blob: dcc1730dd8372850c94af81e1e221b3c398240ce [file] [log] [blame]
Chris Masondc17ff82008-01-08 15:46:30 -05001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/gfp.h>
20#include <linux/slab.h>
Chris Masond6bfde82008-04-30 13:59:35 -040021#include <linux/blkdev.h>
Chris Masonf4219502008-07-22 11:18:09 -040022#include <linux/writeback.h>
23#include <linux/pagevec.h>
Chris Masondc17ff82008-01-08 15:46:30 -050024#include "ctree.h"
25#include "transaction.h"
26#include "btrfs_inode.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040027#include "extent_io.h"
Chris Masondc17ff82008-01-08 15:46:30 -050028
Chris Masone6dcd2d2008-07-17 12:53:50 -040029static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050030{
Chris Masone6dcd2d2008-07-17 12:53:50 -040031 if (entry->file_offset + entry->len < entry->file_offset)
32 return (u64)-1;
33 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050034}
35
Chris Masond352ac62008-09-29 15:18:18 -040036/* returns NULL if the insertion worked, or it returns the node it did find
37 * in the tree
38 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040039static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
40 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050041{
42 struct rb_node ** p = &root->rb_node;
43 struct rb_node * parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040044 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050045
46 while(*p) {
47 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040048 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050049
Chris Masone6dcd2d2008-07-17 12:53:50 -040050 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050051 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040052 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050053 p = &(*p)->rb_right;
54 else
55 return parent;
56 }
57
58 rb_link_node(node, parent, p);
59 rb_insert_color(node, root);
60 return NULL;
61}
62
Chris Masond352ac62008-09-29 15:18:18 -040063/*
64 * look for a given offset in the tree, and if it can't be found return the
65 * first lesser offset
66 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040067static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
68 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050069{
70 struct rb_node * n = root->rb_node;
71 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040072 struct rb_node *test;
73 struct btrfs_ordered_extent *entry;
74 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050075
76 while(n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040077 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050078 prev = n;
79 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050080
Chris Masone6dcd2d2008-07-17 12:53:50 -040081 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050082 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040083 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050084 n = n->rb_right;
85 else
86 return n;
87 }
88 if (!prev_ret)
89 return NULL;
90
Chris Masone6dcd2d2008-07-17 12:53:50 -040091 while(prev && file_offset >= entry_end(prev_entry)) {
92 test = rb_next(prev);
93 if (!test)
94 break;
95 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
96 rb_node);
97 if (file_offset < entry_end(prev_entry))
98 break;
99
100 prev = test;
101 }
102 if (prev)
103 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
104 rb_node);
105 while(prev && file_offset < entry_end(prev_entry)) {
106 test = rb_prev(prev);
107 if (!test)
108 break;
109 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
110 rb_node);
111 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500112 }
113 *prev_ret = prev;
114 return NULL;
115}
116
Chris Masond352ac62008-09-29 15:18:18 -0400117/*
118 * helper to check if a given offset is inside a given entry
119 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400120static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500121{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400122 if (file_offset < entry->file_offset ||
123 entry->file_offset + entry->len <= file_offset)
124 return 0;
125 return 1;
126}
127
Chris Masond352ac62008-09-29 15:18:18 -0400128/*
129 * look find the first ordered struct that has this offset, otherwise
130 * the first one less than this offset
131 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400132static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
133 u64 file_offset)
134{
135 struct rb_root *root = &tree->tree;
Chris Masondc17ff82008-01-08 15:46:30 -0500136 struct rb_node *prev;
137 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400138 struct btrfs_ordered_extent *entry;
139
140 if (tree->last) {
141 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
142 rb_node);
143 if (offset_in_entry(entry, file_offset))
144 return tree->last;
145 }
146 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500147 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400148 ret = prev;
149 if (ret)
150 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500151 return ret;
152}
153
Chris Masoneb84ae02008-07-17 13:53:27 -0400154/* allocate and add a new ordered_extent into the per-inode tree.
155 * file_offset is the logical offset in the file
156 *
157 * start is the disk block number of an extent already reserved in the
158 * extent allocation tree
159 *
160 * len is the length of the extent
161 *
162 * This also sets the EXTENT_ORDERED bit on the range in the inode.
163 *
164 * The tree is given a single reference on the ordered extent that was
165 * inserted.
166 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400167int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
Yan Zheng7ea394f2008-08-05 13:05:02 -0400168 u64 start, u64 len, int nocow)
Chris Masondc17ff82008-01-08 15:46:30 -0500169{
Chris Masondc17ff82008-01-08 15:46:30 -0500170 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400171 struct rb_node *node;
172 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500173
Chris Masone6dcd2d2008-07-17 12:53:50 -0400174 tree = &BTRFS_I(inode)->ordered_tree;
175 entry = kzalloc(sizeof(*entry), GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500176 if (!entry)
177 return -ENOMEM;
178
Chris Masone6dcd2d2008-07-17 12:53:50 -0400179 mutex_lock(&tree->mutex);
180 entry->file_offset = file_offset;
181 entry->start = start;
182 entry->len = len;
Chris Mason3eaa2882008-07-24 11:57:52 -0400183 entry->inode = inode;
Yan Zheng7ea394f2008-08-05 13:05:02 -0400184 if (nocow)
185 set_bit(BTRFS_ORDERED_NOCOW, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400186
Chris Masone6dcd2d2008-07-17 12:53:50 -0400187 /* one ref for the tree */
188 atomic_set(&entry->refs, 1);
189 init_waitqueue_head(&entry->wait);
190 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400191 INIT_LIST_HEAD(&entry->root_extent_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500192
Chris Masone6dcd2d2008-07-17 12:53:50 -0400193 node = tree_insert(&tree->tree, file_offset,
194 &entry->rb_node);
195 if (node) {
Chris Mason3eaa2882008-07-24 11:57:52 -0400196 printk("warning dup entry from add_ordered_extent\n");
197 BUG();
Chris Masone6dcd2d2008-07-17 12:53:50 -0400198 }
199 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
200 entry_end(entry) - 1, GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500201
Chris Mason3eaa2882008-07-24 11:57:52 -0400202 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
203 list_add_tail(&entry->root_extent_list,
204 &BTRFS_I(inode)->root->fs_info->ordered_extents);
205 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
206
Chris Masone6dcd2d2008-07-17 12:53:50 -0400207 mutex_unlock(&tree->mutex);
208 BUG_ON(node);
209 return 0;
210}
Chris Masondc17ff82008-01-08 15:46:30 -0500211
Chris Masoneb84ae02008-07-17 13:53:27 -0400212/*
213 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400214 * when an ordered extent is finished. If the list covers more than one
215 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400216 */
Chris Mason3edf7d32008-07-18 06:17:13 -0400217int btrfs_add_ordered_sum(struct inode *inode,
218 struct btrfs_ordered_extent *entry,
219 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400220{
221 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400222
Chris Masone6dcd2d2008-07-17 12:53:50 -0400223 tree = &BTRFS_I(inode)->ordered_tree;
224 mutex_lock(&tree->mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400225 list_add_tail(&sum->list, &entry->list);
226 mutex_unlock(&tree->mutex);
227 return 0;
228}
229
Chris Masoneb84ae02008-07-17 13:53:27 -0400230/*
231 * this is used to account for finished IO across a given range
232 * of the file. The IO should not span ordered extents. If
233 * a given ordered_extent is completely done, 1 is returned, otherwise
234 * 0.
235 *
236 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
237 * to make sure this function only returns 1 once for a given ordered extent.
238 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400239int btrfs_dec_test_ordered_pending(struct inode *inode,
240 u64 file_offset, u64 io_size)
241{
242 struct btrfs_ordered_inode_tree *tree;
243 struct rb_node *node;
244 struct btrfs_ordered_extent *entry;
245 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
246 int ret;
247
248 tree = &BTRFS_I(inode)->ordered_tree;
249 mutex_lock(&tree->mutex);
250 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
251 GFP_NOFS);
252 node = tree_search(tree, file_offset);
253 if (!node) {
254 ret = 1;
255 goto out;
256 }
257
258 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
259 if (!offset_in_entry(entry, file_offset)) {
260 ret = 1;
261 goto out;
262 }
263
264 ret = test_range_bit(io_tree, entry->file_offset,
265 entry->file_offset + entry->len - 1,
266 EXTENT_ORDERED, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400267 if (ret == 0)
268 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
269out:
270 mutex_unlock(&tree->mutex);
271 return ret == 0;
272}
273
Chris Masoneb84ae02008-07-17 13:53:27 -0400274/*
275 * used to drop a reference on an ordered extent. This will free
276 * the extent if the last reference is dropped
277 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400278int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
279{
Chris Masonba1da2f2008-07-17 12:54:15 -0400280 struct list_head *cur;
281 struct btrfs_ordered_sum *sum;
282
283 if (atomic_dec_and_test(&entry->refs)) {
284 while(!list_empty(&entry->list)) {
285 cur = entry->list.next;
286 sum = list_entry(cur, struct btrfs_ordered_sum, list);
287 list_del(&sum->list);
288 kfree(sum);
289 }
Chris Masondc17ff82008-01-08 15:46:30 -0500290 kfree(entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400291 }
Chris Masondc17ff82008-01-08 15:46:30 -0500292 return 0;
293}
294
Chris Masoneb84ae02008-07-17 13:53:27 -0400295/*
296 * remove an ordered extent from the tree. No references are dropped
297 * but, anyone waiting on this extent is woken up.
298 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400299int btrfs_remove_ordered_extent(struct inode *inode,
300 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500301{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400302 struct btrfs_ordered_inode_tree *tree;
Chris Masondc17ff82008-01-08 15:46:30 -0500303 struct rb_node *node;
304
Chris Masone6dcd2d2008-07-17 12:53:50 -0400305 tree = &BTRFS_I(inode)->ordered_tree;
306 mutex_lock(&tree->mutex);
307 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500308 rb_erase(node, &tree->tree);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400309 tree->last = NULL;
310 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400311
312 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
313 list_del_init(&entry->root_extent_list);
314 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
315
Chris Masone6dcd2d2008-07-17 12:53:50 -0400316 mutex_unlock(&tree->mutex);
317 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400318 return 0;
319}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400320
Chris Masond352ac62008-09-29 15:18:18 -0400321/*
322 * wait for all the ordered extents in a root. This is done when balancing
323 * space between drives.
324 */
Yan Zheng7ea394f2008-08-05 13:05:02 -0400325int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
Chris Mason3eaa2882008-07-24 11:57:52 -0400326{
327 struct list_head splice;
328 struct list_head *cur;
329 struct btrfs_ordered_extent *ordered;
330 struct inode *inode;
331
332 INIT_LIST_HEAD(&splice);
333
334 spin_lock(&root->fs_info->ordered_extent_lock);
335 list_splice_init(&root->fs_info->ordered_extents, &splice);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400336 while (!list_empty(&splice)) {
Chris Mason3eaa2882008-07-24 11:57:52 -0400337 cur = splice.next;
338 ordered = list_entry(cur, struct btrfs_ordered_extent,
339 root_extent_list);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400340 if (nocow_only &&
341 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400342 list_move(&ordered->root_extent_list,
343 &root->fs_info->ordered_extents);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400344 cond_resched_lock(&root->fs_info->ordered_extent_lock);
345 continue;
346 }
347
Chris Mason3eaa2882008-07-24 11:57:52 -0400348 list_del_init(&ordered->root_extent_list);
349 atomic_inc(&ordered->refs);
Chris Mason3eaa2882008-07-24 11:57:52 -0400350
351 /*
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400352 * the inode may be getting freed (in sys_unlink path).
Chris Mason3eaa2882008-07-24 11:57:52 -0400353 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400354 inode = igrab(ordered->inode);
355
Chris Mason3eaa2882008-07-24 11:57:52 -0400356 spin_unlock(&root->fs_info->ordered_extent_lock);
357
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400358 if (inode) {
359 btrfs_start_ordered_extent(inode, ordered, 1);
360 btrfs_put_ordered_extent(ordered);
361 iput(inode);
362 } else {
363 btrfs_put_ordered_extent(ordered);
364 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400365
366 spin_lock(&root->fs_info->ordered_extent_lock);
367 }
368 spin_unlock(&root->fs_info->ordered_extent_lock);
369 return 0;
370}
371
Chris Masoneb84ae02008-07-17 13:53:27 -0400372/*
373 * Used to start IO or wait for a given ordered extent to finish.
374 *
375 * If wait is one, this effectively waits on page writeback for all the pages
376 * in the extent, and it waits on the io completion code to insert
377 * metadata into the btree corresponding to the extent
378 */
379void btrfs_start_ordered_extent(struct inode *inode,
380 struct btrfs_ordered_extent *entry,
381 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400382{
383 u64 start = entry->file_offset;
384 u64 end = start + entry->len - 1;
385
Chris Masoneb84ae02008-07-17 13:53:27 -0400386 /*
387 * pages in the range can be dirty, clean or writeback. We
388 * start IO on any dirty ones so the wait doesn't stall waiting
389 * for pdflush to find them
390 */
Chris Masonf4219502008-07-22 11:18:09 -0400391 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_NONE);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400392 if (wait)
393 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
394 &entry->flags));
395}
396
Chris Masoneb84ae02008-07-17 13:53:27 -0400397/*
398 * Used to wait on ordered extents across a large range of bytes.
399 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400400void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
401{
402 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400403 u64 orig_end;
404 u64 wait_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400405 struct btrfs_ordered_extent *ordered;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400406
Chris Masone5a22172008-07-18 20:42:20 -0400407 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400408 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400409 } else {
410 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400411 if (orig_end > INT_LIMIT(loff_t))
412 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400413 }
Chris Masonf4219502008-07-22 11:18:09 -0400414 wait_end = orig_end;
Chris Mason4a096752008-07-21 10:29:44 -0400415again:
Chris Masone5a22172008-07-18 20:42:20 -0400416 /* start IO across the range first to instantiate any delalloc
417 * extents
418 */
Chris Masonf4219502008-07-22 11:18:09 -0400419 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
Chris Masone5a22172008-07-18 20:42:20 -0400420
Chris Masonf4219502008-07-22 11:18:09 -0400421 btrfs_wait_on_page_writeback_range(inode->i_mapping,
422 start >> PAGE_CACHE_SHIFT,
423 orig_end >> PAGE_CACHE_SHIFT);
424
425 end = orig_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400426 while(1) {
427 ordered = btrfs_lookup_first_ordered_extent(inode, end);
428 if (!ordered) {
429 break;
430 }
Chris Masone5a22172008-07-18 20:42:20 -0400431 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400432 btrfs_put_ordered_extent(ordered);
433 break;
434 }
435 if (ordered->file_offset + ordered->len < start) {
436 btrfs_put_ordered_extent(ordered);
437 break;
438 }
Chris Masone5a22172008-07-18 20:42:20 -0400439 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400440 end = ordered->file_offset;
441 btrfs_put_ordered_extent(ordered);
Chris Masone5a22172008-07-18 20:42:20 -0400442 if (end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400443 break;
444 end--;
445 }
Chris Mason4a096752008-07-21 10:29:44 -0400446 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
447 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
448 printk("inode %lu still ordered or delalloc after wait "
449 "%llu %llu\n", inode->i_ino,
450 (unsigned long long)start,
451 (unsigned long long)orig_end);
452 goto again;
453 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400454}
455
Chris Masoneb84ae02008-07-17 13:53:27 -0400456/*
457 * find an ordered extent corresponding to file_offset. return NULL if
458 * nothing is found, otherwise take a reference on the extent and return it
459 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400460struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
461 u64 file_offset)
462{
463 struct btrfs_ordered_inode_tree *tree;
464 struct rb_node *node;
465 struct btrfs_ordered_extent *entry = NULL;
466
467 tree = &BTRFS_I(inode)->ordered_tree;
468 mutex_lock(&tree->mutex);
469 node = tree_search(tree, file_offset);
470 if (!node)
471 goto out;
472
473 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
474 if (!offset_in_entry(entry, file_offset))
475 entry = NULL;
476 if (entry)
477 atomic_inc(&entry->refs);
478out:
479 mutex_unlock(&tree->mutex);
480 return entry;
481}
482
Chris Masoneb84ae02008-07-17 13:53:27 -0400483/*
484 * lookup and return any extent before 'file_offset'. NULL is returned
485 * if none is found
486 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400487struct btrfs_ordered_extent *
488btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
489{
490 struct btrfs_ordered_inode_tree *tree;
491 struct rb_node *node;
492 struct btrfs_ordered_extent *entry = NULL;
493
494 tree = &BTRFS_I(inode)->ordered_tree;
495 mutex_lock(&tree->mutex);
496 node = tree_search(tree, file_offset);
497 if (!node)
498 goto out;
499
500 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
501 atomic_inc(&entry->refs);
502out:
503 mutex_unlock(&tree->mutex);
504 return entry;
505}
Chris Masondbe674a2008-07-17 12:54:05 -0400506
Chris Masoneb84ae02008-07-17 13:53:27 -0400507/*
508 * After an extent is done, call this to conditionally update the on disk
509 * i_size. i_size is updated to cover any fully written part of the file.
510 */
Chris Masondbe674a2008-07-17 12:54:05 -0400511int btrfs_ordered_update_i_size(struct inode *inode,
512 struct btrfs_ordered_extent *ordered)
513{
514 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
515 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
516 u64 disk_i_size;
517 u64 new_i_size;
518 u64 i_size_test;
519 struct rb_node *node;
520 struct btrfs_ordered_extent *test;
521
522 mutex_lock(&tree->mutex);
523 disk_i_size = BTRFS_I(inode)->disk_i_size;
524
525 /*
526 * if the disk i_size is already at the inode->i_size, or
527 * this ordered extent is inside the disk i_size, we're done
528 */
529 if (disk_i_size >= inode->i_size ||
530 ordered->file_offset + ordered->len <= disk_i_size) {
531 goto out;
532 }
533
534 /*
535 * we can't update the disk_isize if there are delalloc bytes
536 * between disk_i_size and this ordered extent
537 */
538 if (test_range_bit(io_tree, disk_i_size,
539 ordered->file_offset + ordered->len - 1,
540 EXTENT_DELALLOC, 0)) {
541 goto out;
542 }
543 /*
544 * walk backward from this ordered extent to disk_i_size.
545 * if we find an ordered extent then we can't update disk i_size
546 * yet
547 */
Chris Masonba1da2f2008-07-17 12:54:15 -0400548 node = &ordered->rb_node;
Chris Masondbe674a2008-07-17 12:54:05 -0400549 while(1) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400550 node = rb_prev(node);
Chris Masondbe674a2008-07-17 12:54:05 -0400551 if (!node)
552 break;
553 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
554 if (test->file_offset + test->len <= disk_i_size)
555 break;
556 if (test->file_offset >= inode->i_size)
557 break;
558 if (test->file_offset >= disk_i_size)
559 goto out;
560 }
561 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
562
563 /*
564 * at this point, we know we can safely update i_size to at least
565 * the offset from this ordered extent. But, we need to
566 * walk forward and see if ios from higher up in the file have
567 * finished.
568 */
569 node = rb_next(&ordered->rb_node);
570 i_size_test = 0;
571 if (node) {
572 /*
573 * do we have an area where IO might have finished
574 * between our ordered extent and the next one.
575 */
576 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
577 if (test->file_offset > entry_end(ordered)) {
Yan Zhengb48652c2008-08-04 23:23:47 -0400578 i_size_test = test->file_offset;
Chris Masondbe674a2008-07-17 12:54:05 -0400579 }
580 } else {
581 i_size_test = i_size_read(inode);
582 }
583
584 /*
585 * i_size_test is the end of a region after this ordered
586 * extent where there are no ordered extents. As long as there
587 * are no delalloc bytes in this area, it is safe to update
588 * disk_i_size to the end of the region.
589 */
590 if (i_size_test > entry_end(ordered) &&
Yan Zhengb48652c2008-08-04 23:23:47 -0400591 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
Chris Masondbe674a2008-07-17 12:54:05 -0400592 EXTENT_DELALLOC, 0)) {
593 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
594 }
595 BTRFS_I(inode)->disk_i_size = new_i_size;
596out:
597 mutex_unlock(&tree->mutex);
598 return 0;
599}
Chris Masonba1da2f2008-07-17 12:54:15 -0400600
Chris Masoneb84ae02008-07-17 13:53:27 -0400601/*
602 * search the ordered extents for one corresponding to 'offset' and
603 * try to find a checksum. This is used because we allow pages to
604 * be reclaimed before their checksum is actually put into the btree
605 */
Chris Masonba1da2f2008-07-17 12:54:15 -0400606int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
607{
608 struct btrfs_ordered_sum *ordered_sum;
609 struct btrfs_sector_sum *sector_sums;
610 struct btrfs_ordered_extent *ordered;
611 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
612 struct list_head *cur;
Chris Mason3edf7d32008-07-18 06:17:13 -0400613 unsigned long num_sectors;
614 unsigned long i;
615 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -0400616 int ret = 1;
Chris Masonba1da2f2008-07-17 12:54:15 -0400617
618 ordered = btrfs_lookup_ordered_extent(inode, offset);
619 if (!ordered)
620 return 1;
621
622 mutex_lock(&tree->mutex);
623 list_for_each_prev(cur, &ordered->list) {
624 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
Chris Mason3edf7d32008-07-18 06:17:13 -0400625 if (offset >= ordered_sum->file_offset) {
626 num_sectors = ordered_sum->len / sectorsize;
Chris Masoned98b562008-07-22 23:06:42 -0400627 sector_sums = ordered_sum->sums;
Chris Mason3edf7d32008-07-18 06:17:13 -0400628 for (i = 0; i < num_sectors; i++) {
629 if (sector_sums[i].offset == offset) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400630 *sum = sector_sums[i].sum;
631 ret = 0;
632 goto out;
633 }
634 }
Chris Masonba1da2f2008-07-17 12:54:15 -0400635 }
636 }
637out:
638 mutex_unlock(&tree->mutex);
Chris Mason89642222008-07-24 09:41:53 -0400639 btrfs_put_ordered_extent(ordered);
Chris Masonba1da2f2008-07-17 12:54:15 -0400640 return ret;
641}
642
Chris Masonf4219502008-07-22 11:18:09 -0400643
644/**
645 * taken from mm/filemap.c because it isn't exported
646 *
647 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
648 * @mapping: address space structure to write
649 * @start: offset in bytes where the range starts
650 * @end: offset in bytes where the range ends (inclusive)
651 * @sync_mode: enable synchronous operation
652 *
653 * Start writeback against all of a mapping's dirty pages that lie
654 * within the byte offsets <start, end> inclusive.
655 *
656 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
657 * opposed to a regular memory cleansing writeback. The difference between
658 * these two operations is that if a dirty page/buffer is encountered, it must
659 * be waited upon, and not just skipped over.
660 */
661int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
662 loff_t end, int sync_mode)
663{
664 struct writeback_control wbc = {
665 .sync_mode = sync_mode,
666 .nr_to_write = mapping->nrpages * 2,
667 .range_start = start,
668 .range_end = end,
669 .for_writepages = 1,
670 };
671 return btrfs_writepages(mapping, &wbc);
672}
673
674/**
675 * taken from mm/filemap.c because it isn't exported
676 *
677 * wait_on_page_writeback_range - wait for writeback to complete
678 * @mapping: target address_space
679 * @start: beginning page index
680 * @end: ending page index
681 *
682 * Wait for writeback to complete against pages indexed by start->end
683 * inclusive
684 */
685int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
686 pgoff_t start, pgoff_t end)
687{
688 struct pagevec pvec;
689 int nr_pages;
690 int ret = 0;
691 pgoff_t index;
692
693 if (end < start)
694 return 0;
695
696 pagevec_init(&pvec, 0);
697 index = start;
698 while ((index <= end) &&
699 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
700 PAGECACHE_TAG_WRITEBACK,
701 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
702 unsigned i;
703
704 for (i = 0; i < nr_pages; i++) {
705 struct page *page = pvec.pages[i];
706
707 /* until radix tree lookup accepts end_index */
708 if (page->index > end)
709 continue;
710
711 wait_on_page_writeback(page);
712 if (PageError(page))
713 ret = -EIO;
714 }
715 pagevec_release(&pvec);
716 cond_resched();
717 }
718
719 /* Check for outstanding write errors */
720 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
721 ret = -ENOSPC;
722 if (test_and_clear_bit(AS_EIO, &mapping->flags))
723 ret = -EIO;
724
725 return ret;
726}