blob: 0c4ef208b8b91cc54c934d25960b41522c438e54 [file] [log] [blame]
David Sterbac1d7c512018-04-03 19:23:33 +02001// SPDX-License-Identifier: GPL-2.0
Chris Masondc17ff82008-01-08 15:46:30 -05002/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
Chris Masondc17ff82008-01-08 15:46:30 -05004 */
5
Chris Masondc17ff82008-01-08 15:46:30 -05006#include <linux/slab.h>
Chris Masond6bfde82008-04-30 13:59:35 -04007#include <linux/blkdev.h>
Chris Masonf4219502008-07-22 11:18:09 -04008#include <linux/writeback.h>
Chris Masondc17ff82008-01-08 15:46:30 -05009#include "ctree.h"
10#include "transaction.h"
11#include "btrfs_inode.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040012#include "extent_io.h"
Miao Xie199c2a92013-05-15 07:48:23 +000013#include "disk-io.h"
Anand Jainebb87652016-03-10 17:26:59 +080014#include "compression.h"
Chris Masondc17ff82008-01-08 15:46:30 -050015
Miao Xie6352b912012-09-06 04:01:51 -060016static struct kmem_cache *btrfs_ordered_extent_cache;
17
Chris Masone6dcd2d2008-07-17 12:53:50 -040018static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050019{
Chris Masone6dcd2d2008-07-17 12:53:50 -040020 if (entry->file_offset + entry->len < entry->file_offset)
21 return (u64)-1;
22 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050023}
24
Chris Masond352ac62008-09-29 15:18:18 -040025/* returns NULL if the insertion worked, or it returns the node it did find
26 * in the tree
27 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040028static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
29 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050030{
Chris Masond3977122009-01-05 21:25:51 -050031 struct rb_node **p = &root->rb_node;
32 struct rb_node *parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040033 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050034
Chris Masond3977122009-01-05 21:25:51 -050035 while (*p) {
Chris Masondc17ff82008-01-08 15:46:30 -050036 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040037 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050038
Chris Masone6dcd2d2008-07-17 12:53:50 -040039 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050040 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040041 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050042 p = &(*p)->rb_right;
43 else
44 return parent;
45 }
46
47 rb_link_node(node, parent, p);
48 rb_insert_color(node, root);
49 return NULL;
50}
51
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040052static void ordered_data_tree_panic(struct inode *inode, int errno,
53 u64 offset)
54{
55 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Jeff Mahoney5d163e02016-09-20 10:05:00 -040056 btrfs_panic(fs_info, errno,
57 "Inconsistency in ordered tree at offset %llu", offset);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040058}
59
Chris Masond352ac62008-09-29 15:18:18 -040060/*
61 * look for a given offset in the tree, and if it can't be found return the
62 * first lesser offset
63 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040064static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
65 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050066{
Chris Masond3977122009-01-05 21:25:51 -050067 struct rb_node *n = root->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -050068 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040069 struct rb_node *test;
70 struct btrfs_ordered_extent *entry;
71 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050072
Chris Masond3977122009-01-05 21:25:51 -050073 while (n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040074 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050075 prev = n;
76 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050077
Chris Masone6dcd2d2008-07-17 12:53:50 -040078 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050079 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040080 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050081 n = n->rb_right;
82 else
83 return n;
84 }
85 if (!prev_ret)
86 return NULL;
87
Chris Masond3977122009-01-05 21:25:51 -050088 while (prev && file_offset >= entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040089 test = rb_next(prev);
90 if (!test)
91 break;
92 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
93 rb_node);
94 if (file_offset < entry_end(prev_entry))
95 break;
96
97 prev = test;
98 }
99 if (prev)
100 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
101 rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500102 while (prev && file_offset < entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400103 test = rb_prev(prev);
104 if (!test)
105 break;
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 rb_node);
108 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500109 }
110 *prev_ret = prev;
111 return NULL;
112}
113
Chris Masond352ac62008-09-29 15:18:18 -0400114/*
115 * helper to check if a given offset is inside a given entry
116 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400117static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500118{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400119 if (file_offset < entry->file_offset ||
120 entry->file_offset + entry->len <= file_offset)
121 return 0;
122 return 1;
123}
124
Josef Bacik4b46fce2010-05-23 11:00:55 -0400125static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
126 u64 len)
127{
128 if (file_offset + len <= entry->file_offset ||
129 entry->file_offset + entry->len <= file_offset)
130 return 0;
131 return 1;
132}
133
Chris Masond352ac62008-09-29 15:18:18 -0400134/*
135 * look find the first ordered struct that has this offset, otherwise
136 * the first one less than this offset
137 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400138static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
139 u64 file_offset)
140{
141 struct rb_root *root = &tree->tree;
Chris Masonc87fb6f2011-01-31 19:54:59 -0500142 struct rb_node *prev = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -0500143 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400144 struct btrfs_ordered_extent *entry;
145
146 if (tree->last) {
147 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
148 rb_node);
149 if (offset_in_entry(entry, file_offset))
150 return tree->last;
151 }
152 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500153 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400154 ret = prev;
155 if (ret)
156 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500157 return ret;
158}
159
Chris Masoneb84ae02008-07-17 13:53:27 -0400160/* allocate and add a new ordered_extent into the per-inode tree.
161 * file_offset is the logical offset in the file
162 *
163 * start is the disk block number of an extent already reserved in the
164 * extent allocation tree
165 *
166 * len is the length of the extent
167 *
Chris Masoneb84ae02008-07-17 13:53:27 -0400168 * The tree is given a single reference on the ordered extent that was
169 * inserted.
170 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400171static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
172 u64 start, u64 len, u64 disk_len,
Li Zefan261507a02010-12-17 14:21:50 +0800173 int type, int dio, int compress_type)
Chris Masondc17ff82008-01-08 15:46:30 -0500174{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Miao Xie199c2a92013-05-15 07:48:23 +0000176 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500177 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400178 struct rb_node *node;
179 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500180
Chris Masone6dcd2d2008-07-17 12:53:50 -0400181 tree = &BTRFS_I(inode)->ordered_tree;
Miao Xie6352b912012-09-06 04:01:51 -0600182 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500183 if (!entry)
184 return -ENOMEM;
185
Chris Masone6dcd2d2008-07-17 12:53:50 -0400186 entry->file_offset = file_offset;
187 entry->start = start;
188 entry->len = len;
Chris Masonc8b97812008-10-29 14:49:59 -0400189 entry->disk_len = disk_len;
Chris Mason8b62b722009-09-02 16:53:46 -0400190 entry->bytes_left = len;
Josef Bacik5fd02042012-05-02 14:00:54 -0400191 entry->inode = igrab(inode);
Li Zefan261507a02010-12-17 14:21:50 +0800192 entry->compress_type = compress_type;
Josef Bacik77cef2e2013-08-29 13:57:21 -0400193 entry->truncated_len = (u64)-1;
Yan Zhengd899e052008-10-30 14:25:28 -0400194 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400195 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400196
Josef Bacik4b46fce2010-05-23 11:00:55 -0400197 if (dio)
198 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
199
Chris Masone6dcd2d2008-07-17 12:53:50 -0400200 /* one ref for the tree */
Elena Reshetovae76edab2017-03-03 10:55:13 +0200201 refcount_set(&entry->refs, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400202 init_waitqueue_head(&entry->wait);
203 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400204 INIT_LIST_HEAD(&entry->root_extent_list);
Miao Xie9afab882012-10-25 09:41:36 +0000205 INIT_LIST_HEAD(&entry->work_list);
206 init_completion(&entry->completion);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400207 INIT_LIST_HEAD(&entry->log_list);
Josef Bacik50d9aa92014-11-21 14:52:38 -0500208 INIT_LIST_HEAD(&entry->trans_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500209
liubo1abe9b82011-03-24 11:18:59 +0000210 trace_btrfs_ordered_extent_add(inode, entry);
211
Josef Bacik5fd02042012-05-02 14:00:54 -0400212 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400213 node = tree_insert(&tree->tree, file_offset,
214 &entry->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400215 if (node)
216 ordered_data_tree_panic(inode, -EEXIST, file_offset);
Josef Bacik5fd02042012-05-02 14:00:54 -0400217 spin_unlock_irq(&tree->lock);
Chris Masond3977122009-01-05 21:25:51 -0500218
Miao Xie199c2a92013-05-15 07:48:23 +0000219 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400220 list_add_tail(&entry->root_extent_list,
Miao Xie199c2a92013-05-15 07:48:23 +0000221 &root->ordered_extents);
222 root->nr_ordered_extents++;
223 if (root->nr_ordered_extents == 1) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400224 spin_lock(&fs_info->ordered_root_lock);
Miao Xie199c2a92013-05-15 07:48:23 +0000225 BUG_ON(!list_empty(&root->ordered_root));
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400226 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
227 spin_unlock(&fs_info->ordered_root_lock);
Miao Xie199c2a92013-05-15 07:48:23 +0000228 }
229 spin_unlock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400230
Josef Bacik8b62f872017-10-19 14:15:55 -0400231 /*
232 * We don't need the count_max_extents here, we can assume that all of
233 * that work has been done at higher layers, so this is truly the
234 * smallest the extent is going to get.
235 */
236 spin_lock(&BTRFS_I(inode)->lock);
237 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
238 spin_unlock(&BTRFS_I(inode)->lock);
239
Chris Masone6dcd2d2008-07-17 12:53:50 -0400240 return 0;
241}
Chris Masondc17ff82008-01-08 15:46:30 -0500242
Josef Bacik4b46fce2010-05-23 11:00:55 -0400243int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
244 u64 start, u64 len, u64 disk_len, int type)
245{
246 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800247 disk_len, type, 0,
248 BTRFS_COMPRESS_NONE);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400249}
250
251int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
252 u64 start, u64 len, u64 disk_len, int type)
253{
254 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800255 disk_len, type, 1,
256 BTRFS_COMPRESS_NONE);
257}
258
259int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
260 u64 start, u64 len, u64 disk_len,
261 int type, int compress_type)
262{
263 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
264 disk_len, type, 0,
265 compress_type);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400266}
267
Chris Masoneb84ae02008-07-17 13:53:27 -0400268/*
269 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400270 * when an ordered extent is finished. If the list covers more than one
271 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400272 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100273void btrfs_add_ordered_sum(struct inode *inode,
274 struct btrfs_ordered_extent *entry,
275 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400276{
277 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400278
Chris Masone6dcd2d2008-07-17 12:53:50 -0400279 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400280 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400281 list_add_tail(&sum->list, &entry->list);
Josef Bacik5fd02042012-05-02 14:00:54 -0400282 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400283}
284
Chris Masoneb84ae02008-07-17 13:53:27 -0400285/*
286 * this is used to account for finished IO across a given range
Chris Mason163cf092010-11-28 19:56:33 -0500287 * of the file. The IO may span ordered extents. If
288 * a given ordered_extent is completely done, 1 is returned, otherwise
289 * 0.
290 *
291 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
292 * to make sure this function only returns 1 once for a given ordered extent.
293 *
294 * file_offset is updated to one byte past the range that is recorded as
295 * complete. This allows you to walk forward in the file.
296 */
297int btrfs_dec_test_first_ordered_pending(struct inode *inode,
298 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400299 u64 *file_offset, u64 io_size, int uptodate)
Chris Mason163cf092010-11-28 19:56:33 -0500300{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400301 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason163cf092010-11-28 19:56:33 -0500302 struct btrfs_ordered_inode_tree *tree;
303 struct rb_node *node;
304 struct btrfs_ordered_extent *entry = NULL;
305 int ret;
Josef Bacik5fd02042012-05-02 14:00:54 -0400306 unsigned long flags;
Chris Mason163cf092010-11-28 19:56:33 -0500307 u64 dec_end;
308 u64 dec_start;
309 u64 to_dec;
310
311 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400312 spin_lock_irqsave(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500313 node = tree_search(tree, *file_offset);
314 if (!node) {
315 ret = 1;
316 goto out;
317 }
318
319 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
320 if (!offset_in_entry(entry, *file_offset)) {
321 ret = 1;
322 goto out;
323 }
324
325 dec_start = max(*file_offset, entry->file_offset);
326 dec_end = min(*file_offset + io_size, entry->file_offset +
327 entry->len);
328 *file_offset = dec_end;
329 if (dec_start > dec_end) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400330 btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu",
331 dec_start, dec_end);
Chris Mason163cf092010-11-28 19:56:33 -0500332 }
333 to_dec = dec_end - dec_start;
334 if (to_dec > entry->bytes_left) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400335 btrfs_crit(fs_info,
336 "bad ordered accounting left %llu size %llu",
337 entry->bytes_left, to_dec);
Chris Mason163cf092010-11-28 19:56:33 -0500338 }
339 entry->bytes_left -= to_dec;
Josef Bacik5fd02042012-05-02 14:00:54 -0400340 if (!uptodate)
341 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
342
Miao Xieaf7a6502014-03-06 13:54:56 +0800343 if (entry->bytes_left == 0) {
Chris Mason163cf092010-11-28 19:56:33 -0500344 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
David Sterba093258e2018-02-26 16:15:17 +0100345 /* test_and_set_bit implies a barrier */
346 cond_wake_up_nomb(&entry->wait);
Miao Xieaf7a6502014-03-06 13:54:56 +0800347 } else {
Chris Mason163cf092010-11-28 19:56:33 -0500348 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800349 }
Chris Mason163cf092010-11-28 19:56:33 -0500350out:
351 if (!ret && cached && entry) {
352 *cached = entry;
Elena Reshetovae76edab2017-03-03 10:55:13 +0200353 refcount_inc(&entry->refs);
Chris Mason163cf092010-11-28 19:56:33 -0500354 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400355 spin_unlock_irqrestore(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500356 return ret == 0;
357}
358
359/*
360 * this is used to account for finished IO across a given range
Chris Masoneb84ae02008-07-17 13:53:27 -0400361 * of the file. The IO should not span ordered extents. If
362 * a given ordered_extent is completely done, 1 is returned, otherwise
363 * 0.
364 *
365 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
366 * to make sure this function only returns 1 once for a given ordered extent.
367 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400368int btrfs_dec_test_ordered_pending(struct inode *inode,
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000369 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400370 u64 file_offset, u64 io_size, int uptodate)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400371{
372 struct btrfs_ordered_inode_tree *tree;
373 struct rb_node *node;
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000374 struct btrfs_ordered_extent *entry = NULL;
Josef Bacik5fd02042012-05-02 14:00:54 -0400375 unsigned long flags;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400376 int ret;
377
378 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400379 spin_lock_irqsave(&tree->lock, flags);
380 if (cached && *cached) {
381 entry = *cached;
382 goto have_entry;
383 }
384
Chris Masone6dcd2d2008-07-17 12:53:50 -0400385 node = tree_search(tree, file_offset);
386 if (!node) {
387 ret = 1;
388 goto out;
389 }
390
391 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400392have_entry:
Chris Masone6dcd2d2008-07-17 12:53:50 -0400393 if (!offset_in_entry(entry, file_offset)) {
394 ret = 1;
395 goto out;
396 }
397
Chris Mason8b62b722009-09-02 16:53:46 -0400398 if (io_size > entry->bytes_left) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500399 btrfs_crit(BTRFS_I(inode)->root->fs_info,
400 "bad ordered accounting left %llu size %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200401 entry->bytes_left, io_size);
Chris Mason8b62b722009-09-02 16:53:46 -0400402 }
403 entry->bytes_left -= io_size;
Josef Bacik5fd02042012-05-02 14:00:54 -0400404 if (!uptodate)
405 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
406
Miao Xieaf7a6502014-03-06 13:54:56 +0800407 if (entry->bytes_left == 0) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400408 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
David Sterba093258e2018-02-26 16:15:17 +0100409 /* test_and_set_bit implies a barrier */
410 cond_wake_up_nomb(&entry->wait);
Miao Xieaf7a6502014-03-06 13:54:56 +0800411 } else {
Chris Mason8b62b722009-09-02 16:53:46 -0400412 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800413 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400414out:
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000415 if (!ret && cached && entry) {
416 *cached = entry;
Elena Reshetovae76edab2017-03-03 10:55:13 +0200417 refcount_inc(&entry->refs);
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000418 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400419 spin_unlock_irqrestore(&tree->lock, flags);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400420 return ret == 0;
421}
422
Chris Masoneb84ae02008-07-17 13:53:27 -0400423/*
424 * used to drop a reference on an ordered extent. This will free
425 * the extent if the last reference is dropped
426 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100427void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400428{
Chris Masonba1da2f2008-07-17 12:54:15 -0400429 struct list_head *cur;
430 struct btrfs_ordered_sum *sum;
431
liubo1abe9b82011-03-24 11:18:59 +0000432 trace_btrfs_ordered_extent_put(entry->inode, entry);
433
Elena Reshetovae76edab2017-03-03 10:55:13 +0200434 if (refcount_dec_and_test(&entry->refs)) {
Filipe Manana61de7182015-07-01 12:13:10 +0100435 ASSERT(list_empty(&entry->log_list));
436 ASSERT(list_empty(&entry->trans_list));
437 ASSERT(list_empty(&entry->root_extent_list));
438 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
Josef Bacik5fd02042012-05-02 14:00:54 -0400439 if (entry->inode)
440 btrfs_add_delayed_iput(entry->inode);
Chris Masond3977122009-01-05 21:25:51 -0500441 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400442 cur = entry->list.next;
443 sum = list_entry(cur, struct btrfs_ordered_sum, list);
444 list_del(&sum->list);
445 kfree(sum);
446 }
Miao Xie6352b912012-09-06 04:01:51 -0600447 kmem_cache_free(btrfs_ordered_extent_cache, entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400448 }
Chris Masondc17ff82008-01-08 15:46:30 -0500449}
450
Chris Masoneb84ae02008-07-17 13:53:27 -0400451/*
452 * remove an ordered extent from the tree. No references are dropped
Josef Bacik5fd02042012-05-02 14:00:54 -0400453 * and waiters are woken up.
Chris Masoneb84ae02008-07-17 13:53:27 -0400454 */
Josef Bacik5fd02042012-05-02 14:00:54 -0400455void btrfs_remove_ordered_extent(struct inode *inode,
456 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500457{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400458 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400459 struct btrfs_ordered_inode_tree *tree;
Josef Bacik8b62f872017-10-19 14:15:55 -0400460 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
461 struct btrfs_root *root = btrfs_inode->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500462 struct rb_node *node;
Josef Bacik161c35492015-09-24 16:17:39 -0400463 bool dec_pending_ordered = false;
Chris Masondc17ff82008-01-08 15:46:30 -0500464
Josef Bacik8b62f872017-10-19 14:15:55 -0400465 /* This is paired with btrfs_add_ordered_extent. */
466 spin_lock(&btrfs_inode->lock);
467 btrfs_mod_outstanding_extents(btrfs_inode, -1);
468 spin_unlock(&btrfs_inode->lock);
469 if (root != fs_info->tree_root)
Qu Wenruo43b18592017-12-12 15:34:32 +0800470 btrfs_delalloc_release_metadata(btrfs_inode, entry->len, false);
Josef Bacik8b62f872017-10-19 14:15:55 -0400471
472 tree = &btrfs_inode->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400473 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400474 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500475 rb_erase(node, &tree->tree);
Filipe Manana61de7182015-07-01 12:13:10 +0100476 RB_CLEAR_NODE(node);
Filipe David Borba Manana1b8e7e42013-11-22 18:54:58 +0000477 if (tree->last == node)
478 tree->last = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400479 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Josef Bacik161c35492015-09-24 16:17:39 -0400480 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
481 dec_pending_ordered = true;
Josef Bacik5fd02042012-05-02 14:00:54 -0400482 spin_unlock_irq(&tree->lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400483
Josef Bacik161c35492015-09-24 16:17:39 -0400484 /*
485 * The current running transaction is waiting on us, we need to let it
486 * know that we're complete and wake it up.
487 */
488 if (dec_pending_ordered) {
489 struct btrfs_transaction *trans;
490
491 /*
492 * The checks for trans are just a formality, it should be set,
493 * but if it isn't we don't want to deref/assert under the spin
494 * lock, so be nice and check if trans is set, but ASSERT() so
495 * if it isn't set a developer will notice.
496 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400497 spin_lock(&fs_info->trans_lock);
498 trans = fs_info->running_transaction;
Josef Bacik161c35492015-09-24 16:17:39 -0400499 if (trans)
Elena Reshetova9b64f572017-03-03 10:55:11 +0200500 refcount_inc(&trans->use_count);
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400501 spin_unlock(&fs_info->trans_lock);
Josef Bacik161c35492015-09-24 16:17:39 -0400502
503 ASSERT(trans);
504 if (trans) {
505 if (atomic_dec_and_test(&trans->pending_ordered))
506 wake_up(&trans->pending_wait);
507 btrfs_put_transaction(trans);
508 }
509 }
510
Miao Xie199c2a92013-05-15 07:48:23 +0000511 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400512 list_del_init(&entry->root_extent_list);
Miao Xie199c2a92013-05-15 07:48:23 +0000513 root->nr_ordered_extents--;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400514
liubo1abe9b82011-03-24 11:18:59 +0000515 trace_btrfs_ordered_extent_remove(inode, entry);
516
Miao Xie199c2a92013-05-15 07:48:23 +0000517 if (!root->nr_ordered_extents) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400518 spin_lock(&fs_info->ordered_root_lock);
Miao Xie199c2a92013-05-15 07:48:23 +0000519 BUG_ON(list_empty(&root->ordered_root));
520 list_del_init(&root->ordered_root);
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400521 spin_unlock(&fs_info->ordered_root_lock);
Miao Xie199c2a92013-05-15 07:48:23 +0000522 }
523 spin_unlock(&root->ordered_extent_lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400524 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400525}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400526
Qu Wenruod458b052014-02-28 10:46:19 +0800527static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
Miao Xie9afab882012-10-25 09:41:36 +0000528{
529 struct btrfs_ordered_extent *ordered;
530
531 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
532 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
533 complete(&ordered->completion);
534}
535
Chris Masond352ac62008-09-29 15:18:18 -0400536/*
537 * wait for all the ordered extents in a root. This is done when balancing
538 * space between drives.
539 */
Chris Mason6374e57a2017-06-23 09:48:21 -0700540u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
Filipe Manana578def72016-04-26 15:36:38 +0100541 const u64 range_start, const u64 range_len)
Chris Mason3eaa2882008-07-24 11:57:52 -0400542{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400543 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Manana578def72016-04-26 15:36:38 +0100544 LIST_HEAD(splice);
545 LIST_HEAD(skipped);
546 LIST_HEAD(works);
Miao Xie9afab882012-10-25 09:41:36 +0000547 struct btrfs_ordered_extent *ordered, *next;
Chris Mason6374e57a2017-06-23 09:48:21 -0700548 u64 count = 0;
Filipe Manana578def72016-04-26 15:36:38 +0100549 const u64 range_end = range_start + range_len;
Chris Mason3eaa2882008-07-24 11:57:52 -0400550
Miao Xie31f3d252014-03-06 13:55:02 +0800551 mutex_lock(&root->ordered_extent_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000552 spin_lock(&root->ordered_extent_lock);
553 list_splice_init(&root->ordered_extents, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800554 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000555 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
556 root_extent_list);
Filipe Manana578def72016-04-26 15:36:38 +0100557
558 if (range_end <= ordered->start ||
559 ordered->start + ordered->disk_len <= range_start) {
560 list_move_tail(&ordered->root_extent_list, &skipped);
561 cond_resched_lock(&root->ordered_extent_lock);
562 continue;
563 }
564
Miao Xie199c2a92013-05-15 07:48:23 +0000565 list_move_tail(&ordered->root_extent_list,
566 &root->ordered_extents);
Elena Reshetovae76edab2017-03-03 10:55:13 +0200567 refcount_inc(&ordered->refs);
Miao Xie199c2a92013-05-15 07:48:23 +0000568 spin_unlock(&root->ordered_extent_lock);
569
Qu Wenruoa44903a2014-02-28 10:46:09 +0800570 btrfs_init_work(&ordered->flush_work,
Liu Bo9e0af232014-08-15 23:36:53 +0800571 btrfs_flush_delalloc_helper,
Qu Wenruoa44903a2014-02-28 10:46:09 +0800572 btrfs_run_ordered_extent_work, NULL, NULL);
Miao Xie199c2a92013-05-15 07:48:23 +0000573 list_add_tail(&ordered->work_list, &works);
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400574 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
Miao Xie199c2a92013-05-15 07:48:23 +0000575
Miao Xie9afab882012-10-25 09:41:36 +0000576 cond_resched();
Miao Xie199c2a92013-05-15 07:48:23 +0000577 spin_lock(&root->ordered_extent_lock);
Chris Mason6374e57a2017-06-23 09:48:21 -0700578 if (nr != U64_MAX)
Miao Xieb0244192013-11-04 23:13:25 +0800579 nr--;
580 count++;
Chris Mason3eaa2882008-07-24 11:57:52 -0400581 }
Filipe Manana578def72016-04-26 15:36:38 +0100582 list_splice_tail(&skipped, &root->ordered_extents);
Miao Xieb0244192013-11-04 23:13:25 +0800583 list_splice_tail(&splice, &root->ordered_extents);
Miao Xie199c2a92013-05-15 07:48:23 +0000584 spin_unlock(&root->ordered_extent_lock);
Miao Xie9afab882012-10-25 09:41:36 +0000585
586 list_for_each_entry_safe(ordered, next, &works, work_list) {
587 list_del_init(&ordered->work_list);
588 wait_for_completion(&ordered->completion);
Miao Xie9afab882012-10-25 09:41:36 +0000589 btrfs_put_ordered_extent(ordered);
Miao Xie9afab882012-10-25 09:41:36 +0000590 cond_resched();
591 }
Miao Xie31f3d252014-03-06 13:55:02 +0800592 mutex_unlock(&root->ordered_extent_mutex);
Miao Xieb0244192013-11-04 23:13:25 +0800593
594 return count;
Chris Mason3eaa2882008-07-24 11:57:52 -0400595}
596
Chris Mason6374e57a2017-06-23 09:48:21 -0700597u64 btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
598 const u64 range_start, const u64 range_len)
Miao Xie199c2a92013-05-15 07:48:23 +0000599{
600 struct btrfs_root *root;
601 struct list_head splice;
Chris Mason6374e57a2017-06-23 09:48:21 -0700602 u64 total_done = 0;
603 u64 done;
Miao Xie199c2a92013-05-15 07:48:23 +0000604
605 INIT_LIST_HEAD(&splice);
606
Miao Xie8b9d83c2014-03-06 13:54:55 +0800607 mutex_lock(&fs_info->ordered_operations_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000608 spin_lock(&fs_info->ordered_root_lock);
609 list_splice_init(&fs_info->ordered_roots, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800610 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000611 root = list_first_entry(&splice, struct btrfs_root,
612 ordered_root);
613 root = btrfs_grab_fs_root(root);
614 BUG_ON(!root);
615 list_move_tail(&root->ordered_root,
616 &fs_info->ordered_roots);
617 spin_unlock(&fs_info->ordered_root_lock);
618
Filipe Manana578def72016-04-26 15:36:38 +0100619 done = btrfs_wait_ordered_extents(root, nr,
620 range_start, range_len);
Miao Xie199c2a92013-05-15 07:48:23 +0000621 btrfs_put_fs_root(root);
Filipe Mananaf0e9b7d2016-05-14 09:12:53 +0100622 total_done += done;
Miao Xie199c2a92013-05-15 07:48:23 +0000623
624 spin_lock(&fs_info->ordered_root_lock);
Chris Mason6374e57a2017-06-23 09:48:21 -0700625 if (nr != U64_MAX) {
Miao Xieb0244192013-11-04 23:13:25 +0800626 nr -= done;
Miao Xieb0244192013-11-04 23:13:25 +0800627 }
Miao Xie199c2a92013-05-15 07:48:23 +0000628 }
Miao Xie931aa872013-11-14 17:33:21 +0800629 list_splice_tail(&splice, &fs_info->ordered_roots);
Miao Xie199c2a92013-05-15 07:48:23 +0000630 spin_unlock(&fs_info->ordered_root_lock);
Miao Xie8b9d83c2014-03-06 13:54:55 +0800631 mutex_unlock(&fs_info->ordered_operations_mutex);
Filipe Mananaf0e9b7d2016-05-14 09:12:53 +0100632
633 return total_done;
Miao Xie199c2a92013-05-15 07:48:23 +0000634}
635
Chris Masoneb84ae02008-07-17 13:53:27 -0400636/*
637 * Used to start IO or wait for a given ordered extent to finish.
638 *
639 * If wait is one, this effectively waits on page writeback for all the pages
640 * in the extent, and it waits on the io completion code to insert
641 * metadata into the btree corresponding to the extent
642 */
643void btrfs_start_ordered_extent(struct inode *inode,
644 struct btrfs_ordered_extent *entry,
645 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400646{
647 u64 start = entry->file_offset;
648 u64 end = start + entry->len - 1;
649
liubo1abe9b82011-03-24 11:18:59 +0000650 trace_btrfs_ordered_extent_start(inode, entry);
651
Chris Masoneb84ae02008-07-17 13:53:27 -0400652 /*
653 * pages in the range can be dirty, clean or writeback. We
654 * start IO on any dirty ones so the wait doesn't stall waiting
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300655 * for the flusher thread to find them
Chris Masoneb84ae02008-07-17 13:53:27 -0400656 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400657 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
658 filemap_fdatawrite_range(inode->i_mapping, start, end);
Chris Masonc8b97812008-10-29 14:49:59 -0400659 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400660 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
661 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400662 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400663}
664
Chris Masoneb84ae02008-07-17 13:53:27 -0400665/*
666 * Used to wait on ordered extents across a large range of bytes.
667 */
Josef Bacik0ef8b722013-10-25 16:13:35 -0400668int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400669{
Josef Bacik0ef8b722013-10-25 16:13:35 -0400670 int ret = 0;
Filipe Manana28aeeac2015-05-05 19:03:10 +0100671 int ret_wb = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400672 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400673 u64 orig_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400674 struct btrfs_ordered_extent *ordered;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400675
Chris Masone5a22172008-07-18 20:42:20 -0400676 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400677 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400678 } else {
679 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400680 if (orig_end > INT_LIMIT(loff_t))
681 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400682 }
Josef Bacik551ebb22012-04-23 14:41:09 -0400683
Chris Masone5a22172008-07-18 20:42:20 -0400684 /* start IO across the range first to instantiate any delalloc
685 * extents
686 */
Filipe Manana728404d2014-10-10 09:43:11 +0100687 ret = btrfs_fdatawrite_range(inode, start, orig_end);
Josef Bacik0ef8b722013-10-25 16:13:35 -0400688 if (ret)
689 return ret;
Filipe Manana728404d2014-10-10 09:43:11 +0100690
Filipe Manana28aeeac2015-05-05 19:03:10 +0100691 /*
692 * If we have a writeback error don't return immediately. Wait first
693 * for any ordered extents that haven't completed yet. This is to make
694 * sure no one can dirty the same page ranges and call writepages()
695 * before the ordered extents complete - to avoid failures (-EEXIST)
696 * when adding the new ordered extents to the ordered tree.
697 */
698 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
Chris Masonf4219502008-07-22 11:18:09 -0400699
700 end = orig_end;
Chris Masond3977122009-01-05 21:25:51 -0500701 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400702 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500703 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400704 break;
Chris Masone5a22172008-07-18 20:42:20 -0400705 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400706 btrfs_put_ordered_extent(ordered);
707 break;
708 }
Filipe David Borba Mananab52abf12013-11-06 15:12:40 +0000709 if (ordered->file_offset + ordered->len <= start) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400710 btrfs_put_ordered_extent(ordered);
711 break;
712 }
Chris Masone5a22172008-07-18 20:42:20 -0400713 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400714 end = ordered->file_offset;
Josef Bacik0ef8b722013-10-25 16:13:35 -0400715 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
716 ret = -EIO;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400717 btrfs_put_ordered_extent(ordered);
Josef Bacik0ef8b722013-10-25 16:13:35 -0400718 if (ret || end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400719 break;
720 end--;
721 }
Filipe Manana28aeeac2015-05-05 19:03:10 +0100722 return ret_wb ? ret_wb : ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400723}
724
Chris Masoneb84ae02008-07-17 13:53:27 -0400725/*
726 * find an ordered extent corresponding to file_offset. return NULL if
727 * nothing is found, otherwise take a reference on the extent and return it
728 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400729struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
730 u64 file_offset)
731{
732 struct btrfs_ordered_inode_tree *tree;
733 struct rb_node *node;
734 struct btrfs_ordered_extent *entry = NULL;
735
736 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400737 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400738 node = tree_search(tree, file_offset);
739 if (!node)
740 goto out;
741
742 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
743 if (!offset_in_entry(entry, file_offset))
744 entry = NULL;
745 if (entry)
Elena Reshetovae76edab2017-03-03 10:55:13 +0200746 refcount_inc(&entry->refs);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400747out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400748 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400749 return entry;
750}
751
Josef Bacik4b46fce2010-05-23 11:00:55 -0400752/* Since the DIO code tries to lock a wide area we need to look for any ordered
753 * extents that exist in the range, rather than just the start of the range.
754 */
Nikolay Borisova776c6f2017-02-20 13:50:49 +0200755struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
756 struct btrfs_inode *inode, u64 file_offset, u64 len)
Josef Bacik4b46fce2010-05-23 11:00:55 -0400757{
758 struct btrfs_ordered_inode_tree *tree;
759 struct rb_node *node;
760 struct btrfs_ordered_extent *entry = NULL;
761
Nikolay Borisova776c6f2017-02-20 13:50:49 +0200762 tree = &inode->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400763 spin_lock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400764 node = tree_search(tree, file_offset);
765 if (!node) {
766 node = tree_search(tree, file_offset + len);
767 if (!node)
768 goto out;
769 }
770
771 while (1) {
772 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
773 if (range_overlaps(entry, file_offset, len))
774 break;
775
776 if (entry->file_offset >= file_offset + len) {
777 entry = NULL;
778 break;
779 }
780 entry = NULL;
781 node = rb_next(node);
782 if (!node)
783 break;
784 }
785out:
786 if (entry)
Elena Reshetovae76edab2017-03-03 10:55:13 +0200787 refcount_inc(&entry->refs);
Josef Bacik5fd02042012-05-02 14:00:54 -0400788 spin_unlock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400789 return entry;
790}
791
Chris Masoneb84ae02008-07-17 13:53:27 -0400792/*
793 * lookup and return any extent before 'file_offset'. NULL is returned
794 * if none is found
795 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400796struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500797btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400798{
799 struct btrfs_ordered_inode_tree *tree;
800 struct rb_node *node;
801 struct btrfs_ordered_extent *entry = NULL;
802
803 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400804 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400805 node = tree_search(tree, file_offset);
806 if (!node)
807 goto out;
808
809 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Elena Reshetovae76edab2017-03-03 10:55:13 +0200810 refcount_inc(&entry->refs);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400811out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400812 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400813 return entry;
814}
Chris Masondbe674a2008-07-17 12:54:05 -0400815
Chris Masoneb84ae02008-07-17 13:53:27 -0400816/*
817 * After an extent is done, call this to conditionally update the on disk
818 * i_size. i_size is updated to cover any fully written part of the file.
819 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000820int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
Chris Masondbe674a2008-07-17 12:54:05 -0400821 struct btrfs_ordered_extent *ordered)
822{
823 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Masondbe674a2008-07-17 12:54:05 -0400824 u64 disk_i_size;
825 u64 new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +0000826 u64 i_size = i_size_read(inode);
Chris Masondbe674a2008-07-17 12:54:05 -0400827 struct rb_node *node;
Yan, Zhengc2167752009-11-12 09:34:21 +0000828 struct rb_node *prev = NULL;
Chris Masondbe674a2008-07-17 12:54:05 -0400829 struct btrfs_ordered_extent *test;
Yan, Zhengc2167752009-11-12 09:34:21 +0000830 int ret = 1;
Wang Xiaoguangc0d2f612016-06-22 09:57:01 +0800831 u64 orig_offset = offset;
Yan, Zhengc2167752009-11-12 09:34:21 +0000832
Josef Bacik5fd02042012-05-02 14:00:54 -0400833 spin_lock_irq(&tree->lock);
Josef Bacik77cef2e2013-08-29 13:57:21 -0400834 if (ordered) {
835 offset = entry_end(ordered);
836 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
837 offset = min(offset,
838 ordered->file_offset +
839 ordered->truncated_len);
840 } else {
Jeff Mahoneyda170662016-06-15 09:22:56 -0400841 offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
Josef Bacik77cef2e2013-08-29 13:57:21 -0400842 }
Chris Masondbe674a2008-07-17 12:54:05 -0400843 disk_i_size = BTRFS_I(inode)->disk_i_size;
844
Liu Bo19fd2df2016-12-01 13:01:02 -0800845 /*
846 * truncate file.
847 * If ordered is not NULL, then this is called from endio and
848 * disk_i_size will be updated by either truncate itself or any
849 * in-flight IOs which are inside the disk_i_size.
850 *
851 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
852 * fails somehow, we need to make sure we have a precise disk_i_size by
853 * updating it as usual.
854 *
855 */
856 if (!ordered && disk_i_size > i_size) {
Wang Xiaoguangc0d2f612016-06-22 09:57:01 +0800857 BTRFS_I(inode)->disk_i_size = orig_offset;
Yan, Zhengc2167752009-11-12 09:34:21 +0000858 ret = 0;
859 goto out;
860 }
861
Chris Masondbe674a2008-07-17 12:54:05 -0400862 /*
863 * if the disk i_size is already at the inode->i_size, or
864 * this ordered extent is inside the disk i_size, we're done
865 */
Josef Bacik5d1f4022013-01-30 14:17:31 -0500866 if (disk_i_size == i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400867 goto out;
Josef Bacik5d1f4022013-01-30 14:17:31 -0500868
869 /*
870 * We still need to update disk_i_size if outstanding_isize is greater
871 * than disk_i_size.
872 */
873 if (offset <= disk_i_size &&
874 (!ordered || ordered->outstanding_isize <= disk_i_size))
875 goto out;
Chris Masondbe674a2008-07-17 12:54:05 -0400876
877 /*
Chris Masondbe674a2008-07-17 12:54:05 -0400878 * walk backward from this ordered extent to disk_i_size.
879 * if we find an ordered extent then we can't update disk i_size
880 * yet
881 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000882 if (ordered) {
883 node = rb_prev(&ordered->rb_node);
884 } else {
885 prev = tree_search(tree, offset);
886 /*
887 * we insert file extents without involving ordered struct,
888 * so there should be no ordered struct cover this offset
889 */
890 if (prev) {
891 test = rb_entry(prev, struct btrfs_ordered_extent,
892 rb_node);
893 BUG_ON(offset_in_entry(test, offset));
894 }
895 node = prev;
896 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400897 for (; node; node = rb_prev(node)) {
Chris Masondbe674a2008-07-17 12:54:05 -0400898 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400899
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -0800900 /* We treat this entry as if it doesn't exist */
Josef Bacik5fd02042012-05-02 14:00:54 -0400901 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
902 continue;
Liu Bo62c821a2016-12-13 12:51:51 -0800903
904 if (entry_end(test) <= disk_i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400905 break;
Yan, Zhengc2167752009-11-12 09:34:21 +0000906 if (test->file_offset >= i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400907 break;
Liu Bo62c821a2016-12-13 12:51:51 -0800908
909 /*
910 * We don't update disk_i_size now, so record this undealt
911 * i_size. Or we will not know the real i_size.
912 */
913 if (test->outstanding_isize < offset)
914 test->outstanding_isize = offset;
915 if (ordered &&
916 ordered->outstanding_isize > test->outstanding_isize)
917 test->outstanding_isize = ordered->outstanding_isize;
918 goto out;
Chris Masondbe674a2008-07-17 12:54:05 -0400919 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000920 new_i_size = min_t(u64, offset, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -0400921
922 /*
Miao Xieb9a8cc52012-09-06 04:01:21 -0600923 * Some ordered extents may completed before the current one, and
924 * we hold the real i_size in ->outstanding_isize.
Chris Masondbe674a2008-07-17 12:54:05 -0400925 */
Miao Xieb9a8cc52012-09-06 04:01:21 -0600926 if (ordered && ordered->outstanding_isize > new_i_size)
927 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -0400928 BTRFS_I(inode)->disk_i_size = new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +0000929 ret = 0;
Chris Masondbe674a2008-07-17 12:54:05 -0400930out:
Yan, Zhengc2167752009-11-12 09:34:21 +0000931 /*
Josef Bacik5fd02042012-05-02 14:00:54 -0400932 * We need to do this because we can't remove ordered extents until
933 * after the i_disk_size has been updated and then the inode has been
934 * updated to reflect the change, so we need to tell anybody who finds
935 * this ordered extent that we've already done all the real work, we
936 * just haven't completed all the other work.
Yan, Zhengc2167752009-11-12 09:34:21 +0000937 */
938 if (ordered)
Josef Bacik5fd02042012-05-02 14:00:54 -0400939 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
940 spin_unlock_irq(&tree->lock);
Yan, Zhengc2167752009-11-12 09:34:21 +0000941 return ret;
Chris Masondbe674a2008-07-17 12:54:05 -0400942}
Chris Masonba1da2f2008-07-17 12:54:15 -0400943
Chris Masoneb84ae02008-07-17 13:53:27 -0400944/*
945 * search the ordered extents for one corresponding to 'offset' and
946 * try to find a checksum. This is used because we allow pages to
947 * be reclaimed before their checksum is actually put into the btree
948 */
Chris Masond20f7042008-12-08 16:58:54 -0500949int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
Miao Xiee4100d92013-04-05 07:20:56 +0000950 u32 *sum, int len)
Chris Masonba1da2f2008-07-17 12:54:15 -0400951{
952 struct btrfs_ordered_sum *ordered_sum;
Chris Masonba1da2f2008-07-17 12:54:15 -0400953 struct btrfs_ordered_extent *ordered;
954 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -0400955 unsigned long num_sectors;
956 unsigned long i;
Jeff Mahoneyda170662016-06-15 09:22:56 -0400957 u32 sectorsize = btrfs_inode_sectorsize(inode);
Miao Xiee4100d92013-04-05 07:20:56 +0000958 int index = 0;
Chris Masonba1da2f2008-07-17 12:54:15 -0400959
960 ordered = btrfs_lookup_ordered_extent(inode, offset);
961 if (!ordered)
Miao Xiee4100d92013-04-05 07:20:56 +0000962 return 0;
Chris Masonba1da2f2008-07-17 12:54:15 -0400963
Josef Bacik5fd02042012-05-02 14:00:54 -0400964 spin_lock_irq(&tree->lock);
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500965 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Miao Xiee4100d92013-04-05 07:20:56 +0000966 if (disk_bytenr >= ordered_sum->bytenr &&
967 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
968 i = (disk_bytenr - ordered_sum->bytenr) >>
969 inode->i_sb->s_blocksize_bits;
Miao Xiee4100d92013-04-05 07:20:56 +0000970 num_sectors = ordered_sum->len >>
971 inode->i_sb->s_blocksize_bits;
Miao Xief51a4a12013-06-19 10:36:09 +0800972 num_sectors = min_t(int, len - index, num_sectors - i);
973 memcpy(sum + index, ordered_sum->sums + i,
974 num_sectors);
975
976 index += (int)num_sectors;
977 if (index == len)
978 goto out;
979 disk_bytenr += num_sectors * sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -0400980 }
981 }
982out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400983 spin_unlock_irq(&tree->lock);
Chris Mason89642222008-07-24 09:41:53 -0400984 btrfs_put_ordered_extent(ordered);
Miao Xiee4100d92013-04-05 07:20:56 +0000985 return index;
Chris Masonba1da2f2008-07-17 12:54:15 -0400986}
987
Miao Xie6352b912012-09-06 04:01:51 -0600988int __init ordered_data_init(void)
989{
990 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
991 sizeof(struct btrfs_ordered_extent), 0,
Nikolay Borisovfba4b692016-06-23 21:17:08 +0300992 SLAB_MEM_SPREAD,
Miao Xie6352b912012-09-06 04:01:51 -0600993 NULL);
994 if (!btrfs_ordered_extent_cache)
995 return -ENOMEM;
Miao Xie25287e02012-10-25 09:31:03 +0000996
Miao Xie6352b912012-09-06 04:01:51 -0600997 return 0;
998}
999
David Sterbae67c7182018-02-19 17:24:18 +01001000void __cold ordered_data_exit(void)
Miao Xie6352b912012-09-06 04:01:51 -06001001{
Kinglong Mee5598e902016-01-29 21:36:35 +08001002 kmem_cache_destroy(btrfs_ordered_extent_cache);
Miao Xie6352b912012-09-06 04:01:51 -06001003}