blob: 15e5a1cd8764a1d11447408558fddd4163962de0 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason39279cc2007-06-12 06:35:45 -040019#include <linux/fs.h>
20#include <linux/pagemap.h>
21#include <linux/highmem.h>
22#include <linux/time.h>
23#include <linux/init.h>
24#include <linux/string.h>
Chris Mason39279cc2007-06-12 06:35:45 -040025#include <linux/backing-dev.h>
26#include <linux/mpage.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010027#include <linux/falloc.h>
Chris Mason39279cc2007-06-12 06:35:45 -040028#include <linux/swap.h>
29#include <linux/writeback.h>
30#include <linux/statfs.h>
31#include <linux/compat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/slab.h>
Chris Mason39279cc2007-06-12 06:35:45 -040033#include "ctree.h"
34#include "disk-io.h"
35#include "transaction.h"
36#include "btrfs_inode.h"
37#include "ioctl.h"
38#include "print-tree.h"
Chris Masone02119d2008-09-05 16:13:11 -040039#include "tree-log.h"
40#include "locking.h"
Jeff Mahoney12fa8ec2008-05-02 15:03:58 -040041#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040042
Chris Mason4cb53002011-05-24 15:35:30 -040043/*
44 * when auto defrag is enabled we
45 * queue up these defrag structs to remember which
46 * inodes need defragging passes
47 */
48struct inode_defrag {
49 struct rb_node rb_node;
50 /* objectid */
51 u64 ino;
52 /*
53 * transid where the defrag was added, we search for
54 * extents newer than this
55 */
56 u64 transid;
57
58 /* root objectid */
59 u64 root;
60
61 /* last offset we were able to defrag */
62 u64 last_offset;
63
64 /* if we've wrapped around back to zero once already */
65 int cycled;
66};
67
68/* pop a record for an inode into the defrag tree. The lock
69 * must be held already
70 *
71 * If you're inserting a record for an older transid than an
72 * existing record, the transid already in the tree is lowered
73 *
74 * If an existing record is found the defrag item you
75 * pass in is freed
76 */
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +000077static void __btrfs_add_inode_defrag(struct inode *inode,
Chris Mason4cb53002011-05-24 15:35:30 -040078 struct inode_defrag *defrag)
79{
80 struct btrfs_root *root = BTRFS_I(inode)->root;
81 struct inode_defrag *entry;
82 struct rb_node **p;
83 struct rb_node *parent = NULL;
84
85 p = &root->fs_info->defrag_inodes.rb_node;
86 while (*p) {
87 parent = *p;
88 entry = rb_entry(parent, struct inode_defrag, rb_node);
89
90 if (defrag->ino < entry->ino)
91 p = &parent->rb_left;
92 else if (defrag->ino > entry->ino)
93 p = &parent->rb_right;
94 else {
95 /* if we're reinserting an entry for
96 * an old defrag run, make sure to
97 * lower the transid of our existing record
98 */
99 if (defrag->transid < entry->transid)
100 entry->transid = defrag->transid;
101 if (defrag->last_offset > entry->last_offset)
102 entry->last_offset = defrag->last_offset;
103 goto exists;
104 }
105 }
106 BTRFS_I(inode)->in_defrag = 1;
107 rb_link_node(&defrag->rb_node, parent, p);
108 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000109 return;
Chris Mason4cb53002011-05-24 15:35:30 -0400110
111exists:
112 kfree(defrag);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000113 return;
Chris Mason4cb53002011-05-24 15:35:30 -0400114
115}
116
117/*
118 * insert a defrag record for this inode if auto defrag is
119 * enabled
120 */
121int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
122 struct inode *inode)
123{
124 struct btrfs_root *root = BTRFS_I(inode)->root;
125 struct inode_defrag *defrag;
Chris Mason4cb53002011-05-24 15:35:30 -0400126 u64 transid;
127
128 if (!btrfs_test_opt(root, AUTO_DEFRAG))
129 return 0;
130
David Sterba7841cb22011-05-31 18:07:27 +0200131 if (btrfs_fs_closing(root->fs_info))
Chris Mason4cb53002011-05-24 15:35:30 -0400132 return 0;
133
134 if (BTRFS_I(inode)->in_defrag)
135 return 0;
136
137 if (trans)
138 transid = trans->transid;
139 else
140 transid = BTRFS_I(inode)->root->last_trans;
141
142 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
143 if (!defrag)
144 return -ENOMEM;
145
David Sterbaa4689d22011-05-31 17:08:14 +0000146 defrag->ino = btrfs_ino(inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400147 defrag->transid = transid;
148 defrag->root = root->root_key.objectid;
149
150 spin_lock(&root->fs_info->defrag_inodes_lock);
151 if (!BTRFS_I(inode)->in_defrag)
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000152 __btrfs_add_inode_defrag(inode, defrag);
Dan Carpenterf4ac9042011-08-05 14:19:00 +0000153 else
154 kfree(defrag);
Chris Mason4cb53002011-05-24 15:35:30 -0400155 spin_unlock(&root->fs_info->defrag_inodes_lock);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000156 return 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400157}
158
159/*
160 * must be called with the defrag_inodes lock held
161 */
162struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino,
163 struct rb_node **next)
164{
165 struct inode_defrag *entry = NULL;
166 struct rb_node *p;
167 struct rb_node *parent = NULL;
168
169 p = info->defrag_inodes.rb_node;
170 while (p) {
171 parent = p;
172 entry = rb_entry(parent, struct inode_defrag, rb_node);
173
174 if (ino < entry->ino)
175 p = parent->rb_left;
176 else if (ino > entry->ino)
177 p = parent->rb_right;
178 else
179 return entry;
180 }
181
182 if (next) {
183 while (parent && ino > entry->ino) {
184 parent = rb_next(parent);
185 entry = rb_entry(parent, struct inode_defrag, rb_node);
186 }
187 *next = parent;
188 }
189 return NULL;
190}
191
192/*
193 * run through the list of inodes in the FS that need
194 * defragging
195 */
196int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
197{
198 struct inode_defrag *defrag;
199 struct btrfs_root *inode_root;
200 struct inode *inode;
201 struct rb_node *n;
202 struct btrfs_key key;
203 struct btrfs_ioctl_defrag_range_args range;
204 u64 first_ino = 0;
205 int num_defrag;
206 int defrag_batch = 1024;
207
208 memset(&range, 0, sizeof(range));
209 range.len = (u64)-1;
210
211 atomic_inc(&fs_info->defrag_running);
212 spin_lock(&fs_info->defrag_inodes_lock);
213 while(1) {
214 n = NULL;
215
216 /* find an inode to defrag */
217 defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n);
218 if (!defrag) {
219 if (n)
220 defrag = rb_entry(n, struct inode_defrag, rb_node);
221 else if (first_ino) {
222 first_ino = 0;
223 continue;
224 } else {
225 break;
226 }
227 }
228
229 /* remove it from the rbtree */
230 first_ino = defrag->ino + 1;
231 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
232
David Sterba7841cb22011-05-31 18:07:27 +0200233 if (btrfs_fs_closing(fs_info))
Chris Mason4cb53002011-05-24 15:35:30 -0400234 goto next_free;
235
236 spin_unlock(&fs_info->defrag_inodes_lock);
237
238 /* get the inode */
239 key.objectid = defrag->root;
240 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
241 key.offset = (u64)-1;
242 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
243 if (IS_ERR(inode_root))
244 goto next;
245
246 key.objectid = defrag->ino;
247 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
248 key.offset = 0;
249
250 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
251 if (IS_ERR(inode))
252 goto next;
253
254 /* do a chunk of defrag */
255 BTRFS_I(inode)->in_defrag = 0;
256 range.start = defrag->last_offset;
257 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
258 defrag_batch);
259 /*
260 * if we filled the whole defrag batch, there
261 * must be more work to do. Queue this defrag
262 * again
263 */
264 if (num_defrag == defrag_batch) {
265 defrag->last_offset = range.start;
266 __btrfs_add_inode_defrag(inode, defrag);
267 /*
268 * we don't want to kfree defrag, we added it back to
269 * the rbtree
270 */
271 defrag = NULL;
272 } else if (defrag->last_offset && !defrag->cycled) {
273 /*
274 * we didn't fill our defrag batch, but
275 * we didn't start at zero. Make sure we loop
276 * around to the start of the file.
277 */
278 defrag->last_offset = 0;
279 defrag->cycled = 1;
280 __btrfs_add_inode_defrag(inode, defrag);
281 defrag = NULL;
282 }
283
284 iput(inode);
285next:
286 spin_lock(&fs_info->defrag_inodes_lock);
287next_free:
288 kfree(defrag);
289 }
290 spin_unlock(&fs_info->defrag_inodes_lock);
291
292 atomic_dec(&fs_info->defrag_running);
293
294 /*
295 * during unmount, we use the transaction_wait queue to
296 * wait for the defragger to stop
297 */
298 wake_up(&fs_info->transaction_wait);
299 return 0;
300}
Chris Mason39279cc2007-06-12 06:35:45 -0400301
Chris Masond352ac62008-09-29 15:18:18 -0400302/* simple helper to fault in pages and copy. This should go away
303 * and be replaced with calls into generic code.
304 */
Chris Masond3977122009-01-05 21:25:51 -0500305static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -0500306 size_t write_bytes,
Chris Masona1b32a52008-09-05 16:09:51 -0400307 struct page **prepared_pages,
Josef Bacik11c65dc2010-05-23 11:07:21 -0400308 struct iov_iter *i)
Chris Mason39279cc2007-06-12 06:35:45 -0400309{
Xin Zhong914ee292010-12-09 09:30:14 +0000310 size_t copied = 0;
Josef Bacikd0215f32011-01-25 14:57:24 -0500311 size_t total_copied = 0;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400312 int pg = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400313 int offset = pos & (PAGE_CACHE_SIZE - 1);
314
Josef Bacik11c65dc2010-05-23 11:07:21 -0400315 while (write_bytes > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -0400316 size_t count = min_t(size_t,
317 PAGE_CACHE_SIZE - offset, write_bytes);
Josef Bacik11c65dc2010-05-23 11:07:21 -0400318 struct page *page = prepared_pages[pg];
Xin Zhong914ee292010-12-09 09:30:14 +0000319 /*
320 * Copy data from userspace to the current page
321 *
322 * Disable pagefault to avoid recursive lock since
323 * the pages are already locked
324 */
325 pagefault_disable();
326 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
327 pagefault_enable();
Josef Bacik11c65dc2010-05-23 11:07:21 -0400328
Chris Mason39279cc2007-06-12 06:35:45 -0400329 /* Flush processor's dcache for this page */
330 flush_dcache_page(page);
Chris Mason31339ac2011-03-07 11:10:24 -0500331
332 /*
333 * if we get a partial write, we can end up with
334 * partially up to date pages. These add
335 * a lot of complexity, so make sure they don't
336 * happen by forcing this copy to be retried.
337 *
338 * The rest of the btrfs_file_write code will fall
339 * back to page at a time copies after we return 0.
340 */
341 if (!PageUptodate(page) && copied < count)
342 copied = 0;
343
Josef Bacik11c65dc2010-05-23 11:07:21 -0400344 iov_iter_advance(i, copied);
345 write_bytes -= copied;
Xin Zhong914ee292010-12-09 09:30:14 +0000346 total_copied += copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400347
Xin Zhong914ee292010-12-09 09:30:14 +0000348 /* Return to btrfs_file_aio_write to fault page */
Josef Bacik9f570b82011-01-25 12:42:37 -0500349 if (unlikely(copied == 0))
Xin Zhong914ee292010-12-09 09:30:14 +0000350 break;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400351
352 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
353 offset += copied;
354 } else {
355 pg++;
356 offset = 0;
357 }
Chris Mason39279cc2007-06-12 06:35:45 -0400358 }
Xin Zhong914ee292010-12-09 09:30:14 +0000359 return total_copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400360}
361
Chris Masond352ac62008-09-29 15:18:18 -0400362/*
363 * unlocks pages after btrfs_file_write is done with them
364 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400365void btrfs_drop_pages(struct page **pages, size_t num_pages)
Chris Mason39279cc2007-06-12 06:35:45 -0400366{
367 size_t i;
368 for (i = 0; i < num_pages; i++) {
Chris Masond352ac62008-09-29 15:18:18 -0400369 /* page checked is some magic around finding pages that
370 * have been modified without going through btrfs_set_page_dirty
371 * clear it here
372 */
Chris Mason4a096752008-07-21 10:29:44 -0400373 ClearPageChecked(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400374 unlock_page(pages[i]);
375 mark_page_accessed(pages[i]);
376 page_cache_release(pages[i]);
377 }
378}
379
Chris Masond352ac62008-09-29 15:18:18 -0400380/*
381 * after copy_from_user, pages need to be dirtied and we need to make
382 * sure holes are created between the current EOF and the start of
383 * any next extents (if required).
384 *
385 * this also makes the decision about creating an inline extent vs
386 * doing real data extents, marking pages dirty and delalloc as required.
387 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400388int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
389 struct page **pages, size_t num_pages,
390 loff_t pos, size_t write_bytes,
391 struct extent_state **cached)
Chris Mason39279cc2007-06-12 06:35:45 -0400392{
Chris Mason39279cc2007-06-12 06:35:45 -0400393 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400394 int i;
Chris Masondb945352007-10-15 16:15:53 -0400395 u64 num_bytes;
Chris Masona52d9a82007-08-27 16:49:44 -0400396 u64 start_pos;
397 u64 end_of_last_block;
398 u64 end_pos = pos + write_bytes;
399 loff_t isize = i_size_read(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400400
Chris Mason5f39d392007-10-15 16:14:19 -0400401 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -0400402 num_bytes = (write_bytes + pos - start_pos +
403 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
Chris Mason39279cc2007-06-12 06:35:45 -0400404
Chris Masondb945352007-10-15 16:15:53 -0400405 end_of_last_block = start_pos + num_bytes - 1;
Josef Bacik2ac55d42010-02-03 19:33:23 +0000406 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400407 cached);
Josef Bacikd0215f32011-01-25 14:57:24 -0500408 if (err)
409 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -0400410
Chris Masonc8b97812008-10-29 14:49:59 -0400411 for (i = 0; i < num_pages; i++) {
412 struct page *p = pages[i];
413 SetPageUptodate(p);
414 ClearPageChecked(p);
415 set_page_dirty(p);
Chris Masona52d9a82007-08-27 16:49:44 -0400416 }
Josef Bacik9f570b82011-01-25 12:42:37 -0500417
418 /*
419 * we've only changed i_size in ram, and we haven't updated
420 * the disk i_size. There is no need to log the inode
421 * at this time.
422 */
423 if (end_pos > isize)
Chris Masona52d9a82007-08-27 16:49:44 -0400424 i_size_write(inode, end_pos);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400425 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400426}
427
Chris Masond352ac62008-09-29 15:18:18 -0400428/*
429 * this drops all the extents in the cache that intersect the range
430 * [start, end]. Existing extents are split as required.
431 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400432int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
433 int skip_pinned)
Chris Masona52d9a82007-08-27 16:49:44 -0400434{
435 struct extent_map *em;
Chris Mason3b951512008-04-17 11:29:12 -0400436 struct extent_map *split = NULL;
437 struct extent_map *split2 = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -0400438 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan39b56372008-02-15 10:40:50 -0500439 u64 len = end - start + 1;
Chris Mason3b951512008-04-17 11:29:12 -0400440 int ret;
441 int testend = 1;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400442 unsigned long flags;
Chris Masonc8b97812008-10-29 14:49:59 -0400443 int compressed = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400444
Chris Masone6dcd2d2008-07-17 12:53:50 -0400445 WARN_ON(end < start);
Chris Mason3b951512008-04-17 11:29:12 -0400446 if (end == (u64)-1) {
Yan39b56372008-02-15 10:40:50 -0500447 len = (u64)-1;
Chris Mason3b951512008-04-17 11:29:12 -0400448 testend = 0;
449 }
Chris Masond3977122009-01-05 21:25:51 -0500450 while (1) {
Chris Mason3b951512008-04-17 11:29:12 -0400451 if (!split)
David Sterba172ddd62011-04-21 00:48:27 +0200452 split = alloc_extent_map();
Chris Mason3b951512008-04-17 11:29:12 -0400453 if (!split2)
David Sterba172ddd62011-04-21 00:48:27 +0200454 split2 = alloc_extent_map();
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000455 BUG_ON(!split || !split2);
Chris Mason3b951512008-04-17 11:29:12 -0400456
Chris Mason890871b2009-09-02 16:24:52 -0400457 write_lock(&em_tree->lock);
Yan39b56372008-02-15 10:40:50 -0500458 em = lookup_extent_mapping(em_tree, start, len);
Chris Masond1310b22008-01-24 16:13:08 -0500459 if (!em) {
Chris Mason890871b2009-09-02 16:24:52 -0400460 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -0400461 break;
Chris Masond1310b22008-01-24 16:13:08 -0500462 }
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400463 flags = em->flags;
464 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
Yan, Zheng55ef6892009-11-12 09:36:44 +0000465 if (testend && em->start + em->len >= start + len) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400466 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400467 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400468 break;
469 }
Yan, Zheng55ef6892009-11-12 09:36:44 +0000470 start = em->start + em->len;
471 if (testend)
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400472 len = start + len - (em->start + em->len);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400473 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400474 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400475 continue;
476 }
Chris Masonc8b97812008-10-29 14:49:59 -0400477 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Chris Mason3ce7e672008-07-31 15:42:54 -0400478 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -0400479 remove_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -0400480
481 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
482 em->start < start) {
483 split->start = em->start;
484 split->len = start - em->start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500485 split->orig_start = em->orig_start;
Chris Mason3b951512008-04-17 11:29:12 -0400486 split->block_start = em->block_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400487
488 if (compressed)
489 split->block_len = em->block_len;
490 else
491 split->block_len = split->len;
492
Chris Mason3b951512008-04-17 11:29:12 -0400493 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400494 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800495 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400496 ret = add_extent_mapping(em_tree, split);
497 BUG_ON(ret);
498 free_extent_map(split);
499 split = split2;
500 split2 = NULL;
501 }
502 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
503 testend && em->start + em->len > start + len) {
504 u64 diff = start + len - em->start;
505
506 split->start = start + len;
507 split->len = em->start + em->len - (start + len);
508 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400509 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800510 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400511
Chris Masonc8b97812008-10-29 14:49:59 -0400512 if (compressed) {
513 split->block_len = em->block_len;
514 split->block_start = em->block_start;
Chris Mason445a6942008-11-10 11:53:33 -0500515 split->orig_start = em->orig_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400516 } else {
517 split->block_len = split->len;
518 split->block_start = em->block_start + diff;
Chris Mason445a6942008-11-10 11:53:33 -0500519 split->orig_start = split->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400520 }
Chris Mason3b951512008-04-17 11:29:12 -0400521
522 ret = add_extent_mapping(em_tree, split);
523 BUG_ON(ret);
524 free_extent_map(split);
525 split = NULL;
526 }
Chris Mason890871b2009-09-02 16:24:52 -0400527 write_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -0500528
Chris Masona52d9a82007-08-27 16:49:44 -0400529 /* once for us */
530 free_extent_map(em);
531 /* once for the tree*/
532 free_extent_map(em);
533 }
Chris Mason3b951512008-04-17 11:29:12 -0400534 if (split)
535 free_extent_map(split);
536 if (split2)
537 free_extent_map(split2);
Chris Masona52d9a82007-08-27 16:49:44 -0400538 return 0;
539}
540
Chris Mason39279cc2007-06-12 06:35:45 -0400541/*
542 * this is very complex, but the basic idea is to drop all extents
543 * in the range start - end. hint_block is filled in with a block number
544 * that would be a good hint to the block allocator for this file.
545 *
546 * If an extent intersects the range but is not entirely inside the range
547 * it is either truncated or split. Anything entirely inside the range
548 * is deleted from the tree.
549 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000550int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
551 u64 start, u64 end, u64 *hint_byte, int drop_cache)
Chris Mason39279cc2007-06-12 06:35:45 -0400552{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000553 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason00f5c792007-11-30 10:09:33 -0500554 struct extent_buffer *leaf;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000555 struct btrfs_file_extent_item *fi;
Chris Mason00f5c792007-11-30 10:09:33 -0500556 struct btrfs_path *path;
557 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000558 struct btrfs_key new_key;
Li Zefan33345d012011-04-20 10:31:50 +0800559 u64 ino = btrfs_ino(inode);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000560 u64 search_start = start;
561 u64 disk_bytenr = 0;
562 u64 num_bytes = 0;
563 u64 extent_offset = 0;
564 u64 extent_end = 0;
565 int del_nr = 0;
566 int del_slot = 0;
567 int extent_type;
Chris Masonccd467d2007-06-28 15:57:36 -0400568 int recow;
Chris Mason00f5c792007-11-30 10:09:33 -0500569 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -0400570
Chris Masona1ed8352009-09-11 12:27:37 -0400571 if (drop_cache)
572 btrfs_drop_extent_cache(inode, start, end - 1, 0);
Chris Masona52d9a82007-08-27 16:49:44 -0400573
Chris Mason39279cc2007-06-12 06:35:45 -0400574 path = btrfs_alloc_path();
575 if (!path)
576 return -ENOMEM;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000577
Chris Masond3977122009-01-05 21:25:51 -0500578 while (1) {
Chris Masonccd467d2007-06-28 15:57:36 -0400579 recow = 0;
Li Zefan33345d012011-04-20 10:31:50 +0800580 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Chris Mason39279cc2007-06-12 06:35:45 -0400581 search_start, -1);
582 if (ret < 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000583 break;
584 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
585 leaf = path->nodes[0];
586 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +0800587 if (key.objectid == ino &&
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000588 key.type == BTRFS_EXTENT_DATA_KEY)
589 path->slots[0]--;
Chris Mason39279cc2007-06-12 06:35:45 -0400590 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400591 ret = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000592next_slot:
593 leaf = path->nodes[0];
594 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
595 BUG_ON(del_nr > 0);
596 ret = btrfs_next_leaf(root, path);
597 if (ret < 0)
598 break;
599 if (ret > 0) {
600 ret = 0;
601 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400602 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000603 leaf = path->nodes[0];
604 recow = 1;
605 }
606
607 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800608 if (key.objectid > ino ||
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000609 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
610 break;
611
612 fi = btrfs_item_ptr(leaf, path->slots[0],
613 struct btrfs_file_extent_item);
614 extent_type = btrfs_file_extent_type(leaf, fi);
615
616 if (extent_type == BTRFS_FILE_EXTENT_REG ||
617 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
618 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
619 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
620 extent_offset = btrfs_file_extent_offset(leaf, fi);
621 extent_end = key.offset +
622 btrfs_file_extent_num_bytes(leaf, fi);
623 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
624 extent_end = key.offset +
625 btrfs_file_extent_inline_len(leaf, fi);
Chris Mason8c2383c2007-06-18 09:57:58 -0400626 } else {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000627 WARN_ON(1);
Chris Mason8c2383c2007-06-18 09:57:58 -0400628 extent_end = search_start;
Chris Mason39279cc2007-06-12 06:35:45 -0400629 }
630
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000631 if (extent_end <= search_start) {
632 path->slots[0]++;
Chris Mason8c2383c2007-06-18 09:57:58 -0400633 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400634 }
635
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000636 search_start = max(key.offset, start);
637 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200638 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000639 continue;
Chris Mason39279cc2007-06-12 06:35:45 -0400640 }
Chris Mason771ed682008-11-06 22:02:51 -0500641
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000642 /*
643 * | - range to drop - |
644 * | -------- extent -------- |
645 */
646 if (start > key.offset && end < extent_end) {
647 BUG_ON(del_nr > 0);
648 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
649
650 memcpy(&new_key, &key, sizeof(new_key));
651 new_key.offset = start;
652 ret = btrfs_duplicate_item(trans, root, path,
653 &new_key);
654 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200655 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000656 continue;
657 }
658 if (ret < 0)
659 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400660
Chris Mason5f39d392007-10-15 16:14:19 -0400661 leaf = path->nodes[0];
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000662 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
663 struct btrfs_file_extent_item);
664 btrfs_set_file_extent_num_bytes(leaf, fi,
665 start - key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400666
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000667 fi = btrfs_item_ptr(leaf, path->slots[0],
668 struct btrfs_file_extent_item);
Chris Masonc8b97812008-10-29 14:49:59 -0400669
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000670 extent_offset += start - key.offset;
671 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
672 btrfs_set_file_extent_num_bytes(leaf, fi,
673 extent_end - start);
674 btrfs_mark_buffer_dirty(leaf);
Chris Masondb945352007-10-15 16:15:53 -0400675
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000676 if (disk_bytenr > 0) {
677 ret = btrfs_inc_extent_ref(trans, root,
678 disk_bytenr, num_bytes, 0,
679 root->root_key.objectid,
680 new_key.objectid,
681 start - extent_offset);
Zheng Yan31840ae2008-09-23 13:14:14 -0400682 BUG_ON(ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000683 *hint_byte = disk_bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -0400684 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000685 key.offset = start;
686 }
687 /*
688 * | ---- range to drop ----- |
689 * | -------- extent -------- |
690 */
691 if (start <= key.offset && end < extent_end) {
692 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
693
694 memcpy(&new_key, &key, sizeof(new_key));
695 new_key.offset = end;
696 btrfs_set_item_key_safe(trans, root, path, &new_key);
697
698 extent_offset += end - key.offset;
699 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
700 btrfs_set_file_extent_num_bytes(leaf, fi,
701 extent_end - end);
702 btrfs_mark_buffer_dirty(leaf);
703 if (disk_bytenr > 0) {
704 inode_sub_bytes(inode, end - key.offset);
705 *hint_byte = disk_bytenr;
706 }
707 break;
Zheng Yan31840ae2008-09-23 13:14:14 -0400708 }
709
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000710 search_start = extent_end;
711 /*
712 * | ---- range to drop ----- |
713 * | -------- extent -------- |
714 */
715 if (start > key.offset && end >= extent_end) {
716 BUG_ON(del_nr > 0);
717 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
718
719 btrfs_set_file_extent_num_bytes(leaf, fi,
720 start - key.offset);
721 btrfs_mark_buffer_dirty(leaf);
722 if (disk_bytenr > 0) {
723 inode_sub_bytes(inode, extent_end - start);
724 *hint_byte = disk_bytenr;
725 }
726 if (end == extent_end)
727 break;
728
729 path->slots[0]++;
730 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400731 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000732
733 /*
734 * | ---- range to drop ----- |
735 * | ------ extent ------ |
736 */
737 if (start <= key.offset && end >= extent_end) {
738 if (del_nr == 0) {
739 del_slot = path->slots[0];
740 del_nr = 1;
741 } else {
742 BUG_ON(del_slot + del_nr != path->slots[0]);
743 del_nr++;
744 }
745
746 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
747 inode_sub_bytes(inode,
748 extent_end - key.offset);
749 extent_end = ALIGN(extent_end,
750 root->sectorsize);
751 } else if (disk_bytenr > 0) {
752 ret = btrfs_free_extent(trans, root,
753 disk_bytenr, num_bytes, 0,
754 root->root_key.objectid,
755 key.objectid, key.offset -
756 extent_offset);
757 BUG_ON(ret);
758 inode_sub_bytes(inode,
759 extent_end - key.offset);
760 *hint_byte = disk_bytenr;
761 }
762
763 if (end == extent_end)
764 break;
765
766 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
767 path->slots[0]++;
768 goto next_slot;
769 }
770
771 ret = btrfs_del_items(trans, root, path, del_slot,
772 del_nr);
773 BUG_ON(ret);
774
775 del_nr = 0;
776 del_slot = 0;
777
David Sterbab3b4aa72011-04-21 01:20:15 +0200778 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000779 continue;
780 }
781
782 BUG_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -0400783 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000784
785 if (del_nr > 0) {
786 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
787 BUG_ON(ret);
788 }
789
Chris Mason39279cc2007-06-12 06:35:45 -0400790 btrfs_free_path(path);
791 return ret;
792}
793
Yan Zhengd899e052008-10-30 14:25:28 -0400794static int extent_mergeable(struct extent_buffer *leaf, int slot,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000795 u64 objectid, u64 bytenr, u64 orig_offset,
796 u64 *start, u64 *end)
Yan Zhengd899e052008-10-30 14:25:28 -0400797{
798 struct btrfs_file_extent_item *fi;
799 struct btrfs_key key;
800 u64 extent_end;
801
802 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
803 return 0;
804
805 btrfs_item_key_to_cpu(leaf, &key, slot);
806 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
807 return 0;
808
809 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
810 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
811 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000812 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Yan Zhengd899e052008-10-30 14:25:28 -0400813 btrfs_file_extent_compression(leaf, fi) ||
814 btrfs_file_extent_encryption(leaf, fi) ||
815 btrfs_file_extent_other_encoding(leaf, fi))
816 return 0;
817
818 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
819 if ((*start && *start != key.offset) || (*end && *end != extent_end))
820 return 0;
821
822 *start = key.offset;
823 *end = extent_end;
824 return 1;
825}
826
827/*
828 * Mark extent in the range start - end as written.
829 *
830 * This changes extent type from 'pre-allocated' to 'regular'. If only
831 * part of extent is marked as written, the extent will be split into
832 * two or three.
833 */
834int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
Yan Zhengd899e052008-10-30 14:25:28 -0400835 struct inode *inode, u64 start, u64 end)
836{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000837 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zhengd899e052008-10-30 14:25:28 -0400838 struct extent_buffer *leaf;
839 struct btrfs_path *path;
840 struct btrfs_file_extent_item *fi;
841 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000842 struct btrfs_key new_key;
Yan Zhengd899e052008-10-30 14:25:28 -0400843 u64 bytenr;
844 u64 num_bytes;
845 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400846 u64 orig_offset;
Yan Zhengd899e052008-10-30 14:25:28 -0400847 u64 other_start;
848 u64 other_end;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000849 u64 split;
850 int del_nr = 0;
851 int del_slot = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000852 int recow;
Yan Zhengd899e052008-10-30 14:25:28 -0400853 int ret;
Li Zefan33345d012011-04-20 10:31:50 +0800854 u64 ino = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -0400855
856 btrfs_drop_extent_cache(inode, start, end - 1, 0);
857
858 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -0700859 if (!path)
860 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -0400861again:
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000862 recow = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000863 split = start;
Li Zefan33345d012011-04-20 10:31:50 +0800864 key.objectid = ino;
Yan Zhengd899e052008-10-30 14:25:28 -0400865 key.type = BTRFS_EXTENT_DATA_KEY;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000866 key.offset = split;
Yan Zhengd899e052008-10-30 14:25:28 -0400867
868 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Josef Bacik41415732011-03-16 13:59:32 -0400869 if (ret < 0)
870 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -0400871 if (ret > 0 && path->slots[0] > 0)
872 path->slots[0]--;
873
874 leaf = path->nodes[0];
875 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800876 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
Yan Zhengd899e052008-10-30 14:25:28 -0400877 fi = btrfs_item_ptr(leaf, path->slots[0],
878 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000879 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
880 BTRFS_FILE_EXTENT_PREALLOC);
Yan Zhengd899e052008-10-30 14:25:28 -0400881 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
882 BUG_ON(key.offset > start || extent_end < end);
883
884 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
885 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400886 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000887 memcpy(&new_key, &key, sizeof(new_key));
888
889 if (start == key.offset && end < extent_end) {
890 other_start = 0;
891 other_end = start;
892 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +0800893 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000894 &other_start, &other_end)) {
895 new_key.offset = end;
896 btrfs_set_item_key_safe(trans, root, path, &new_key);
897 fi = btrfs_item_ptr(leaf, path->slots[0],
898 struct btrfs_file_extent_item);
899 btrfs_set_file_extent_num_bytes(leaf, fi,
900 extent_end - end);
901 btrfs_set_file_extent_offset(leaf, fi,
902 end - orig_offset);
903 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
904 struct btrfs_file_extent_item);
905 btrfs_set_file_extent_num_bytes(leaf, fi,
906 end - other_start);
907 btrfs_mark_buffer_dirty(leaf);
908 goto out;
909 }
910 }
911
912 if (start > key.offset && end == extent_end) {
913 other_start = end;
914 other_end = 0;
915 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +0800916 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000917 &other_start, &other_end)) {
918 fi = btrfs_item_ptr(leaf, path->slots[0],
919 struct btrfs_file_extent_item);
920 btrfs_set_file_extent_num_bytes(leaf, fi,
921 start - key.offset);
922 path->slots[0]++;
923 new_key.offset = start;
924 btrfs_set_item_key_safe(trans, root, path, &new_key);
925
926 fi = btrfs_item_ptr(leaf, path->slots[0],
927 struct btrfs_file_extent_item);
928 btrfs_set_file_extent_num_bytes(leaf, fi,
929 other_end - start);
930 btrfs_set_file_extent_offset(leaf, fi,
931 start - orig_offset);
932 btrfs_mark_buffer_dirty(leaf);
933 goto out;
934 }
935 }
Yan Zhengd899e052008-10-30 14:25:28 -0400936
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000937 while (start > key.offset || end < extent_end) {
938 if (key.offset == start)
939 split = end;
Yan Zhengd899e052008-10-30 14:25:28 -0400940
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000941 new_key.offset = split;
942 ret = btrfs_duplicate_item(trans, root, path, &new_key);
943 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200944 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000945 goto again;
Yan Zhengd899e052008-10-30 14:25:28 -0400946 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000947 BUG_ON(ret < 0);
Yan Zhengd899e052008-10-30 14:25:28 -0400948
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000949 leaf = path->nodes[0];
950 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Yan Zhengd899e052008-10-30 14:25:28 -0400951 struct btrfs_file_extent_item);
Yan Zhengd899e052008-10-30 14:25:28 -0400952 btrfs_set_file_extent_num_bytes(leaf, fi,
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000953 split - key.offset);
954
955 fi = btrfs_item_ptr(leaf, path->slots[0],
956 struct btrfs_file_extent_item);
957
958 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
959 btrfs_set_file_extent_num_bytes(leaf, fi,
960 extent_end - split);
Yan Zhengd899e052008-10-30 14:25:28 -0400961 btrfs_mark_buffer_dirty(leaf);
962
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000963 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
964 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +0800965 ino, orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400966 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -0400967
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000968 if (split == start) {
969 key.offset = start;
970 } else {
971 BUG_ON(start != key.offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400972 path->slots[0]--;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000973 extent_end = end;
Yan Zhengd899e052008-10-30 14:25:28 -0400974 }
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000975 recow = 1;
Yan Zhengd899e052008-10-30 14:25:28 -0400976 }
977
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000978 other_start = end;
979 other_end = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000980 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +0800981 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000982 &other_start, &other_end)) {
983 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200984 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000985 goto again;
986 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000987 extent_end = other_end;
988 del_slot = path->slots[0] + 1;
989 del_nr++;
990 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
991 0, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +0800992 ino, orig_offset);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000993 BUG_ON(ret);
994 }
995 other_start = 0;
996 other_end = start;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000997 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +0800998 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000999 &other_start, &other_end)) {
1000 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001001 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001002 goto again;
1003 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001004 key.offset = other_start;
1005 del_slot = path->slots[0];
1006 del_nr++;
1007 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1008 0, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001009 ino, orig_offset);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001010 BUG_ON(ret);
1011 }
1012 if (del_nr == 0) {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001013 fi = btrfs_item_ptr(leaf, path->slots[0],
1014 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001015 btrfs_set_file_extent_type(leaf, fi,
1016 BTRFS_FILE_EXTENT_REG);
1017 btrfs_mark_buffer_dirty(leaf);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001018 } else {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001019 fi = btrfs_item_ptr(leaf, del_slot - 1,
1020 struct btrfs_file_extent_item);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001021 btrfs_set_file_extent_type(leaf, fi,
1022 BTRFS_FILE_EXTENT_REG);
1023 btrfs_set_file_extent_num_bytes(leaf, fi,
1024 extent_end - key.offset);
1025 btrfs_mark_buffer_dirty(leaf);
1026
1027 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1028 BUG_ON(ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001029 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001030out:
Yan Zhengd899e052008-10-30 14:25:28 -04001031 btrfs_free_path(path);
1032 return 0;
1033}
1034
Chris Mason39279cc2007-06-12 06:35:45 -04001035/*
Chris Masonb1bf8622011-02-28 09:52:08 -05001036 * on error we return an unlocked page and the error value
1037 * on success we return a locked page and 0
1038 */
1039static int prepare_uptodate_page(struct page *page, u64 pos)
1040{
1041 int ret = 0;
1042
1043 if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
1044 ret = btrfs_readpage(NULL, page);
1045 if (ret)
1046 return ret;
1047 lock_page(page);
1048 if (!PageUptodate(page)) {
1049 unlock_page(page);
1050 return -EIO;
1051 }
1052 }
1053 return 0;
1054}
1055
1056/*
Chris Masond352ac62008-09-29 15:18:18 -04001057 * this gets pages into the page cache and locks them down, it also properly
1058 * waits for data=ordered extents to finish before allowing the pages to be
1059 * modified.
Chris Mason39279cc2007-06-12 06:35:45 -04001060 */
Chris Masond3977122009-01-05 21:25:51 -05001061static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
Chris Mason98ed5172008-01-03 10:01:48 -05001062 struct page **pages, size_t num_pages,
1063 loff_t pos, unsigned long first_index,
Mitch Harder341d14f2011-07-12 19:43:45 +00001064 size_t write_bytes)
Chris Mason39279cc2007-06-12 06:35:45 -04001065{
Josef Bacik2ac55d42010-02-03 19:33:23 +00001066 struct extent_state *cached_state = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001067 int i;
1068 unsigned long index = pos >> PAGE_CACHE_SHIFT;
Chris Mason6da6aba2007-12-18 16:15:09 -05001069 struct inode *inode = fdentry(file)->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04001070 int err = 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001071 int faili = 0;
Chris Mason8c2383c2007-06-18 09:57:58 -04001072 u64 start_pos;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001073 u64 last_pos;
Chris Mason8c2383c2007-06-18 09:57:58 -04001074
Chris Mason5f39d392007-10-15 16:14:19 -04001075 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001076 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001077
Yan Zheng9036c102008-10-30 14:19:41 -04001078 if (start_pos > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -05001079 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
Yan Zheng9036c102008-10-30 14:19:41 -04001080 if (err)
1081 return err;
1082 }
1083
Chris Masone6dcd2d2008-07-17 12:53:50 -04001084again:
Chris Mason39279cc2007-06-12 06:35:45 -04001085 for (i = 0; i < num_pages; i++) {
Josef Bacika94733d2011-07-11 10:47:06 -04001086 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1087 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04001088 if (!pages[i]) {
Chris Masonb1bf8622011-02-28 09:52:08 -05001089 faili = i - 1;
1090 err = -ENOMEM;
1091 goto fail;
1092 }
1093
1094 if (i == 0)
1095 err = prepare_uptodate_page(pages[i], pos);
1096 if (i == num_pages - 1)
1097 err = prepare_uptodate_page(pages[i],
1098 pos + write_bytes);
1099 if (err) {
1100 page_cache_release(pages[i]);
1101 faili = i - 1;
1102 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -04001103 }
Chris Masonccd467d2007-06-28 15:57:36 -04001104 wait_on_page_writeback(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -04001105 }
Chris Masonb1bf8622011-02-28 09:52:08 -05001106 err = 0;
Chris Mason07627042008-02-19 11:29:24 -05001107 if (start_pos < inode->i_size) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04001108 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001109 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1110 start_pos, last_pos - 1, 0, &cached_state,
1111 GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001112 ordered = btrfs_lookup_first_ordered_extent(inode,
1113 last_pos - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001114 if (ordered &&
1115 ordered->file_offset + ordered->len > start_pos &&
1116 ordered->file_offset < last_pos) {
1117 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001118 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1119 start_pos, last_pos - 1,
1120 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001121 for (i = 0; i < num_pages; i++) {
1122 unlock_page(pages[i]);
1123 page_cache_release(pages[i]);
1124 }
1125 btrfs_wait_ordered_range(inode, start_pos,
1126 last_pos - start_pos);
1127 goto again;
1128 }
1129 if (ordered)
1130 btrfs_put_ordered_extent(ordered);
1131
Josef Bacik2ac55d42010-02-03 19:33:23 +00001132 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
Josef Bacik32c00af2009-10-08 13:34:05 -04001133 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00001134 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
Chris Mason07627042008-02-19 11:29:24 -05001135 GFP_NOFS);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001136 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1137 start_pos, last_pos - 1, &cached_state,
1138 GFP_NOFS);
Chris Mason07627042008-02-19 11:29:24 -05001139 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001140 for (i = 0; i < num_pages; i++) {
Chris Masonf87f0572008-08-01 11:27:23 -04001141 clear_page_dirty_for_io(pages[i]);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001142 set_page_extent_mapped(pages[i]);
1143 WARN_ON(!PageLocked(pages[i]));
1144 }
Chris Mason39279cc2007-06-12 06:35:45 -04001145 return 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001146fail:
1147 while (faili >= 0) {
1148 unlock_page(pages[faili]);
1149 page_cache_release(pages[faili]);
1150 faili--;
1151 }
1152 return err;
1153
Chris Mason39279cc2007-06-12 06:35:45 -04001154}
1155
Josef Bacikd0215f32011-01-25 14:57:24 -05001156static noinline ssize_t __btrfs_buffered_write(struct file *file,
1157 struct iov_iter *i,
1158 loff_t pos)
Josef Bacik4b46fce2010-05-23 11:00:55 -04001159{
Josef Bacik11c65dc2010-05-23 11:07:21 -04001160 struct inode *inode = fdentry(file)->d_inode;
1161 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik11c65dc2010-05-23 11:07:21 -04001162 struct page **pages = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001163 unsigned long first_index;
Josef Bacikd0215f32011-01-25 14:57:24 -05001164 size_t num_written = 0;
1165 int nrptrs;
Tsutomu Itohc9149232011-03-30 00:57:23 +00001166 int ret = 0;
Chris Masoncb843a62008-10-03 12:30:02 -04001167
Josef Bacikd0215f32011-01-25 14:57:24 -05001168 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
Josef Bacik11c65dc2010-05-23 11:07:21 -04001169 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1170 (sizeof(struct page *)));
Chris Mason8c2383c2007-06-18 09:57:58 -04001171 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001172 if (!pages)
1173 return -ENOMEM;
Chris Masonab93dbe2009-10-01 12:29:10 -04001174
Chris Mason39279cc2007-06-12 06:35:45 -04001175 first_index = pos >> PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001176
Josef Bacikd0215f32011-01-25 14:57:24 -05001177 while (iov_iter_count(i) > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -04001178 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
Josef Bacikd0215f32011-01-25 14:57:24 -05001179 size_t write_bytes = min(iov_iter_count(i),
Josef Bacik11c65dc2010-05-23 11:07:21 -04001180 nrptrs * (size_t)PAGE_CACHE_SIZE -
Chris Mason8c2383c2007-06-18 09:57:58 -04001181 offset);
Yan, Zheng3a909832011-01-18 13:34:40 +08001182 size_t num_pages = (write_bytes + offset +
1183 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Josef Bacikd0215f32011-01-25 14:57:24 -05001184 size_t dirty_pages;
1185 size_t copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001186
Chris Mason8c2383c2007-06-18 09:57:58 -04001187 WARN_ON(num_pages > nrptrs);
Chris Mason1832a6d2007-12-21 16:27:21 -05001188
Xin Zhong914ee292010-12-09 09:30:14 +00001189 /*
1190 * Fault pages before locking them in prepare_pages
1191 * to avoid recursive lock
1192 */
Josef Bacikd0215f32011-01-25 14:57:24 -05001193 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
Xin Zhong914ee292010-12-09 09:30:14 +00001194 ret = -EFAULT;
Josef Bacikd0215f32011-01-25 14:57:24 -05001195 break;
Xin Zhong914ee292010-12-09 09:30:14 +00001196 }
1197
1198 ret = btrfs_delalloc_reserve_space(inode,
1199 num_pages << PAGE_CACHE_SHIFT);
Chris Mason1832a6d2007-12-21 16:27:21 -05001200 if (ret)
Josef Bacikd0215f32011-01-25 14:57:24 -05001201 break;
Chris Mason1832a6d2007-12-21 16:27:21 -05001202
Josef Bacik4a640012011-01-25 15:10:08 -05001203 /*
1204 * This is going to setup the pages array with the number of
1205 * pages we want, so we don't really need to worry about the
1206 * contents of pages from loop to loop
1207 */
Chris Mason39279cc2007-06-12 06:35:45 -04001208 ret = prepare_pages(root, file, pages, num_pages,
Mitch Harder341d14f2011-07-12 19:43:45 +00001209 pos, first_index, write_bytes);
Josef Bacik6a632092009-02-20 11:00:09 -05001210 if (ret) {
Xin Zhong914ee292010-12-09 09:30:14 +00001211 btrfs_delalloc_release_space(inode,
1212 num_pages << PAGE_CACHE_SHIFT);
Josef Bacikd0215f32011-01-25 14:57:24 -05001213 break;
Josef Bacik6a632092009-02-20 11:00:09 -05001214 }
Chris Mason39279cc2007-06-12 06:35:45 -04001215
Xin Zhong914ee292010-12-09 09:30:14 +00001216 copied = btrfs_copy_from_user(pos, num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -05001217 write_bytes, pages, i);
Chris Masonb1bf8622011-02-28 09:52:08 -05001218
1219 /*
1220 * if we have trouble faulting in the pages, fall
1221 * back to one page at a time
1222 */
1223 if (copied < write_bytes)
1224 nrptrs = 1;
1225
1226 if (copied == 0)
1227 dirty_pages = 0;
1228 else
1229 dirty_pages = (copied + offset +
1230 PAGE_CACHE_SIZE - 1) >>
1231 PAGE_CACHE_SHIFT;
Xin Zhong914ee292010-12-09 09:30:14 +00001232
Josef Bacikd0215f32011-01-25 14:57:24 -05001233 /*
1234 * If we had a short copy we need to release the excess delaloc
1235 * bytes we reserved. We need to increment outstanding_extents
1236 * because btrfs_delalloc_release_space will decrement it, but
1237 * we still have an outstanding extent for the chunk we actually
1238 * managed to copy.
1239 */
Xin Zhong914ee292010-12-09 09:30:14 +00001240 if (num_pages > dirty_pages) {
Josef Bacik9e0baf62011-07-15 15:16:44 +00001241 if (copied > 0) {
1242 spin_lock(&BTRFS_I(inode)->lock);
1243 BTRFS_I(inode)->outstanding_extents++;
1244 spin_unlock(&BTRFS_I(inode)->lock);
1245 }
Xin Zhong914ee292010-12-09 09:30:14 +00001246 btrfs_delalloc_release_space(inode,
1247 (num_pages - dirty_pages) <<
1248 PAGE_CACHE_SHIFT);
1249 }
1250
1251 if (copied > 0) {
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001252 ret = btrfs_dirty_pages(root, inode, pages,
1253 dirty_pages, pos, copied,
1254 NULL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001255 if (ret) {
1256 btrfs_delalloc_release_space(inode,
1257 dirty_pages << PAGE_CACHE_SHIFT);
1258 btrfs_drop_pages(pages, num_pages);
1259 break;
1260 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001261 }
Chris Mason39279cc2007-06-12 06:35:45 -04001262
Chris Mason39279cc2007-06-12 06:35:45 -04001263 btrfs_drop_pages(pages, num_pages);
Xin Zhong914ee292010-12-09 09:30:14 +00001264
Josef Bacikd0215f32011-01-25 14:57:24 -05001265 cond_resched();
1266
1267 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1268 dirty_pages);
1269 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1270 btrfs_btree_balance_dirty(root, 1);
1271 btrfs_throttle(root);
Chris Mason39279cc2007-06-12 06:35:45 -04001272
Xin Zhong914ee292010-12-09 09:30:14 +00001273 pos += copied;
1274 num_written += copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001275 }
Chris Mason5b92ee72008-01-03 13:46:11 -05001276
Chris Mason8c2383c2007-06-18 09:57:58 -04001277 kfree(pages);
Josef Bacikd0215f32011-01-25 14:57:24 -05001278
1279 return num_written ? num_written : ret;
1280}
1281
1282static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1283 const struct iovec *iov,
1284 unsigned long nr_segs, loff_t pos,
1285 loff_t *ppos, size_t count, size_t ocount)
1286{
1287 struct file *file = iocb->ki_filp;
1288 struct inode *inode = fdentry(file)->d_inode;
1289 struct iov_iter i;
1290 ssize_t written;
1291 ssize_t written_buffered;
1292 loff_t endbyte;
1293 int err;
1294
1295 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1296 count, ocount);
1297
1298 /*
1299 * the generic O_DIRECT will update in-memory i_size after the
1300 * DIOs are done. But our endio handlers that update the on
1301 * disk i_size never update past the in memory i_size. So we
1302 * need one more update here to catch any additions to the
1303 * file
1304 */
1305 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1306 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1307 mark_inode_dirty(inode);
1308 }
1309
1310 if (written < 0 || written == count)
1311 return written;
1312
1313 pos += written;
1314 count -= written;
1315 iov_iter_init(&i, iov, nr_segs, count, written);
1316 written_buffered = __btrfs_buffered_write(file, &i, pos);
1317 if (written_buffered < 0) {
1318 err = written_buffered;
1319 goto out;
1320 }
1321 endbyte = pos + written_buffered - 1;
1322 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1323 if (err)
1324 goto out;
1325 written += written_buffered;
1326 *ppos = pos + written_buffered;
1327 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1328 endbyte >> PAGE_CACHE_SHIFT);
1329out:
1330 return written ? written : err;
1331}
1332
1333static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1334 const struct iovec *iov,
1335 unsigned long nr_segs, loff_t pos)
1336{
1337 struct file *file = iocb->ki_filp;
1338 struct inode *inode = fdentry(file)->d_inode;
1339 struct btrfs_root *root = BTRFS_I(inode)->root;
1340 loff_t *ppos = &iocb->ki_pos;
1341 ssize_t num_written = 0;
1342 ssize_t err = 0;
1343 size_t count, ocount;
1344
1345 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1346
1347 mutex_lock(&inode->i_mutex);
1348
1349 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1350 if (err) {
1351 mutex_unlock(&inode->i_mutex);
1352 goto out;
1353 }
1354 count = ocount;
1355
1356 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1357 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1358 if (err) {
1359 mutex_unlock(&inode->i_mutex);
1360 goto out;
1361 }
1362
1363 if (count == 0) {
1364 mutex_unlock(&inode->i_mutex);
1365 goto out;
1366 }
1367
1368 err = file_remove_suid(file);
1369 if (err) {
1370 mutex_unlock(&inode->i_mutex);
1371 goto out;
1372 }
1373
1374 /*
1375 * If BTRFS flips readonly due to some impossible error
1376 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1377 * although we have opened a file as writable, we have
1378 * to stop this write operation to ensure FS consistency.
1379 */
1380 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1381 mutex_unlock(&inode->i_mutex);
1382 err = -EROFS;
1383 goto out;
1384 }
1385
1386 file_update_time(file);
1387 BTRFS_I(inode)->sequence++;
1388
1389 if (unlikely(file->f_flags & O_DIRECT)) {
1390 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1391 pos, ppos, count, ocount);
1392 } else {
1393 struct iov_iter i;
1394
1395 iov_iter_init(&i, iov, nr_segs, count, num_written);
1396
1397 num_written = __btrfs_buffered_write(file, &i, pos);
1398 if (num_written > 0)
1399 *ppos = pos + num_written;
1400 }
1401
1402 mutex_unlock(&inode->i_mutex);
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001403
Chris Mason5a3f23d2009-03-31 13:27:11 -04001404 /*
1405 * we want to make sure fsync finds this change
1406 * but we haven't joined a transaction running right now.
1407 *
1408 * Later on, someone is sure to update the inode and get the
1409 * real transid recorded.
1410 *
1411 * We set last_trans now to the fs_info generation + 1,
1412 * this will either be one more than the running transaction
1413 * or the generation used for the next transaction if there isn't
1414 * one running right now.
1415 */
1416 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
Josef Bacikd0215f32011-01-25 14:57:24 -05001417 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1418 err = generic_write_sync(file, pos, num_written);
1419 if (err < 0 && num_written > 0)
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001420 num_written = err;
1421 }
Josef Bacikd0215f32011-01-25 14:57:24 -05001422out:
Chris Mason39279cc2007-06-12 06:35:45 -04001423 current->backing_dev_info = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001424 return num_written ? num_written : err;
1425}
1426
Chris Masond3977122009-01-05 21:25:51 -05001427int btrfs_release_file(struct inode *inode, struct file *filp)
Mingminge1b81e62008-05-27 10:55:43 -04001428{
Chris Mason5a3f23d2009-03-31 13:27:11 -04001429 /*
1430 * ordered_data_close is set by settattr when we are about to truncate
1431 * a file from a non-zero size to a zero size. This tries to
1432 * flush down new bytes that may have been written if the
1433 * application were using truncate to replace a file in place.
1434 */
1435 if (BTRFS_I(inode)->ordered_data_close) {
1436 BTRFS_I(inode)->ordered_data_close = 0;
1437 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1438 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1439 filemap_flush(inode->i_mapping);
1440 }
Sage Weil6bf13c02008-06-10 10:07:39 -04001441 if (filp->private_data)
1442 btrfs_ioctl_trans_end(filp);
Mingminge1b81e62008-05-27 10:55:43 -04001443 return 0;
1444}
1445
Chris Masond352ac62008-09-29 15:18:18 -04001446/*
1447 * fsync call for both files and directories. This logs the inode into
1448 * the tree log instead of forcing full commits whenever possible.
1449 *
1450 * It needs to call filemap_fdatawait so that all ordered extent updates are
1451 * in the metadata btree are up to date for copying to the log.
1452 *
1453 * It drops the inode mutex before doing the tree log commit. This is an
1454 * important optimization for directories because holding the mutex prevents
1455 * new operations on the dir while we write to disk.
1456 */
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001457int btrfs_sync_file(struct file *file, int datasync)
Chris Mason39279cc2007-06-12 06:35:45 -04001458{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001459 struct dentry *dentry = file->f_path.dentry;
Chris Mason39279cc2007-06-12 06:35:45 -04001460 struct inode *inode = dentry->d_inode;
1461 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001462 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001463 struct btrfs_trans_handle *trans;
1464
liubo1abe9b82011-03-24 11:18:59 +00001465 trace_btrfs_sync_file(file, datasync);
Chris Mason257c62e2009-10-13 13:21:08 -04001466
1467 /* we wait first, since the writeback may change the inode */
1468 root->log_batch++;
1469 /* the VFS called filemap_fdatawrite for us */
1470 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1471 root->log_batch++;
1472
Chris Mason39279cc2007-06-12 06:35:45 -04001473 /*
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001474 * check the transaction that last modified this inode
1475 * and see if its already been committed
Chris Mason39279cc2007-06-12 06:35:45 -04001476 */
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001477 if (!BTRFS_I(inode)->last_trans)
1478 goto out;
Chris Masona2135012008-06-25 16:01:30 -04001479
Chris Mason257c62e2009-10-13 13:21:08 -04001480 /*
1481 * if the last transaction that changed this file was before
1482 * the current transaction, we can bail out now without any
1483 * syncing
1484 */
Josef Bacika4abeea2011-04-11 17:25:13 -04001485 smp_mb();
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001486 if (BTRFS_I(inode)->last_trans <=
1487 root->fs_info->last_trans_committed) {
1488 BTRFS_I(inode)->last_trans = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001489 goto out;
1490 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001491
1492 /*
Chris Masona52d9a82007-08-27 16:49:44 -04001493 * ok we haven't committed the transaction yet, lets do a commit
1494 */
Dan Carpenter6f902af2010-05-29 09:49:07 +00001495 if (file->private_data)
Sage Weil6bf13c02008-06-10 10:07:39 -04001496 btrfs_ioctl_trans_end(file);
1497
Yan, Zhenga22285a2010-05-16 10:48:46 -04001498 trans = btrfs_start_transaction(root, 0);
1499 if (IS_ERR(trans)) {
1500 ret = PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04001501 goto out;
1502 }
Chris Masone02119d2008-09-05 16:13:11 -04001503
Chris Mason2cfbd502009-02-20 10:55:10 -05001504 ret = btrfs_log_dentry_safe(trans, root, dentry);
Chris Masond3977122009-01-05 21:25:51 -05001505 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04001506 goto out;
Chris Mason49eb7e42008-09-11 15:53:12 -04001507
1508 /* we've logged all the items and now have a consistent
1509 * version of the file in the log. It is possible that
1510 * someone will come in and modify the file, but that's
1511 * fine because the log is consistent on disk, and we
1512 * have references to all of the file's extents
1513 *
1514 * It is possible that someone will come in and log the
1515 * file again, but that will end up using the synchronization
1516 * inside btrfs_sync_log to keep things safe.
1517 */
Chris Mason2cfbd502009-02-20 10:55:10 -05001518 mutex_unlock(&dentry->d_inode->i_mutex);
Chris Mason49eb7e42008-09-11 15:53:12 -04001519
Chris Mason257c62e2009-10-13 13:21:08 -04001520 if (ret != BTRFS_NO_LOG_SYNC) {
1521 if (ret > 0) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001522 ret = btrfs_commit_transaction(trans, root);
Chris Mason257c62e2009-10-13 13:21:08 -04001523 } else {
1524 ret = btrfs_sync_log(trans, root);
1525 if (ret == 0)
1526 ret = btrfs_end_transaction(trans, root);
1527 else
1528 ret = btrfs_commit_transaction(trans, root);
1529 }
1530 } else {
1531 ret = btrfs_end_transaction(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001532 }
Chris Mason2cfbd502009-02-20 10:55:10 -05001533 mutex_lock(&dentry->d_inode->i_mutex);
Chris Mason39279cc2007-06-12 06:35:45 -04001534out:
Roel Kluin014e4ac2010-01-29 10:42:11 +00001535 return ret > 0 ? -EIO : ret;
Chris Mason39279cc2007-06-12 06:35:45 -04001536}
1537
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001538static const struct vm_operations_struct btrfs_file_vm_ops = {
Chris Mason92fee662007-07-25 12:31:35 -04001539 .fault = filemap_fault,
Chris Mason9ebefb182007-06-15 13:50:00 -04001540 .page_mkwrite = btrfs_page_mkwrite,
1541};
1542
1543static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1544{
Miao Xie058a4572010-05-20 07:21:50 +00001545 struct address_space *mapping = filp->f_mapping;
1546
1547 if (!mapping->a_ops->readpage)
1548 return -ENOEXEC;
1549
Chris Mason9ebefb182007-06-15 13:50:00 -04001550 file_accessed(filp);
Miao Xie058a4572010-05-20 07:21:50 +00001551 vma->vm_ops = &btrfs_file_vm_ops;
1552 vma->vm_flags |= VM_CAN_NONLINEAR;
1553
Chris Mason9ebefb182007-06-15 13:50:00 -04001554 return 0;
1555}
1556
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001557static long btrfs_fallocate(struct file *file, int mode,
1558 loff_t offset, loff_t len)
1559{
1560 struct inode *inode = file->f_path.dentry->d_inode;
1561 struct extent_state *cached_state = NULL;
1562 u64 cur_offset;
1563 u64 last_byte;
1564 u64 alloc_start;
1565 u64 alloc_end;
1566 u64 alloc_hint = 0;
1567 u64 locked_end;
1568 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1569 struct extent_map *em;
1570 int ret;
1571
1572 alloc_start = offset & ~mask;
1573 alloc_end = (offset + len + mask) & ~mask;
1574
1575 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1576 if (mode & ~FALLOC_FL_KEEP_SIZE)
1577 return -EOPNOTSUPP;
1578
1579 /*
1580 * wait for ordered IO before we have any locks. We'll loop again
1581 * below with the locks held.
1582 */
1583 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1584
1585 mutex_lock(&inode->i_mutex);
1586 ret = inode_newsize_ok(inode, alloc_end);
1587 if (ret)
1588 goto out;
1589
1590 if (alloc_start > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -05001591 ret = btrfs_cont_expand(inode, i_size_read(inode),
1592 alloc_start);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001593 if (ret)
1594 goto out;
1595 }
1596
1597 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1598 if (ret)
1599 goto out;
1600
1601 locked_end = alloc_end - 1;
1602 while (1) {
1603 struct btrfs_ordered_extent *ordered;
1604
1605 /* the extent lock is ordered inside the running
1606 * transaction
1607 */
1608 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1609 locked_end, 0, &cached_state, GFP_NOFS);
1610 ordered = btrfs_lookup_first_ordered_extent(inode,
1611 alloc_end - 1);
1612 if (ordered &&
1613 ordered->file_offset + ordered->len > alloc_start &&
1614 ordered->file_offset < alloc_end) {
1615 btrfs_put_ordered_extent(ordered);
1616 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1617 alloc_start, locked_end,
1618 &cached_state, GFP_NOFS);
1619 /*
1620 * we can't wait on the range with the transaction
1621 * running or with the extent lock held
1622 */
1623 btrfs_wait_ordered_range(inode, alloc_start,
1624 alloc_end - alloc_start);
1625 } else {
1626 if (ordered)
1627 btrfs_put_ordered_extent(ordered);
1628 break;
1629 }
1630 }
1631
1632 cur_offset = alloc_start;
1633 while (1) {
Josef Bacikf1e490a2011-08-18 10:36:39 -04001634 u64 actual_end;
1635
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001636 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1637 alloc_end - cur_offset, 0);
David Sterbac7040052011-04-19 18:00:01 +02001638 BUG_ON(IS_ERR_OR_NULL(em));
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001639 last_byte = min(extent_map_end(em), alloc_end);
Josef Bacikf1e490a2011-08-18 10:36:39 -04001640 actual_end = min_t(u64, extent_map_end(em), offset + len);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001641 last_byte = (last_byte + mask) & ~mask;
Josef Bacikf1e490a2011-08-18 10:36:39 -04001642
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001643 if (em->block_start == EXTENT_MAP_HOLE ||
1644 (cur_offset >= inode->i_size &&
1645 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1646 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1647 last_byte - cur_offset,
1648 1 << inode->i_blkbits,
1649 offset + len,
1650 &alloc_hint);
1651 if (ret < 0) {
1652 free_extent_map(em);
1653 break;
1654 }
Josef Bacikf1e490a2011-08-18 10:36:39 -04001655 } else if (actual_end > inode->i_size &&
1656 !(mode & FALLOC_FL_KEEP_SIZE)) {
1657 /*
1658 * We didn't need to allocate any more space, but we
1659 * still extended the size of the file so we need to
1660 * update i_size.
1661 */
1662 inode->i_ctime = CURRENT_TIME;
1663 i_size_write(inode, actual_end);
1664 btrfs_ordered_update_i_size(inode, actual_end, NULL);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001665 }
1666 free_extent_map(em);
1667
1668 cur_offset = last_byte;
1669 if (cur_offset >= alloc_end) {
1670 ret = 0;
1671 break;
1672 }
1673 }
1674 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1675 &cached_state, GFP_NOFS);
1676
1677 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1678out:
1679 mutex_unlock(&inode->i_mutex);
1680 return ret;
1681}
1682
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001683const struct file_operations btrfs_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04001684 .llseek = generic_file_llseek,
1685 .read = do_sync_read,
Miao Xie4a0010712010-06-07 03:38:51 +00001686 .write = do_sync_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001687 .aio_read = generic_file_aio_read,
Chris Masone9906a92007-12-14 12:56:58 -05001688 .splice_read = generic_file_splice_read,
Josef Bacik11c65dc2010-05-23 11:07:21 -04001689 .aio_write = btrfs_file_aio_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001690 .mmap = btrfs_file_mmap,
Chris Mason39279cc2007-06-12 06:35:45 -04001691 .open = generic_file_open,
Mingminge1b81e62008-05-27 10:55:43 -04001692 .release = btrfs_release_file,
Chris Mason39279cc2007-06-12 06:35:45 -04001693 .fsync = btrfs_sync_file,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001694 .fallocate = btrfs_fallocate,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001695 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001696#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001697 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001698#endif
1699};