +0200362 complete_all(&thi->stop);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700363 spin_unlock_irqrestore(&thi->t_lock, flags);
364
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200365 drbd_info(resource, "Terminating %s\n", current->comm);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700366
367 /* Release mod reference taken when thread was started */
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +0200368
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200369 if (thi->connection)
370 kref_put(&thi->connection->kref, drbd_destroy_connection);
371 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700372 module_put(THIS_MODULE);
373 return retval;
374}
375
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200376static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
Andreas Gruenbacherc60b0252011-08-10 15:05:02 +0200377 int (*func) (struct drbd_thread *), const char *name)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700378{
379 spin_lock_init(&thi->t_lock);
380 thi->task = NULL;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100381 thi->t_state = NONE;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700382 thi->function = func;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200383 thi->resource = resource;
384 thi->connection = NULL;
Andreas Gruenbacherc60b0252011-08-10 15:05:02 +0200385 thi->name = name;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700386}
387
388int drbd_thread_start(struct drbd_thread *thi)
389{
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200390 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700391 struct task_struct *nt;
392 unsigned long flags;
393
Philipp Reisnerb411b362009-09-25 16:07:19 -0700394 /* is used from state engine doing drbd_thread_stop_nowait,
395 * while holding the req lock irqsave */
396 spin_lock_irqsave(&thi->t_lock, flags);
397
398 switch (thi->t_state) {
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100399 case NONE:
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200400 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100401 thi->name, current->comm, current->pid);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700402
403 /* Get ref on module for thread - this is released when thread exits */
404 if (!try_module_get(THIS_MODULE)) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200405 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -0700406 spin_unlock_irqrestore(&thi->t_lock, flags);
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100407 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700408 }
409
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200410 kref_get(&resource->kref);
411 if (thi->connection)
412 kref_get(&thi->connection->kref);
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +0200413
Philipp Reisnerb411b362009-09-25 16:07:19 -0700414 init_completion(&thi->stop);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700415 thi->reset_cpu_mask = 1;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100416 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700417 spin_unlock_irqrestore(&thi->t_lock, flags);
418 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
419
420 nt = kthread_create(drbd_thread_setup, (void *) thi,
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200421 "drbd_%c_%s", thi->name[0], thi->resource->name);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700422
423 if (IS_ERR(nt)) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200424 drbd_err(resource, "Couldn't start thread\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -0700425
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200426 if (thi->connection)
427 kref_put(&thi->connection->kref, drbd_destroy_connection);
428 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700429 module_put(THIS_MODULE);
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100430 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700431 }
432 spin_lock_irqsave(&thi->t_lock, flags);
433 thi->task = nt;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100434 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700435 spin_unlock_irqrestore(&thi->t_lock, flags);
436 wake_up_process(nt);
437 break;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100438 case EXITING:
439 thi->t_state = RESTARTING;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200440 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100441 thi->name, current->comm, current->pid);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700442 /* fall through */
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100443 case RUNNING:
444 case RESTARTING:
Philipp Reisnerb411b362009-09-25 16:07:19 -0700445 default:
446 spin_unlock_irqrestore(&thi->t_lock, flags);
447 break;
448 }
449
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100450 return true;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700451}
452
453
454void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
455{
456 unsigned long flags;
457
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100458 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700459
460 /* may be called from state engine, holding the req lock irqsave */
461 spin_lock_irqsave(&thi->t_lock, flags);
462
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100463 if (thi->t_state == NONE) {
Philipp Reisnerb411b362009-09-25 16:07:19 -0700464 spin_unlock_irqrestore(&thi->t_lock, flags);
465 if (restart)
466 drbd_thread_start(thi);
467 return;
468 }
469
470 if (thi->t_state != ns) {
471 if (thi->task == NULL) {
472 spin_unlock_irqrestore(&thi->t_lock, flags);
473 return;
474 }
475
476 thi->t_state = ns;
477 smp_mb();
478 init_completion(&thi->stop);
479 if (thi->task != current)
480 force_sig(DRBD_SIGKILL, thi->task);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700481 }
482
483 spin_unlock_irqrestore(&thi->t_lock, flags);
484
485 if (wait)
486 wait_for_completion(&thi->stop);
487}
488
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200489int conn_lowest_minor(struct drbd_connection *connection)
Philipp Reisner80822282011-02-08 12:46:30 +0100490{
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200491 struct drbd_peer_device *peer_device;
492 int vnr = 0, minor = -1;
Philipp Reisner774b3052011-02-22 02:07:03 -0500493
Philipp Reisner695d08f2011-04-11 22:53:32 -0700494 rcu_read_lock();
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200495 peer_device = idr_get_next(&connection->peer_devices, &vnr);
496 if (peer_device)
497 minor = device_to_minor(peer_device->device);
Philipp Reisner695d08f2011-04-11 22:53:32 -0700498 rcu_read_unlock();
499
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200500 return minor;
Philipp Reisner80822282011-02-08 12:46:30 +0100501}
Philipp Reisner774b3052011-02-22 02:07:03 -0500502
Philipp Reisnerb411b362009-09-25 16:07:19 -0700503#ifdef CONFIG_SMP
504/**
505 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
Philipp Reisnerb411b362009-09-25 16:07:19 -0700506 *
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200507 * Forces all threads of a resource onto the same CPU. This is beneficial for
Philipp Reisnerb411b362009-09-25 16:07:19 -0700508 * DRBD's performance. May be overwritten by user's configuration.
509 */
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200510static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700511{
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200512 unsigned int *resources_per_cpu, min_index = ~0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700513
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200514 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
515 if (resources_per_cpu) {
516 struct drbd_resource *resource;
517 unsigned int cpu, min = ~0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700518
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200519 rcu_read_lock();
520 for_each_resource_rcu(resource, &drbd_resources) {
521 for_each_cpu(cpu, resource->cpu_mask)
522 resources_per_cpu[cpu]++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700523 }
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200524 rcu_read_unlock();
525 for_each_online_cpu(cpu) {
526 if (resources_per_cpu[cpu] < min) {
527 min = resources_per_cpu[cpu];
528 min_index = cpu;
529 }
530 }
531 kfree(resources_per_cpu);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700532 }
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200533 if (min_index == ~0) {
534 cpumask_setall(*cpu_mask);
535 return;
536 }
537 cpumask_set_cpu(min_index, *cpu_mask);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700538}
539
540/**
541 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200542 * @device: DRBD device.
Philipp Reisnerbc31fe32011-02-07 11:14:38 +0100543 * @thi: drbd_thread object
Philipp Reisnerb411b362009-09-25 16:07:19 -0700544 *
545 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
546 * prematurely.
547 */
Philipp Reisner80822282011-02-08 12:46:30 +0100548void drbd_thread_current_set_cpu(struct drbd_thread *thi)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700549{
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200550 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700551 struct task_struct *p = current;
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100552
Philipp Reisnerb411b362009-09-25 16:07:19 -0700553 if (!thi->reset_cpu_mask)
554 return;
555 thi->reset_cpu_mask = 0;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200556 set_cpus_allowed_ptr(p, resource->cpu_mask);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700557}
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200558#else
559#define drbd_calc_cpu_mask(A) ({})
Philipp Reisnerb411b362009-09-25 16:07:19 -0700560#endif
561
Andreas Gruenbacher52b061a2011-03-30 11:38:49 +0200562/**
563 * drbd_header_size - size of a packet header
564 *
565 * The header size is a multiple of 8, so any payload following the header is
566 * word aligned on 64-bit architectures. (The bitmap send and receive code
567 * relies on this.)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700568 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200569unsigned int drbd_header_size(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700570{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200571 if (connection->agreed_pro_version >= 100) {
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200572 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
573 return sizeof(struct p_header100);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700574 } else {
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200575 BUILD_BUG_ON(sizeof(struct p_header80) !=
576 sizeof(struct p_header95));
577 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
578 return sizeof(struct p_header80);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700579 }
Andreas Gruenbacher52b061a2011-03-30 11:38:49 +0200580}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700581
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200582static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100583{
584 h->magic = cpu_to_be32(DRBD_MAGIC);
585 h->command = cpu_to_be16(cmd);
586 h->length = cpu_to_be16(size);
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200587 return sizeof(struct p_header80);
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100588}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700589
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200590static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100591{
592 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
593 h->command = cpu_to_be16(cmd);
Andreas Gruenbacherb55d84b2011-03-22 13:17:47 +0100594 h->length = cpu_to_be32(size);
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200595 return sizeof(struct p_header95);
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100596}
597
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200598static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
599 int size, int vnr)
Philipp Reisnerd38e7872011-02-07 15:32:04 +0100600{
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200601 h->magic = cpu_to_be32(DRBD_MAGIC_100);
602 h->volume = cpu_to_be16(vnr);
603 h->command = cpu_to_be16(cmd);
604 h->length = cpu_to_be32(size);
605 h->pad = 0;
606 return sizeof(struct p_header100);
607}
608
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200609static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200610 void *buffer, enum drbd_packet cmd, int size)
611{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200612 if (connection->agreed_pro_version >= 100)
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200613 return prepare_header100(buffer, cmd, size, vnr);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200614 else if (connection->agreed_pro_version >= 95 &&
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200615 size > DRBD_MAX_SIZE_H80_PACKET)
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200616 return prepare_header95(buffer, cmd, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700617 else
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200618 return prepare_header80(buffer, cmd, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700619}
620
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200621static void *__conn_prepare_command(struct drbd_connection *connection,
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200622 struct drbd_socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700623{
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200624 if (!sock->socket)
625 return NULL;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200626 return sock->sbuf + drbd_header_size(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700627}
628
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200629void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700630{
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200631 void *p;
632
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200633 mutex_lock(&sock->mutex);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200634 p = __conn_prepare_command(connection, sock);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200635 if (!p)
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200636 mutex_unlock(&sock->mutex);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200637
638 return p;
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200639}
640
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200641void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200642{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200643 return conn_prepare_command(peer_device->connection, sock);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200644}
645
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200646static int __send_command(struct drbd_connection *connection, int vnr,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200647 struct drbd_socket *sock, enum drbd_packet cmd,
648 unsigned int header_size, void *data,
649 unsigned int size)
650{
651 int msg_flags;
652 int err;
653
654 /*
655 * Called with @data == NULL and the size of the data blocks in @size
656 * for commands that send data blocks. For those commands, omit the
657 * MSG_MORE flag: this will increase the likelihood that data blocks
658 * which are page aligned on the sender will end up page aligned on the
659 * receiver.
660 */
661 msg_flags = data ? MSG_MORE : 0;
662
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200663 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200664 header_size + size);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200665 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200666 msg_flags);
667 if (data && !err)
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200668 err = drbd_send_all(connection, sock->socket, data, size, 0);
Lars Ellenberg123ff122014-02-05 06:13:53 +0100669 /* DRBD protocol "pings" are latency critical.
670 * This is supposed to trigger tcp_push_pending_frames() */
671 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
672 drbd_tcp_nodelay(sock->socket);
673
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200674 return err;
675}
676
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200677static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200678 enum drbd_packet cmd, unsigned int header_size,
679 void *data, unsigned int size)
680{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200681 return __send_command(connection, 0, sock, cmd, header_size, data, size);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200682}
683
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200684int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200685 enum drbd_packet cmd, unsigned int header_size,
686 void *data, unsigned int size)
687{
688 int err;
689
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200690 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200691 mutex_unlock(&sock->mutex);
692 return err;
693}
694
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200695int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200696 enum drbd_packet cmd, unsigned int header_size,
697 void *data, unsigned int size)
698{
699 int err;
700
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200701 err = __send_command(peer_device->connection, peer_device->device->vnr,
702 sock, cmd, header_size, data, size);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200703 mutex_unlock(&sock->mutex);
704 return err;
705}
706
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200707int drbd_send_ping(struct drbd_connection *connection)
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100708{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200709 struct drbd_socket *sock;
710
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200711 sock = &connection->meta;
712 if (!conn_prepare_command(connection, sock))
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200713 return -EIO;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200714 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100715}
716
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200717int drbd_send_ping_ack(struct drbd_connection *connection)
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100718{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200719 struct drbd_socket *sock;
720
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200721 sock = &connection->meta;
722 if (!conn_prepare_command(connection, sock))
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200723 return -EIO;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200724 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100725}
726
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200727int drbd_send_sync_param(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700728{
Andreas Gruenbacher7c967152011-03-22 00:49:36 +0100729 struct drbd_socket *sock;
Philipp Reisner8e26f9c2010-07-06 17:25:54 +0200730 struct p_rs_param_95 *p;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200731 int size;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200732 const int apv = peer_device->connection->agreed_pro_version;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200733 enum drbd_packet cmd;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200734 struct net_conf *nc;
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200735 struct disk_conf *dc;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200736
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200737 sock = &peer_device->connection->data;
738 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200739 if (!p)
740 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700741
Philipp Reisner44ed1672011-04-19 17:10:19 +0200742 rcu_read_lock();
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200743 nc = rcu_dereference(peer_device->connection->net_conf);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700744
745 size = apv <= 87 ? sizeof(struct p_rs_param)
746 : apv == 88 ? sizeof(struct p_rs_param)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200747 + strlen(nc->verify_alg) + 1
Philipp Reisner8e26f9c2010-07-06 17:25:54 +0200748 : apv <= 94 ? sizeof(struct p_rs_param_89)
749 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700750
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200751 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700752
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200753 /* initialize verify_alg and csums_alg */
754 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700755
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200756 if (get_ldev(peer_device->device)) {
757 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
Andreas Gruenbacher6394b932011-05-11 14:29:52 +0200758 p->resync_rate = cpu_to_be32(dc->resync_rate);
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200759 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
760 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
761 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
762 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200763 put_ldev(peer_device->device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200764 } else {
Andreas Gruenbacher6394b932011-05-11 14:29:52 +0200765 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200766 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
767 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
768 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
769 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
770 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700771
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200772 if (apv >= 88)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200773 strcpy(p->verify_alg, nc->verify_alg);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200774 if (apv >= 89)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200775 strcpy(p->csums_alg, nc->csums_alg);
776 rcu_read_unlock();
Philipp Reisnerb411b362009-09-25 16:07:19 -0700777
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200778 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700779}
780
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200781int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700782{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200783 struct drbd_socket *sock;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700784 struct p_protocol *p;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200785 struct net_conf *nc;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200786 int size, cf;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700787
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200788 sock = &connection->data;
789 p = __conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200790 if (!p)
791 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700792
Philipp Reisner44ed1672011-04-19 17:10:19 +0200793 rcu_read_lock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200794 nc = rcu_dereference(connection->net_conf);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700795
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200796 if (nc->tentative && connection->agreed_pro_version < 92) {
Philipp Reisner44ed1672011-04-19 17:10:19 +0200797 rcu_read_unlock();
798 mutex_unlock(&sock->mutex);
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200799 drbd_err(connection, "--dry-run is not supported by peer");
Philipp Reisner44ed1672011-04-19 17:10:19 +0200800 return -EOPNOTSUPP;
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100801 }
Philipp Reisner44ed1672011-04-19 17:10:19 +0200802
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200803 size = sizeof(*p);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200804 if (connection->agreed_pro_version >= 87)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200805 size += strlen(nc->integrity_alg) + 1;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700806
Philipp Reisner44ed1672011-04-19 17:10:19 +0200807 p->protocol = cpu_to_be32(nc->wire_protocol);
808 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
809 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
810 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
811 p->two_primaries = cpu_to_be32(nc->two_primaries);
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100812 cf = 0;
Andreas Gruenbacher6139f602011-05-06 20:00:02 +0200813 if (nc->discard_my_data)
814 cf |= CF_DISCARD_MY_DATA;
Andreas Gruenbacher6dff2902011-06-28 14:18:12 +0200815 if (nc->tentative)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200816 cf |= CF_DRY_RUN;
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100817 p->conn_flags = cpu_to_be32(cf);
818
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200819 if (connection->agreed_pro_version >= 87)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200820 strcpy(p->integrity_alg, nc->integrity_alg);
821 rcu_read_unlock();
Philipp Reisnerb411b362009-09-25 16:07:19 -0700822
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200823 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200824}
825
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200826int drbd_send_protocol(struct drbd_connection *connection)
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200827{
828 int err;
829
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200830 mutex_lock(&connection->data.mutex);
831 err = __drbd_send_protocol(connection, P_PROTOCOL);
832 mutex_unlock(&connection->data.mutex);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200833
834 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700835}
836
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200837static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700838{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200839 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200840 struct drbd_socket *sock;
841 struct p_uuids *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700842 int i;
843
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200844 if (!get_ldev_if_state(device, D_NEGOTIATING))
Andreas Gruenbacher2ae5f952011-03-16 01:07:20 +0100845 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700846
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200847 sock = &peer_device->connection->data;
848 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200849 if (!p) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200850 put_ldev(device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200851 return -EIO;
852 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200853 spin_lock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700854 for (i = UI_CURRENT; i < UI_SIZE; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200855 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
856 spin_unlock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700857
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200858 device->comm_bm_set = drbd_bm_total_weight(device);
859 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
Philipp Reisner44ed1672011-04-19 17:10:19 +0200860 rcu_read_lock();
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200861 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200862 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200863 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
864 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200865 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700866
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200867 put_ldev(device);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200868 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700869}
870
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200871int drbd_send_uuids(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700872{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200873 return _drbd_send_uuids(peer_device, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700874}
875
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200876int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700877{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200878 return _drbd_send_uuids(peer_device, 8);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700879}
880
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200881void drbd_print_uuids(struct drbd_device *device, const char *text)
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100882{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200883 if (get_ldev_if_state(device, D_NEGOTIATING)) {
884 u64 *uuid = device->ldev->md.uuid;
Andreas Gruenbacherd0180172011-07-03 17:53:52 +0200885 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100886 text,
887 (unsigned long long)uuid[UI_CURRENT],
888 (unsigned long long)uuid[UI_BITMAP],
889 (unsigned long long)uuid[UI_HISTORY_START],
890 (unsigned long long)uuid[UI_HISTORY_END]);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200891 put_ldev(device);
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100892 } else {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +0200893 drbd_info(device, "%s effective data uuid: %016llX\n",
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100894 text,
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200895 (unsigned long long)device->ed_uuid);
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100896 }
897}
898
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200899void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700900{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200901 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200902 struct drbd_socket *sock;
903 struct p_rs_uuid *p;
Lars Ellenberg5a22db82010-12-17 21:14:23 +0100904 u64 uuid;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700905
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +0200906 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
Lars Ellenberg5a22db82010-12-17 21:14:23 +0100907
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200908 uuid = device->ldev->md.uuid[UI_BITMAP];
Philipp Reisner5ba3dac2011-10-05 15:54:18 +0200909 if (uuid && uuid != UUID_JUST_CREATED)
910 uuid = uuid + UUID_NEW_BM_OFFSET;
911 else
912 get_random_bytes(&uuid, sizeof(u64));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200913 drbd_uuid_set(device, UI_BITMAP, uuid);
914 drbd_print_uuids(device, "updated sync UUID");
915 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700916
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200917 sock = &peer_device->connection->data;
918 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200919 if (p) {
920 p->uuid = cpu_to_be64(uuid);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200921 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200922 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700923}
924
Lars Ellenberg9104d312016-06-14 00:26:31 +0200925/* communicated if (agreed_features & DRBD_FF_WSAME) */
Baoyou Xie1ffa7bf2017-08-29 10:20:36 +0200926static void
927assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
928 struct request_queue *q)
Lars Ellenberg9104d312016-06-14 00:26:31 +0200929{
930 if (q) {
931 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
932 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
933 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
934 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
935 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
936 p->qlim->discard_enabled = blk_queue_discard(q);
Lars Ellenberg9104d312016-06-14 00:26:31 +0200937 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
938 } else {
939 q = device->rq_queue;
940 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
941 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
942 p->qlim->alignment_offset = 0;
943 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
944 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
945 p->qlim->discard_enabled = 0;
Lars Ellenberg9104d312016-06-14 00:26:31 +0200946 p->qlim->write_same_capable = 0;
947 }
948}
949
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200950int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700951{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200952 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200953 struct drbd_socket *sock;
954 struct p_sizes *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700955 sector_t d_size, u_size;
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200956 int q_order_type;
957 unsigned int max_bio_size;
Lars Ellenberg9104d312016-06-14 00:26:31 +0200958 unsigned int packet_size;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700959
Lars Ellenberg9104d312016-06-14 00:26:31 +0200960 sock = &peer_device->connection->data;
961 p = drbd_prepare_command(peer_device, sock);
962 if (!p)
963 return -EIO;
964
965 packet_size = sizeof(*p);
966 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
967 packet_size += sizeof(p->qlim[0]);
968
969 memset(p, 0, packet_size);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200970 if (get_ldev_if_state(device, D_NEGOTIATING)) {
Lars Ellenberg9104d312016-06-14 00:26:31 +0200971 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200972 d_size = drbd_get_max_capacity(device->ldev);
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200973 rcu_read_lock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200974 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200975 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200976 q_order_type = drbd_queue_order_type(device);
Lars Ellenberg9104d312016-06-14 00:26:31 +0200977 max_bio_size = queue_max_hw_sectors(q) << 9;
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200978 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
Lars Ellenberg9104d312016-06-14 00:26:31 +0200979 assign_p_sizes_qlim(device, p, q);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200980 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700981 } else {
982 d_size = 0;
983 u_size = 0;
984 q_order_type = QUEUE_ORDERED_NONE;
Philipp Reisner99432fc2011-05-20 16:39:13 +0200985 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
Lars Ellenberg9104d312016-06-14 00:26:31 +0200986 assign_p_sizes_qlim(device, p, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700987 }
988
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200989 if (peer_device->connection->agreed_pro_version <= 94)
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200990 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200991 else if (peer_device->connection->agreed_pro_version < 100)
Philipp Reisner98683652012-11-09 14:18:43 +0100992 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
Philipp Reisner68093842011-06-30 15:43:06 +0200993
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200994 p->d_size = cpu_to_be64(d_size);
995 p->u_size = cpu_to_be64(u_size);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200996 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200997 p->max_bio_size = cpu_to_be32(max_bio_size);
998 p->queue_order_type = cpu_to_be16(q_order_type);
999 p->dds_flags = cpu_to_be16(flags);
Lars Ellenberg9104d312016-06-14 00:26:31 +02001000
1001 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001002}
1003
1004/**
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001005 * drbd_send_current_state() - Sends the drbd state to the peer
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001006 * @peer_device: DRBD peer device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001007 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001008int drbd_send_current_state(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001009{
Andreas Gruenbacher7c967152011-03-22 00:49:36 +01001010 struct drbd_socket *sock;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001011 struct p_state *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001012
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001013 sock = &peer_device->connection->data;
1014 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001015 if (!p)
1016 return -EIO;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001017 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1018 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001019}
1020
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001021/**
1022 * drbd_send_state() - After a state change, sends the new state to the peer
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001023 * @peer_device: DRBD peer device.
Philipp Reisner43de7c82011-11-10 13:16:13 +01001024 * @state: the state to send, not necessarily the current state.
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001025 *
1026 * Each state change queues an "after_state_ch" work, which will eventually
1027 * send the resulting new state to the peer. If more state changes happen
1028 * between queuing and processing of the after_state_ch work, we still
1029 * want to send each intermediary state in the order it occurred.
1030 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001031int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001032{
Philipp Reisner43de7c82011-11-10 13:16:13 +01001033 struct drbd_socket *sock;
1034 struct p_state *p;
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001035
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001036 sock = &peer_device->connection->data;
1037 p = drbd_prepare_command(peer_device, sock);
Philipp Reisner43de7c82011-11-10 13:16:13 +01001038 if (!p)
1039 return -EIO;
1040 p->state = cpu_to_be32(state.i); /* Within the send mutex */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001041 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
Philipp Reisner43de7c82011-11-10 13:16:13 +01001042}
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001043
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001044int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001045{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001046 struct drbd_socket *sock;
1047 struct p_req_state *p;
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001048
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001049 sock = &peer_device->connection->data;
1050 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001051 if (!p)
1052 return -EIO;
1053 p->mask = cpu_to_be32(mask.i);
1054 p->val = cpu_to_be32(val.i);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001055 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001056}
1057
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001058int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001059{
1060 enum drbd_packet cmd;
1061 struct drbd_socket *sock;
1062 struct p_req_state *p;
1063
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001064 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1065 sock = &connection->data;
1066 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001067 if (!p)
1068 return -EIO;
1069 p->mask = cpu_to_be32(mask.i);
1070 p->val = cpu_to_be32(val.i);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001071 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001072}
1073
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001074void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001075{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001076 struct drbd_socket *sock;
1077 struct p_req_state_reply *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001078
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001079 sock = &peer_device->connection->meta;
1080 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001081 if (p) {
1082 p->retcode = cpu_to_be32(retcode);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001083 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001084 }
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001085}
1086
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001087void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001088{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001089 struct drbd_socket *sock;
1090 struct p_req_state_reply *p;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001091 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001092
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001093 sock = &connection->meta;
1094 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001095 if (p) {
1096 p->retcode = cpu_to_be32(retcode);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001097 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001098 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001099}
1100
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001101static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001102{
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001103 BUG_ON(code & ~0xf);
1104 p->encoding = (p->encoding & ~0xf) | code;
1105}
Philipp Reisnerb411b362009-09-25 16:07:19 -07001106
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001107static void dcbp_set_start(struct p_compressed_bm *p, int set)
1108{
1109 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1110}
Philipp Reisnerb411b362009-09-25 16:07:19 -07001111
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001112static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1113{
1114 BUG_ON(n & ~0x7);
1115 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001116}
1117
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001118static int fill_bitmap_rle_bits(struct drbd_device *device,
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001119 struct p_compressed_bm *p,
1120 unsigned int size,
1121 struct bm_xfer_ctx *c)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001122{
1123 struct bitstream bs;
1124 unsigned long plain_bits;
1125 unsigned long tmp;
1126 unsigned long rl;
1127 unsigned len;
1128 unsigned toggle;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001129 int bits, use_rle;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001130
1131 /* may we use this feature? */
Philipp Reisner44ed1672011-04-19 17:10:19 +02001132 rcu_read_lock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001133 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001134 rcu_read_unlock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001135 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
Philipp Reisner44ed1672011-04-19 17:10:19 +02001136 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001137
1138 if (c->bit_offset >= c->bm_bits)
1139 return 0; /* nothing to do. */
1140
1141 /* use at most thus many bytes */
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001142 bitstream_init(&bs, p->code, size, 0);
1143 memset(p->code, 0, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001144 /* plain bits covered in this code string */
1145 plain_bits = 0;
1146
1147 /* p->encoding & 0x80 stores whether the first run length is set.
1148 * bit offset is implicit.
1149 * start with toggle == 2 to be able to tell the first iteration */
1150 toggle = 2;
1151
1152 /* see how much plain bits we can stuff into one packet
1153 * using RLE and VLI. */
1154 do {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001155 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1156 : _drbd_bm_find_next(device, c->bit_offset);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001157 if (tmp == -1UL)
1158 tmp = c->bm_bits;
1159 rl = tmp - c->bit_offset;
1160
1161 if (toggle == 2) { /* first iteration */
1162 if (rl == 0) {
1163 /* the first checked bit was set,
1164 * store start value, */
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001165 dcbp_set_start(p, 1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001166 /* but skip encoding of zero run length */
1167 toggle = !toggle;
1168 continue;
1169 }
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001170 dcbp_set_start(p, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001171 }
1172
1173 /* paranoia: catch zero runlength.
1174 * can only happen if bitmap is modified while we scan it. */
1175 if (rl == 0) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001176 drbd_err(device, "unexpected zero runlength while encoding bitmap "
Philipp Reisnerb411b362009-09-25 16:07:19 -07001177 "t:%u bo:%lu\n", toggle, c->bit_offset);
1178 return -1;
1179 }
1180
1181 bits = vli_encode_bits(&bs, rl);
1182 if (bits == -ENOBUFS) /* buffer full */
1183 break;
1184 if (bits <= 0) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001185 drbd_err(device, "error while encoding bitmap: %d\n", bits);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001186 return 0;
1187 }
1188
1189 toggle = !toggle;
1190 plain_bits += rl;
1191 c->bit_offset = tmp;
1192 } while (c->bit_offset < c->bm_bits);
1193
1194 len = bs.cur.b - p->code + !!bs.cur.bit;
1195
1196 if (plain_bits < (len << 3)) {
1197 /* incompressible with this method.
1198 * we need to rewind both word and bit position. */
1199 c->bit_offset -= plain_bits;
1200 bm_xfer_ctx_bit_to_word_offset(c);
1201 c->bit_offset = c->word_offset * BITS_PER_LONG;
1202 return 0;
1203 }
1204
1205 /* RLE + VLI was able to compress it just fine.
1206 * update c->word_offset. */
1207 bm_xfer_ctx_bit_to_word_offset(c);
1208
1209 /* store pad_bits */
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001210 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001211
1212 return len;
1213}
1214
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001215/**
1216 * send_bitmap_rle_or_plain
1217 *
1218 * Return 0 when done, 1 when another iteration is needed, and a negative error
1219 * code upon failure.
1220 */
1221static int
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001222send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001223{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001224 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1225 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001226 struct p_compressed_bm *p = sock->sbuf + header_size;
Andreas Gruenbachera982dd52010-12-10 00:45:25 +01001227 int len, err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001228
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001229 len = fill_bitmap_rle_bits(device, p,
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001230 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001231 if (len < 0)
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001232 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001233
1234 if (len) {
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001235 dcbp_set_code(p, RLE_VLI_Bits);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001236 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001237 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1238 NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001239 c->packets[0]++;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001240 c->bytes[0] += header_size + sizeof(*p) + len;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001241
1242 if (c->bit_offset >= c->bm_bits)
1243 len = 0; /* DONE */
1244 } else {
1245 /* was not compressible.
1246 * send a buffer full of plain text bits instead. */
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001247 unsigned int data_size;
1248 unsigned long num_words;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001249 unsigned long *p = sock->sbuf + header_size;
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001250
1251 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001252 num_words = min_t(size_t, data_size / sizeof(*p),
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001253 c->bm_words - c->word_offset);
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001254 len = num_words * sizeof(*p);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001255 if (len)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001256 drbd_bm_get_lel(device, c->word_offset, num_words, p);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001257 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001258 c->word_offset += num_words;
1259 c->bit_offset = c->word_offset * BITS_PER_LONG;
1260
1261 c->packets[1]++;
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001262 c->bytes[1] += header_size + len;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001263
1264 if (c->bit_offset > c->bm_bits)
1265 c->bit_offset = c->bm_bits;
1266 }
Andreas Gruenbachera982dd52010-12-10 00:45:25 +01001267 if (!err) {
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001268 if (len == 0) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001269 INFO_bm_xfer_stats(device, "send", c);
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001270 return 0;
1271 } else
1272 return 1;
1273 }
1274 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001275}
1276
1277/* See the comment at receive_bitmap() */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001278static int _drbd_send_bitmap(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001279{
1280 struct bm_xfer_ctx c;
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001281 int err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001282
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001283 if (!expect(device->bitmap))
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001284 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001285
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001286 if (get_ldev(device)) {
1287 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001288 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001289 drbd_bm_set_all(device);
1290 if (drbd_bm_write(device)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001291 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1292 * but otherwise process as per normal - need to tell other
1293 * side that a full resync is required! */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001294 drbd_err(device, "Failed to write bitmap to disk!\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07001295 } else {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001296 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1297 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001298 }
1299 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001300 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001301 }
1302
1303 c = (struct bm_xfer_ctx) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001304 .bm_bits = drbd_bm_bits(device),
1305 .bm_words = drbd_bm_words(device),
Philipp Reisnerb411b362009-09-25 16:07:19 -07001306 };
1307
1308 do {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001309 err = send_bitmap_rle_or_plain(device, &c);
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001310 } while (err > 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001311
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001312 return err == 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001313}
1314
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001315int drbd_send_bitmap(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001316{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001317 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001318 int err = -1;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001319
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001320 mutex_lock(&sock->mutex);
1321 if (sock->socket)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001322 err = !_drbd_send_bitmap(device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001323 mutex_unlock(&sock->mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001324 return err;
1325}
1326
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001327void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001328{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001329 struct drbd_socket *sock;
1330 struct p_barrier_ack *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001331
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001332 if (connection->cstate < C_WF_REPORT_PARAMS)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001333 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001334
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001335 sock = &connection->meta;
1336 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001337 if (!p)
1338 return;
1339 p->barrier = barrier_nr;
1340 p->set_size = cpu_to_be32(set_size);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001341 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001342}
1343
1344/**
1345 * _drbd_send_ack() - Sends an ack packet
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001346 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001347 * @cmd: Packet command code.
1348 * @sector: sector, needs to be in big endian byte order
1349 * @blksize: size in byte, needs to be in big endian byte order
1350 * @block_id: Id, big endian byte order
1351 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001352static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01001353 u64 sector, u32 blksize, u64 block_id)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001354{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001355 struct drbd_socket *sock;
1356 struct p_block_ack *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001357
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001358 if (peer_device->device->state.conn < C_CONNECTED)
Andreas Gruenbachera8c32aa2011-03-16 01:27:22 +01001359 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001360
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001361 sock = &peer_device->connection->meta;
1362 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001363 if (!p)
1364 return -EIO;
1365 p->sector = sector;
1366 p->block_id = block_id;
1367 p->blksize = blksize;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001368 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1369 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001370}
1371
Lars Ellenberg2b2bf212010-10-06 11:46:55 +02001372/* dp->sector and dp->block_id already/still in network byte order,
1373 * data_size is payload size according to dp->head,
1374 * and may need to be corrected for digest size. */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001375void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001376 struct p_data *dp, int data_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001377{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001378 if (peer_device->connection->peer_integrity_tfm)
Herbert Xu9534d672016-01-24 21:19:21 +08001379 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001380 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001381 dp->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001382}
1383
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001384void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001385 struct p_block_req *rp)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001386{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001387 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001388}
1389
1390/**
1391 * drbd_send_ack() - Sends an ack packet
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001392 * @device: DRBD device
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001393 * @cmd: packet command code
1394 * @peer_req: peer request
Philipp Reisnerb411b362009-09-25 16:07:19 -07001395 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001396int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001397 struct drbd_peer_request *peer_req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001398{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001399 return _drbd_send_ack(peer_device, cmd,
Andreas Gruenbacherdd516122011-03-16 15:39:08 +01001400 cpu_to_be64(peer_req->i.sector),
1401 cpu_to_be32(peer_req->i.size),
1402 peer_req->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001403}
1404
1405/* This function misuses the block_id field to signal if the blocks
1406 * are is sync or not. */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001407int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001408 sector_t sector, int blksize, u64 block_id)
1409{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001410 return _drbd_send_ack(peer_device, cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001411 cpu_to_be64(sector),
1412 cpu_to_be32(blksize),
1413 cpu_to_be64(block_id));
1414}
1415
Philipp Reisner700ca8c2016-06-14 00:26:13 +02001416int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1417 struct drbd_peer_request *peer_req)
1418{
1419 struct drbd_socket *sock;
1420 struct p_block_desc *p;
1421
1422 sock = &peer_device->connection->data;
1423 p = drbd_prepare_command(peer_device, sock);
1424 if (!p)
1425 return -EIO;
1426 p->sector = cpu_to_be64(peer_req->i.sector);
1427 p->blksize = cpu_to_be32(peer_req->i.size);
1428 p->pad = 0;
1429 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1430}
1431
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001432int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001433 sector_t sector, int size, u64 block_id)
1434{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001435 struct drbd_socket *sock;
1436 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001437
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001438 sock = &peer_device->connection->data;
1439 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001440 if (!p)
1441 return -EIO;
1442 p->sector = cpu_to_be64(sector);
1443 p->block_id = block_id;
1444 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001445 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001446}
1447
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001448int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01001449 void *digest, int digest_size, enum drbd_packet cmd)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001450{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001451 struct drbd_socket *sock;
1452 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001453
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001454 /* FIXME: Put the digest into the preallocated socket buffer. */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001455
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001456 sock = &peer_device->connection->data;
1457 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001458 if (!p)
1459 return -EIO;
1460 p->sector = cpu_to_be64(sector);
1461 p->block_id = ID_SYNCER /* unused */;
1462 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001463 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001464}
1465
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001466int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001467{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001468 struct drbd_socket *sock;
1469 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001470
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001471 sock = &peer_device->connection->data;
1472 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001473 if (!p)
1474 return -EIO;
1475 p->sector = cpu_to_be64(sector);
1476 p->block_id = ID_SYNCER /* unused */;
1477 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001478 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001479}
1480
1481/* called on sndtimeo
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001482 * returns false if we should retry,
1483 * true if we think connection is dead
Philipp Reisnerb411b362009-09-25 16:07:19 -07001484 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001485static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001486{
1487 int drop_it;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001488 /* long elapsed = (long)(jiffies - device->last_received); */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001489
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001490 drop_it = connection->meta.socket == sock
Philipp Reisner1c03e522015-03-16 15:01:00 +01001491 || !connection->ack_receiver.task
1492 || get_t_state(&connection->ack_receiver) != RUNNING
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001493 || connection->cstate < C_WF_REPORT_PARAMS;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001494
1495 if (drop_it)
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001496 return true;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001497
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001498 drop_it = !--connection->ko_count;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001499 if (!drop_it) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02001500 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001501 current->comm, current->pid, connection->ko_count);
1502 request_ping(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001503 }
1504
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001505 return drop_it; /* && (device->state == R_PRIMARY) */;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001506}
1507
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001508static void drbd_update_congested(struct drbd_connection *connection)
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001509{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001510 struct sock *sk = connection->data.socket->sk;
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001511 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001512 set_bit(NET_CONGESTED, &connection->flags);
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001513}
1514
Philipp Reisnerb411b362009-09-25 16:07:19 -07001515/* The idea of sendpage seems to be to put some kind of reference
1516 * to the page into the skb, and to hand it over to the NIC. In
1517 * this process get_page() gets called.
1518 *
1519 * As soon as the page was really sent over the network put_page()
1520 * gets called by some part of the network layer. [ NIC driver? ]
1521 *
1522 * [ get_page() / put_page() increment/decrement the count. If count
1523 * reaches 0 the page will be freed. ]
1524 *
1525 * This works nicely with pages from FSs.
1526 * But this means that in protocol A we might signal IO completion too early!
1527 *
1528 * In order not to corrupt data during a resync we must make sure
1529 * that we do not reuse our own buffer pages (EEs) to early, therefore
1530 * we have the net_ee list.
1531 *
1532 * XFS seems to have problems, still, it submits pages with page_count == 0!
1533 * As a workaround, we disable sendpage on pages
1534 * with page_count == 0 or PageSlab.
1535 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001536static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001537 int offset, size_t size, unsigned msg_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001538{
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001539 struct socket *socket;
1540 void *addr;
1541 int err;
1542
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001543 socket = peer_device->connection->data.socket;
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001544 addr = kmap(page) + offset;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001545 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001546 kunmap(page);
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001547 if (!err)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001548 peer_device->device->send_cnt += size >> 9;
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001549 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001550}
1551
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001552static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001553 int offset, size_t size, unsigned msg_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001554{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001555 struct socket *socket = peer_device->connection->data.socket;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001556 int len = size;
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001557 int err = -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001558
1559 /* e.g. XFS meta- & log-data is in slab pages, which have a
1560 * page_count of 0 and/or have PageSlab() set.
1561 * we cannot use send_page for those, as that does get_page();
1562 * put_page(); and would cause either a VM_BUG directly, or
1563 * __page_cache_release a page that would actually still be referenced
1564 * by someone, leading to some obscure delayed Oops somewhere else. */
1565 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001566 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001567
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001568 msg_flags |= MSG_NOSIGNAL;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001569 drbd_update_congested(peer_device->connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001570 do {
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001571 int sent;
1572
1573 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001574 if (sent <= 0) {
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001575 if (sent == -EAGAIN) {
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001576 if (we_should_drop_the_connection(peer_device->connection, socket))
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001577 break;
1578 continue;
1579 }
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001580 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07001581 __func__, (int)size, len, sent);
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001582 if (sent < 0)
1583 err = sent;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001584 break;
1585 }
1586 len -= sent;
1587 offset += sent;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001588 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001589 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001590
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001591 if (len == 0) {
1592 err = 0;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001593 peer_device->device->send_cnt += size >> 9;
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001594 }
1595 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001596}
1597
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001598static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001599{
Kent Overstreet79886132013-11-23 17:19:00 -08001600 struct bio_vec bvec;
1601 struct bvec_iter iter;
1602
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001603 /* hint all but last page with MSG_MORE */
Kent Overstreet79886132013-11-23 17:19:00 -08001604 bio_for_each_segment(bvec, bio, iter) {
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001605 int err;
1606
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001607 err = _drbd_no_send_page(peer_device, bvec.bv_page,
Kent Overstreet79886132013-11-23 17:19:00 -08001608 bvec.bv_offset, bvec.bv_len,
Kent Overstreet4550dd62013-08-07 14:26:21 -07001609 bio_iter_last(bvec, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08001610 ? 0 : MSG_MORE);
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001611 if (err)
1612 return err;
Lars Ellenberg9104d312016-06-14 00:26:31 +02001613 /* REQ_OP_WRITE_SAME has only one segment */
1614 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1615 break;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001616 }
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001617 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001618}
1619
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001620static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001621{
Kent Overstreet79886132013-11-23 17:19:00 -08001622 struct bio_vec bvec;
1623 struct bvec_iter iter;
1624
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001625 /* hint all but last page with MSG_MORE */
Kent Overstreet79886132013-11-23 17:19:00 -08001626 bio_for_each_segment(bvec, bio, iter) {
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001627 int err;
1628
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001629 err = _drbd_send_page(peer_device, bvec.bv_page,
Kent Overstreet79886132013-11-23 17:19:00 -08001630 bvec.bv_offset, bvec.bv_len,
Kent Overstreet4550dd62013-08-07 14:26:21 -07001631 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001632 if (err)
1633 return err;
Lars Ellenberg9104d312016-06-14 00:26:31 +02001634 /* REQ_OP_WRITE_SAME has only one segment */
1635 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1636 break;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001637 }
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001638 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001639}
1640
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001641static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001642 struct drbd_peer_request *peer_req)
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001643{
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001644 struct page *page = peer_req->pages;
1645 unsigned len = peer_req->i.size;
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001646 int err;
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001647
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001648 /* hint all but last page with MSG_MORE */
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001649 page_chain_for_each(page) {
1650 unsigned l = min_t(unsigned, len, PAGE_SIZE);
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001651
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001652 err = _drbd_send_page(peer_device, page, 0, l,
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001653 page_chain_next(page) ? MSG_MORE : 0);
1654 if (err)
1655 return err;
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001656 len -= l;
1657 }
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001658 return 0;
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001659}
1660
Mike Christiebb3cc852016-06-05 14:32:06 -05001661static u32 bio_flags_to_wire(struct drbd_connection *connection,
1662 struct bio *bio)
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001663{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001664 if (connection->agreed_pro_version >= 95)
Jens Axboe1eff9d32016-08-05 15:35:16 -06001665 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1666 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1667 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
Lars Ellenberg9104d312016-06-14 00:26:31 +02001668 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
Christoph Hellwig45c21792017-04-05 19:21:22 +02001669 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1670 (bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001671 else
Jens Axboe1eff9d32016-08-05 15:35:16 -06001672 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001673}
1674
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001675/* Used to send write or TRIM aka REQ_DISCARD requests
1676 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001677 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001678int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001679{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001680 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001681 struct drbd_socket *sock;
1682 struct p_data *p;
Lars Ellenberg9104d312016-06-14 00:26:31 +02001683 struct p_wsame *wsame = NULL;
1684 void *digest_out;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001685 unsigned int dp_flags = 0;
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001686 int digest_size;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001687 int err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001688
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001689 sock = &peer_device->connection->data;
1690 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001691 digest_size = peer_device->connection->integrity_tfm ?
Herbert Xu9534d672016-01-24 21:19:21 +08001692 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001693
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001694 if (!p)
1695 return -EIO;
1696 p->sector = cpu_to_be64(req->i.sector);
1697 p->block_id = (unsigned long)req;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001698 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
Mike Christiebb3cc852016-06-05 14:32:06 -05001699 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001700 if (device->state.conn >= C_SYNC_SOURCE &&
1701 device->state.conn <= C_PAUSED_SYNC_T)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001702 dp_flags |= DP_MAY_SET_IN_SYNC;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001703 if (peer_device->connection->agreed_pro_version >= 100) {
Philipp Reisner303d1442011-04-13 16:24:47 -07001704 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1705 dp_flags |= DP_SEND_RECEIVE_ACK;
Lars Ellenberg08d0dab2014-03-20 11:19:22 +01001706 /* During resync, request an explicit write ack,
1707 * even in protocol != C */
1708 if (req->rq_state & RQ_EXP_WRITE_ACK
1709 || (dp_flags & DP_MAY_SET_IN_SYNC))
Philipp Reisner303d1442011-04-13 16:24:47 -07001710 dp_flags |= DP_SEND_WRITE_ACK;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001711 }
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001712 p->dp_flags = cpu_to_be32(dp_flags);
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001713
1714 if (dp_flags & DP_DISCARD) {
1715 struct p_trim *t = (struct p_trim*)p;
1716 t->size = cpu_to_be32(req->i.size);
1717 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1718 goto out;
1719 }
Lars Ellenberg9104d312016-06-14 00:26:31 +02001720 if (dp_flags & DP_WSAME) {
1721 /* this will only work if DRBD_FF_WSAME is set AND the
1722 * handshake agreed that all nodes and backend devices are
1723 * WRITE_SAME capable and agree on logical_block_size */
1724 wsame = (struct p_wsame*)p;
1725 digest_out = wsame + 1;
1726 wsame->size = cpu_to_be32(req->i.size);
1727 } else
1728 digest_out = p + 1;
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001729
1730 /* our digest is still only over the payload.
1731 * TRIM does not carry any payload. */
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001732 if (digest_size)
Lars Ellenberg9104d312016-06-14 00:26:31 +02001733 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1734 if (wsame) {
1735 err =
1736 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1737 sizeof(*wsame) + digest_size, NULL,
1738 bio_iovec(req->master_bio).bv_len);
1739 } else
1740 err =
1741 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1742 sizeof(*p) + digest_size, NULL, req->i.size);
Andreas Gruenbacher6bdb9b02011-03-16 11:52:58 +01001743 if (!err) {
Lars Ellenberg470be442010-11-10 10:36:52 +01001744 /* For protocol A, we have to memcpy the payload into
1745 * socket buffers, as we may complete right away
1746 * as soon as we handed it over to tcp, at which point the data
1747 * pages may become invalid.
1748 *
1749 * For data-integrity enabled, we copy it as well, so we can be
1750 * sure that even if the bio pages may still be modified, it
1751 * won't change the data on the wire, thus if the digest checks
1752 * out ok after sending on this side, but does not fit on the
1753 * receiving side, we sure have detected corruption elsewhere.
1754 */
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001755 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001756 err = _drbd_send_bio(peer_device, req->master_bio);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001757 else
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001758 err = _drbd_send_zc_bio(peer_device, req->master_bio);
Lars Ellenberg470be442010-11-10 10:36:52 +01001759
1760 /* double check digest, sometimes buffers have been modified in flight. */
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001761 if (digest_size > 0 && digest_size <= 64) {
Bart Van Assche24c48302011-05-21 18:32:29 +02001762 /* 64 byte, 512 bit, is the largest digest size
Lars Ellenberg470be442010-11-10 10:36:52 +01001763 * currently supported in kernel crypto. */
1764 unsigned char digest[64];
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001765 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001766 if (memcmp(p + 1, digest, digest_size)) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001767 drbd_warn(device,
Lars Ellenberg470be442010-11-10 10:36:52 +01001768 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
Andreas Gruenbacherace652a2011-01-03 17:09:58 +01001769 (unsigned long long)req->i.sector, req->i.size);
Lars Ellenberg470be442010-11-10 10:36:52 +01001770 }
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001771 } /* else if (digest_size > 64) {
Lars Ellenberg470be442010-11-10 10:36:52 +01001772 ... Be noisy about digest too large ...
1773 } */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001774 }
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001775out:
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001776 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001777
Andreas Gruenbacher6bdb9b02011-03-16 11:52:58 +01001778 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001779}
1780
1781/* answer packet, used to send data back for read requests:
1782 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1783 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1784 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001785int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001786 struct drbd_peer_request *peer_req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001787{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001788 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001789 struct drbd_socket *sock;
1790 struct p_data *p;
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001791 int err;
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001792 int digest_size;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001793
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001794 sock = &peer_device->connection->data;
1795 p = drbd_prepare_command(peer_device, sock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001796
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001797 digest_size = peer_device->connection->integrity_tfm ?
Herbert Xu9534d672016-01-24 21:19:21 +08001798 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001799
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001800 if (!p)
1801 return -EIO;
1802 p->sector = cpu_to_be64(peer_req->i.sector);
1803 p->block_id = peer_req->block_id;
1804 p->seq_num = 0; /* unused */
Lars Ellenbergb17f33c2012-02-08 15:32:51 +01001805 p->dp_flags = 0;
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001806 if (digest_size)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001807 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
Andreas Gruenbacher11f8b2b2014-09-11 14:29:05 +02001808 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001809 if (!err)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001810 err = _drbd_send_zc_ee(peer_device, peer_req);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001811 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001812
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001813 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001814}
1815
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001816int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
Philipp Reisner73a01a12010-10-27 14:33:00 +02001817{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001818 struct drbd_socket *sock;
1819 struct p_block_desc *p;
Philipp Reisner73a01a12010-10-27 14:33:00 +02001820
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001821 sock = &peer_device->connection->data;
1822 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001823 if (!p)
1824 return -EIO;
1825 p->sector = cpu_to_be64(req->i.sector);
1826 p->blksize = cpu_to_be32(req->i.size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001827 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
Philipp Reisner73a01a12010-10-27 14:33:00 +02001828}
1829
Philipp Reisnerb411b362009-09-25 16:07:19 -07001830/*
1831 drbd_send distinguishes two cases:
1832
1833 Packets sent via the data socket "sock"
1834 and packets sent via the meta data socket "msock"
1835
1836 sock msock
1837 -----------------+-------------------------+------------------------------
1838 timeout conf.timeout / 2 conf.timeout / 2
1839 timeout action send a ping via msock Abort communication
1840 and close all sockets
1841*/
1842
1843/*
1844 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1845 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001846int drbd_send(struct drbd_connection *connection, struct socket *sock,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001847 void *buf, size_t size, unsigned msg_flags)
1848{
Al Viro79ab80b2014-12-10 17:06:38 -05001849 struct kvec iov = {.iov_base = buf, .iov_len = size};
Philipp Reisnerb411b362009-09-25 16:07:19 -07001850 struct msghdr msg;
1851 int rv, sent = 0;
1852
1853 if (!sock)
Andreas Gruenbacherc0d42c82010-12-09 23:52:22 +01001854 return -EBADR;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001855
1856 /* THINK if (signal_pending) return ... ? */
1857
Philipp Reisnerb411b362009-09-25 16:07:19 -07001858 msg.msg_name = NULL;
1859 msg.msg_namelen = 0;
1860 msg.msg_control = NULL;
1861 msg.msg_controllen = 0;
1862 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1863
Al Viro79ab80b2014-12-10 17:06:38 -05001864 iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
1865
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001866 if (sock == connection->data.socket) {
Philipp Reisner44ed1672011-04-19 17:10:19 +02001867 rcu_read_lock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001868 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001869 rcu_read_unlock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001870 drbd_update_congested(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001871 }
1872 do {
Al Viro79ab80b2014-12-10 17:06:38 -05001873 rv = sock_sendmsg(sock, &msg);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001874 if (rv == -EAGAIN) {
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001875 if (we_should_drop_the_connection(connection, sock))
Philipp Reisnerb411b362009-09-25 16:07:19 -07001876 break;
1877 else
1878 continue;
1879 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001880 if (rv == -EINTR) {
1881 flush_signals(current);
1882 rv = 0;
1883 }
1884 if (rv < 0)
1885 break;
1886 sent += rv;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001887 } while (sent < size);
1888
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001889 if (sock == connection->data.socket)
1890 clear_bit(NET_CONGESTED, &connection->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001891
1892 if (rv <= 0) {
1893 if (rv != -EAGAIN) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02001894 drbd_err(connection, "%s_sendmsg returned %d\n",
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001895 sock == connection->meta.socket ? "msock" : "sock",
Philipp Reisnerbedbd2a2011-02-07 15:08:48 +01001896 rv);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001897 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001898 } else
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001899 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001900 }
1901
1902 return sent;
1903}
1904
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001905/**
1906 * drbd_send_all - Send an entire buffer
1907 *
1908 * Returns 0 upon success and a negative error value otherwise.
1909 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001910int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001911 size_t size, unsigned msg_flags)
1912{
1913 int err;
1914
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001915 err = drbd_send(connection, sock, buffer, size, msg_flags);
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001916 if (err < 0)
1917 return err;
1918 if (err != size)
1919 return -EIO;
1920 return 0;
1921}
1922
Philipp Reisnerb411b362009-09-25 16:07:19 -07001923static int drbd_open(struct block_device *bdev, fmode_t mode)
1924{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001925 struct drbd_device *device = bdev->bd_disk->private_data;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001926 unsigned long flags;
1927 int rv = 0;
1928
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001929 mutex_lock(&drbd_main_mutex);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02001930 spin_lock_irqsave(&device->resource->req_lock, flags);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001931 /* to have a stable device->state.role
Philipp Reisnerb411b362009-09-25 16:07:19 -07001932 * and no race with updating open_cnt */
1933
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001934 if (device->state.role != R_PRIMARY) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001935 if (mode & FMODE_WRITE)
1936 rv = -EROFS;
1937 else if (!allow_oos)
1938 rv = -EMEDIUMTYPE;
1939 }
1940
1941 if (!rv)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001942 device->open_cnt++;
Andreas Gruenbacher05008132011-07-07 14:19:42 +02001943 spin_unlock_irqrestore(&device->resource->req_lock, flags);
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001944 mutex_unlock(&drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001945
1946 return rv;
1947}
1948
Al Virodb2a1442013-05-05 21:52:57 -04001949static void drbd_release(struct gendisk *gd, fmode_t mode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001950{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001951 struct drbd_device *device = gd->private_data;
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001952 mutex_lock(&drbd_main_mutex);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001953 device->open_cnt--;
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001954 mutex_unlock(&drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001955}
1956
Lars Ellenbergc51a0ef2017-08-29 10:20:32 +02001957/* need to hold resource->req_lock */
1958void drbd_queue_unplug(struct drbd_device *device)
1959{
1960 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1961 D_ASSERT(device, device->state.role == R_PRIMARY);
1962 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1963 drbd_queue_work_if_unqueued(
1964 &first_peer_device(device)->connection->sender_work,
1965 &device->unplug_work);
1966 }
1967 }
1968}
1969
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001970static void drbd_set_defaults(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001971{
Lars Ellenbergf3990022011-03-23 14:31:09 +01001972 /* Beware! The actual layout differs
1973 * between big endian and little endian */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001974 device->state = (union drbd_dev_state) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001975 { .role = R_SECONDARY,
1976 .peer = R_UNKNOWN,
1977 .conn = C_STANDALONE,
1978 .disk = D_DISKLESS,
1979 .pdsk = D_UNKNOWN,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001980 } };
1981}
1982
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001983void drbd_init_set_defaults(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001984{
1985 /* the memset(,0,) did most of this.
1986 * note: only assignments, no allocation in here */
1987
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001988 drbd_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001989
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001990 atomic_set(&device->ap_bio_cnt, 0);
Lars Ellenbergad3fee72013-12-20 11:22:13 +01001991 atomic_set(&device->ap_actlog_cnt, 0);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001992 atomic_set(&device->ap_pending_cnt, 0);
1993 atomic_set(&device->rs_pending_cnt, 0);
1994 atomic_set(&device->unacked_cnt, 0);
1995 atomic_set(&device->local_cnt, 0);
1996 atomic_set(&device->pp_in_use_by_net, 0);
1997 atomic_set(&device->rs_sect_in, 0);
1998 atomic_set(&device->rs_sect_ev, 0);
1999 atomic_set(&device->ap_in_flight, 0);
Lars Ellenberge37d2432014-04-01 23:53:30 +02002000 atomic_set(&device->md_io.in_use, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002001
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002002 mutex_init(&device->own_state_mutex);
2003 device->state_mutex = &device->own_state_mutex;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002004
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002005 spin_lock_init(&device->al_lock);
2006 spin_lock_init(&device->peer_seq_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002007
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002008 INIT_LIST_HEAD(&device->active_ee);
2009 INIT_LIST_HEAD(&device->sync_ee);
2010 INIT_LIST_HEAD(&device->done_ee);
2011 INIT_LIST_HEAD(&device->read_ee);
2012 INIT_LIST_HEAD(&device->net_ee);
2013 INIT_LIST_HEAD(&device->resync_reads);
2014 INIT_LIST_HEAD(&device->resync_work.list);
2015 INIT_LIST_HEAD(&device->unplug_work.list);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002016 INIT_LIST_HEAD(&device->bm_io_work.w.list);
Lars Ellenberg844a6ae72013-11-22 12:52:03 +01002017 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2018 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2019 INIT_LIST_HEAD(&device->pending_completion[0]);
2020 INIT_LIST_HEAD(&device->pending_completion[1]);
Philipp Reisner0ced55a2010-04-30 15:26:20 +02002021
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002022 device->resync_work.cb = w_resync_timer;
2023 device->unplug_work.cb = w_send_write_hint;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002024 device->bm_io_work.w.cb = w_bitmap_io;
Philipp Reisnera21e9292011-02-08 15:08:49 +01002025
Geliang Tangbe7445a2017-08-29 10:20:40 +02002026 setup_timer(&device->resync_timer, resync_timer_fn,
2027 (unsigned long)device);
2028 setup_timer(&device->md_sync_timer, md_sync_timer_fn,
2029 (unsigned long)device);
2030 setup_timer(&device->start_resync_timer, start_resync_timer_fn,
2031 (unsigned long)device);
2032 setup_timer(&device->request_timer, request_timer_fn,
2033 (unsigned long)device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002034
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002035 init_waitqueue_head(&device->misc_wait);
2036 init_waitqueue_head(&device->state_wait);
2037 init_waitqueue_head(&device->ee_wait);
2038 init_waitqueue_head(&device->al_wait);
2039 init_waitqueue_head(&device->seq_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002040
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002041 device->resync_wenr = LC_FREE;
2042 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2043 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002044}
2045
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002046void drbd_device_cleanup(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002047{
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02002048 int i;
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002049 if (first_peer_device(device)->connection->receiver.t_state != NONE)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002050 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002051 first_peer_device(device)->connection->receiver.t_state);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002052
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002053 device->al_writ_cnt =
2054 device->bm_writ_cnt =
2055 device->read_cnt =
2056 device->recv_cnt =
2057 device->send_cnt =
2058 device->writ_cnt =
2059 device->p_size =
2060 device->rs_start =
2061 device->rs_total =
2062 device->rs_failed = 0;
2063 device->rs_last_events = 0;
2064 device->rs_last_sect_ev = 0;
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02002065 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002066 device->rs_mark_left[i] = 0;
2067 device->rs_mark_time[i] = 0;
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02002068 }
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002069 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002070
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002071 drbd_set_my_capacity(device, 0);
2072 if (device->bitmap) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07002073 /* maybe never allocated. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002074 drbd_bm_resize(device, 0, 1);
2075 drbd_bm_cleanup(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002076 }
2077
Lars Ellenberg63a7c8a2015-03-26 20:53:55 +01002078 drbd_backing_dev_free(device, device->ldev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002079 device->ldev = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002080
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002081 clear_bit(AL_SUSPENDED, &device->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002082
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002083 D_ASSERT(device, list_empty(&device->active_ee));
2084 D_ASSERT(device, list_empty(&device->sync_ee));
2085 D_ASSERT(device, list_empty(&device->done_ee));
2086 D_ASSERT(device, list_empty(&device->read_ee));
2087 D_ASSERT(device, list_empty(&device->net_ee));
2088 D_ASSERT(device, list_empty(&device->resync_reads));
2089 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2090 D_ASSERT(device, list_empty(&device->resync_work.list));
2091 D_ASSERT(device, list_empty(&device->unplug_work.list));
Lars Ellenberg2265b472010-12-16 15:41:26 +01002092
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002093 drbd_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002094}
2095
2096
2097static void drbd_destroy_mempools(void)
2098{
2099 struct page *page;
2100
2101 while (drbd_pp_pool) {
2102 page = drbd_pp_pool;
2103 drbd_pp_pool = (struct page *)page_private(page);
2104 __free_page(page);
2105 drbd_pp_vacant--;
2106 }
2107
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002108 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002109
NeilBrown8cb0def2017-06-18 14:38:58 +10002110 if (drbd_io_bio_set)
2111 bioset_free(drbd_io_bio_set);
Lars Ellenberg9476f392011-02-23 17:02:01 +01002112 if (drbd_md_io_bio_set)
2113 bioset_free(drbd_md_io_bio_set);
Lars Ellenberg42818082011-02-23 12:39:46 +01002114 if (drbd_md_io_page_pool)
2115 mempool_destroy(drbd_md_io_page_pool);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002116 if (drbd_ee_mempool)
2117 mempool_destroy(drbd_ee_mempool);
2118 if (drbd_request_mempool)
2119 mempool_destroy(drbd_request_mempool);
2120 if (drbd_ee_cache)
2121 kmem_cache_destroy(drbd_ee_cache);
2122 if (drbd_request_cache)
2123 kmem_cache_destroy(drbd_request_cache);
2124 if (drbd_bm_ext_cache)
2125 kmem_cache_destroy(drbd_bm_ext_cache);
2126 if (drbd_al_ext_cache)
2127 kmem_cache_destroy(drbd_al_ext_cache);
2128
NeilBrown8cb0def2017-06-18 14:38:58 +10002129 drbd_io_bio_set = NULL;
Lars Ellenberg9476f392011-02-23 17:02:01 +01002130 drbd_md_io_bio_set = NULL;
Lars Ellenberg42818082011-02-23 12:39:46 +01002131 drbd_md_io_page_pool = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002132 drbd_ee_mempool = NULL;
2133 drbd_request_mempool = NULL;
2134 drbd_ee_cache = NULL;
2135 drbd_request_cache = NULL;
2136 drbd_bm_ext_cache = NULL;
2137 drbd_al_ext_cache = NULL;
2138
2139 return;
2140}
2141
2142static int drbd_create_mempools(void)
2143{
2144 struct page *page;
Lars Ellenberg1816a2b2010-11-11 15:19:07 +01002145 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002146 int i;
2147
2148 /* prepare our caches and mempools */
2149 drbd_request_mempool = NULL;
2150 drbd_ee_cache = NULL;
2151 drbd_request_cache = NULL;
2152 drbd_bm_ext_cache = NULL;
2153 drbd_al_ext_cache = NULL;
2154 drbd_pp_pool = NULL;
Lars Ellenberg42818082011-02-23 12:39:46 +01002155 drbd_md_io_page_pool = NULL;
Lars Ellenberg9476f392011-02-23 17:02:01 +01002156 drbd_md_io_bio_set = NULL;
NeilBrown8cb0def2017-06-18 14:38:58 +10002157 drbd_io_bio_set = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002158
2159 /* caches */
2160 drbd_request_cache = kmem_cache_create(
2161 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2162 if (drbd_request_cache == NULL)
2163 goto Enomem;
2164
2165 drbd_ee_cache = kmem_cache_create(
Andreas Gruenbacherf6ffca92011-02-04 15:30:34 +01002166 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002167 if (drbd_ee_cache == NULL)
2168 goto Enomem;
2169
2170 drbd_bm_ext_cache = kmem_cache_create(
2171 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2172 if (drbd_bm_ext_cache == NULL)
2173 goto Enomem;
2174
2175 drbd_al_ext_cache = kmem_cache_create(
2176 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2177 if (drbd_al_ext_cache == NULL)
2178 goto Enomem;
2179
2180 /* mempools */
NeilBrown8cb0def2017-06-18 14:38:58 +10002181 drbd_io_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_RESCUER);
2182 if (drbd_io_bio_set == NULL)
2183 goto Enomem;
2184
NeilBrown47e0fb42017-06-18 14:38:57 +10002185 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0,
2186 BIOSET_NEED_BVECS |
2187 BIOSET_NEED_RESCUER);
Lars Ellenberg9476f392011-02-23 17:02:01 +01002188 if (drbd_md_io_bio_set == NULL)
2189 goto Enomem;
Lars Ellenberg9476f392011-02-23 17:02:01 +01002190
Lars Ellenberg42818082011-02-23 12:39:46 +01002191 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2192 if (drbd_md_io_page_pool == NULL)
2193 goto Enomem;
2194
David Rientjescbc4ffd2015-03-24 16:22:54 -07002195 drbd_request_mempool = mempool_create_slab_pool(number,
2196 drbd_request_cache);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002197 if (drbd_request_mempool == NULL)
2198 goto Enomem;
2199
David Rientjescbc4ffd2015-03-24 16:22:54 -07002200 drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
Nicolas Kaiser2027ae12010-10-28 06:15:26 -06002201 if (drbd_ee_mempool == NULL)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002202 goto Enomem;
2203
2204 /* drbd's page pool */
2205 spin_lock_init(&drbd_pp_lock);
2206
2207 for (i = 0; i < number; i++) {
2208 page = alloc_page(GFP_HIGHUSER);
2209 if (!page)
2210 goto Enomem;
2211 set_page_private(page, (unsigned long)drbd_pp_pool);
2212 drbd_pp_pool = page;
2213 }
2214 drbd_pp_vacant = number;
2215
2216 return 0;
2217
2218Enomem:
2219 drbd_destroy_mempools(); /* in case we allocated some */
2220 return -ENOMEM;
2221}
2222
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002223static void drbd_release_all_peer_reqs(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002224{
2225 int rr;
2226
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002227 rr = drbd_free_peer_reqs(device, &device->active_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002228 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002229 drbd_err(device, "%d EEs in active list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002230
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002231 rr = drbd_free_peer_reqs(device, &device->sync_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002232 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002233 drbd_err(device, "%d EEs in sync list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002234
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002235 rr = drbd_free_peer_reqs(device, &device->read_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002236 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002237 drbd_err(device, "%d EEs in read list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002238
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002239 rr = drbd_free_peer_reqs(device, &device->done_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002240 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002241 drbd_err(device, "%d EEs in done list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002242
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002243 rr = drbd_free_peer_reqs(device, &device->net_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002244 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002245 drbd_err(device, "%d EEs in net list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002246}
2247
Philipp Reisner774b3052011-02-22 02:07:03 -05002248/* caution. no locking. */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002249void drbd_destroy_device(struct kref *kref)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002250{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002251 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002252 struct drbd_resource *resource = device->resource;
Lars Ellenberga8ba0d62014-05-14 21:34:47 +02002253 struct drbd_peer_device *peer_device, *tmp_peer_device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002254
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002255 del_timer_sync(&device->request_timer);
Philipp Reisnerdfa8bed2011-06-29 14:06:08 +02002256
Philipp Reisnerb411b362009-09-25 16:07:19 -07002257 /* paranoia asserts */
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002258 D_ASSERT(device, device->open_cnt == 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002259 /* end paranoia asserts */
2260
Philipp Reisnerb411b362009-09-25 16:07:19 -07002261 /* cleanup stuff that may have been allocated during
2262 * device (re-)configuration or state changes */
2263
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002264 if (device->this_bdev)
2265 bdput(device->this_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002266
Lars Ellenberg63a7c8a2015-03-26 20:53:55 +01002267 drbd_backing_dev_free(device, device->ldev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002268 device->ldev = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002269
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002270 drbd_release_all_peer_reqs(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002271
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002272 lc_destroy(device->act_log);
2273 lc_destroy(device->resync);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002274
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002275 kfree(device->p_uuid);
2276 /* device->p_uuid = NULL; */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002277
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002278 if (device->bitmap) /* should no longer be there. */
2279 drbd_bm_cleanup(device);
Lars Ellenberge37d2432014-04-01 23:53:30 +02002280 __free_page(device->md_io.page);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002281 put_disk(device->vdisk);
2282 blk_cleanup_queue(device->rq_queue);
2283 kfree(device->rs_plan_s);
Lars Ellenberga8ba0d62014-05-14 21:34:47 +02002284
2285 /* not for_each_connection(connection, resource):
2286 * those may have been cleaned up and disassociated already.
2287 */
2288 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2289 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2290 kfree(peer_device);
2291 }
Lars Ellenbergc2258ff2014-05-14 21:35:21 +02002292 memset(device, 0xfd, sizeof(*device));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002293 kfree(device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002294 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002295}
2296
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002297/* One global retry thread, if we need to push back some bio and have it
2298 * reinserted through our make request function.
2299 */
2300static struct retry_worker {
2301 struct workqueue_struct *wq;
2302 struct work_struct worker;
2303
2304 spinlock_t lock;
2305 struct list_head writes;
2306} retry;
2307
2308static void do_retry(struct work_struct *ws)
2309{
2310 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2311 LIST_HEAD(writes);
2312 struct drbd_request *req, *tmp;
2313
2314 spin_lock_irq(&retry->lock);
2315 list_splice_init(&retry->writes, &writes);
2316 spin_unlock_irq(&retry->lock);
2317
2318 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002319 struct drbd_device *device = req->device;
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002320 struct bio *bio = req->master_bio;
Lars Ellenberge5f891b2013-11-22 12:32:01 +01002321 unsigned long start_jif = req->start_jif;
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002322 bool expected;
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002323
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002324 expected =
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002325 expect(atomic_read(&req->completion_ref) == 0) &&
2326 expect(req->rq_state & RQ_POSTPONED) &&
2327 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2328 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2329
2330 if (!expected)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002331 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002332 req, atomic_read(&req->completion_ref),
2333 req->rq_state);
2334
2335 /* We still need to put one kref associated with the
2336 * "completion_ref" going zero in the code path that queued it
2337 * here. The request object may still be referenced by a
2338 * frozen local req->private_bio, in case we force-detached.
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002339 */
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002340 kref_put(&req->kref, drbd_req_destroy);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002341
2342 /* A single suspended or otherwise blocking device may stall
2343 * all others as well. Fortunately, this code path is to
2344 * recover from a situation that "should not happen":
2345 * concurrent writes in multi-primary setup.
2346 * In a "normal" lifecycle, this workqueue is supposed to be
2347 * destroyed without ever doing anything.
2348 * If it turns out to be an issue anyways, we can do per
2349 * resource (replication group) or per device (minor) retry
2350 * workqueues instead.
2351 */
2352
2353 /* We are not just doing generic_make_request(),
2354 * as we want to keep the start_time information. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002355 inc_ap_bio(device);
Lars Ellenberge5f891b2013-11-22 12:32:01 +01002356 __drbd_make_request(device, bio, start_jif);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002357 }
2358}
2359
Lars Ellenberg844a6ae72013-11-22 12:52:03 +01002360/* called via drbd_req_put_completion_ref(),
2361 * holds resource->req_lock */
Lars Ellenberg9d05e7c2012-07-17 10:05:04 +02002362void drbd_restart_request(struct drbd_request *req)
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002363{
2364 unsigned long flags;
2365 spin_lock_irqsave(&retry.lock, flags);
2366 list_move_tail(&req->tl_requests, &retry.writes);
2367 spin_unlock_irqrestore(&retry.lock, flags);
2368
2369 /* Drop the extra reference that would otherwise
2370 * have been dropped by complete_master_bio.
2371 * do_retry() needs to grab a new one. */
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002372 dec_ap_bio(req->device);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002373
2374 queue_work(retry.wq, &retry.worker);
2375}
2376
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002377void drbd_destroy_resource(struct kref *kref)
2378{
2379 struct drbd_resource *resource =
2380 container_of(kref, struct drbd_resource, kref);
2381
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002382 idr_destroy(&resource->devices);
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002383 free_cpumask_var(resource->cpu_mask);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002384 kfree(resource->name);
Lars Ellenbergc2258ff2014-05-14 21:35:21 +02002385 memset(resource, 0xf2, sizeof(*resource));
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002386 kfree(resource);
2387}
2388
2389void drbd_free_resource(struct drbd_resource *resource)
2390{
2391 struct drbd_connection *connection, *tmp;
2392
2393 for_each_connection_safe(connection, tmp, resource) {
2394 list_del(&connection->connections);
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002395 drbd_debugfs_connection_cleanup(connection);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002396 kref_put(&connection->kref, drbd_destroy_connection);
2397 }
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002398 drbd_debugfs_resource_cleanup(resource);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002399 kref_put(&resource->kref, drbd_destroy_resource);
2400}
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002401
Philipp Reisnerb411b362009-09-25 16:07:19 -07002402static void drbd_cleanup(void)
2403{
2404 unsigned int i;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002405 struct drbd_device *device;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002406 struct drbd_resource *resource, *tmp;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002407
Lars Ellenberg17a93f32010-11-24 10:37:35 +01002408 /* first remove proc,
2409 * drbdsetup uses it's presence to detect
2410 * whether DRBD is loaded.
2411 * If we would get stuck in proc removal,
2412 * but have netlink already deregistered,
2413 * some drbdsetup commands may wait forever
2414 * for an answer.
2415 */
2416 if (drbd_proc)
2417 remove_proc_entry("drbd", NULL);
2418
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002419 if (retry.wq)
2420 destroy_workqueue(retry.wq);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002421
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002422 drbd_genl_unregister();
Philipp Reisnerb411b362009-09-25 16:07:19 -07002423
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002424 idr_for_each_entry(&drbd_devices, device, i)
Andreas Gruenbacherf82795d2011-07-03 23:32:26 +02002425 drbd_delete_device(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002426
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002427 /* not _rcu since, no other updater anymore. Genl already unregistered */
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002428 for_each_resource_safe(resource, tmp, &drbd_resources) {
2429 list_del(&resource->resources);
2430 drbd_free_resource(resource);
Philipp Reisner81fa2e62011-05-04 15:10:30 +02002431 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07002432
Lars Ellenberg3f1a1b72017-08-29 10:20:41 +02002433 drbd_debugfs_cleanup();
2434
Philipp Reisner81a5d602011-02-22 19:53:16 -05002435 drbd_destroy_mempools();
Philipp Reisnerb411b362009-09-25 16:07:19 -07002436 unregister_blkdev(DRBD_MAJOR, "drbd");
2437
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002438 idr_destroy(&drbd_devices);
Philipp Reisner81a5d602011-02-22 19:53:16 -05002439
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002440 pr_info("module cleanup done.\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07002441}
2442
2443/**
Artem Bityutskiyd97482e2012-07-25 18:12:12 +03002444 * drbd_congested() - Callback for the flusher thread
Philipp Reisnerb411b362009-09-25 16:07:19 -07002445 * @congested_data: User data
Artem Bityutskiyd97482e2012-07-25 18:12:12 +03002446 * @bdi_bits: Bits the BDI flusher thread is currently interested in
Philipp Reisnerb411b362009-09-25 16:07:19 -07002447 *
Tejun Heo44522262015-05-22 17:13:26 -04002448 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
Philipp Reisnerb411b362009-09-25 16:07:19 -07002449 */
2450static int drbd_congested(void *congested_data, int bdi_bits)
2451{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002452 struct drbd_device *device = congested_data;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002453 struct request_queue *q;
2454 char reason = '-';
2455 int r = 0;
2456
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002457 if (!may_inc_ap_bio(device)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07002458 /* DRBD has frozen IO */
2459 r = bdi_bits;
2460 reason = 'd';
2461 goto out;
2462 }
2463
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002464 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
Tejun Heo44522262015-05-22 17:13:26 -04002465 r |= (1 << WB_async_congested);
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002466 /* Without good local data, we would need to read from remote,
2467 * and that would need the worker thread as well, which is
2468 * currently blocked waiting for that usermode helper to
2469 * finish.
2470 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002471 if (!get_ldev_if_state(device, D_UP_TO_DATE))
Tejun Heo44522262015-05-22 17:13:26 -04002472 r |= (1 << WB_sync_congested);
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002473 else
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002474 put_ldev(device);
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002475 r &= bdi_bits;
2476 reason = 'c';
2477 goto out;
2478 }
2479
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002480 if (get_ldev(device)) {
2481 q = bdev_get_queue(device->ldev->backing_bdev);
Jan Karadc3b17c2017-02-02 15:56:50 +01002482 r = bdi_congested(q->backing_dev_info, bdi_bits);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002483 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002484 if (r)
2485 reason = 'b';
2486 }
2487
Tejun Heo44522262015-05-22 17:13:26 -04002488 if (bdi_bits & (1 << WB_async_congested) &&
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002489 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
Tejun Heo44522262015-05-22 17:13:26 -04002490 r |= (1 << WB_async_congested);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002491 reason = reason == 'b' ? 'a' : 'n';
2492 }
2493
2494out:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002495 device->congestion_reason = reason;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002496 return r;
2497}
2498
Philipp Reisner6699b652011-02-09 11:10:24 +01002499static void drbd_init_workqueue(struct drbd_work_queue* wq)
2500{
Philipp Reisner6699b652011-02-09 11:10:24 +01002501 spin_lock_init(&wq->q_lock);
2502 INIT_LIST_HEAD(&wq->q);
Lars Ellenberg8c0785a2011-10-19 11:50:57 +02002503 init_waitqueue_head(&wq->q_wait);
Philipp Reisner6699b652011-02-09 11:10:24 +01002504}
2505
Andreas Gruenbacherb5043c52011-07-28 15:56:02 +02002506struct completion_work {
2507 struct drbd_work w;
2508 struct completion done;
2509};
2510
2511static int w_complete(struct drbd_work *w, int cancel)
2512{
2513 struct completion_work *completion_work =
2514 container_of(w, struct completion_work, w);
2515
2516 complete(&completion_work->done);
2517 return 0;
2518}
2519
2520void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2521{
2522 struct completion_work completion_work;
2523
2524 completion_work.w.cb = w_complete;
2525 init_completion(&completion_work.done);
2526 drbd_queue_work(work_queue, &completion_work.w);
2527 wait_for_completion(&completion_work.done);
2528}
2529
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002530struct drbd_resource *drbd_find_resource(const char *name)
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002531{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002532 struct drbd_resource *resource;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002533
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002534 if (!name || !name[0])
2535 return NULL;
2536
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002537 rcu_read_lock();
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002538 for_each_resource_rcu(resource, &drbd_resources) {
2539 if (!strcmp(resource->name, name)) {
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002540 kref_get(&resource->kref);
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002541 goto found;
Philipp Reisner0ace9df2011-04-24 10:53:19 +02002542 }
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002543 }
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002544 resource = NULL;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002545found:
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002546 rcu_read_unlock();
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002547 return resource;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002548}
2549
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002550struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002551 void *peer_addr, int peer_addr_len)
2552{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002553 struct drbd_resource *resource;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002554 struct drbd_connection *connection;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002555
2556 rcu_read_lock();
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002557 for_each_resource_rcu(resource, &drbd_resources) {
2558 for_each_connection_rcu(connection, resource) {
2559 if (connection->my_addr_len == my_addr_len &&
2560 connection->peer_addr_len == peer_addr_len &&
2561 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2562 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2563 kref_get(&connection->kref);
2564 goto found;
2565 }
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002566 }
2567 }
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002568 connection = NULL;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002569found:
2570 rcu_read_unlock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002571 return connection;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002572}
2573
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002574static int drbd_alloc_socket(struct drbd_socket *socket)
2575{
2576 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2577 if (!socket->rbuf)
2578 return -ENOMEM;
Andreas Gruenbacher5a87d922011-03-24 21:17:52 +01002579 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2580 if (!socket->sbuf)
2581 return -ENOMEM;
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002582 return 0;
2583}
2584
2585static void drbd_free_socket(struct drbd_socket *socket)
2586{
Andreas Gruenbacher5a87d922011-03-24 21:17:52 +01002587 free_page((unsigned long) socket->sbuf);
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002588 free_page((unsigned long) socket->rbuf);
2589}
2590
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002591void conn_free_crypto(struct drbd_connection *connection)
Philipp Reisner91fd4da2011-04-20 17:47:29 +02002592{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002593 drbd_free_sock(connection);
Philipp Reisner1d041222011-04-22 15:20:23 +02002594
Herbert Xu9534d672016-01-24 21:19:21 +08002595 crypto_free_ahash(connection->csums_tfm);
2596 crypto_free_ahash(connection->verify_tfm);
2597 crypto_free_shash(connection->cram_hmac_tfm);
2598 crypto_free_ahash(connection->integrity_tfm);
2599 crypto_free_ahash(connection->peer_integrity_tfm);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002600 kfree(connection->int_dig_in);
2601 kfree(connection->int_dig_vv);
Philipp Reisner1d041222011-04-22 15:20:23 +02002602
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002603 connection->csums_tfm = NULL;
2604 connection->verify_tfm = NULL;
2605 connection->cram_hmac_tfm = NULL;
2606 connection->integrity_tfm = NULL;
2607 connection->peer_integrity_tfm = NULL;
2608 connection->int_dig_in = NULL;
2609 connection->int_dig_vv = NULL;
Philipp Reisner91fd4da2011-04-20 17:47:29 +02002610}
2611
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002612int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002613{
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002614 struct drbd_connection *connection;
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002615 cpumask_var_t new_cpu_mask;
2616 int err;
2617
2618 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2619 return -ENOMEM;
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002620
2621 /* silently ignore cpu mask on UP kernel */
2622 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
Andreas Gruenbacherf44d0432011-07-22 13:53:19 +02002623 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
Philipp Reisnerc5b005a2012-04-30 12:53:52 +02002624 cpumask_bits(new_cpu_mask), nr_cpu_ids);
Lars Ellenberg1e391522014-05-19 09:52:02 +02002625 if (err == -EOVERFLOW) {
2626 /* So what. mask it out. */
2627 cpumask_var_t tmp_cpu_mask;
2628 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2629 cpumask_setall(tmp_cpu_mask);
2630 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2631 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2632 res_opts->cpu_mask,
2633 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2634 nr_cpu_ids);
2635 free_cpumask_var(tmp_cpu_mask);
2636 err = 0;
2637 }
2638 }
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002639 if (err) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02002640 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002641 /* retcode = ERR_CPU_MASK_PARSE; */
2642 goto fail;
2643 }
2644 }
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002645 resource->res_opts = *res_opts;
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002646 if (cpumask_empty(new_cpu_mask))
2647 drbd_calc_cpu_mask(&new_cpu_mask);
2648 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2649 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2650 for_each_connection_rcu(connection, resource) {
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002651 connection->receiver.reset_cpu_mask = 1;
Philipp Reisner1c03e522015-03-16 15:01:00 +01002652 connection->ack_receiver.reset_cpu_mask = 1;
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002653 connection->worker.reset_cpu_mask = 1;
2654 }
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002655 }
2656 err = 0;
2657
2658fail:
2659 free_cpumask_var(new_cpu_mask);
2660 return err;
2661
2662}
2663
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002664struct drbd_resource *drbd_create_resource(const char *name)
2665{
2666 struct drbd_resource *resource;
2667
Andreas Gruenbacher6bbf53c2011-07-08 01:19:44 +02002668 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002669 if (!resource)
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002670 goto fail;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002671 resource->name = kstrdup(name, GFP_KERNEL);
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002672 if (!resource->name)
2673 goto fail_free_resource;
2674 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2675 goto fail_free_name;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002676 kref_init(&resource->kref);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002677 idr_init(&resource->devices);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002678 INIT_LIST_HEAD(&resource->connections);
Andreas Gruenbacherf6ba8632014-08-13 18:33:55 +02002679 resource->write_ordering = WO_BDEV_FLUSH;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002680 list_add_tail_rcu(&resource->resources, &drbd_resources);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002681 mutex_init(&resource->conf_update);
Lars Ellenberg9e276872014-04-28 18:43:22 +02002682 mutex_init(&resource->adm_mutex);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002683 spin_lock_init(&resource->req_lock);
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002684 drbd_debugfs_resource_add(resource);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002685 return resource;
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002686
2687fail_free_name:
2688 kfree(resource->name);
2689fail_free_resource:
2690 kfree(resource);
2691fail:
2692 return NULL;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002693}
2694
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002695/* caller must be under adm_mutex */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002696struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
Philipp Reisner21114382011-01-19 12:26:59 +01002697{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002698 struct drbd_resource *resource;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002699 struct drbd_connection *connection;
Philipp Reisner21114382011-01-19 12:26:59 +01002700
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002701 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2702 if (!connection)
Philipp Reisner21114382011-01-19 12:26:59 +01002703 return NULL;
2704
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002705 if (drbd_alloc_socket(&connection->data))
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002706 goto fail;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002707 if (drbd_alloc_socket(&connection->meta))
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002708 goto fail;
2709
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002710 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2711 if (!connection->current_epoch)
Philipp Reisner12038a32011-11-09 19:18:00 +01002712 goto fail;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002713
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002714 INIT_LIST_HEAD(&connection->transfer_log);
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002715
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002716 INIT_LIST_HEAD(&connection->current_epoch->list);
2717 connection->epochs = 1;
2718 spin_lock_init(&connection->epoch_lock);
Philipp Reisner4b0007c2011-11-09 20:12:34 +01002719
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002720 connection->send.seen_any_write_yet = false;
2721 connection->send.current_epoch_nr = 0;
2722 connection->send.current_epoch_writes = 0;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002723
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002724 resource = drbd_create_resource(name);
2725 if (!resource)
2726 goto fail;
2727
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002728 connection->cstate = C_STANDALONE;
2729 mutex_init(&connection->cstate_mutex);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002730 init_waitqueue_head(&connection->ping_wait);
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002731 idr_init(&connection->peer_devices);
Philipp Reisnerb2fb6dbe2011-01-19 13:48:44 +01002732
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002733 drbd_init_workqueue(&connection->sender_work);
2734 mutex_init(&connection->data.mutex);
2735 mutex_init(&connection->meta.mutex);
Philipp Reisner6699b652011-02-09 11:10:24 +01002736
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +02002737 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2738 connection->receiver.connection = connection;
2739 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2740 connection->worker.connection = connection;
Philipp Reisner1c03e522015-03-16 15:01:00 +01002741 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2742 connection->ack_receiver.connection = connection;
Philipp Reisner392c8802011-02-09 10:33:31 +01002743
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002744 kref_init(&connection->kref);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002745
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002746 connection->resource = resource;
Philipp Reisner21114382011-01-19 12:26:59 +01002747
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002748 if (set_resource_options(resource, res_opts))
2749 goto fail_resource;
2750
2751 kref_get(&resource->kref);
2752 list_add_tail_rcu(&connection->connections, &resource->connections);
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002753 drbd_debugfs_connection_add(connection);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002754 return connection;
Philipp Reisner21114382011-01-19 12:26:59 +01002755
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002756fail_resource:
2757 list_del(&resource->resources);
2758 drbd_free_resource(resource);
Philipp Reisner21114382011-01-19 12:26:59 +01002759fail:
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002760 kfree(connection->current_epoch);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002761 drbd_free_socket(&connection->meta);
2762 drbd_free_socket(&connection->data);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002763 kfree(connection);
Philipp Reisner21114382011-01-19 12:26:59 +01002764 return NULL;
2765}
2766
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002767void drbd_destroy_connection(struct kref *kref)
Philipp Reisner21114382011-01-19 12:26:59 +01002768{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002769 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002770 struct drbd_resource *resource = connection->resource;
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +02002771
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002772 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02002773 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002774 kfree(connection->current_epoch);
Philipp Reisner12038a32011-11-09 19:18:00 +01002775
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002776 idr_destroy(&connection->peer_devices);
Philipp Reisner21114382011-01-19 12:26:59 +01002777
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002778 drbd_free_socket(&connection->meta);
2779 drbd_free_socket(&connection->data);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002780 kfree(connection->int_dig_in);
2781 kfree(connection->int_dig_vv);
Lars Ellenbergc2258ff2014-05-14 21:35:21 +02002782 memset(connection, 0xfc, sizeof(*connection));
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002783 kfree(connection);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002784 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisner21114382011-01-19 12:26:59 +01002785}
2786
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002787static int init_submitter(struct drbd_device *device)
Lars Ellenberg113fef92013-03-22 18:14:40 -06002788{
2789 /* opencoded create_singlethread_workqueue(),
2790 * to be able to say "drbd%d", ..., minor */
Lars Ellenberg39e91a62015-03-24 10:40:26 +01002791 device->submit.wq =
2792 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002793 if (!device->submit.wq)
Lars Ellenberg113fef92013-03-22 18:14:40 -06002794 return -ENOMEM;
2795
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002796 INIT_WORK(&device->submit.worker, do_submit);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002797 INIT_LIST_HEAD(&device->submit.writes);
Lars Ellenberg113fef92013-03-22 18:14:40 -06002798 return 0;
2799}
2800
Lars Ellenberga910b122014-04-28 18:43:21 +02002801enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002802{
Lars Ellenberga910b122014-04-28 18:43:21 +02002803 struct drbd_resource *resource = adm_ctx->resource;
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002804 struct drbd_connection *connection;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002805 struct drbd_device *device;
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002806 struct drbd_peer_device *peer_device, *tmp_peer_device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002807 struct gendisk *disk;
2808 struct request_queue *q;
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002809 int id;
Lars Ellenberga910b122014-04-28 18:43:21 +02002810 int vnr = adm_ctx->volume;
Lars Ellenberg8432b312011-03-08 16:11:16 +01002811 enum drbd_ret_code err = ERR_NOMEM;
Philipp Reisner774b3052011-02-22 02:07:03 -05002812
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002813 device = minor_to_device(minor);
2814 if (device)
Andreas Gruenbacher179e20b82014-11-10 17:21:09 +01002815 return ERR_MINOR_OR_VOLUME_EXISTS;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002816
2817 /* GFP_KERNEL, we are outside of all write-out paths */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002818 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2819 if (!device)
Philipp Reisner774b3052011-02-22 02:07:03 -05002820 return ERR_NOMEM;
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002821 kref_init(&device->kref);
2822
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002823 kref_get(&resource->kref);
2824 device->resource = resource;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002825 device->minor = minor;
2826 device->vnr = vnr;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002827
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002828 drbd_init_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002829
2830 q = blk_alloc_queue(GFP_KERNEL);
2831 if (!q)
2832 goto out_no_q;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002833 device->rq_queue = q;
2834 q->queuedata = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002835
2836 disk = alloc_disk(1);
2837 if (!disk)
2838 goto out_no_disk;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002839 device->vdisk = disk;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002840
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01002841 set_disk_ro(disk, true);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002842
2843 disk->queue = q;
2844 disk->major = DRBD_MAJOR;
2845 disk->first_minor = minor;
2846 disk->fops = &drbd_ops;
2847 sprintf(disk->disk_name, "drbd%d", minor);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002848 disk->private_data = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002849
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002850 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
Philipp Reisnerb411b362009-09-25 16:07:19 -07002851 /* we have no partitions. we contain only ourselves. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002852 device->this_bdev->bd_contains = device->this_bdev;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002853
Jan Karadc3b17c2017-02-02 15:56:50 +01002854 q->backing_dev_info->congested_fn = drbd_congested;
2855 q->backing_dev_info->congested_data = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002856
Andreas Gruenbacher2f58dcf2010-12-13 17:48:19 +01002857 blk_queue_make_request(q, drbd_make_request);
Jens Axboefe8fb752016-03-30 10:09:01 -06002858 blk_queue_write_cache(q, true, true);
Philipp Reisner99432fc2011-05-20 16:39:13 +02002859 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2860 This triggers a max_bio_size message upon first attach or connect */
2861 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002862 q->queue_lock = &resource->req_lock;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002863
Lars Ellenberge37d2432014-04-01 23:53:30 +02002864 device->md_io.page = alloc_page(GFP_KERNEL);
2865 if (!device->md_io.page)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002866 goto out_no_io_page;
2867
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002868 if (drbd_bm_init(device))
Philipp Reisnerb411b362009-09-25 16:07:19 -07002869 goto out_no_bitmap;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002870 device->read_requests = RB_ROOT;
2871 device->write_requests = RB_ROOT;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002872
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002873 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2874 if (id < 0) {
Andreas Gruenbacherf221f4b2014-11-10 17:21:10 +01002875 if (id == -ENOSPC)
Andreas Gruenbacher179e20b82014-11-10 17:21:09 +01002876 err = ERR_MINOR_OR_VOLUME_EXISTS;
Lars Ellenberg8432b312011-03-08 16:11:16 +01002877 goto out_no_minor_idr;
Tejun Heo56de2102013-02-27 17:04:01 -08002878 }
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002879 kref_get(&device->kref);
2880
2881 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2882 if (id < 0) {
Andreas Gruenbacherf221f4b2014-11-10 17:21:10 +01002883 if (id == -ENOSPC)
Andreas Gruenbacher179e20b82014-11-10 17:21:09 +01002884 err = ERR_MINOR_OR_VOLUME_EXISTS;
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002885 goto out_idr_remove_minor;
2886 }
2887 kref_get(&device->kref);
Tejun Heo56de2102013-02-27 17:04:01 -08002888
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002889 INIT_LIST_HEAD(&device->peer_devices);
Lars Ellenberg4ce49262014-05-06 00:44:59 +02002890 INIT_LIST_HEAD(&device->pending_bitmap_io);
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002891 for_each_connection(connection, resource) {
2892 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2893 if (!peer_device)
2894 goto out_idr_remove_from_resource;
2895 peer_device->connection = connection;
2896 peer_device->device = device;
2897
2898 list_add(&peer_device->peer_devices, &device->peer_devices);
2899 kref_get(&device->kref);
2900
2901 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2902 if (id < 0) {
Andreas Gruenbacherf221f4b2014-11-10 17:21:10 +01002903 if (id == -ENOSPC)
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002904 err = ERR_INVALID_REQUEST;
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002905 goto out_idr_remove_from_resource;
Tejun Heo56de2102013-02-27 17:04:01 -08002906 }
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002907 kref_get(&connection->kref);
Philipp Reisner668700b2015-03-16 16:08:29 +01002908 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
Lars Ellenberg569083c2011-03-07 09:49:02 +01002909 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07002910
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002911 if (init_submitter(device)) {
Lars Ellenberg113fef92013-03-22 18:14:40 -06002912 err = ERR_NOMEM;
Lars Ellenberg113fef92013-03-22 18:14:40 -06002913 goto out_idr_remove_vol;
2914 }
2915
Philipp Reisner774b3052011-02-22 02:07:03 -05002916 add_disk(disk);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002917
Philipp Reisner2325eb62011-03-15 16:56:18 +01002918 /* inherit the connection state */
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002919 device->state.conn = first_connection(resource)->cstate;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02002920 if (device->state.conn == C_WF_REPORT_PARAMS) {
2921 for_each_peer_device(peer_device, device)
2922 drbd_connected(peer_device);
2923 }
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002924 /* move to create_peer_device() */
2925 for_each_peer_device(peer_device, device)
2926 drbd_debugfs_peer_device_add(peer_device);
2927 drbd_debugfs_device_add(device);
Philipp Reisner774b3052011-02-22 02:07:03 -05002928 return NO_ERROR;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002929
Lars Ellenberg113fef92013-03-22 18:14:40 -06002930out_idr_remove_vol:
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002931 idr_remove(&connection->peer_devices, vnr);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002932out_idr_remove_from_resource:
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002933 for_each_connection(connection, resource) {
Matthew Wilcoxd3e709e2016-12-22 13:30:22 -05002934 peer_device = idr_remove(&connection->peer_devices, vnr);
2935 if (peer_device)
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002936 kref_put(&connection->kref, drbd_destroy_connection);
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002937 }
2938 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2939 list_del(&peer_device->peer_devices);
2940 kfree(peer_device);
2941 }
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002942 idr_remove(&resource->devices, vnr);
Lars Ellenberg8432b312011-03-08 16:11:16 +01002943out_idr_remove_minor:
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002944 idr_remove(&drbd_devices, minor);
Lars Ellenberg569083c2011-03-07 09:49:02 +01002945 synchronize_rcu();
Lars Ellenberg8432b312011-03-08 16:11:16 +01002946out_no_minor_idr:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002947 drbd_bm_cleanup(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002948out_no_bitmap:
Lars Ellenberge37d2432014-04-01 23:53:30 +02002949 __free_page(device->md_io.page);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002950out_no_io_page:
2951 put_disk(disk);
2952out_no_disk:
2953 blk_cleanup_queue(q);
2954out_no_q:
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002955 kref_put(&resource->kref, drbd_destroy_resource);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002956 kfree(device);
Lars Ellenberg8432b312011-03-08 16:11:16 +01002957 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002958}
2959
Andreas Gruenbacherf82795d2011-07-03 23:32:26 +02002960void drbd_delete_device(struct drbd_device *device)
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002961{
2962 struct drbd_resource *resource = device->resource;
2963 struct drbd_connection *connection;
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002964 struct drbd_peer_device *peer_device;
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002965
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02002966 /* move to free_peer_device() */
2967 for_each_peer_device(peer_device, device)
2968 drbd_debugfs_peer_device_cleanup(peer_device);
2969 drbd_debugfs_device_cleanup(device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002970 for_each_connection(connection, resource) {
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002971 idr_remove(&connection->peer_devices, device->vnr);
Peter Zijlstrabdfafc42016-11-14 17:34:19 +01002972 kref_put(&device->kref, drbd_destroy_device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002973 }
2974 idr_remove(&resource->devices, device->vnr);
Peter Zijlstrabdfafc42016-11-14 17:34:19 +01002975 kref_put(&device->kref, drbd_destroy_device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002976 idr_remove(&drbd_devices, device_to_minor(device));
Peter Zijlstrabdfafc42016-11-14 17:34:19 +01002977 kref_put(&device->kref, drbd_destroy_device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002978 del_gendisk(device->vdisk);
2979 synchronize_rcu();
Peter Zijlstrabdfafc42016-11-14 17:34:19 +01002980 kref_put(&device->kref, drbd_destroy_device);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002981}
2982
Lars Ellenberg8ce953a2014-02-27 09:46:18 +01002983static int __init drbd_init(void)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002984{
2985 int err;
2986
Philipp Reisner2b8a90b2011-01-10 11:15:17 +01002987 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002988 pr_err("invalid minor_count (%d)\n", minor_count);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002989#ifdef MODULE
2990 return -EINVAL;
2991#else
Andreas Gruenbacher46530e82011-05-31 13:08:53 +02002992 minor_count = DRBD_MINOR_COUNT_DEF;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002993#endif
2994 }
2995
Philipp Reisnerb411b362009-09-25 16:07:19 -07002996 err = register_blkdev(DRBD_MAJOR, "drbd");
2997 if (err) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002998 pr_err("unable to register block device major %d\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07002999 DRBD_MAJOR);
3000 return err;
3001 }
3002
Philipp Reisnerb411b362009-09-25 16:07:19 -07003003 /*
3004 * allocate all necessary structs
3005 */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003006 init_waitqueue_head(&drbd_pp_wait);
3007
3008 drbd_proc = NULL; /* play safe for drbd_cleanup */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02003009 idr_init(&drbd_devices);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003010
Andreas Gruenbacher28bc3b82014-08-14 18:33:30 +02003011 mutex_init(&resources_mutex);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02003012 INIT_LIST_HEAD(&drbd_resources);
Lars Ellenberg69babf02013-10-23 10:59:15 +02003013
3014 err = drbd_genl_register();
3015 if (err) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003016 pr_err("unable to register generic netlink family\n");
Lars Ellenberg69babf02013-10-23 10:59:15 +02003017 goto fail;
3018 }
3019
Philipp Reisnerb411b362009-09-25 16:07:19 -07003020 err = drbd_create_mempools();
3021 if (err)
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01003022 goto fail;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003023
Wei Yongjun6110d702013-06-25 16:50:04 +02003024 err = -ENOMEM;
Lars Ellenberg8c484ee2010-03-11 16:47:58 +01003025 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003026 if (!drbd_proc) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003027 pr_err("unable to register proc file\n");
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01003028 goto fail;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003029 }
3030
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01003031 retry.wq = create_singlethread_workqueue("drbd-reissue");
3032 if (!retry.wq) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003033 pr_err("unable to create retry workqueue\n");
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01003034 goto fail;
3035 }
3036 INIT_WORK(&retry.worker, do_retry);
3037 spin_lock_init(&retry.lock);
3038 INIT_LIST_HEAD(&retry.writes);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003039
Lars Ellenberg4d3d5aa2014-05-02 13:19:51 +02003040 if (drbd_debugfs_init())
3041 pr_notice("failed to initialize debugfs -- will not be available\n");
3042
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003043 pr_info("initialized. "
Philipp Reisnerb411b362009-09-25 16:07:19 -07003044 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3045 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003046 pr_info("%s\n", drbd_buildtag());
3047 pr_info("registered as block device major %d\n", DRBD_MAJOR);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003048 return 0; /* Success! */
3049
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01003050fail:
Philipp Reisnerb411b362009-09-25 16:07:19 -07003051 drbd_cleanup();
3052 if (err == -ENOMEM)
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003053 pr_err("ran out of memory\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003054 else
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01003055 pr_err("initialization failure\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003056 return err;
3057}
3058
Lars Ellenbergf4188152014-05-05 23:05:47 +02003059static void drbd_free_one_sock(struct drbd_socket *ds)
3060{
3061 struct socket *s;
3062 mutex_lock(&ds->mutex);
3063 s = ds->socket;
3064 ds->socket = NULL;
3065 mutex_unlock(&ds->mutex);
3066 if (s) {
3067 /* so debugfs does not need to mutex_lock() */
3068 synchronize_rcu();
3069 kernel_sock_shutdown(s, SHUT_RDWR);
3070 sock_release(s);
3071 }
3072}
3073
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02003074void drbd_free_sock(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003075{
Lars Ellenbergf4188152014-05-05 23:05:47 +02003076 if (connection->data.socket)
3077 drbd_free_one_sock(&connection->data);
3078 if (connection->meta.socket)
3079 drbd_free_one_sock(&connection->meta);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003080}
3081
Philipp Reisnerb411b362009-09-25 16:07:19 -07003082/* meta data management */
3083
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02003084void conn_md_sync(struct drbd_connection *connection)
Philipp Reisner19fffd72012-08-28 16:48:03 +02003085{
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02003086 struct drbd_peer_device *peer_device;
Philipp Reisner19fffd72012-08-28 16:48:03 +02003087 int vnr;
3088
3089 rcu_read_lock();
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02003090 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3091 struct drbd_device *device = peer_device->device;
3092
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003093 kref_get(&device->kref);
Philipp Reisner19fffd72012-08-28 16:48:03 +02003094 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003095 drbd_md_sync(device);
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02003096 kref_put(&device->kref, drbd_destroy_device);
Philipp Reisner19fffd72012-08-28 16:48:03 +02003097 rcu_read_lock();
3098 }
3099 rcu_read_unlock();
3100}
3101
Lars Ellenbergae8bf312013-03-19 18:16:43 +01003102/* aligned 4kByte */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003103struct meta_data_on_disk {
Lars Ellenbergcccac982013-03-19 18:16:46 +01003104 u64 la_size_sect; /* last agreed size. */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003105 u64 uuid[UI_SIZE]; /* UUIDs. */
3106 u64 device_uuid;
3107 u64 reserved_u64_1;
3108 u32 flags; /* MDF */
3109 u32 magic;
3110 u32 md_size_sect;
3111 u32 al_offset; /* offset to this block */
Lars Ellenbergae8bf312013-03-19 18:16:43 +01003112 u32 al_nr_extents; /* important for restoring the AL (userspace) */
Lars Ellenbergf3990022011-03-23 14:31:09 +01003113 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003114 u32 bm_offset; /* offset to the bitmap, from here */
3115 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
Philipp Reisner99432fc2011-05-20 16:39:13 +02003116 u32 la_peer_max_bio_size; /* last peer max_bio_size */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003117
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003118 /* see al_tr_number_to_on_disk_sector() */
3119 u32 al_stripes;
3120 u32 al_stripe_size_4k;
3121
3122 u8 reserved_u8[4096 - (7*8 + 10*4)];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003123} __packed;
3124
Philipp Reisnerd752b262013-06-25 16:50:08 +02003125
3126
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003127void drbd_md_write(struct drbd_device *device, void *b)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003128{
Philipp Reisnerd752b262013-06-25 16:50:08 +02003129 struct meta_data_on_disk *buffer = b;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003130 sector_t sector;
3131 int i;
3132
Lars Ellenbergae8bf312013-03-19 18:16:43 +01003133 memset(buffer, 0, sizeof(*buffer));
Philipp Reisnerb411b362009-09-25 16:07:19 -07003134
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003135 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
Philipp Reisnerb411b362009-09-25 16:07:19 -07003136 for (i = UI_CURRENT; i < UI_SIZE; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003137 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3138 buffer->flags = cpu_to_be32(device->ldev->md.flags);
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003139 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003140
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003141 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3142 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3143 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003144 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003145 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003146
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003147 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3148 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003149
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003150 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3151 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003152
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003153 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003154 sector = device->ldev->md.md_offset;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003155
Mike Christiebb3cc852016-06-05 14:32:06 -05003156 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07003157 /* this was a try anyways ... */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003158 drbd_err(device, "meta data update failed!\n");
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003159 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003160 }
Philipp Reisnerd752b262013-06-25 16:50:08 +02003161}
3162
3163/**
3164 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003165 * @device: DRBD device.
Philipp Reisnerd752b262013-06-25 16:50:08 +02003166 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003167void drbd_md_sync(struct drbd_device *device)
Philipp Reisnerd752b262013-06-25 16:50:08 +02003168{
3169 struct meta_data_on_disk *buffer;
3170
3171 /* Don't accidentally change the DRBD meta data layout. */
3172 BUILD_BUG_ON(UI_SIZE != 4);
3173 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3174
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003175 del_timer(&device->md_sync_timer);
Philipp Reisnerd752b262013-06-25 16:50:08 +02003176 /* timer may be rearmed by drbd_md_mark_dirty() now. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003177 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
Philipp Reisnerd752b262013-06-25 16:50:08 +02003178 return;
3179
3180 /* We use here D_FAILED and not D_ATTACHING because we try to write
3181 * metadata even if we detach due to a disk failure! */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003182 if (!get_ldev_if_state(device, D_FAILED))
Philipp Reisnerd752b262013-06-25 16:50:08 +02003183 return;
3184
Lars Ellenberge37d2432014-04-01 23:53:30 +02003185 buffer = drbd_md_get_buffer(device, __func__);
Philipp Reisnerd752b262013-06-25 16:50:08 +02003186 if (!buffer)
3187 goto out;
3188
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003189 drbd_md_write(device, buffer);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003190
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003191 /* Update device->ldev->md.la_size_sect,
Philipp Reisnerb411b362009-09-25 16:07:19 -07003192 * since we updated it on metadata. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003193 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003194
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003195 drbd_md_put_buffer(device);
Philipp Reisnere1711732011-06-27 11:51:46 +02003196out:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003197 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003198}
3199
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003200static int check_activity_log_stripe_size(struct drbd_device *device,
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003201 struct meta_data_on_disk *on_disk,
3202 struct drbd_md *in_core)
3203{
3204 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3205 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3206 u64 al_size_4k;
3207
3208 /* both not set: default to old fixed size activity log */
3209 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3210 al_stripes = 1;
3211 al_stripe_size_4k = MD_32kB_SECT/8;
3212 }
3213
3214 /* some paranoia plausibility checks */
3215
3216 /* we need both values to be set */
3217 if (al_stripes == 0 || al_stripe_size_4k == 0)
3218 goto err;
3219
3220 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3221
3222 /* Upper limit of activity log area, to avoid potential overflow
3223 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3224 * than 72 * 4k blocks total only increases the amount of history,
3225 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3226 if (al_size_4k > (16 * 1024 * 1024/4))
3227 goto err;
3228
3229 /* Lower limit: we need at least 8 transaction slots (32kB)
3230 * to not break existing setups */
3231 if (al_size_4k < MD_32kB_SECT/8)
3232 goto err;
3233
3234 in_core->al_stripe_size_4k = al_stripe_size_4k;
3235 in_core->al_stripes = al_stripes;
3236 in_core->al_size_4k = al_size_4k;
3237
3238 return 0;
3239err:
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003240 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003241 al_stripes, al_stripe_size_4k);
3242 return -EINVAL;
3243}
3244
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003245static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003246{
3247 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3248 struct drbd_md *in_core = &bdev->md;
3249 s32 on_disk_al_sect;
3250 s32 on_disk_bm_sect;
3251
3252 /* The on-disk size of the activity log, calculated from offsets, and
3253 * the size of the activity log calculated from the stripe settings,
3254 * should match.
3255 * Though we could relax this a bit: it is ok, if the striped activity log
3256 * fits in the available on-disk activity log size.
3257 * Right now, that would break how resize is implemented.
3258 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3259 * of possible unused padding space in the on disk layout. */
3260 if (in_core->al_offset < 0) {
3261 if (in_core->bm_offset > in_core->al_offset)
3262 goto err;
3263 on_disk_al_sect = -in_core->al_offset;
3264 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3265 } else {
3266 if (in_core->al_offset != MD_4kB_SECT)
3267 goto err;
3268 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3269 goto err;
3270
3271 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3272 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3273 }
3274
3275 /* old fixed size meta data is exactly that: fixed. */
3276 if (in_core->meta_dev_idx >= 0) {
3277 if (in_core->md_size_sect != MD_128MB_SECT
3278 || in_core->al_offset != MD_4kB_SECT
3279 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3280 || in_core->al_stripes != 1
3281 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3282 goto err;
3283 }
3284
3285 if (capacity < in_core->md_size_sect)
3286 goto err;
3287 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3288 goto err;
3289
3290 /* should be aligned, and at least 32k */
3291 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3292 goto err;
3293
3294 /* should fit (for now: exactly) into the available on-disk space;
3295 * overflow prevention is in check_activity_log_stripe_size() above. */
3296 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3297 goto err;
3298
3299 /* again, should be aligned */
3300 if (in_core->bm_offset & 7)
3301 goto err;
3302
3303 /* FIXME check for device grow with flex external meta data? */
3304
3305 /* can the available bitmap space cover the last agreed device size? */
3306 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3307 goto err;
3308
3309 return 0;
3310
3311err:
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003312 drbd_err(device, "meta data offsets don't make sense: idx=%d "
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003313 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3314 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3315 in_core->meta_dev_idx,
3316 in_core->al_stripes, in_core->al_stripe_size_4k,
3317 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3318 (unsigned long long)in_core->la_size_sect,
3319 (unsigned long long)capacity);
3320
3321 return -EINVAL;
3322}
3323
3324
Philipp Reisnerb411b362009-09-25 16:07:19 -07003325/**
3326 * drbd_md_read() - Reads in the meta data super block
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003327 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003328 * @bdev: Device from which the meta data should be read in.
3329 *
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003330 * Return NO_ERROR on success, and an enum drbd_ret_code in case
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003331 * something goes wrong.
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003332 *
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003333 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003334 * even before @bdev is assigned to @device->ldev.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003335 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003336int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003337{
3338 struct meta_data_on_disk *buffer;
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003339 u32 magic, flags;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003340 int i, rv = NO_ERROR;
3341
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003342 if (device->state.disk != D_DISKLESS)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003343 return ERR_DISK_CONFIGURED;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003344
Lars Ellenberge37d2432014-04-01 23:53:30 +02003345 buffer = drbd_md_get_buffer(device, __func__);
Philipp Reisnere1711732011-06-27 11:51:46 +02003346 if (!buffer)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003347 return ERR_NOMEM;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003348
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003349 /* First, figure out where our meta data superblock is located,
3350 * and read it. */
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003351 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3352 bdev->md.md_offset = drbd_md_ss(bdev);
Lars Ellenbergedb5e5f2015-03-24 21:35:26 +01003353 /* Even for (flexible or indexed) external meta data,
3354 * initially restrict us to the 4k superblock for now.
3355 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3356 bdev->md.md_size_sect = 8;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003357
Mike Christiebb3cc852016-06-05 14:32:06 -05003358 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3359 REQ_OP_READ)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003360 /* NOTE: can't do normal error processing here as this is
Philipp Reisnerb411b362009-09-25 16:07:19 -07003361 called BEFORE disk is attached */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003362 drbd_err(device, "Error while reading metadata.\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003363 rv = ERR_IO_MD_DISK;
3364 goto err;
3365 }
3366
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003367 magic = be32_to_cpu(buffer->magic);
3368 flags = be32_to_cpu(buffer->flags);
3369 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3370 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3371 /* btw: that's Activity Log clean, not "all" clean. */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003372 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003373 rv = ERR_MD_UNCLEAN;
3374 goto err;
3375 }
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003376
3377 rv = ERR_MD_INVALID;
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003378 if (magic != DRBD_MD_MAGIC_08) {
Philipp Reisner43de7c82011-11-10 13:16:13 +01003379 if (magic == DRBD_MD_MAGIC_07)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003380 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003381 else
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003382 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003383 goto err;
3384 }
3385
3386 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003387 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003388 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003389 goto err;
3390 }
3391
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003392
3393 /* convert to in_core endian */
3394 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003395 for (i = UI_CURRENT; i < UI_SIZE; i++)
3396 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3397 bdev->md.flags = be32_to_cpu(buffer->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003398 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3399
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003400 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3401 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3402 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3403
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003404 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003405 goto err;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003406 if (check_offsets_and_sizes(device, bdev))
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003407 goto err;
3408
Philipp Reisnerb411b362009-09-25 16:07:19 -07003409 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003410 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003411 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003412 goto err;
3413 }
3414 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003415 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003416 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003417 goto err;
3418 }
3419
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003420 rv = NO_ERROR;
3421
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003422 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003423 if (device->state.conn < C_CONNECTED) {
Lars Ellenbergdb141b22012-06-25 19:15:58 +02003424 unsigned int peer;
Philipp Reisner99432fc2011-05-20 16:39:13 +02003425 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
Lars Ellenbergdb141b22012-06-25 19:15:58 +02003426 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003427 device->peer_max_bio_size = peer;
Philipp Reisner99432fc2011-05-20 16:39:13 +02003428 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003429 spin_unlock_irq(&device->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003430
3431 err:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003432 drbd_md_put_buffer(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003433
3434 return rv;
3435}
3436
3437/**
3438 * drbd_md_mark_dirty() - Mark meta data super block as dirty
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003439 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003440 *
3441 * Call this function if you change anything that should be written to
3442 * the meta-data super block. This function sets MD_DIRTY, and starts a
3443 * timer that ensures that within five seconds you have to call drbd_md_sync().
3444 */
Lars Ellenbergca0e6092010-10-14 15:01:21 +02003445#ifdef DEBUG
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003446void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
Lars Ellenbergee15b032010-09-03 10:00:09 +02003447{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003448 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3449 mod_timer(&device->md_sync_timer, jiffies + HZ);
3450 device->last_md_mark_dirty.line = line;
3451 device->last_md_mark_dirty.func = func;
Lars Ellenbergee15b032010-09-03 10:00:09 +02003452 }
3453}
3454#else
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003455void drbd_md_mark_dirty(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003456{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003457 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3458 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003459}
Lars Ellenbergee15b032010-09-03 10:00:09 +02003460#endif
Philipp Reisnerb411b362009-09-25 16:07:19 -07003461
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003462void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003463{
3464 int i;
3465
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003466 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003467 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003468}
3469
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003470void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003471{
3472 if (idx == UI_CURRENT) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003473 if (device->state.role == R_PRIMARY)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003474 val |= 1;
3475 else
3476 val &= ~((u64)1);
3477
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003478 drbd_set_ed_uuid(device, val);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003479 }
3480
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003481 device->ldev->md.uuid[idx] = val;
3482 drbd_md_mark_dirty(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003483}
3484
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003485void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003486{
3487 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003488 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3489 __drbd_uuid_set(device, idx, val);
3490 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003491}
Philipp Reisnerb411b362009-09-25 16:07:19 -07003492
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003493void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003494{
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003495 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003496 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3497 if (device->ldev->md.uuid[idx]) {
3498 drbd_uuid_move_history(device);
3499 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003500 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003501 __drbd_uuid_set(device, idx, val);
3502 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003503}
3504
3505/**
3506 * drbd_uuid_new_current() - Creates a new current UUID
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003507 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003508 *
3509 * Creates a new current UUID, and rotates the old current UUID into
3510 * the bitmap slot. Causes an incremental resync upon next connect.
3511 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003512void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003513{
3514 u64 val;
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003515 unsigned long long bm_uuid;
3516
3517 get_random_bytes(&val, sizeof(u64));
3518
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003519 spin_lock_irq(&device->ldev->md.uuid_lock);
3520 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003521
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003522 if (bm_uuid)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003523 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003524
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003525 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3526 __drbd_uuid_set(device, UI_CURRENT, val);
3527 spin_unlock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003528
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003529 drbd_print_uuids(device, "new current UUID");
Lars Ellenbergaaa8e2b2010-10-15 13:16:53 +02003530 /* get it to stable storage _now_ */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003531 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003532}
3533
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003534void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003535{
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003536 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003537 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003538 return;
3539
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003540 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003541 if (val == 0) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003542 drbd_uuid_move_history(device);
3543 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3544 device->ldev->md.uuid[UI_BITMAP] = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003545 } else {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003546 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003547 if (bm_uuid)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003548 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003549
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003550 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003551 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003552 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003553
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003554 drbd_md_mark_dirty(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003555}
3556
3557/**
3558 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003559 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003560 *
3561 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3562 */
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003563int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003564{
3565 int rv = -EIO;
3566
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003567 drbd_md_set_flag(device, MDF_FULL_SYNC);
3568 drbd_md_sync(device);
3569 drbd_bm_set_all(device);
3570
3571 rv = drbd_bm_write(device);
3572
3573 if (!rv) {
3574 drbd_md_clear_flag(device, MDF_FULL_SYNC);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003575 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003576 }
3577
3578 return rv;
3579}
3580
3581/**
3582 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003583 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003584 *
3585 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3586 */
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003587int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003588{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003589 drbd_resume_al(device);
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003590 drbd_bm_clear_all(device);
3591 return drbd_bm_write(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003592}
3593
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +01003594static int w_bitmap_io(struct drbd_work *w, int unused)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003595{
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02003596 struct drbd_device *device =
3597 container_of(w, struct drbd_device, bm_io_work.w);
3598 struct bm_io_work *work = &device->bm_io_work;
Lars Ellenberg02851e92010-12-16 14:47:39 +01003599 int rv = -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003600
Lars Ellenbergbca1cba2016-06-14 00:26:16 +02003601 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3602 int cnt = atomic_read(&device->ap_bio_cnt);
3603 if (cnt)
3604 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3605 cnt, work->why);
3606 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07003607
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003608 if (get_ldev(device)) {
3609 drbd_bm_lock(device, work->why, work->flags);
3610 rv = work->io_fn(device);
3611 drbd_bm_unlock(device);
3612 put_ldev(device);
Lars Ellenberg02851e92010-12-16 14:47:39 +01003613 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07003614
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003615 clear_bit_unlock(BITMAP_IO, &device->flags);
3616 wake_up(&device->misc_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003617
3618 if (work->done)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003619 work->done(device, rv);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003620
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003621 clear_bit(BITMAP_IO_QUEUED, &device->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003622 work->why = NULL;
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003623 work->flags = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003624
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +01003625 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003626}
3627
3628/**
3629 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003630 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003631 * @io_fn: IO callback to be called when bitmap IO is possible
3632 * @done: callback to be called after the bitmap IO was performed
3633 * @why: Descriptive text of the reason for doing the IO
3634 *
3635 * While IO on the bitmap happens we freeze application IO thus we ensure
3636 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3637 * called from worker context. It MUST NOT be used while a previous such
3638 * work is still pending!
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003639 *
3640 * Its worker function encloses the call of io_fn() by get_ldev() and
3641 * put_ldev().
Philipp Reisnerb411b362009-09-25 16:07:19 -07003642 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003643void drbd_queue_bitmap_io(struct drbd_device *device,
Andreas Gruenbacher54761692011-05-30 16:15:21 +02003644 int (*io_fn)(struct drbd_device *),
3645 void (*done)(struct drbd_device *, int),
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003646 char *why, enum bm_flag flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003647{
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003648 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003649
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003650 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3651 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3652 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003653 if (device->bm_io_work.why)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003654 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003655 why, device->bm_io_work.why);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003656
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003657 device->bm_io_work.io_fn = io_fn;
3658 device->bm_io_work.done = done;
3659 device->bm_io_work.why = why;
3660 device->bm_io_work.flags = flags;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003661
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003662 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003663 set_bit(BITMAP_IO, &device->flags);
Lars Ellenberg5bded4e2015-04-16 16:51:34 +02003664 /* don't wait for pending application IO if the caller indicates that
3665 * application IO does not conflict anyways. */
3666 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003667 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02003668 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3669 &device->bm_io_work.w);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003670 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003671 spin_unlock_irq(&device->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003672}
3673
3674/**
3675 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003676 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003677 * @io_fn: IO callback to be called when bitmap IO is possible
3678 * @why: Descriptive text of the reason for doing the IO
3679 *
3680 * freezes application IO while that the actual IO operations runs. This
3681 * functions MAY NOT be called from worker context.
3682 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003683int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003684 char *why, enum bm_flag flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003685{
Lars Ellenbergc0065f92016-06-14 00:26:10 +02003686 /* Only suspend io, if some operation is supposed to be locked out */
3687 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003688 int rv;
3689
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003690 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003691
Lars Ellenbergc0065f92016-06-14 00:26:10 +02003692 if (do_suspend_io)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003693 drbd_suspend_io(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003694
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003695 drbd_bm_lock(device, why, flags);
3696 rv = io_fn(device);
3697 drbd_bm_unlock(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003698
Lars Ellenbergc0065f92016-06-14 00:26:10 +02003699 if (do_suspend_io)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003700 drbd_resume_io(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003701
3702 return rv;
3703}
3704
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003705void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003706{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003707 if ((device->ldev->md.flags & flag) != flag) {
3708 drbd_md_mark_dirty(device);
3709 device->ldev->md.flags |= flag;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003710 }
3711}
3712
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003713void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003714{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003715 if ((device->ldev->md.flags & flag) != 0) {
3716 drbd_md_mark_dirty(device);
3717 device->ldev->md.flags &= ~flag;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003718 }
3719}
3720int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3721{
3722 return (bdev->md.flags & flag) != 0;
3723}
3724
3725static void md_sync_timer_fn(unsigned long data)
3726{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003727 struct drbd_device *device = (struct drbd_device *) data;
Lars Ellenbergac0acb92014-02-11 09:47:58 +01003728 drbd_device_post_work(device, MD_SYNC);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003729}
3730
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01003731const char *cmdname(enum drbd_packet cmd)
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003732{
3733 /* THINK may need to become several global tables
3734 * when we want to support more than
3735 * one PRO_VERSION */
3736 static const char *cmdnames[] = {
3737 [P_DATA] = "Data",
Lars Ellenberg9104d312016-06-14 00:26:31 +02003738 [P_WSAME] = "WriteSame",
3739 [P_TRIM] = "Trim",
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003740 [P_DATA_REPLY] = "DataReply",
3741 [P_RS_DATA_REPLY] = "RSDataReply",
3742 [P_BARRIER] = "Barrier",
3743 [P_BITMAP] = "ReportBitMap",
3744 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3745 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3746 [P_UNPLUG_REMOTE] = "UnplugRemote",
3747 [P_DATA_REQUEST] = "DataRequest",
3748 [P_RS_DATA_REQUEST] = "RSDataRequest",
3749 [P_SYNC_PARAM] = "SyncParam",
3750 [P_SYNC_PARAM89] = "SyncParam89",
3751 [P_PROTOCOL] = "ReportProtocol",
3752 [P_UUIDS] = "ReportUUIDs",
3753 [P_SIZES] = "ReportSizes",
3754 [P_STATE] = "ReportState",
3755 [P_SYNC_UUID] = "ReportSyncUUID",
3756 [P_AUTH_CHALLENGE] = "AuthChallenge",
3757 [P_AUTH_RESPONSE] = "AuthResponse",
3758 [P_PING] = "Ping",
3759 [P_PING_ACK] = "PingAck",
3760 [P_RECV_ACK] = "RecvAck",
3761 [P_WRITE_ACK] = "WriteAck",
3762 [P_RS_WRITE_ACK] = "RSWriteAck",
Lars Ellenbergd4dabbe2012-08-01 12:33:51 +02003763 [P_SUPERSEDED] = "Superseded",
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003764 [P_NEG_ACK] = "NegAck",
3765 [P_NEG_DREPLY] = "NegDReply",
3766 [P_NEG_RS_DREPLY] = "NegRSDReply",
3767 [P_BARRIER_ACK] = "BarrierAck",
3768 [P_STATE_CHG_REQ] = "StateChgRequest",
3769 [P_STATE_CHG_REPLY] = "StateChgReply",
3770 [P_OV_REQUEST] = "OVRequest",
3771 [P_OV_REPLY] = "OVReply",
3772 [P_OV_RESULT] = "OVResult",
3773 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3774 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3775 [P_COMPRESSED_BITMAP] = "CBitmap",
3776 [P_DELAY_PROBE] = "DelayProbe",
3777 [P_OUT_OF_SYNC] = "OutOfSync",
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003778 [P_RETRY_WRITE] = "RetryWrite",
Lars Ellenbergae25b332011-04-24 00:01:16 +02003779 [P_RS_CANCEL] = "RSCancel",
3780 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3781 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
Philipp Reisner036b17e2011-05-16 17:38:11 +02003782 [P_RETRY_WRITE] = "retry_write",
3783 [P_PROTOCOL_UPDATE] = "protocol_update",
Philipp Reisner700ca8c2016-06-14 00:26:13 +02003784 [P_RS_THIN_REQ] = "rs_thin_req",
3785 [P_RS_DEALLOCATED] = "rs_deallocated",
Lars Ellenbergae25b332011-04-24 00:01:16 +02003786
3787 /* enum drbd_packet, but not commands - obsoleted flags:
3788 * P_MAY_IGNORE
3789 * P_MAX_OPT_CMD
3790 */
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003791 };
3792
Lars Ellenbergae25b332011-04-24 00:01:16 +02003793 /* too big for the array: 0xfffX */
Andreas Gruenbachere5d6f332011-03-28 16:44:40 +02003794 if (cmd == P_INITIAL_META)
3795 return "InitialMeta";
3796 if (cmd == P_INITIAL_DATA)
3797 return "InitialData";
Andreas Gruenbacher60381782011-03-28 17:05:50 +02003798 if (cmd == P_CONNECTION_FEATURES)
3799 return "ConnectionFeatures";
Andreas Gruenbacher6e849ce2011-03-14 17:27:45 +01003800 if (cmd >= ARRAY_SIZE(cmdnames))
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003801 return "Unknown";
3802 return cmdnames[cmd];
3803}
3804
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003805/**
3806 * drbd_wait_misc - wait for a request to make progress
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003807 * @device: device associated with the request
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003808 * @i: the struct drbd_interval embedded in struct drbd_request or
3809 * struct drbd_peer_request
3810 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003811int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003812{
Philipp Reisner44ed1672011-04-19 17:10:19 +02003813 struct net_conf *nc;
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003814 DEFINE_WAIT(wait);
3815 long timeout;
3816
Philipp Reisner44ed1672011-04-19 17:10:19 +02003817 rcu_read_lock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02003818 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
Philipp Reisner44ed1672011-04-19 17:10:19 +02003819 if (!nc) {
3820 rcu_read_unlock();
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003821 return -ETIMEDOUT;
Philipp Reisner44ed1672011-04-19 17:10:19 +02003822 }
3823 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3824 rcu_read_unlock();
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003825
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003826 /* Indicate to wake up device->misc_wait on progress. */
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003827 i->waiting = true;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003828 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003829 spin_unlock_irq(&device->resource->req_lock);
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003830 timeout = schedule_timeout(timeout);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003831 finish_wait(&device->misc_wait, &wait);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003832 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003833 if (!timeout || device->state.conn < C_CONNECTED)
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003834 return -ETIMEDOUT;
3835 if (signal_pending(current))
3836 return -ERESTARTSYS;
3837 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003838}
3839
Andreas Gruenbacher28bc3b82014-08-14 18:33:30 +02003840void lock_all_resources(void)
3841{
3842 struct drbd_resource *resource;
3843 int __maybe_unused i = 0;
3844
3845 mutex_lock(&resources_mutex);
3846 local_irq_disable();
3847 for_each_resource(resource, &drbd_resources)
3848 spin_lock_nested(&resource->req_lock, i++);
3849}
3850
3851void unlock_all_resources(void)
3852{
3853 struct drbd_resource *resource;
3854
3855 for_each_resource(resource, &drbd_resources)
3856 spin_unlock(&resource->req_lock);
3857 local_irq_enable();
3858 mutex_unlock(&resources_mutex);
3859}
3860
Philipp Reisnerb411b362009-09-25 16:07:19 -07003861#ifdef CONFIG_DRBD_FAULT_INJECTION
3862/* Fault insertion support including random number generator shamelessly
3863 * stolen from kernel/rcutorture.c */
3864struct fault_random_state {
3865 unsigned long state;
3866 unsigned long count;
3867};
3868
3869#define FAULT_RANDOM_MULT 39916801 /* prime */
3870#define FAULT_RANDOM_ADD 479001701 /* prime */
3871#define FAULT_RANDOM_REFRESH 10000
3872
3873/*
3874 * Crude but fast random-number generator. Uses a linear congruential
3875 * generator, with occasional help from get_random_bytes().
3876 */
3877static unsigned long
3878_drbd_fault_random(struct fault_random_state *rsp)
3879{
3880 long refresh;
3881
Roel Kluin49829ea2009-12-15 22:55:44 +01003882 if (!rsp->count--) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07003883 get_random_bytes(&refresh, sizeof(refresh));
3884 rsp->state += refresh;
3885 rsp->count = FAULT_RANDOM_REFRESH;
3886 }
3887 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3888 return swahw32(rsp->state);
3889}
3890
3891static char *
3892_drbd_fault_str(unsigned int type) {
3893 static char *_faults[] = {
3894 [DRBD_FAULT_MD_WR] = "Meta-data write",
3895 [DRBD_FAULT_MD_RD] = "Meta-data read",
3896 [DRBD_FAULT_RS_WR] = "Resync write",
3897 [DRBD_FAULT_RS_RD] = "Resync read",
3898 [DRBD_FAULT_DT_WR] = "Data write",
3899 [DRBD_FAULT_DT_RD] = "Data read",
3900 [DRBD_FAULT_DT_RA] = "Data read ahead",
3901 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
Philipp Reisner6b4388a2010-04-26 14:11:45 +02003902 [DRBD_FAULT_AL_EE] = "EE allocation",
3903 [DRBD_FAULT_RECEIVE] = "receive data corruption",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003904 };
3905
3906 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3907}
3908
3909unsigned int
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003910_drbd_insert_fault(struct drbd_device *device, unsigned int type)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003911{
3912 static struct fault_random_state rrs = {0, 0};
3913
3914 unsigned int ret = (
3915 (fault_devs == 0 ||
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003916 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
Philipp Reisnerb411b362009-09-25 16:07:19 -07003917 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3918
3919 if (ret) {
3920 fault_count++;
3921
Lars Ellenberg73835062010-05-27 11:51:56 +02003922 if (__ratelimit(&drbd_ratelimit_state))
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003923 drbd_warn(device, "***Simulating %s failure\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003924 _drbd_fault_str(type));
3925 }
3926
3927 return ret;
3928}
3929#endif
3930
3931const char *drbd_buildtag(void)
3932{
3933 /* DRBD built from external sources has here a reference to the
3934 git hash of the source code. */
3935
3936 static char buildtag[38] = "\0uilt-in";
3937
3938 if (buildtag[0] == 0) {
Cong Wangbc4854b2012-04-03 14:13:36 +08003939#ifdef MODULE
3940 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3941#else
3942 buildtag[0] = 'b';
Philipp Reisnerb411b362009-09-25 16:07:19 -07003943#endif
Philipp Reisnerb411b362009-09-25 16:07:19 -07003944 }
3945
3946 return buildtag;
3947}
3948
3949module_init(drbd_init)
3950module_exit(drbd_cleanup)
3951
Philipp Reisnerb411b362009-09-25 16:07:19 -07003952EXPORT_SYMBOL(drbd_conn_str);
3953EXPORT_SYMBOL(drbd_role_str);
3954EXPORT_SYMBOL(drbd_disk_str);
3955EXPORT_SYMBOL(drbd_set_st_err_str);