blob: 4d2eb6417145964c8731bf5e56673a0e816d59aa [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070027#include <linux/aio.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010028#include <linux/falloc.h>
Chris Mason39279cc2007-06-12 06:35:45 -040029#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/statfs.h>
32#include <linux/compat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/slab.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000034#include <linux/btrfs.h>
Chris Mason39279cc2007-06-12 06:35:45 -040035#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
Chris Mason39279cc2007-06-12 06:35:45 -040039#include "print-tree.h"
Chris Masone02119d2008-09-05 16:13:11 -040040#include "tree-log.h"
41#include "locking.h"
Jeff Mahoney12fa8ec2008-05-02 15:03:58 -040042#include "compat.h"
Josef Bacik2aaa6652012-08-29 14:27:18 -040043#include "volumes.h"
Chris Mason39279cc2007-06-12 06:35:45 -040044
Miao Xie9247f312012-11-26 09:24:43 +000045static struct kmem_cache *btrfs_inode_defrag_cachep;
Chris Mason4cb53002011-05-24 15:35:30 -040046/*
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
50 */
51struct inode_defrag {
52 struct rb_node rb_node;
53 /* objectid */
54 u64 ino;
55 /*
56 * transid where the defrag was added, we search for
57 * extents newer than this
58 */
59 u64 transid;
60
61 /* root objectid */
62 u64 root;
63
64 /* last offset we were able to defrag */
65 u64 last_offset;
66
67 /* if we've wrapped around back to zero once already */
68 int cycled;
69};
70
Miao Xie762f2262012-05-24 18:58:27 +080071static int __compare_inode_defrag(struct inode_defrag *defrag1,
72 struct inode_defrag *defrag2)
73{
74 if (defrag1->root > defrag2->root)
75 return 1;
76 else if (defrag1->root < defrag2->root)
77 return -1;
78 else if (defrag1->ino > defrag2->ino)
79 return 1;
80 else if (defrag1->ino < defrag2->ino)
81 return -1;
82 else
83 return 0;
84}
85
Chris Mason4cb53002011-05-24 15:35:30 -040086/* pop a record for an inode into the defrag tree. The lock
87 * must be held already
88 *
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
91 *
92 * If an existing record is found the defrag item you
93 * pass in is freed
94 */
Miao Xie8ddc4732012-11-26 09:25:38 +000095static int __btrfs_add_inode_defrag(struct inode *inode,
Chris Mason4cb53002011-05-24 15:35:30 -040096 struct inode_defrag *defrag)
97{
98 struct btrfs_root *root = BTRFS_I(inode)->root;
99 struct inode_defrag *entry;
100 struct rb_node **p;
101 struct rb_node *parent = NULL;
Miao Xie762f2262012-05-24 18:58:27 +0800102 int ret;
Chris Mason4cb53002011-05-24 15:35:30 -0400103
104 p = &root->fs_info->defrag_inodes.rb_node;
105 while (*p) {
106 parent = *p;
107 entry = rb_entry(parent, struct inode_defrag, rb_node);
108
Miao Xie762f2262012-05-24 18:58:27 +0800109 ret = __compare_inode_defrag(defrag, entry);
110 if (ret < 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400111 p = &parent->rb_left;
Miao Xie762f2262012-05-24 18:58:27 +0800112 else if (ret > 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400113 p = &parent->rb_right;
114 else {
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
118 */
119 if (defrag->transid < entry->transid)
120 entry->transid = defrag->transid;
121 if (defrag->last_offset > entry->last_offset)
122 entry->last_offset = defrag->last_offset;
Miao Xie8ddc4732012-11-26 09:25:38 +0000123 return -EEXIST;
Chris Mason4cb53002011-05-24 15:35:30 -0400124 }
125 }
Josef Bacik72ac3c02012-05-23 14:13:11 -0400126 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
Chris Mason4cb53002011-05-24 15:35:30 -0400127 rb_link_node(&defrag->rb_node, parent, p);
128 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
Miao Xie8ddc4732012-11-26 09:25:38 +0000129 return 0;
130}
Chris Mason4cb53002011-05-24 15:35:30 -0400131
Miao Xie8ddc4732012-11-26 09:25:38 +0000132static inline int __need_auto_defrag(struct btrfs_root *root)
133{
134 if (!btrfs_test_opt(root, AUTO_DEFRAG))
135 return 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400136
Miao Xie8ddc4732012-11-26 09:25:38 +0000137 if (btrfs_fs_closing(root->fs_info))
138 return 0;
139
140 return 1;
Chris Mason4cb53002011-05-24 15:35:30 -0400141}
142
143/*
144 * insert a defrag record for this inode if auto defrag is
145 * enabled
146 */
147int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
148 struct inode *inode)
149{
150 struct btrfs_root *root = BTRFS_I(inode)->root;
151 struct inode_defrag *defrag;
Chris Mason4cb53002011-05-24 15:35:30 -0400152 u64 transid;
Miao Xie8ddc4732012-11-26 09:25:38 +0000153 int ret;
Chris Mason4cb53002011-05-24 15:35:30 -0400154
Miao Xie8ddc4732012-11-26 09:25:38 +0000155 if (!__need_auto_defrag(root))
Chris Mason4cb53002011-05-24 15:35:30 -0400156 return 0;
157
Josef Bacik72ac3c02012-05-23 14:13:11 -0400158 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
Chris Mason4cb53002011-05-24 15:35:30 -0400159 return 0;
160
161 if (trans)
162 transid = trans->transid;
163 else
164 transid = BTRFS_I(inode)->root->last_trans;
165
Miao Xie9247f312012-11-26 09:24:43 +0000166 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
Chris Mason4cb53002011-05-24 15:35:30 -0400167 if (!defrag)
168 return -ENOMEM;
169
David Sterbaa4689d22011-05-31 17:08:14 +0000170 defrag->ino = btrfs_ino(inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400171 defrag->transid = transid;
172 defrag->root = root->root_key.objectid;
173
174 spin_lock(&root->fs_info->defrag_inodes_lock);
Miao Xie8ddc4732012-11-26 09:25:38 +0000175 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
176 /*
177 * If we set IN_DEFRAG flag and evict the inode from memory,
178 * and then re-read this inode, this new inode doesn't have
179 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 */
181 ret = __btrfs_add_inode_defrag(inode, defrag);
182 if (ret)
183 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
184 } else {
Miao Xie9247f312012-11-26 09:24:43 +0000185 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
Miao Xie8ddc4732012-11-26 09:25:38 +0000186 }
Chris Mason4cb53002011-05-24 15:35:30 -0400187 spin_unlock(&root->fs_info->defrag_inodes_lock);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000188 return 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400189}
190
191/*
Miao Xie8ddc4732012-11-26 09:25:38 +0000192 * Requeue the defrag object. If there is a defrag object that points to
193 * the same inode in the tree, we will merge them together (by
194 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
Chris Mason4cb53002011-05-24 15:35:30 -0400195 */
Eric Sandeen48a3b632013-04-25 20:41:01 +0000196static void btrfs_requeue_inode_defrag(struct inode *inode,
197 struct inode_defrag *defrag)
Miao Xie8ddc4732012-11-26 09:25:38 +0000198{
199 struct btrfs_root *root = BTRFS_I(inode)->root;
200 int ret;
201
202 if (!__need_auto_defrag(root))
203 goto out;
204
205 /*
206 * Here we don't check the IN_DEFRAG flag, because we need merge
207 * them together.
208 */
209 spin_lock(&root->fs_info->defrag_inodes_lock);
210 ret = __btrfs_add_inode_defrag(inode, defrag);
211 spin_unlock(&root->fs_info->defrag_inodes_lock);
212 if (ret)
213 goto out;
214 return;
215out:
216 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
217}
218
219/*
Miao Xie26176e72012-11-26 09:26:20 +0000220 * pick the defragable inode that we want, if it doesn't exist, we will get
221 * the next one.
Chris Mason4cb53002011-05-24 15:35:30 -0400222 */
Miao Xie26176e72012-11-26 09:26:20 +0000223static struct inode_defrag *
224btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
Chris Mason4cb53002011-05-24 15:35:30 -0400225{
226 struct inode_defrag *entry = NULL;
Miao Xie762f2262012-05-24 18:58:27 +0800227 struct inode_defrag tmp;
Chris Mason4cb53002011-05-24 15:35:30 -0400228 struct rb_node *p;
229 struct rb_node *parent = NULL;
Miao Xie762f2262012-05-24 18:58:27 +0800230 int ret;
231
232 tmp.ino = ino;
233 tmp.root = root;
Chris Mason4cb53002011-05-24 15:35:30 -0400234
Miao Xie26176e72012-11-26 09:26:20 +0000235 spin_lock(&fs_info->defrag_inodes_lock);
236 p = fs_info->defrag_inodes.rb_node;
Chris Mason4cb53002011-05-24 15:35:30 -0400237 while (p) {
238 parent = p;
239 entry = rb_entry(parent, struct inode_defrag, rb_node);
240
Miao Xie762f2262012-05-24 18:58:27 +0800241 ret = __compare_inode_defrag(&tmp, entry);
242 if (ret < 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400243 p = parent->rb_left;
Miao Xie762f2262012-05-24 18:58:27 +0800244 else if (ret > 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400245 p = parent->rb_right;
246 else
Miao Xie26176e72012-11-26 09:26:20 +0000247 goto out;
Chris Mason4cb53002011-05-24 15:35:30 -0400248 }
249
Miao Xie26176e72012-11-26 09:26:20 +0000250 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
251 parent = rb_next(parent);
252 if (parent)
Chris Mason4cb53002011-05-24 15:35:30 -0400253 entry = rb_entry(parent, struct inode_defrag, rb_node);
Miao Xie26176e72012-11-26 09:26:20 +0000254 else
255 entry = NULL;
Chris Mason4cb53002011-05-24 15:35:30 -0400256 }
Miao Xie26176e72012-11-26 09:26:20 +0000257out:
258 if (entry)
259 rb_erase(parent, &fs_info->defrag_inodes);
260 spin_unlock(&fs_info->defrag_inodes_lock);
261 return entry;
262}
263
264void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
265{
266 struct inode_defrag *defrag;
267 struct rb_node *node;
268
269 spin_lock(&fs_info->defrag_inodes_lock);
270 node = rb_first(&fs_info->defrag_inodes);
271 while (node) {
272 rb_erase(node, &fs_info->defrag_inodes);
273 defrag = rb_entry(node, struct inode_defrag, rb_node);
274 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
275
276 if (need_resched()) {
277 spin_unlock(&fs_info->defrag_inodes_lock);
278 cond_resched();
279 spin_lock(&fs_info->defrag_inodes_lock);
280 }
281
282 node = rb_first(&fs_info->defrag_inodes);
283 }
284 spin_unlock(&fs_info->defrag_inodes_lock);
285}
286
287#define BTRFS_DEFRAG_BATCH 1024
288
289static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
290 struct inode_defrag *defrag)
291{
292 struct btrfs_root *inode_root;
293 struct inode *inode;
294 struct btrfs_key key;
295 struct btrfs_ioctl_defrag_range_args range;
296 int num_defrag;
Liu Bo6f1c3602013-01-29 03:22:10 +0000297 int index;
298 int ret;
Miao Xie26176e72012-11-26 09:26:20 +0000299
300 /* get the inode */
301 key.objectid = defrag->root;
302 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
303 key.offset = (u64)-1;
Liu Bo6f1c3602013-01-29 03:22:10 +0000304
305 index = srcu_read_lock(&fs_info->subvol_srcu);
306
Miao Xie26176e72012-11-26 09:26:20 +0000307 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
308 if (IS_ERR(inode_root)) {
Liu Bo6f1c3602013-01-29 03:22:10 +0000309 ret = PTR_ERR(inode_root);
310 goto cleanup;
311 }
Miao Xie26176e72012-11-26 09:26:20 +0000312
313 key.objectid = defrag->ino;
314 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
315 key.offset = 0;
316 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
317 if (IS_ERR(inode)) {
Liu Bo6f1c3602013-01-29 03:22:10 +0000318 ret = PTR_ERR(inode);
319 goto cleanup;
Miao Xie26176e72012-11-26 09:26:20 +0000320 }
Liu Bo6f1c3602013-01-29 03:22:10 +0000321 srcu_read_unlock(&fs_info->subvol_srcu, index);
Miao Xie26176e72012-11-26 09:26:20 +0000322
323 /* do a chunk of defrag */
324 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
325 memset(&range, 0, sizeof(range));
326 range.len = (u64)-1;
327 range.start = defrag->last_offset;
Miao Xieb66f00d2012-11-26 09:27:29 +0000328
329 sb_start_write(fs_info->sb);
Miao Xie26176e72012-11-26 09:26:20 +0000330 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
331 BTRFS_DEFRAG_BATCH);
Miao Xieb66f00d2012-11-26 09:27:29 +0000332 sb_end_write(fs_info->sb);
Miao Xie26176e72012-11-26 09:26:20 +0000333 /*
334 * if we filled the whole defrag batch, there
335 * must be more work to do. Queue this defrag
336 * again
337 */
338 if (num_defrag == BTRFS_DEFRAG_BATCH) {
339 defrag->last_offset = range.start;
340 btrfs_requeue_inode_defrag(inode, defrag);
341 } else if (defrag->last_offset && !defrag->cycled) {
342 /*
343 * we didn't fill our defrag batch, but
344 * we didn't start at zero. Make sure we loop
345 * around to the start of the file.
346 */
347 defrag->last_offset = 0;
348 defrag->cycled = 1;
349 btrfs_requeue_inode_defrag(inode, defrag);
350 } else {
351 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
352 }
353
354 iput(inode);
355 return 0;
Liu Bo6f1c3602013-01-29 03:22:10 +0000356cleanup:
357 srcu_read_unlock(&fs_info->subvol_srcu, index);
358 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
359 return ret;
Chris Mason4cb53002011-05-24 15:35:30 -0400360}
361
362/*
363 * run through the list of inodes in the FS that need
364 * defragging
365 */
366int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
367{
368 struct inode_defrag *defrag;
Chris Mason4cb53002011-05-24 15:35:30 -0400369 u64 first_ino = 0;
Miao Xie762f2262012-05-24 18:58:27 +0800370 u64 root_objectid = 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400371
372 atomic_inc(&fs_info->defrag_running);
Chris Mason4cb53002011-05-24 15:35:30 -0400373 while(1) {
Miao Xiedc81cdc2013-02-20 23:32:52 -0700374 /* Pause the auto defragger. */
375 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
376 &fs_info->fs_state))
377 break;
378
Miao Xie26176e72012-11-26 09:26:20 +0000379 if (!__need_auto_defrag(fs_info->tree_root))
380 break;
Chris Mason4cb53002011-05-24 15:35:30 -0400381
382 /* find an inode to defrag */
Miao Xie26176e72012-11-26 09:26:20 +0000383 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
384 first_ino);
Chris Mason4cb53002011-05-24 15:35:30 -0400385 if (!defrag) {
Miao Xie26176e72012-11-26 09:26:20 +0000386 if (root_objectid || first_ino) {
Miao Xie762f2262012-05-24 18:58:27 +0800387 root_objectid = 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400388 first_ino = 0;
389 continue;
390 } else {
391 break;
392 }
393 }
394
Chris Mason4cb53002011-05-24 15:35:30 -0400395 first_ino = defrag->ino + 1;
Miao Xie762f2262012-05-24 18:58:27 +0800396 root_objectid = defrag->root;
Chris Mason4cb53002011-05-24 15:35:30 -0400397
Miao Xie26176e72012-11-26 09:26:20 +0000398 __btrfs_run_defrag_inode(fs_info, defrag);
Chris Mason4cb53002011-05-24 15:35:30 -0400399 }
Chris Mason4cb53002011-05-24 15:35:30 -0400400 atomic_dec(&fs_info->defrag_running);
401
402 /*
403 * during unmount, we use the transaction_wait queue to
404 * wait for the defragger to stop
405 */
406 wake_up(&fs_info->transaction_wait);
407 return 0;
408}
Chris Mason39279cc2007-06-12 06:35:45 -0400409
Chris Masond352ac62008-09-29 15:18:18 -0400410/* simple helper to fault in pages and copy. This should go away
411 * and be replaced with calls into generic code.
412 */
Chris Masond3977122009-01-05 21:25:51 -0500413static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -0500414 size_t write_bytes,
Chris Masona1b32a52008-09-05 16:09:51 -0400415 struct page **prepared_pages,
Josef Bacik11c65dc2010-05-23 11:07:21 -0400416 struct iov_iter *i)
Chris Mason39279cc2007-06-12 06:35:45 -0400417{
Xin Zhong914ee292010-12-09 09:30:14 +0000418 size_t copied = 0;
Josef Bacikd0215f32011-01-25 14:57:24 -0500419 size_t total_copied = 0;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400420 int pg = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400421 int offset = pos & (PAGE_CACHE_SIZE - 1);
422
Josef Bacik11c65dc2010-05-23 11:07:21 -0400423 while (write_bytes > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -0400424 size_t count = min_t(size_t,
425 PAGE_CACHE_SIZE - offset, write_bytes);
Josef Bacik11c65dc2010-05-23 11:07:21 -0400426 struct page *page = prepared_pages[pg];
Xin Zhong914ee292010-12-09 09:30:14 +0000427 /*
428 * Copy data from userspace to the current page
429 *
430 * Disable pagefault to avoid recursive lock since
431 * the pages are already locked
432 */
433 pagefault_disable();
434 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
435 pagefault_enable();
Josef Bacik11c65dc2010-05-23 11:07:21 -0400436
Chris Mason39279cc2007-06-12 06:35:45 -0400437 /* Flush processor's dcache for this page */
438 flush_dcache_page(page);
Chris Mason31339ac2011-03-07 11:10:24 -0500439
440 /*
441 * if we get a partial write, we can end up with
442 * partially up to date pages. These add
443 * a lot of complexity, so make sure they don't
444 * happen by forcing this copy to be retried.
445 *
446 * The rest of the btrfs_file_write code will fall
447 * back to page at a time copies after we return 0.
448 */
449 if (!PageUptodate(page) && copied < count)
450 copied = 0;
451
Josef Bacik11c65dc2010-05-23 11:07:21 -0400452 iov_iter_advance(i, copied);
453 write_bytes -= copied;
Xin Zhong914ee292010-12-09 09:30:14 +0000454 total_copied += copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400455
Xin Zhong914ee292010-12-09 09:30:14 +0000456 /* Return to btrfs_file_aio_write to fault page */
Josef Bacik9f570b82011-01-25 12:42:37 -0500457 if (unlikely(copied == 0))
Xin Zhong914ee292010-12-09 09:30:14 +0000458 break;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400459
460 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
461 offset += copied;
462 } else {
463 pg++;
464 offset = 0;
465 }
Chris Mason39279cc2007-06-12 06:35:45 -0400466 }
Xin Zhong914ee292010-12-09 09:30:14 +0000467 return total_copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400468}
469
Chris Masond352ac62008-09-29 15:18:18 -0400470/*
471 * unlocks pages after btrfs_file_write is done with them
472 */
Eric Sandeen48a3b632013-04-25 20:41:01 +0000473static void btrfs_drop_pages(struct page **pages, size_t num_pages)
Chris Mason39279cc2007-06-12 06:35:45 -0400474{
475 size_t i;
476 for (i = 0; i < num_pages; i++) {
Chris Masond352ac62008-09-29 15:18:18 -0400477 /* page checked is some magic around finding pages that
478 * have been modified without going through btrfs_set_page_dirty
479 * clear it here
480 */
Chris Mason4a096752008-07-21 10:29:44 -0400481 ClearPageChecked(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400482 unlock_page(pages[i]);
483 mark_page_accessed(pages[i]);
484 page_cache_release(pages[i]);
485 }
486}
487
Chris Masond352ac62008-09-29 15:18:18 -0400488/*
489 * after copy_from_user, pages need to be dirtied and we need to make
490 * sure holes are created between the current EOF and the start of
491 * any next extents (if required).
492 *
493 * this also makes the decision about creating an inline extent vs
494 * doing real data extents, marking pages dirty and delalloc as required.
495 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400496int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
Eric Sandeen48a3b632013-04-25 20:41:01 +0000497 struct page **pages, size_t num_pages,
498 loff_t pos, size_t write_bytes,
499 struct extent_state **cached)
Chris Mason39279cc2007-06-12 06:35:45 -0400500{
Chris Mason39279cc2007-06-12 06:35:45 -0400501 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400502 int i;
Chris Masondb945352007-10-15 16:15:53 -0400503 u64 num_bytes;
Chris Masona52d9a82007-08-27 16:49:44 -0400504 u64 start_pos;
505 u64 end_of_last_block;
506 u64 end_pos = pos + write_bytes;
507 loff_t isize = i_size_read(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400508
Chris Mason5f39d392007-10-15 16:14:19 -0400509 start_pos = pos & ~((u64)root->sectorsize - 1);
Qu Wenruofda28322013-02-26 08:10:22 +0000510 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
Chris Mason39279cc2007-06-12 06:35:45 -0400511
Chris Masondb945352007-10-15 16:15:53 -0400512 end_of_last_block = start_pos + num_bytes - 1;
Josef Bacik2ac55d42010-02-03 19:33:23 +0000513 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400514 cached);
Josef Bacikd0215f32011-01-25 14:57:24 -0500515 if (err)
516 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -0400517
Chris Masonc8b97812008-10-29 14:49:59 -0400518 for (i = 0; i < num_pages; i++) {
519 struct page *p = pages[i];
520 SetPageUptodate(p);
521 ClearPageChecked(p);
522 set_page_dirty(p);
Chris Masona52d9a82007-08-27 16:49:44 -0400523 }
Josef Bacik9f570b82011-01-25 12:42:37 -0500524
525 /*
526 * we've only changed i_size in ram, and we haven't updated
527 * the disk i_size. There is no need to log the inode
528 * at this time.
529 */
530 if (end_pos > isize)
Chris Masona52d9a82007-08-27 16:49:44 -0400531 i_size_write(inode, end_pos);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400532 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400533}
534
Chris Masond352ac62008-09-29 15:18:18 -0400535/*
536 * this drops all the extents in the cache that intersect the range
537 * [start, end]. Existing extents are split as required.
538 */
Josef Bacik7014cdb2012-08-30 20:06:49 -0400539void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
540 int skip_pinned)
Chris Masona52d9a82007-08-27 16:49:44 -0400541{
542 struct extent_map *em;
Chris Mason3b951512008-04-17 11:29:12 -0400543 struct extent_map *split = NULL;
544 struct extent_map *split2 = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -0400545 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan39b56372008-02-15 10:40:50 -0500546 u64 len = end - start + 1;
Josef Bacik5dc562c2012-08-17 13:14:17 -0400547 u64 gen;
Chris Mason3b951512008-04-17 11:29:12 -0400548 int ret;
549 int testend = 1;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400550 unsigned long flags;
Chris Masonc8b97812008-10-29 14:49:59 -0400551 int compressed = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400552 bool modified;
Chris Masona52d9a82007-08-27 16:49:44 -0400553
Chris Masone6dcd2d2008-07-17 12:53:50 -0400554 WARN_ON(end < start);
Chris Mason3b951512008-04-17 11:29:12 -0400555 if (end == (u64)-1) {
Yan39b56372008-02-15 10:40:50 -0500556 len = (u64)-1;
Chris Mason3b951512008-04-17 11:29:12 -0400557 testend = 0;
558 }
Chris Masond3977122009-01-05 21:25:51 -0500559 while (1) {
Josef Bacik7014cdb2012-08-30 20:06:49 -0400560 int no_splits = 0;
561
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400562 modified = false;
Chris Mason3b951512008-04-17 11:29:12 -0400563 if (!split)
David Sterba172ddd62011-04-21 00:48:27 +0200564 split = alloc_extent_map();
Chris Mason3b951512008-04-17 11:29:12 -0400565 if (!split2)
David Sterba172ddd62011-04-21 00:48:27 +0200566 split2 = alloc_extent_map();
Josef Bacik7014cdb2012-08-30 20:06:49 -0400567 if (!split || !split2)
568 no_splits = 1;
Chris Mason3b951512008-04-17 11:29:12 -0400569
Chris Mason890871b2009-09-02 16:24:52 -0400570 write_lock(&em_tree->lock);
Yan39b56372008-02-15 10:40:50 -0500571 em = lookup_extent_mapping(em_tree, start, len);
Chris Masond1310b22008-01-24 16:13:08 -0500572 if (!em) {
Chris Mason890871b2009-09-02 16:24:52 -0400573 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -0400574 break;
Chris Masond1310b22008-01-24 16:13:08 -0500575 }
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400576 flags = em->flags;
Josef Bacik5dc562c2012-08-17 13:14:17 -0400577 gen = em->generation;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400578 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
Yan, Zheng55ef6892009-11-12 09:36:44 +0000579 if (testend && em->start + em->len >= start + len) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400580 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400581 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400582 break;
583 }
Yan, Zheng55ef6892009-11-12 09:36:44 +0000584 start = em->start + em->len;
585 if (testend)
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400586 len = start + len - (em->start + em->len);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400587 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400588 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400589 continue;
590 }
Chris Masonc8b97812008-10-29 14:49:59 -0400591 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Chris Mason3ce7e672008-07-31 15:42:54 -0400592 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
Liu Bo3b277592013-03-15 08:46:39 -0600593 clear_bit(EXTENT_FLAG_LOGGING, &flags);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400594 modified = !list_empty(&em->list);
Chris Masona52d9a82007-08-27 16:49:44 -0400595 remove_extent_mapping(em_tree, em);
Josef Bacik7014cdb2012-08-30 20:06:49 -0400596 if (no_splits)
597 goto next;
Chris Mason3b951512008-04-17 11:29:12 -0400598
Josef Bacikee20a982013-07-11 10:34:59 -0400599 if (em->start < start) {
Chris Mason3b951512008-04-17 11:29:12 -0400600 split->start = em->start;
601 split->len = start - em->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400602
Josef Bacikee20a982013-07-11 10:34:59 -0400603 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
604 split->orig_start = em->orig_start;
605 split->block_start = em->block_start;
606
607 if (compressed)
608 split->block_len = em->block_len;
609 else
610 split->block_len = split->len;
611 split->orig_block_len = max(split->block_len,
612 em->orig_block_len);
613 split->ram_bytes = em->ram_bytes;
614 } else {
615 split->orig_start = split->start;
616 split->block_len = 0;
617 split->block_start = em->block_start;
618 split->orig_block_len = 0;
619 split->ram_bytes = split->len;
620 }
621
Josef Bacik5dc562c2012-08-17 13:14:17 -0400622 split->generation = gen;
Chris Mason3b951512008-04-17 11:29:12 -0400623 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400624 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800625 split->compress_type = em->compress_type;
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400626 ret = add_extent_mapping(em_tree, split, modified);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100627 BUG_ON(ret); /* Logic error */
Chris Mason3b951512008-04-17 11:29:12 -0400628 free_extent_map(split);
629 split = split2;
630 split2 = NULL;
631 }
Josef Bacikee20a982013-07-11 10:34:59 -0400632 if (testend && em->start + em->len > start + len) {
Chris Mason3b951512008-04-17 11:29:12 -0400633 u64 diff = start + len - em->start;
634
635 split->start = start + len;
636 split->len = em->start + em->len - (start + len);
637 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400638 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800639 split->compress_type = em->compress_type;
Josef Bacik5dc562c2012-08-17 13:14:17 -0400640 split->generation = gen;
Chris Mason3b951512008-04-17 11:29:12 -0400641
Josef Bacikee20a982013-07-11 10:34:59 -0400642 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
643 split->orig_block_len = max(em->block_len,
644 em->orig_block_len);
645
646 split->ram_bytes = em->ram_bytes;
647 if (compressed) {
648 split->block_len = em->block_len;
649 split->block_start = em->block_start;
650 split->orig_start = em->orig_start;
651 } else {
652 split->block_len = split->len;
653 split->block_start = em->block_start
654 + diff;
655 split->orig_start = em->orig_start;
656 }
Chris Masonc8b97812008-10-29 14:49:59 -0400657 } else {
Josef Bacikee20a982013-07-11 10:34:59 -0400658 split->ram_bytes = split->len;
659 split->orig_start = split->start;
660 split->block_len = 0;
661 split->block_start = em->block_start;
662 split->orig_block_len = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400663 }
Chris Mason3b951512008-04-17 11:29:12 -0400664
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400665 ret = add_extent_mapping(em_tree, split, modified);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100666 BUG_ON(ret); /* Logic error */
Chris Mason3b951512008-04-17 11:29:12 -0400667 free_extent_map(split);
668 split = NULL;
669 }
Josef Bacik7014cdb2012-08-30 20:06:49 -0400670next:
Chris Mason890871b2009-09-02 16:24:52 -0400671 write_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -0500672
Chris Masona52d9a82007-08-27 16:49:44 -0400673 /* once for us */
674 free_extent_map(em);
675 /* once for the tree*/
676 free_extent_map(em);
677 }
Chris Mason3b951512008-04-17 11:29:12 -0400678 if (split)
679 free_extent_map(split);
680 if (split2)
681 free_extent_map(split2);
Chris Masona52d9a82007-08-27 16:49:44 -0400682}
683
Chris Mason39279cc2007-06-12 06:35:45 -0400684/*
685 * this is very complex, but the basic idea is to drop all extents
686 * in the range start - end. hint_block is filled in with a block number
687 * that would be a good hint to the block allocator for this file.
688 *
689 * If an extent intersects the range but is not entirely inside the range
690 * it is either truncated or split. Anything entirely inside the range
691 * is deleted from the tree.
692 */
Josef Bacik5dc562c2012-08-17 13:14:17 -0400693int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root, struct inode *inode,
695 struct btrfs_path *path, u64 start, u64 end,
Josef Bacik2aaa6652012-08-29 14:27:18 -0400696 u64 *drop_end, int drop_cache)
Chris Mason39279cc2007-06-12 06:35:45 -0400697{
Chris Mason00f5c792007-11-30 10:09:33 -0500698 struct extent_buffer *leaf;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000699 struct btrfs_file_extent_item *fi;
Chris Mason00f5c792007-11-30 10:09:33 -0500700 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000701 struct btrfs_key new_key;
Li Zefan33345d012011-04-20 10:31:50 +0800702 u64 ino = btrfs_ino(inode);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000703 u64 search_start = start;
704 u64 disk_bytenr = 0;
705 u64 num_bytes = 0;
706 u64 extent_offset = 0;
707 u64 extent_end = 0;
708 int del_nr = 0;
709 int del_slot = 0;
710 int extent_type;
Chris Masonccd467d2007-06-28 15:57:36 -0400711 int recow;
Chris Mason00f5c792007-11-30 10:09:33 -0500712 int ret;
Chris Masondc7fdde2012-04-27 14:31:29 -0400713 int modify_tree = -1;
Josef Bacik5dc562c2012-08-17 13:14:17 -0400714 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
Josef Bacikc3308f82012-09-14 14:51:22 -0400715 int found = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400716
Chris Masona1ed8352009-09-11 12:27:37 -0400717 if (drop_cache)
718 btrfs_drop_extent_cache(inode, start, end - 1, 0);
Chris Masona52d9a82007-08-27 16:49:44 -0400719
Chris Masondc7fdde2012-04-27 14:31:29 -0400720 if (start >= BTRFS_I(inode)->disk_i_size)
721 modify_tree = 0;
722
Chris Masond3977122009-01-05 21:25:51 -0500723 while (1) {
Chris Masonccd467d2007-06-28 15:57:36 -0400724 recow = 0;
Li Zefan33345d012011-04-20 10:31:50 +0800725 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Chris Masondc7fdde2012-04-27 14:31:29 -0400726 search_start, modify_tree);
Chris Mason39279cc2007-06-12 06:35:45 -0400727 if (ret < 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000728 break;
729 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
730 leaf = path->nodes[0];
731 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +0800732 if (key.objectid == ino &&
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000733 key.type == BTRFS_EXTENT_DATA_KEY)
734 path->slots[0]--;
Chris Mason39279cc2007-06-12 06:35:45 -0400735 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400736 ret = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000737next_slot:
738 leaf = path->nodes[0];
739 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
740 BUG_ON(del_nr > 0);
741 ret = btrfs_next_leaf(root, path);
742 if (ret < 0)
743 break;
744 if (ret > 0) {
745 ret = 0;
746 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400747 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000748 leaf = path->nodes[0];
749 recow = 1;
750 }
751
752 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800753 if (key.objectid > ino ||
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000754 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
755 break;
756
757 fi = btrfs_item_ptr(leaf, path->slots[0],
758 struct btrfs_file_extent_item);
759 extent_type = btrfs_file_extent_type(leaf, fi);
760
761 if (extent_type == BTRFS_FILE_EXTENT_REG ||
762 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
763 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
764 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
765 extent_offset = btrfs_file_extent_offset(leaf, fi);
766 extent_end = key.offset +
767 btrfs_file_extent_num_bytes(leaf, fi);
768 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
769 extent_end = key.offset +
770 btrfs_file_extent_inline_len(leaf, fi);
Chris Mason8c2383c2007-06-18 09:57:58 -0400771 } else {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000772 WARN_ON(1);
Chris Mason8c2383c2007-06-18 09:57:58 -0400773 extent_end = search_start;
Chris Mason39279cc2007-06-12 06:35:45 -0400774 }
775
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000776 if (extent_end <= search_start) {
777 path->slots[0]++;
Chris Mason8c2383c2007-06-18 09:57:58 -0400778 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400779 }
780
Josef Bacikc3308f82012-09-14 14:51:22 -0400781 found = 1;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000782 search_start = max(key.offset, start);
Chris Masondc7fdde2012-04-27 14:31:29 -0400783 if (recow || !modify_tree) {
784 modify_tree = -1;
David Sterbab3b4aa72011-04-21 01:20:15 +0200785 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000786 continue;
Chris Mason39279cc2007-06-12 06:35:45 -0400787 }
Chris Mason771ed682008-11-06 22:02:51 -0500788
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000789 /*
790 * | - range to drop - |
791 * | -------- extent -------- |
792 */
793 if (start > key.offset && end < extent_end) {
794 BUG_ON(del_nr > 0);
795 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
796
797 memcpy(&new_key, &key, sizeof(new_key));
798 new_key.offset = start;
799 ret = btrfs_duplicate_item(trans, root, path,
800 &new_key);
801 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200802 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000803 continue;
804 }
805 if (ret < 0)
806 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400807
Chris Mason5f39d392007-10-15 16:14:19 -0400808 leaf = path->nodes[0];
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000809 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
810 struct btrfs_file_extent_item);
811 btrfs_set_file_extent_num_bytes(leaf, fi,
812 start - key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400813
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000814 fi = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_file_extent_item);
Chris Masonc8b97812008-10-29 14:49:59 -0400816
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000817 extent_offset += start - key.offset;
818 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
819 btrfs_set_file_extent_num_bytes(leaf, fi,
820 extent_end - start);
821 btrfs_mark_buffer_dirty(leaf);
Chris Masondb945352007-10-15 16:15:53 -0400822
Josef Bacik5dc562c2012-08-17 13:14:17 -0400823 if (update_refs && disk_bytenr > 0) {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000824 ret = btrfs_inc_extent_ref(trans, root,
825 disk_bytenr, num_bytes, 0,
826 root->root_key.objectid,
827 new_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200828 start - extent_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100829 BUG_ON(ret); /* -ENOMEM */
Zheng Yan31840ae2008-09-23 13:14:14 -0400830 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000831 key.offset = start;
832 }
833 /*
834 * | ---- range to drop ----- |
835 * | -------- extent -------- |
836 */
837 if (start <= key.offset && end < extent_end) {
838 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
839
840 memcpy(&new_key, &key, sizeof(new_key));
841 new_key.offset = end;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000842 btrfs_set_item_key_safe(root, path, &new_key);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000843
844 extent_offset += end - key.offset;
845 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
846 btrfs_set_file_extent_num_bytes(leaf, fi,
847 extent_end - end);
848 btrfs_mark_buffer_dirty(leaf);
Josef Bacik26714852012-08-29 12:24:27 -0400849 if (update_refs && disk_bytenr > 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000850 inode_sub_bytes(inode, end - key.offset);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000851 break;
Zheng Yan31840ae2008-09-23 13:14:14 -0400852 }
853
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000854 search_start = extent_end;
855 /*
856 * | ---- range to drop ----- |
857 * | -------- extent -------- |
858 */
859 if (start > key.offset && end >= extent_end) {
860 BUG_ON(del_nr > 0);
861 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
862
863 btrfs_set_file_extent_num_bytes(leaf, fi,
864 start - key.offset);
865 btrfs_mark_buffer_dirty(leaf);
Josef Bacik26714852012-08-29 12:24:27 -0400866 if (update_refs && disk_bytenr > 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000867 inode_sub_bytes(inode, extent_end - start);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000868 if (end == extent_end)
869 break;
870
871 path->slots[0]++;
872 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400873 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000874
875 /*
876 * | ---- range to drop ----- |
877 * | ------ extent ------ |
878 */
879 if (start <= key.offset && end >= extent_end) {
880 if (del_nr == 0) {
881 del_slot = path->slots[0];
882 del_nr = 1;
883 } else {
884 BUG_ON(del_slot + del_nr != path->slots[0]);
885 del_nr++;
886 }
887
Josef Bacik5dc562c2012-08-17 13:14:17 -0400888 if (update_refs &&
889 extent_type == BTRFS_FILE_EXTENT_INLINE) {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000890 inode_sub_bytes(inode,
891 extent_end - key.offset);
892 extent_end = ALIGN(extent_end,
893 root->sectorsize);
Josef Bacik5dc562c2012-08-17 13:14:17 -0400894 } else if (update_refs && disk_bytenr > 0) {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000895 ret = btrfs_free_extent(trans, root,
896 disk_bytenr, num_bytes, 0,
897 root->root_key.objectid,
898 key.objectid, key.offset -
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200899 extent_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100900 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000901 inode_sub_bytes(inode,
902 extent_end - key.offset);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000903 }
904
905 if (end == extent_end)
906 break;
907
908 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
909 path->slots[0]++;
910 goto next_slot;
911 }
912
913 ret = btrfs_del_items(trans, root, path, del_slot,
914 del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100915 if (ret) {
916 btrfs_abort_transaction(trans, root, ret);
Josef Bacik5dc562c2012-08-17 13:14:17 -0400917 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100918 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000919
920 del_nr = 0;
921 del_slot = 0;
922
David Sterbab3b4aa72011-04-21 01:20:15 +0200923 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000924 continue;
925 }
926
927 BUG_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -0400928 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000929
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100930 if (!ret && del_nr > 0) {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000931 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100932 if (ret)
933 btrfs_abort_transaction(trans, root, ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000934 }
935
Josef Bacik2aaa6652012-08-29 14:27:18 -0400936 if (drop_end)
Josef Bacikc3308f82012-09-14 14:51:22 -0400937 *drop_end = found ? min(end, extent_end) : end;
Josef Bacik5dc562c2012-08-17 13:14:17 -0400938 btrfs_release_path(path);
939 return ret;
940}
941
942int btrfs_drop_extents(struct btrfs_trans_handle *trans,
943 struct btrfs_root *root, struct inode *inode, u64 start,
Josef Bacik26714852012-08-29 12:24:27 -0400944 u64 end, int drop_cache)
Josef Bacik5dc562c2012-08-17 13:14:17 -0400945{
946 struct btrfs_path *path;
947 int ret;
948
949 path = btrfs_alloc_path();
950 if (!path)
951 return -ENOMEM;
Josef Bacik2aaa6652012-08-29 14:27:18 -0400952 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
Josef Bacik26714852012-08-29 12:24:27 -0400953 drop_cache);
Chris Mason39279cc2007-06-12 06:35:45 -0400954 btrfs_free_path(path);
955 return ret;
956}
957
Yan Zhengd899e052008-10-30 14:25:28 -0400958static int extent_mergeable(struct extent_buffer *leaf, int slot,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000959 u64 objectid, u64 bytenr, u64 orig_offset,
960 u64 *start, u64 *end)
Yan Zhengd899e052008-10-30 14:25:28 -0400961{
962 struct btrfs_file_extent_item *fi;
963 struct btrfs_key key;
964 u64 extent_end;
965
966 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
967 return 0;
968
969 btrfs_item_key_to_cpu(leaf, &key, slot);
970 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
971 return 0;
972
973 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
974 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
975 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000976 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Yan Zhengd899e052008-10-30 14:25:28 -0400977 btrfs_file_extent_compression(leaf, fi) ||
978 btrfs_file_extent_encryption(leaf, fi) ||
979 btrfs_file_extent_other_encoding(leaf, fi))
980 return 0;
981
982 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
983 if ((*start && *start != key.offset) || (*end && *end != extent_end))
984 return 0;
985
986 *start = key.offset;
987 *end = extent_end;
988 return 1;
989}
990
991/*
992 * Mark extent in the range start - end as written.
993 *
994 * This changes extent type from 'pre-allocated' to 'regular'. If only
995 * part of extent is marked as written, the extent will be split into
996 * two or three.
997 */
998int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
Yan Zhengd899e052008-10-30 14:25:28 -0400999 struct inode *inode, u64 start, u64 end)
1000{
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001001 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zhengd899e052008-10-30 14:25:28 -04001002 struct extent_buffer *leaf;
1003 struct btrfs_path *path;
1004 struct btrfs_file_extent_item *fi;
1005 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001006 struct btrfs_key new_key;
Yan Zhengd899e052008-10-30 14:25:28 -04001007 u64 bytenr;
1008 u64 num_bytes;
1009 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001010 u64 orig_offset;
Yan Zhengd899e052008-10-30 14:25:28 -04001011 u64 other_start;
1012 u64 other_end;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001013 u64 split;
1014 int del_nr = 0;
1015 int del_slot = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001016 int recow;
Yan Zhengd899e052008-10-30 14:25:28 -04001017 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08001018 u64 ino = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04001019
Yan Zhengd899e052008-10-30 14:25:28 -04001020 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001021 if (!path)
1022 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001023again:
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001024 recow = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001025 split = start;
Li Zefan33345d012011-04-20 10:31:50 +08001026 key.objectid = ino;
Yan Zhengd899e052008-10-30 14:25:28 -04001027 key.type = BTRFS_EXTENT_DATA_KEY;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001028 key.offset = split;
Yan Zhengd899e052008-10-30 14:25:28 -04001029
1030 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Josef Bacik41415732011-03-16 13:59:32 -04001031 if (ret < 0)
1032 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001033 if (ret > 0 && path->slots[0] > 0)
1034 path->slots[0]--;
1035
1036 leaf = path->nodes[0];
1037 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001038 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
Yan Zhengd899e052008-10-30 14:25:28 -04001039 fi = btrfs_item_ptr(leaf, path->slots[0],
1040 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001041 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1042 BTRFS_FILE_EXTENT_PREALLOC);
Yan Zhengd899e052008-10-30 14:25:28 -04001043 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1044 BUG_ON(key.offset > start || extent_end < end);
1045
1046 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1047 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001048 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001049 memcpy(&new_key, &key, sizeof(new_key));
1050
1051 if (start == key.offset && end < extent_end) {
1052 other_start = 0;
1053 other_end = start;
1054 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +08001055 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001056 &other_start, &other_end)) {
1057 new_key.offset = end;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001058 btrfs_set_item_key_safe(root, path, &new_key);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001059 fi = btrfs_item_ptr(leaf, path->slots[0],
1060 struct btrfs_file_extent_item);
Josef Bacik224ecce2012-08-16 16:32:06 -04001061 btrfs_set_file_extent_generation(leaf, fi,
1062 trans->transid);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001063 btrfs_set_file_extent_num_bytes(leaf, fi,
1064 extent_end - end);
1065 btrfs_set_file_extent_offset(leaf, fi,
1066 end - orig_offset);
1067 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1068 struct btrfs_file_extent_item);
Josef Bacik224ecce2012-08-16 16:32:06 -04001069 btrfs_set_file_extent_generation(leaf, fi,
1070 trans->transid);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001071 btrfs_set_file_extent_num_bytes(leaf, fi,
1072 end - other_start);
1073 btrfs_mark_buffer_dirty(leaf);
1074 goto out;
1075 }
1076 }
1077
1078 if (start > key.offset && end == extent_end) {
1079 other_start = end;
1080 other_end = 0;
1081 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +08001082 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001083 &other_start, &other_end)) {
1084 fi = btrfs_item_ptr(leaf, path->slots[0],
1085 struct btrfs_file_extent_item);
1086 btrfs_set_file_extent_num_bytes(leaf, fi,
1087 start - key.offset);
Josef Bacik224ecce2012-08-16 16:32:06 -04001088 btrfs_set_file_extent_generation(leaf, fi,
1089 trans->transid);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001090 path->slots[0]++;
1091 new_key.offset = start;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001092 btrfs_set_item_key_safe(root, path, &new_key);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001093
1094 fi = btrfs_item_ptr(leaf, path->slots[0],
1095 struct btrfs_file_extent_item);
Josef Bacik224ecce2012-08-16 16:32:06 -04001096 btrfs_set_file_extent_generation(leaf, fi,
1097 trans->transid);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001098 btrfs_set_file_extent_num_bytes(leaf, fi,
1099 other_end - start);
1100 btrfs_set_file_extent_offset(leaf, fi,
1101 start - orig_offset);
1102 btrfs_mark_buffer_dirty(leaf);
1103 goto out;
1104 }
1105 }
Yan Zhengd899e052008-10-30 14:25:28 -04001106
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001107 while (start > key.offset || end < extent_end) {
1108 if (key.offset == start)
1109 split = end;
Yan Zhengd899e052008-10-30 14:25:28 -04001110
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001111 new_key.offset = split;
1112 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1113 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001114 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001115 goto again;
Yan Zhengd899e052008-10-30 14:25:28 -04001116 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001117 if (ret < 0) {
1118 btrfs_abort_transaction(trans, root, ret);
1119 goto out;
1120 }
Yan Zhengd899e052008-10-30 14:25:28 -04001121
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001122 leaf = path->nodes[0];
1123 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Yan Zhengd899e052008-10-30 14:25:28 -04001124 struct btrfs_file_extent_item);
Josef Bacik224ecce2012-08-16 16:32:06 -04001125 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Yan Zhengd899e052008-10-30 14:25:28 -04001126 btrfs_set_file_extent_num_bytes(leaf, fi,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001127 split - key.offset);
1128
1129 fi = btrfs_item_ptr(leaf, path->slots[0],
1130 struct btrfs_file_extent_item);
1131
Josef Bacik224ecce2012-08-16 16:32:06 -04001132 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001133 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1134 btrfs_set_file_extent_num_bytes(leaf, fi,
1135 extent_end - split);
Yan Zhengd899e052008-10-30 14:25:28 -04001136 btrfs_mark_buffer_dirty(leaf);
1137
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001138 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1139 root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001140 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001141 BUG_ON(ret); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001142
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001143 if (split == start) {
1144 key.offset = start;
1145 } else {
1146 BUG_ON(start != key.offset);
Yan Zhengd899e052008-10-30 14:25:28 -04001147 path->slots[0]--;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001148 extent_end = end;
Yan Zhengd899e052008-10-30 14:25:28 -04001149 }
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001150 recow = 1;
Yan Zhengd899e052008-10-30 14:25:28 -04001151 }
1152
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001153 other_start = end;
1154 other_end = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001155 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +08001156 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001157 &other_start, &other_end)) {
1158 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001159 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001160 goto again;
1161 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001162 extent_end = other_end;
1163 del_slot = path->slots[0] + 1;
1164 del_nr++;
1165 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1166 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001167 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001168 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001169 }
1170 other_start = 0;
1171 other_end = start;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001172 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +08001173 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001174 &other_start, &other_end)) {
1175 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001176 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001177 goto again;
1178 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001179 key.offset = other_start;
1180 del_slot = path->slots[0];
1181 del_nr++;
1182 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1183 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001184 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001185 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001186 }
1187 if (del_nr == 0) {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001188 fi = btrfs_item_ptr(leaf, path->slots[0],
1189 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001190 btrfs_set_file_extent_type(leaf, fi,
1191 BTRFS_FILE_EXTENT_REG);
Josef Bacik224ecce2012-08-16 16:32:06 -04001192 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001193 btrfs_mark_buffer_dirty(leaf);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001194 } else {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001195 fi = btrfs_item_ptr(leaf, del_slot - 1,
1196 struct btrfs_file_extent_item);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001197 btrfs_set_file_extent_type(leaf, fi,
1198 BTRFS_FILE_EXTENT_REG);
Josef Bacik224ecce2012-08-16 16:32:06 -04001199 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001200 btrfs_set_file_extent_num_bytes(leaf, fi,
1201 extent_end - key.offset);
1202 btrfs_mark_buffer_dirty(leaf);
1203
1204 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001205 if (ret < 0) {
1206 btrfs_abort_transaction(trans, root, ret);
1207 goto out;
1208 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001209 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001210out:
Yan Zhengd899e052008-10-30 14:25:28 -04001211 btrfs_free_path(path);
1212 return 0;
1213}
1214
Chris Mason39279cc2007-06-12 06:35:45 -04001215/*
Chris Masonb1bf8622011-02-28 09:52:08 -05001216 * on error we return an unlocked page and the error value
1217 * on success we return a locked page and 0
1218 */
Josef Bacikb63164292011-09-30 15:23:54 -04001219static int prepare_uptodate_page(struct page *page, u64 pos,
1220 bool force_uptodate)
Chris Masonb1bf8622011-02-28 09:52:08 -05001221{
1222 int ret = 0;
1223
Josef Bacikb63164292011-09-30 15:23:54 -04001224 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1225 !PageUptodate(page)) {
Chris Masonb1bf8622011-02-28 09:52:08 -05001226 ret = btrfs_readpage(NULL, page);
1227 if (ret)
1228 return ret;
1229 lock_page(page);
1230 if (!PageUptodate(page)) {
1231 unlock_page(page);
1232 return -EIO;
1233 }
1234 }
1235 return 0;
1236}
1237
1238/*
Chris Masond352ac62008-09-29 15:18:18 -04001239 * this gets pages into the page cache and locks them down, it also properly
1240 * waits for data=ordered extents to finish before allowing the pages to be
1241 * modified.
Chris Mason39279cc2007-06-12 06:35:45 -04001242 */
Chris Masond3977122009-01-05 21:25:51 -05001243static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
Chris Mason98ed5172008-01-03 10:01:48 -05001244 struct page **pages, size_t num_pages,
1245 loff_t pos, unsigned long first_index,
Josef Bacikb63164292011-09-30 15:23:54 -04001246 size_t write_bytes, bool force_uptodate)
Chris Mason39279cc2007-06-12 06:35:45 -04001247{
Josef Bacik2ac55d42010-02-03 19:33:23 +00001248 struct extent_state *cached_state = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001249 int i;
1250 unsigned long index = pos >> PAGE_CACHE_SHIFT;
Al Viro496ad9a2013-01-23 17:07:38 -05001251 struct inode *inode = file_inode(file);
Josef Bacik3b16a4e2011-09-21 15:05:58 -04001252 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
Chris Mason39279cc2007-06-12 06:35:45 -04001253 int err = 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001254 int faili = 0;
Chris Mason8c2383c2007-06-18 09:57:58 -04001255 u64 start_pos;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001256 u64 last_pos;
Chris Mason8c2383c2007-06-18 09:57:58 -04001257
Chris Mason5f39d392007-10-15 16:14:19 -04001258 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001259 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001260
Chris Masone6dcd2d2008-07-17 12:53:50 -04001261again:
Chris Mason39279cc2007-06-12 06:35:45 -04001262 for (i = 0; i < num_pages; i++) {
Josef Bacika94733d2011-07-11 10:47:06 -04001263 pages[i] = find_or_create_page(inode->i_mapping, index + i,
Johannes Weinere3a41a52012-01-10 15:07:55 -08001264 mask | __GFP_WRITE);
Chris Mason39279cc2007-06-12 06:35:45 -04001265 if (!pages[i]) {
Chris Masonb1bf8622011-02-28 09:52:08 -05001266 faili = i - 1;
1267 err = -ENOMEM;
1268 goto fail;
1269 }
1270
1271 if (i == 0)
Josef Bacikb63164292011-09-30 15:23:54 -04001272 err = prepare_uptodate_page(pages[i], pos,
1273 force_uptodate);
Chris Masonb1bf8622011-02-28 09:52:08 -05001274 if (i == num_pages - 1)
1275 err = prepare_uptodate_page(pages[i],
Josef Bacikb63164292011-09-30 15:23:54 -04001276 pos + write_bytes, false);
Chris Masonb1bf8622011-02-28 09:52:08 -05001277 if (err) {
1278 page_cache_release(pages[i]);
1279 faili = i - 1;
1280 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -04001281 }
Chris Masonccd467d2007-06-28 15:57:36 -04001282 wait_on_page_writeback(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -04001283 }
Chris Masonb1bf8622011-02-28 09:52:08 -05001284 err = 0;
Chris Mason07627042008-02-19 11:29:24 -05001285 if (start_pos < inode->i_size) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04001286 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001287 lock_extent_bits(&BTRFS_I(inode)->io_tree,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001288 start_pos, last_pos - 1, 0, &cached_state);
Chris Masond3977122009-01-05 21:25:51 -05001289 ordered = btrfs_lookup_first_ordered_extent(inode,
1290 last_pos - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001291 if (ordered &&
1292 ordered->file_offset + ordered->len > start_pos &&
1293 ordered->file_offset < last_pos) {
1294 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001295 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1296 start_pos, last_pos - 1,
1297 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001298 for (i = 0; i < num_pages; i++) {
1299 unlock_page(pages[i]);
1300 page_cache_release(pages[i]);
1301 }
1302 btrfs_wait_ordered_range(inode, start_pos,
1303 last_pos - start_pos);
1304 goto again;
1305 }
1306 if (ordered)
1307 btrfs_put_ordered_extent(ordered);
1308
Josef Bacik2ac55d42010-02-03 19:33:23 +00001309 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
Josef Bacik32c00af2009-10-08 13:34:05 -04001310 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
Liu Bo9e8a4a82012-09-05 19:10:51 -06001311 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1312 0, 0, &cached_state, GFP_NOFS);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001313 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1314 start_pos, last_pos - 1, &cached_state,
1315 GFP_NOFS);
Chris Mason07627042008-02-19 11:29:24 -05001316 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001317 for (i = 0; i < num_pages; i++) {
Wu Fengguang32c7f202011-08-08 15:19:47 -06001318 if (clear_page_dirty_for_io(pages[i]))
1319 account_page_redirty(pages[i]);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001320 set_page_extent_mapped(pages[i]);
1321 WARN_ON(!PageLocked(pages[i]));
1322 }
Chris Mason39279cc2007-06-12 06:35:45 -04001323 return 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001324fail:
1325 while (faili >= 0) {
1326 unlock_page(pages[faili]);
1327 page_cache_release(pages[faili]);
1328 faili--;
1329 }
1330 return err;
1331
Chris Mason39279cc2007-06-12 06:35:45 -04001332}
1333
Josef Bacik7ee9e442013-06-21 16:37:03 -04001334static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1335 size_t *write_bytes)
1336{
1337 struct btrfs_trans_handle *trans;
1338 struct btrfs_root *root = BTRFS_I(inode)->root;
1339 struct btrfs_ordered_extent *ordered;
1340 u64 lockstart, lockend;
1341 u64 num_bytes;
1342 int ret;
1343
1344 lockstart = round_down(pos, root->sectorsize);
1345 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1346
1347 while (1) {
1348 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1349 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1350 lockend - lockstart + 1);
1351 if (!ordered) {
1352 break;
1353 }
1354 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1355 btrfs_start_ordered_extent(inode, ordered, 1);
1356 btrfs_put_ordered_extent(ordered);
1357 }
1358
1359 trans = btrfs_join_transaction(root);
1360 if (IS_ERR(trans)) {
1361 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1362 return PTR_ERR(trans);
1363 }
1364
1365 num_bytes = lockend - lockstart + 1;
1366 ret = can_nocow_extent(trans, inode, lockstart, &num_bytes, NULL, NULL,
1367 NULL);
1368 btrfs_end_transaction(trans, root);
1369 if (ret <= 0) {
1370 ret = 0;
1371 } else {
1372 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1373 EXTENT_DIRTY | EXTENT_DELALLOC |
1374 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1375 NULL, GFP_NOFS);
1376 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1377 }
1378
1379 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1380
1381 return ret;
1382}
1383
Josef Bacikd0215f32011-01-25 14:57:24 -05001384static noinline ssize_t __btrfs_buffered_write(struct file *file,
1385 struct iov_iter *i,
1386 loff_t pos)
Josef Bacik4b46fce2010-05-23 11:00:55 -04001387{
Al Viro496ad9a2013-01-23 17:07:38 -05001388 struct inode *inode = file_inode(file);
Josef Bacik11c65dc2010-05-23 11:07:21 -04001389 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik11c65dc2010-05-23 11:07:21 -04001390 struct page **pages = NULL;
Josef Bacik7ee9e442013-06-21 16:37:03 -04001391 u64 release_bytes = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001392 unsigned long first_index;
Josef Bacikd0215f32011-01-25 14:57:24 -05001393 size_t num_written = 0;
1394 int nrptrs;
Tsutomu Itohc9149232011-03-30 00:57:23 +00001395 int ret = 0;
Josef Bacik7ee9e442013-06-21 16:37:03 -04001396 bool only_release_metadata = false;
Josef Bacikb63164292011-09-30 15:23:54 -04001397 bool force_page_uptodate = false;
Chris Masoncb843a62008-10-03 12:30:02 -04001398
Josef Bacikd0215f32011-01-25 14:57:24 -05001399 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
Josef Bacik11c65dc2010-05-23 11:07:21 -04001400 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1401 (sizeof(struct page *)));
Wu Fengguang142349f2011-12-16 12:32:57 -05001402 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1403 nrptrs = max(nrptrs, 8);
Chris Mason8c2383c2007-06-18 09:57:58 -04001404 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001405 if (!pages)
1406 return -ENOMEM;
Chris Masonab93dbe2009-10-01 12:29:10 -04001407
Chris Mason39279cc2007-06-12 06:35:45 -04001408 first_index = pos >> PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001409
Josef Bacikd0215f32011-01-25 14:57:24 -05001410 while (iov_iter_count(i) > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -04001411 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
Josef Bacikd0215f32011-01-25 14:57:24 -05001412 size_t write_bytes = min(iov_iter_count(i),
Josef Bacik11c65dc2010-05-23 11:07:21 -04001413 nrptrs * (size_t)PAGE_CACHE_SIZE -
Chris Mason8c2383c2007-06-18 09:57:58 -04001414 offset);
Yan, Zheng3a909832011-01-18 13:34:40 +08001415 size_t num_pages = (write_bytes + offset +
1416 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Josef Bacik7ee9e442013-06-21 16:37:03 -04001417 size_t reserve_bytes;
Josef Bacikd0215f32011-01-25 14:57:24 -05001418 size_t dirty_pages;
1419 size_t copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001420
Chris Mason8c2383c2007-06-18 09:57:58 -04001421 WARN_ON(num_pages > nrptrs);
Chris Mason1832a6d2007-12-21 16:27:21 -05001422
Xin Zhong914ee292010-12-09 09:30:14 +00001423 /*
1424 * Fault pages before locking them in prepare_pages
1425 * to avoid recursive lock
1426 */
Josef Bacikd0215f32011-01-25 14:57:24 -05001427 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
Xin Zhong914ee292010-12-09 09:30:14 +00001428 ret = -EFAULT;
Josef Bacikd0215f32011-01-25 14:57:24 -05001429 break;
Xin Zhong914ee292010-12-09 09:30:14 +00001430 }
1431
Josef Bacik7ee9e442013-06-21 16:37:03 -04001432 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1433 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1434 if (ret == -ENOSPC &&
1435 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1436 BTRFS_INODE_PREALLOC))) {
1437 ret = check_can_nocow(inode, pos, &write_bytes);
1438 if (ret > 0) {
1439 only_release_metadata = true;
1440 /*
1441 * our prealloc extent may be smaller than
1442 * write_bytes, so scale down.
1443 */
1444 num_pages = (write_bytes + offset +
1445 PAGE_CACHE_SIZE - 1) >>
1446 PAGE_CACHE_SHIFT;
1447 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1448 ret = 0;
1449 } else {
1450 ret = -ENOSPC;
1451 }
1452 }
1453
Chris Mason1832a6d2007-12-21 16:27:21 -05001454 if (ret)
Josef Bacikd0215f32011-01-25 14:57:24 -05001455 break;
Chris Mason1832a6d2007-12-21 16:27:21 -05001456
Josef Bacik7ee9e442013-06-21 16:37:03 -04001457 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1458 if (ret) {
1459 if (!only_release_metadata)
1460 btrfs_free_reserved_data_space(inode,
1461 reserve_bytes);
1462 break;
1463 }
1464
1465 release_bytes = reserve_bytes;
1466
Josef Bacik4a640012011-01-25 15:10:08 -05001467 /*
1468 * This is going to setup the pages array with the number of
1469 * pages we want, so we don't really need to worry about the
1470 * contents of pages from loop to loop
1471 */
Chris Mason39279cc2007-06-12 06:35:45 -04001472 ret = prepare_pages(root, file, pages, num_pages,
Josef Bacikb63164292011-09-30 15:23:54 -04001473 pos, first_index, write_bytes,
1474 force_page_uptodate);
Josef Bacik7ee9e442013-06-21 16:37:03 -04001475 if (ret)
Josef Bacikd0215f32011-01-25 14:57:24 -05001476 break;
Chris Mason39279cc2007-06-12 06:35:45 -04001477
Xin Zhong914ee292010-12-09 09:30:14 +00001478 copied = btrfs_copy_from_user(pos, num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -05001479 write_bytes, pages, i);
Chris Masonb1bf8622011-02-28 09:52:08 -05001480
1481 /*
1482 * if we have trouble faulting in the pages, fall
1483 * back to one page at a time
1484 */
1485 if (copied < write_bytes)
1486 nrptrs = 1;
1487
Josef Bacikb63164292011-09-30 15:23:54 -04001488 if (copied == 0) {
1489 force_page_uptodate = true;
Chris Masonb1bf8622011-02-28 09:52:08 -05001490 dirty_pages = 0;
Josef Bacikb63164292011-09-30 15:23:54 -04001491 } else {
1492 force_page_uptodate = false;
Chris Masonb1bf8622011-02-28 09:52:08 -05001493 dirty_pages = (copied + offset +
1494 PAGE_CACHE_SIZE - 1) >>
1495 PAGE_CACHE_SHIFT;
Josef Bacikb63164292011-09-30 15:23:54 -04001496 }
Xin Zhong914ee292010-12-09 09:30:14 +00001497
Josef Bacikd0215f32011-01-25 14:57:24 -05001498 /*
1499 * If we had a short copy we need to release the excess delaloc
1500 * bytes we reserved. We need to increment outstanding_extents
1501 * because btrfs_delalloc_release_space will decrement it, but
1502 * we still have an outstanding extent for the chunk we actually
1503 * managed to copy.
1504 */
Xin Zhong914ee292010-12-09 09:30:14 +00001505 if (num_pages > dirty_pages) {
Josef Bacik7ee9e442013-06-21 16:37:03 -04001506 release_bytes = (num_pages - dirty_pages) <<
1507 PAGE_CACHE_SHIFT;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001508 if (copied > 0) {
1509 spin_lock(&BTRFS_I(inode)->lock);
1510 BTRFS_I(inode)->outstanding_extents++;
1511 spin_unlock(&BTRFS_I(inode)->lock);
1512 }
Josef Bacik7ee9e442013-06-21 16:37:03 -04001513 if (only_release_metadata)
1514 btrfs_delalloc_release_metadata(inode,
1515 release_bytes);
1516 else
1517 btrfs_delalloc_release_space(inode,
1518 release_bytes);
Xin Zhong914ee292010-12-09 09:30:14 +00001519 }
1520
Josef Bacik7ee9e442013-06-21 16:37:03 -04001521 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
Xin Zhong914ee292010-12-09 09:30:14 +00001522 if (copied > 0) {
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001523 ret = btrfs_dirty_pages(root, inode, pages,
1524 dirty_pages, pos, copied,
1525 NULL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001526 if (ret) {
Josef Bacikd0215f32011-01-25 14:57:24 -05001527 btrfs_drop_pages(pages, num_pages);
1528 break;
1529 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001530 }
Chris Mason39279cc2007-06-12 06:35:45 -04001531
Josef Bacik7ee9e442013-06-21 16:37:03 -04001532 release_bytes = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001533 btrfs_drop_pages(pages, num_pages);
Xin Zhong914ee292010-12-09 09:30:14 +00001534
Josef Bacik7ee9e442013-06-21 16:37:03 -04001535 if (only_release_metadata && copied > 0) {
1536 u64 lockstart = round_down(pos, root->sectorsize);
1537 u64 lockend = lockstart +
1538 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1539
1540 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1541 lockend, EXTENT_NORESERVE, NULL,
1542 NULL, GFP_NOFS);
1543 only_release_metadata = false;
1544 }
1545
Josef Bacikd0215f32011-01-25 14:57:24 -05001546 cond_resched();
1547
Namjae Jeond0e1d662012-12-11 16:00:21 -08001548 balance_dirty_pages_ratelimited(inode->i_mapping);
Josef Bacikd0215f32011-01-25 14:57:24 -05001549 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
Liu Bob53d3f52012-11-14 14:34:34 +00001550 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04001551
Xin Zhong914ee292010-12-09 09:30:14 +00001552 pos += copied;
1553 num_written += copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001554 }
Chris Mason5b92ee72008-01-03 13:46:11 -05001555
Chris Mason8c2383c2007-06-18 09:57:58 -04001556 kfree(pages);
Josef Bacikd0215f32011-01-25 14:57:24 -05001557
Josef Bacik7ee9e442013-06-21 16:37:03 -04001558 if (release_bytes) {
1559 if (only_release_metadata)
1560 btrfs_delalloc_release_metadata(inode, release_bytes);
1561 else
1562 btrfs_delalloc_release_space(inode, release_bytes);
1563 }
1564
Josef Bacikd0215f32011-01-25 14:57:24 -05001565 return num_written ? num_written : ret;
1566}
1567
1568static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1569 const struct iovec *iov,
1570 unsigned long nr_segs, loff_t pos,
1571 loff_t *ppos, size_t count, size_t ocount)
1572{
1573 struct file *file = iocb->ki_filp;
Josef Bacikd0215f32011-01-25 14:57:24 -05001574 struct iov_iter i;
1575 ssize_t written;
1576 ssize_t written_buffered;
1577 loff_t endbyte;
1578 int err;
1579
1580 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1581 count, ocount);
1582
Josef Bacikd0215f32011-01-25 14:57:24 -05001583 if (written < 0 || written == count)
1584 return written;
1585
1586 pos += written;
1587 count -= written;
1588 iov_iter_init(&i, iov, nr_segs, count, written);
1589 written_buffered = __btrfs_buffered_write(file, &i, pos);
1590 if (written_buffered < 0) {
1591 err = written_buffered;
1592 goto out;
1593 }
1594 endbyte = pos + written_buffered - 1;
1595 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1596 if (err)
1597 goto out;
1598 written += written_buffered;
1599 *ppos = pos + written_buffered;
1600 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1601 endbyte >> PAGE_CACHE_SHIFT);
1602out:
1603 return written ? written : err;
1604}
1605
Josef Bacik6c760c02012-11-09 10:53:21 -05001606static void update_time_for_write(struct inode *inode)
1607{
1608 struct timespec now;
1609
1610 if (IS_NOCMTIME(inode))
1611 return;
1612
1613 now = current_fs_time(inode->i_sb);
1614 if (!timespec_equal(&inode->i_mtime, &now))
1615 inode->i_mtime = now;
1616
1617 if (!timespec_equal(&inode->i_ctime, &now))
1618 inode->i_ctime = now;
1619
1620 if (IS_I_VERSION(inode))
1621 inode_inc_iversion(inode);
1622}
1623
Josef Bacikd0215f32011-01-25 14:57:24 -05001624static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1625 const struct iovec *iov,
1626 unsigned long nr_segs, loff_t pos)
1627{
1628 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -05001629 struct inode *inode = file_inode(file);
Josef Bacikd0215f32011-01-25 14:57:24 -05001630 struct btrfs_root *root = BTRFS_I(inode)->root;
1631 loff_t *ppos = &iocb->ki_pos;
Miao Xie0c1a98c2011-09-11 10:52:24 -04001632 u64 start_pos;
Josef Bacikd0215f32011-01-25 14:57:24 -05001633 ssize_t num_written = 0;
1634 ssize_t err = 0;
1635 size_t count, ocount;
Josef Bacikb812ce22012-11-16 13:56:32 -05001636 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
Josef Bacikd0215f32011-01-25 14:57:24 -05001637
Josef Bacikd0215f32011-01-25 14:57:24 -05001638 mutex_lock(&inode->i_mutex);
1639
1640 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1641 if (err) {
1642 mutex_unlock(&inode->i_mutex);
1643 goto out;
1644 }
1645 count = ocount;
1646
1647 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1648 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1649 if (err) {
1650 mutex_unlock(&inode->i_mutex);
1651 goto out;
1652 }
1653
1654 if (count == 0) {
1655 mutex_unlock(&inode->i_mutex);
1656 goto out;
1657 }
1658
1659 err = file_remove_suid(file);
1660 if (err) {
1661 mutex_unlock(&inode->i_mutex);
1662 goto out;
1663 }
1664
1665 /*
1666 * If BTRFS flips readonly due to some impossible error
1667 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1668 * although we have opened a file as writable, we have
1669 * to stop this write operation to ensure FS consistency.
1670 */
Miao Xie87533c42013-01-29 10:14:48 +00001671 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
Josef Bacikd0215f32011-01-25 14:57:24 -05001672 mutex_unlock(&inode->i_mutex);
1673 err = -EROFS;
1674 goto out;
1675 }
1676
Josef Bacik6c760c02012-11-09 10:53:21 -05001677 /*
1678 * We reserve space for updating the inode when we reserve space for the
1679 * extent we are going to write, so we will enospc out there. We don't
1680 * need to start yet another transaction to update the inode as we will
1681 * update the inode when we finish writing whatever data we write.
1682 */
1683 update_time_for_write(inode);
Josef Bacikd0215f32011-01-25 14:57:24 -05001684
Miao Xie0c1a98c2011-09-11 10:52:24 -04001685 start_pos = round_down(pos, root->sectorsize);
1686 if (start_pos > i_size_read(inode)) {
1687 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1688 if (err) {
1689 mutex_unlock(&inode->i_mutex);
1690 goto out;
1691 }
1692 }
1693
Josef Bacikb812ce22012-11-16 13:56:32 -05001694 if (sync)
1695 atomic_inc(&BTRFS_I(inode)->sync_writers);
1696
Josef Bacikd0215f32011-01-25 14:57:24 -05001697 if (unlikely(file->f_flags & O_DIRECT)) {
1698 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1699 pos, ppos, count, ocount);
1700 } else {
1701 struct iov_iter i;
1702
1703 iov_iter_init(&i, iov, nr_segs, count, num_written);
1704
1705 num_written = __btrfs_buffered_write(file, &i, pos);
1706 if (num_written > 0)
1707 *ppos = pos + num_written;
1708 }
1709
1710 mutex_unlock(&inode->i_mutex);
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001711
Chris Mason5a3f23d2009-03-31 13:27:11 -04001712 /*
1713 * we want to make sure fsync finds this change
1714 * but we haven't joined a transaction running right now.
1715 *
1716 * Later on, someone is sure to update the inode and get the
1717 * real transid recorded.
1718 *
1719 * We set last_trans now to the fs_info generation + 1,
1720 * this will either be one more than the running transaction
1721 * or the generation used for the next transaction if there isn't
1722 * one running right now.
Josef Bacik6c760c02012-11-09 10:53:21 -05001723 *
1724 * We also have to set last_sub_trans to the current log transid,
1725 * otherwise subsequent syncs to a file that's been synced in this
1726 * transaction will appear to have already occured.
Chris Mason5a3f23d2009-03-31 13:27:11 -04001727 */
1728 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
Josef Bacik6c760c02012-11-09 10:53:21 -05001729 BTRFS_I(inode)->last_sub_trans = root->log_transid;
Christoph Hellwig02afc272013-09-04 15:04:40 +02001730 if (num_written > 0) {
Josef Bacikd0215f32011-01-25 14:57:24 -05001731 err = generic_write_sync(file, pos, num_written);
1732 if (err < 0 && num_written > 0)
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001733 num_written = err;
1734 }
Miao Xie0a3404d2013-01-28 12:34:55 +00001735
Josef Bacikb812ce22012-11-16 13:56:32 -05001736 if (sync)
1737 atomic_dec(&BTRFS_I(inode)->sync_writers);
Miao Xie0a3404d2013-01-28 12:34:55 +00001738out:
Chris Mason39279cc2007-06-12 06:35:45 -04001739 current->backing_dev_info = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001740 return num_written ? num_written : err;
1741}
1742
Chris Masond3977122009-01-05 21:25:51 -05001743int btrfs_release_file(struct inode *inode, struct file *filp)
Mingminge1b81e62008-05-27 10:55:43 -04001744{
Chris Mason5a3f23d2009-03-31 13:27:11 -04001745 /*
1746 * ordered_data_close is set by settattr when we are about to truncate
1747 * a file from a non-zero size to a zero size. This tries to
1748 * flush down new bytes that may have been written if the
1749 * application were using truncate to replace a file in place.
1750 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04001751 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1752 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacik569e0f32013-02-13 11:09:14 -05001753 struct btrfs_trans_handle *trans;
1754 struct btrfs_root *root = BTRFS_I(inode)->root;
1755
1756 /*
1757 * We need to block on a committing transaction to keep us from
1758 * throwing a ordered operation on to the list and causing
1759 * something like sync to deadlock trying to flush out this
1760 * inode.
1761 */
1762 trans = btrfs_start_transaction(root, 0);
1763 if (IS_ERR(trans))
1764 return PTR_ERR(trans);
1765 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1766 btrfs_end_transaction(trans, root);
Chris Mason5a3f23d2009-03-31 13:27:11 -04001767 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1768 filemap_flush(inode->i_mapping);
1769 }
Sage Weil6bf13c02008-06-10 10:07:39 -04001770 if (filp->private_data)
1771 btrfs_ioctl_trans_end(filp);
Mingminge1b81e62008-05-27 10:55:43 -04001772 return 0;
1773}
1774
Chris Masond352ac62008-09-29 15:18:18 -04001775/*
1776 * fsync call for both files and directories. This logs the inode into
1777 * the tree log instead of forcing full commits whenever possible.
1778 *
1779 * It needs to call filemap_fdatawait so that all ordered extent updates are
1780 * in the metadata btree are up to date for copying to the log.
1781 *
1782 * It drops the inode mutex before doing the tree log commit. This is an
1783 * important optimization for directories because holding the mutex prevents
1784 * new operations on the dir while we write to disk.
1785 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001786int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
Chris Mason39279cc2007-06-12 06:35:45 -04001787{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001788 struct dentry *dentry = file->f_path.dentry;
Chris Mason39279cc2007-06-12 06:35:45 -04001789 struct inode *inode = dentry->d_inode;
1790 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001791 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001792 struct btrfs_trans_handle *trans;
Josef Bacik2ab28f32012-10-12 15:27:49 -04001793 bool full_sync = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001794
liubo1abe9b82011-03-24 11:18:59 +00001795 trace_btrfs_sync_file(file, datasync);
Chris Mason257c62e2009-10-13 13:21:08 -04001796
Miao Xie90abccf2012-09-13 04:53:47 -06001797 /*
1798 * We write the dirty pages in the range and wait until they complete
1799 * out of the ->i_mutex. If so, we can flush the dirty pages by
Josef Bacik2ab28f32012-10-12 15:27:49 -04001800 * multi-task, and make the performance up. See
1801 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
Miao Xie90abccf2012-09-13 04:53:47 -06001802 */
Josef Bacikb812ce22012-11-16 13:56:32 -05001803 atomic_inc(&BTRFS_I(inode)->sync_writers);
Josef Bacik2ab28f32012-10-12 15:27:49 -04001804 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1805 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1806 &BTRFS_I(inode)->runtime_flags))
1807 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
Josef Bacikb812ce22012-11-16 13:56:32 -05001808 atomic_dec(&BTRFS_I(inode)->sync_writers);
Miao Xie90abccf2012-09-13 04:53:47 -06001809 if (ret)
1810 return ret;
1811
Josef Bacik02c24a82011-07-16 20:44:56 -04001812 mutex_lock(&inode->i_mutex);
1813
Josef Bacik0885ef52012-04-23 15:09:39 -04001814 /*
Miao Xie90abccf2012-09-13 04:53:47 -06001815 * We flush the dirty pages again to avoid some dirty pages in the
1816 * range being left.
Josef Bacik0885ef52012-04-23 15:09:39 -04001817 */
Miao Xie2ecb7922012-09-06 04:04:27 -06001818 atomic_inc(&root->log_batch);
Josef Bacik2ab28f32012-10-12 15:27:49 -04001819 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1820 &BTRFS_I(inode)->runtime_flags);
1821 if (full_sync)
1822 btrfs_wait_ordered_range(inode, start, end - start + 1);
Miao Xie2ecb7922012-09-06 04:04:27 -06001823 atomic_inc(&root->log_batch);
Chris Mason257c62e2009-10-13 13:21:08 -04001824
Chris Mason39279cc2007-06-12 06:35:45 -04001825 /*
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001826 * check the transaction that last modified this inode
1827 * and see if its already been committed
Chris Mason39279cc2007-06-12 06:35:45 -04001828 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001829 if (!BTRFS_I(inode)->last_trans) {
1830 mutex_unlock(&inode->i_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001831 goto out;
Josef Bacik02c24a82011-07-16 20:44:56 -04001832 }
Chris Masona2135012008-06-25 16:01:30 -04001833
Chris Mason257c62e2009-10-13 13:21:08 -04001834 /*
1835 * if the last transaction that changed this file was before
1836 * the current transaction, we can bail out now without any
1837 * syncing
1838 */
Josef Bacika4abeea2011-04-11 17:25:13 -04001839 smp_mb();
Josef Bacik22ee6982012-05-29 16:57:49 -04001840 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1841 BTRFS_I(inode)->last_trans <=
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001842 root->fs_info->last_trans_committed) {
1843 BTRFS_I(inode)->last_trans = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04001844
1845 /*
1846 * We'v had everything committed since the last time we were
1847 * modified so clear this flag in case it was set for whatever
1848 * reason, it's no longer relevant.
1849 */
1850 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1851 &BTRFS_I(inode)->runtime_flags);
Josef Bacik02c24a82011-07-16 20:44:56 -04001852 mutex_unlock(&inode->i_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001853 goto out;
1854 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001855
1856 /*
Chris Masona52d9a82007-08-27 16:49:44 -04001857 * ok we haven't committed the transaction yet, lets do a commit
1858 */
Dan Carpenter6f902af2010-05-29 09:49:07 +00001859 if (file->private_data)
Sage Weil6bf13c02008-06-10 10:07:39 -04001860 btrfs_ioctl_trans_end(file);
1861
Yan, Zhenga22285a2010-05-16 10:48:46 -04001862 trans = btrfs_start_transaction(root, 0);
1863 if (IS_ERR(trans)) {
1864 ret = PTR_ERR(trans);
Josef Bacik02c24a82011-07-16 20:44:56 -04001865 mutex_unlock(&inode->i_mutex);
Chris Mason39279cc2007-06-12 06:35:45 -04001866 goto out;
1867 }
Chris Masone02119d2008-09-05 16:13:11 -04001868
Chris Mason2cfbd502009-02-20 10:55:10 -05001869 ret = btrfs_log_dentry_safe(trans, root, dentry);
Josef Bacik02c24a82011-07-16 20:44:56 -04001870 if (ret < 0) {
1871 mutex_unlock(&inode->i_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04001872 goto out;
Josef Bacik02c24a82011-07-16 20:44:56 -04001873 }
Chris Mason49eb7e42008-09-11 15:53:12 -04001874
1875 /* we've logged all the items and now have a consistent
1876 * version of the file in the log. It is possible that
1877 * someone will come in and modify the file, but that's
1878 * fine because the log is consistent on disk, and we
1879 * have references to all of the file's extents
1880 *
1881 * It is possible that someone will come in and log the
1882 * file again, but that will end up using the synchronization
1883 * inside btrfs_sync_log to keep things safe.
1884 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001885 mutex_unlock(&inode->i_mutex);
Chris Mason49eb7e42008-09-11 15:53:12 -04001886
Chris Mason257c62e2009-10-13 13:21:08 -04001887 if (ret != BTRFS_NO_LOG_SYNC) {
1888 if (ret > 0) {
Josef Bacik2ab28f32012-10-12 15:27:49 -04001889 /*
1890 * If we didn't already wait for ordered extents we need
1891 * to do that now.
1892 */
1893 if (!full_sync)
1894 btrfs_wait_ordered_range(inode, start,
1895 end - start + 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04001896 ret = btrfs_commit_transaction(trans, root);
Chris Mason257c62e2009-10-13 13:21:08 -04001897 } else {
1898 ret = btrfs_sync_log(trans, root);
Josef Bacik2ab28f32012-10-12 15:27:49 -04001899 if (ret == 0) {
Chris Mason257c62e2009-10-13 13:21:08 -04001900 ret = btrfs_end_transaction(trans, root);
Josef Bacik2ab28f32012-10-12 15:27:49 -04001901 } else {
1902 if (!full_sync)
1903 btrfs_wait_ordered_range(inode, start,
1904 end -
1905 start + 1);
Chris Mason257c62e2009-10-13 13:21:08 -04001906 ret = btrfs_commit_transaction(trans, root);
Josef Bacik2ab28f32012-10-12 15:27:49 -04001907 }
Chris Mason257c62e2009-10-13 13:21:08 -04001908 }
1909 } else {
1910 ret = btrfs_end_transaction(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001911 }
Chris Mason39279cc2007-06-12 06:35:45 -04001912out:
Roel Kluin014e4ac2010-01-29 10:42:11 +00001913 return ret > 0 ? -EIO : ret;
Chris Mason39279cc2007-06-12 06:35:45 -04001914}
1915
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001916static const struct vm_operations_struct btrfs_file_vm_ops = {
Chris Mason92fee662007-07-25 12:31:35 -04001917 .fault = filemap_fault,
Chris Mason9ebefb182007-06-15 13:50:00 -04001918 .page_mkwrite = btrfs_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001919 .remap_pages = generic_file_remap_pages,
Chris Mason9ebefb182007-06-15 13:50:00 -04001920};
1921
1922static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1923{
Miao Xie058a4572010-05-20 07:21:50 +00001924 struct address_space *mapping = filp->f_mapping;
1925
1926 if (!mapping->a_ops->readpage)
1927 return -ENOEXEC;
1928
Chris Mason9ebefb182007-06-15 13:50:00 -04001929 file_accessed(filp);
Miao Xie058a4572010-05-20 07:21:50 +00001930 vma->vm_ops = &btrfs_file_vm_ops;
Miao Xie058a4572010-05-20 07:21:50 +00001931
Chris Mason9ebefb182007-06-15 13:50:00 -04001932 return 0;
1933}
1934
Josef Bacik2aaa6652012-08-29 14:27:18 -04001935static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1936 int slot, u64 start, u64 end)
1937{
1938 struct btrfs_file_extent_item *fi;
1939 struct btrfs_key key;
1940
1941 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1942 return 0;
1943
1944 btrfs_item_key_to_cpu(leaf, &key, slot);
1945 if (key.objectid != btrfs_ino(inode) ||
1946 key.type != BTRFS_EXTENT_DATA_KEY)
1947 return 0;
1948
1949 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1950
1951 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1952 return 0;
1953
1954 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1955 return 0;
1956
1957 if (key.offset == end)
1958 return 1;
1959 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1960 return 1;
1961 return 0;
1962}
1963
1964static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1965 struct btrfs_path *path, u64 offset, u64 end)
1966{
1967 struct btrfs_root *root = BTRFS_I(inode)->root;
1968 struct extent_buffer *leaf;
1969 struct btrfs_file_extent_item *fi;
1970 struct extent_map *hole_em;
1971 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1972 struct btrfs_key key;
1973 int ret;
1974
1975 key.objectid = btrfs_ino(inode);
1976 key.type = BTRFS_EXTENT_DATA_KEY;
1977 key.offset = offset;
1978
1979
1980 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1981 if (ret < 0)
1982 return ret;
1983 BUG_ON(!ret);
1984
1985 leaf = path->nodes[0];
1986 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1987 u64 num_bytes;
1988
1989 path->slots[0]--;
1990 fi = btrfs_item_ptr(leaf, path->slots[0],
1991 struct btrfs_file_extent_item);
1992 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1993 end - offset;
1994 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1995 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1996 btrfs_set_file_extent_offset(leaf, fi, 0);
1997 btrfs_mark_buffer_dirty(leaf);
1998 goto out;
1999 }
2000
2001 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2002 u64 num_bytes;
2003
2004 path->slots[0]++;
2005 key.offset = offset;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00002006 btrfs_set_item_key_safe(root, path, &key);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002007 fi = btrfs_item_ptr(leaf, path->slots[0],
2008 struct btrfs_file_extent_item);
2009 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2010 offset;
2011 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2012 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2013 btrfs_set_file_extent_offset(leaf, fi, 0);
2014 btrfs_mark_buffer_dirty(leaf);
2015 goto out;
2016 }
2017 btrfs_release_path(path);
2018
2019 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2020 0, 0, end - offset, 0, end - offset,
2021 0, 0, 0);
2022 if (ret)
2023 return ret;
2024
2025out:
2026 btrfs_release_path(path);
2027
2028 hole_em = alloc_extent_map();
2029 if (!hole_em) {
2030 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2031 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2032 &BTRFS_I(inode)->runtime_flags);
2033 } else {
2034 hole_em->start = offset;
2035 hole_em->len = end - offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -04002036 hole_em->ram_bytes = hole_em->len;
Josef Bacik2aaa6652012-08-29 14:27:18 -04002037 hole_em->orig_start = offset;
2038
2039 hole_em->block_start = EXTENT_MAP_HOLE;
2040 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05002041 hole_em->orig_block_len = 0;
Josef Bacik2aaa6652012-08-29 14:27:18 -04002042 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2043 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2044 hole_em->generation = trans->transid;
2045
2046 do {
2047 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2048 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04002049 ret = add_extent_mapping(em_tree, hole_em, 1);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002050 write_unlock(&em_tree->lock);
2051 } while (ret == -EEXIST);
2052 free_extent_map(hole_em);
2053 if (ret)
2054 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2055 &BTRFS_I(inode)->runtime_flags);
2056 }
2057
2058 return 0;
2059}
2060
2061static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2062{
2063 struct btrfs_root *root = BTRFS_I(inode)->root;
2064 struct extent_state *cached_state = NULL;
2065 struct btrfs_path *path;
2066 struct btrfs_block_rsv *rsv;
2067 struct btrfs_trans_handle *trans;
Miao Xie00612802012-12-05 10:54:12 +00002068 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2069 u64 lockend = round_down(offset + len,
2070 BTRFS_I(inode)->root->sectorsize) - 1;
Josef Bacik2aaa6652012-08-29 14:27:18 -04002071 u64 cur_offset = lockstart;
2072 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2073 u64 drop_end;
Josef Bacik2aaa6652012-08-29 14:27:18 -04002074 int ret = 0;
2075 int err = 0;
Miao Xie6347b3c2012-12-05 10:53:45 +00002076 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2077 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
Josef Bacik2aaa6652012-08-29 14:27:18 -04002078
2079 btrfs_wait_ordered_range(inode, offset, len);
2080
2081 mutex_lock(&inode->i_mutex);
Miao Xie7426cc02012-12-05 10:54:52 +00002082 /*
2083 * We needn't truncate any page which is beyond the end of the file
2084 * because we are sure there is no data there.
2085 */
Josef Bacik2aaa6652012-08-29 14:27:18 -04002086 /*
2087 * Only do this if we are in the same page and we aren't doing the
2088 * entire page.
2089 */
2090 if (same_page && len < PAGE_CACHE_SIZE) {
Miao Xie7426cc02012-12-05 10:54:52 +00002091 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2092 ret = btrfs_truncate_page(inode, offset, len, 0);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002093 mutex_unlock(&inode->i_mutex);
2094 return ret;
2095 }
2096
2097 /* zero back part of the first page */
Miao Xie7426cc02012-12-05 10:54:52 +00002098 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2099 ret = btrfs_truncate_page(inode, offset, 0, 0);
2100 if (ret) {
2101 mutex_unlock(&inode->i_mutex);
2102 return ret;
2103 }
Josef Bacik2aaa6652012-08-29 14:27:18 -04002104 }
2105
2106 /* zero the front end of the last page */
Miao Xie00612802012-12-05 10:54:12 +00002107 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2108 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2109 if (ret) {
2110 mutex_unlock(&inode->i_mutex);
2111 return ret;
2112 }
Josef Bacik2aaa6652012-08-29 14:27:18 -04002113 }
2114
2115 if (lockend < lockstart) {
2116 mutex_unlock(&inode->i_mutex);
2117 return 0;
2118 }
2119
2120 while (1) {
2121 struct btrfs_ordered_extent *ordered;
2122
2123 truncate_pagecache_range(inode, lockstart, lockend);
2124
2125 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2126 0, &cached_state);
2127 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2128
2129 /*
2130 * We need to make sure we have no ordered extents in this range
2131 * and nobody raced in and read a page in this range, if we did
2132 * we need to try again.
2133 */
2134 if ((!ordered ||
2135 (ordered->file_offset + ordered->len < lockstart ||
2136 ordered->file_offset > lockend)) &&
2137 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2138 lockend, EXTENT_UPTODATE, 0,
2139 cached_state)) {
2140 if (ordered)
2141 btrfs_put_ordered_extent(ordered);
2142 break;
2143 }
2144 if (ordered)
2145 btrfs_put_ordered_extent(ordered);
2146 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2147 lockend, &cached_state, GFP_NOFS);
2148 btrfs_wait_ordered_range(inode, lockstart,
2149 lockend - lockstart + 1);
2150 }
2151
2152 path = btrfs_alloc_path();
2153 if (!path) {
2154 ret = -ENOMEM;
2155 goto out;
2156 }
2157
Miao Xie66d8f3d2012-09-06 04:02:28 -06002158 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002159 if (!rsv) {
2160 ret = -ENOMEM;
2161 goto out_free;
2162 }
2163 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2164 rsv->failfast = 1;
2165
2166 /*
2167 * 1 - update the inode
2168 * 1 - removing the extents in the range
2169 * 1 - adding the hole extent
2170 */
2171 trans = btrfs_start_transaction(root, 3);
2172 if (IS_ERR(trans)) {
2173 err = PTR_ERR(trans);
2174 goto out_free;
2175 }
2176
2177 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2178 min_size);
2179 BUG_ON(ret);
2180 trans->block_rsv = rsv;
2181
2182 while (cur_offset < lockend) {
2183 ret = __btrfs_drop_extents(trans, root, inode, path,
2184 cur_offset, lockend + 1,
2185 &drop_end, 1);
2186 if (ret != -ENOSPC)
2187 break;
2188
2189 trans->block_rsv = &root->fs_info->trans_block_rsv;
2190
2191 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2192 if (ret) {
2193 err = ret;
2194 break;
2195 }
2196
2197 cur_offset = drop_end;
2198
2199 ret = btrfs_update_inode(trans, root, inode);
2200 if (ret) {
2201 err = ret;
2202 break;
2203 }
2204
Josef Bacik2aaa6652012-08-29 14:27:18 -04002205 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00002206 btrfs_btree_balance_dirty(root);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002207
2208 trans = btrfs_start_transaction(root, 3);
2209 if (IS_ERR(trans)) {
2210 ret = PTR_ERR(trans);
2211 trans = NULL;
2212 break;
2213 }
2214
2215 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2216 rsv, min_size);
2217 BUG_ON(ret); /* shouldn't happen */
2218 trans->block_rsv = rsv;
2219 }
2220
2221 if (ret) {
2222 err = ret;
2223 goto out_trans;
2224 }
2225
2226 trans->block_rsv = &root->fs_info->trans_block_rsv;
2227 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2228 if (ret) {
2229 err = ret;
2230 goto out_trans;
2231 }
2232
2233out_trans:
2234 if (!trans)
2235 goto out_free;
2236
Tsutomu Itohe1f57902012-11-08 04:47:33 +00002237 inode_inc_iversion(inode);
2238 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2239
Josef Bacik2aaa6652012-08-29 14:27:18 -04002240 trans->block_rsv = &root->fs_info->trans_block_rsv;
2241 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002242 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00002243 btrfs_btree_balance_dirty(root);
Josef Bacik2aaa6652012-08-29 14:27:18 -04002244out_free:
2245 btrfs_free_path(path);
2246 btrfs_free_block_rsv(root, rsv);
2247out:
2248 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2249 &cached_state, GFP_NOFS);
2250 mutex_unlock(&inode->i_mutex);
2251 if (ret && !err)
2252 err = ret;
2253 return err;
2254}
2255
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002256static long btrfs_fallocate(struct file *file, int mode,
2257 loff_t offset, loff_t len)
2258{
Al Viro496ad9a2013-01-23 17:07:38 -05002259 struct inode *inode = file_inode(file);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002260 struct extent_state *cached_state = NULL;
Wang Shilong61130772013-03-19 10:57:14 +00002261 struct btrfs_root *root = BTRFS_I(inode)->root;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002262 u64 cur_offset;
2263 u64 last_byte;
2264 u64 alloc_start;
2265 u64 alloc_end;
2266 u64 alloc_hint = 0;
2267 u64 locked_end;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002268 struct extent_map *em;
Miao Xie797f4272012-11-28 10:28:07 +00002269 int blocksize = BTRFS_I(inode)->root->sectorsize;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002270 int ret;
2271
Miao Xie797f4272012-11-28 10:28:07 +00002272 alloc_start = round_down(offset, blocksize);
2273 alloc_end = round_up(offset + len, blocksize);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002274
Josef Bacik2aaa6652012-08-29 14:27:18 -04002275 /* Make sure we aren't being give some crap mode */
2276 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002277 return -EOPNOTSUPP;
2278
Josef Bacik2aaa6652012-08-29 14:27:18 -04002279 if (mode & FALLOC_FL_PUNCH_HOLE)
2280 return btrfs_punch_hole(inode, offset, len);
2281
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002282 /*
Chris Masond98456f2012-01-31 20:27:41 -05002283 * Make sure we have enough space before we do the
2284 * allocation.
2285 */
Miao Xie0ff6fab2012-11-28 10:28:54 +00002286 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
Chris Masond98456f2012-01-31 20:27:41 -05002287 if (ret)
2288 return ret;
Wang Shilong61130772013-03-19 10:57:14 +00002289 if (root->fs_info->quota_enabled) {
2290 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2291 if (ret)
2292 goto out_reserve_fail;
2293 }
Chris Masond98456f2012-01-31 20:27:41 -05002294
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002295 mutex_lock(&inode->i_mutex);
2296 ret = inode_newsize_ok(inode, alloc_end);
2297 if (ret)
2298 goto out;
2299
2300 if (alloc_start > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -05002301 ret = btrfs_cont_expand(inode, i_size_read(inode),
2302 alloc_start);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002303 if (ret)
2304 goto out;
Josef Bacika71754f2013-06-17 17:14:39 -04002305 } else {
2306 /*
2307 * If we are fallocating from the end of the file onward we
2308 * need to zero out the end of the page if i_size lands in the
2309 * middle of a page.
2310 */
2311 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2312 if (ret)
2313 goto out;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002314 }
2315
Josef Bacika71754f2013-06-17 17:14:39 -04002316 /*
2317 * wait for ordered IO before we have any locks. We'll loop again
2318 * below with the locks held.
2319 */
2320 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2321
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002322 locked_end = alloc_end - 1;
2323 while (1) {
2324 struct btrfs_ordered_extent *ordered;
2325
2326 /* the extent lock is ordered inside the running
2327 * transaction
2328 */
2329 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002330 locked_end, 0, &cached_state);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002331 ordered = btrfs_lookup_first_ordered_extent(inode,
2332 alloc_end - 1);
2333 if (ordered &&
2334 ordered->file_offset + ordered->len > alloc_start &&
2335 ordered->file_offset < alloc_end) {
2336 btrfs_put_ordered_extent(ordered);
2337 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2338 alloc_start, locked_end,
2339 &cached_state, GFP_NOFS);
2340 /*
2341 * we can't wait on the range with the transaction
2342 * running or with the extent lock held
2343 */
2344 btrfs_wait_ordered_range(inode, alloc_start,
2345 alloc_end - alloc_start);
2346 } else {
2347 if (ordered)
2348 btrfs_put_ordered_extent(ordered);
2349 break;
2350 }
2351 }
2352
2353 cur_offset = alloc_start;
2354 while (1) {
Josef Bacikf1e490a2011-08-18 10:36:39 -04002355 u64 actual_end;
2356
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002357 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2358 alloc_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002359 if (IS_ERR_OR_NULL(em)) {
2360 if (!em)
2361 ret = -ENOMEM;
2362 else
2363 ret = PTR_ERR(em);
2364 break;
2365 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002366 last_byte = min(extent_map_end(em), alloc_end);
Josef Bacikf1e490a2011-08-18 10:36:39 -04002367 actual_end = min_t(u64, extent_map_end(em), offset + len);
Miao Xie797f4272012-11-28 10:28:07 +00002368 last_byte = ALIGN(last_byte, blocksize);
Josef Bacikf1e490a2011-08-18 10:36:39 -04002369
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002370 if (em->block_start == EXTENT_MAP_HOLE ||
2371 (cur_offset >= inode->i_size &&
2372 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2373 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2374 last_byte - cur_offset,
2375 1 << inode->i_blkbits,
2376 offset + len,
2377 &alloc_hint);
Josef Bacik1b9c3322011-08-17 10:19:52 -04002378
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002379 if (ret < 0) {
2380 free_extent_map(em);
2381 break;
2382 }
Josef Bacikf1e490a2011-08-18 10:36:39 -04002383 } else if (actual_end > inode->i_size &&
2384 !(mode & FALLOC_FL_KEEP_SIZE)) {
2385 /*
2386 * We didn't need to allocate any more space, but we
2387 * still extended the size of the file so we need to
2388 * update i_size.
2389 */
2390 inode->i_ctime = CURRENT_TIME;
2391 i_size_write(inode, actual_end);
2392 btrfs_ordered_update_i_size(inode, actual_end, NULL);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002393 }
2394 free_extent_map(em);
2395
2396 cur_offset = last_byte;
2397 if (cur_offset >= alloc_end) {
2398 ret = 0;
2399 break;
2400 }
2401 }
2402 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2403 &cached_state, GFP_NOFS);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002404out:
2405 mutex_unlock(&inode->i_mutex);
Wang Shilong61130772013-03-19 10:57:14 +00002406 if (root->fs_info->quota_enabled)
2407 btrfs_qgroup_free(root, alloc_end - alloc_start);
2408out_reserve_fail:
Chris Masond98456f2012-01-31 20:27:41 -05002409 /* Let go of our reservation. */
Miao Xie0ff6fab2012-11-28 10:28:54 +00002410 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002411 return ret;
2412}
2413
Andrew Morton965c8e52012-12-17 15:59:39 -08002414static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
Josef Bacikb2675152011-07-18 13:21:36 -04002415{
2416 struct btrfs_root *root = BTRFS_I(inode)->root;
2417 struct extent_map *em;
2418 struct extent_state *cached_state = NULL;
2419 u64 lockstart = *offset;
2420 u64 lockend = i_size_read(inode);
2421 u64 start = *offset;
2422 u64 orig_start = *offset;
2423 u64 len = i_size_read(inode);
2424 u64 last_end = 0;
2425 int ret = 0;
2426
2427 lockend = max_t(u64, root->sectorsize, lockend);
2428 if (lockend <= lockstart)
2429 lockend = lockstart + root->sectorsize;
2430
Liu Bo1214b532013-01-07 03:53:08 +00002431 lockend--;
Josef Bacikb2675152011-07-18 13:21:36 -04002432 len = lockend - lockstart + 1;
2433
2434 len = max_t(u64, len, root->sectorsize);
2435 if (inode->i_size == 0)
2436 return -ENXIO;
2437
2438 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002439 &cached_state);
Josef Bacikb2675152011-07-18 13:21:36 -04002440
2441 /*
2442 * Delalloc is such a pain. If we have a hole and we have pending
2443 * delalloc for a portion of the hole we will get back a hole that
2444 * exists for the entire range since it hasn't been actually written
2445 * yet. So to take care of this case we need to look for an extent just
2446 * before the position we want in case there is outstanding delalloc
2447 * going on here.
2448 */
Andrew Morton965c8e52012-12-17 15:59:39 -08002449 if (whence == SEEK_HOLE && start != 0) {
Josef Bacikb2675152011-07-18 13:21:36 -04002450 if (start <= root->sectorsize)
2451 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2452 root->sectorsize, 0);
2453 else
2454 em = btrfs_get_extent_fiemap(inode, NULL, 0,
2455 start - root->sectorsize,
2456 root->sectorsize, 0);
2457 if (IS_ERR(em)) {
Jeff Liu6af021d2012-02-09 14:25:50 +08002458 ret = PTR_ERR(em);
Josef Bacikb2675152011-07-18 13:21:36 -04002459 goto out;
2460 }
2461 last_end = em->start + em->len;
2462 if (em->block_start == EXTENT_MAP_DELALLOC)
2463 last_end = min_t(u64, last_end, inode->i_size);
2464 free_extent_map(em);
2465 }
2466
2467 while (1) {
2468 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2469 if (IS_ERR(em)) {
Jeff Liu6af021d2012-02-09 14:25:50 +08002470 ret = PTR_ERR(em);
Josef Bacikb2675152011-07-18 13:21:36 -04002471 break;
2472 }
2473
2474 if (em->block_start == EXTENT_MAP_HOLE) {
2475 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2476 if (last_end <= orig_start) {
2477 free_extent_map(em);
2478 ret = -ENXIO;
2479 break;
2480 }
2481 }
2482
Andrew Morton965c8e52012-12-17 15:59:39 -08002483 if (whence == SEEK_HOLE) {
Josef Bacikb2675152011-07-18 13:21:36 -04002484 *offset = start;
2485 free_extent_map(em);
2486 break;
2487 }
2488 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -08002489 if (whence == SEEK_DATA) {
Josef Bacikb2675152011-07-18 13:21:36 -04002490 if (em->block_start == EXTENT_MAP_DELALLOC) {
2491 if (start >= inode->i_size) {
2492 free_extent_map(em);
2493 ret = -ENXIO;
2494 break;
2495 }
2496 }
2497
Liu Bof9e4fb52013-01-07 10:10:12 +00002498 if (!test_bit(EXTENT_FLAG_PREALLOC,
2499 &em->flags)) {
2500 *offset = start;
2501 free_extent_map(em);
2502 break;
2503 }
Josef Bacikb2675152011-07-18 13:21:36 -04002504 }
2505 }
2506
2507 start = em->start + em->len;
2508 last_end = em->start + em->len;
2509
2510 if (em->block_start == EXTENT_MAP_DELALLOC)
2511 last_end = min_t(u64, last_end, inode->i_size);
2512
2513 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2514 free_extent_map(em);
2515 ret = -ENXIO;
2516 break;
2517 }
2518 free_extent_map(em);
2519 cond_resched();
2520 }
2521 if (!ret)
2522 *offset = min(*offset, inode->i_size);
2523out:
2524 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2525 &cached_state, GFP_NOFS);
2526 return ret;
2527}
2528
Andrew Morton965c8e52012-12-17 15:59:39 -08002529static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
Josef Bacikb2675152011-07-18 13:21:36 -04002530{
2531 struct inode *inode = file->f_mapping->host;
2532 int ret;
2533
2534 mutex_lock(&inode->i_mutex);
Andrew Morton965c8e52012-12-17 15:59:39 -08002535 switch (whence) {
Josef Bacikb2675152011-07-18 13:21:36 -04002536 case SEEK_END:
2537 case SEEK_CUR:
Andrew Morton965c8e52012-12-17 15:59:39 -08002538 offset = generic_file_llseek(file, offset, whence);
Josef Bacikb2675152011-07-18 13:21:36 -04002539 goto out;
2540 case SEEK_DATA:
2541 case SEEK_HOLE:
Jeff Liu48802c82011-09-18 10:34:02 -04002542 if (offset >= i_size_read(inode)) {
2543 mutex_unlock(&inode->i_mutex);
2544 return -ENXIO;
2545 }
2546
Andrew Morton965c8e52012-12-17 15:59:39 -08002547 ret = find_desired_extent(inode, &offset, whence);
Josef Bacikb2675152011-07-18 13:21:36 -04002548 if (ret) {
2549 mutex_unlock(&inode->i_mutex);
2550 return ret;
2551 }
2552 }
2553
Jie Liu46a1c2c2013-06-25 12:02:13 +08002554 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Josef Bacikb2675152011-07-18 13:21:36 -04002555out:
2556 mutex_unlock(&inode->i_mutex);
2557 return offset;
2558}
2559
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002560const struct file_operations btrfs_file_operations = {
Josef Bacikb2675152011-07-18 13:21:36 -04002561 .llseek = btrfs_file_llseek,
Chris Mason39279cc2007-06-12 06:35:45 -04002562 .read = do_sync_read,
Miao Xie4a0010712010-06-07 03:38:51 +00002563 .write = do_sync_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04002564 .aio_read = generic_file_aio_read,
Chris Masone9906a92007-12-14 12:56:58 -05002565 .splice_read = generic_file_splice_read,
Josef Bacik11c65dc2010-05-23 11:07:21 -04002566 .aio_write = btrfs_file_aio_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04002567 .mmap = btrfs_file_mmap,
Chris Mason39279cc2007-06-12 06:35:45 -04002568 .open = generic_file_open,
Mingminge1b81e62008-05-27 10:55:43 -04002569 .release = btrfs_release_file,
Chris Mason39279cc2007-06-12 06:35:45 -04002570 .fsync = btrfs_sync_file,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01002571 .fallocate = btrfs_fallocate,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04002572 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04002573#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04002574 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04002575#endif
2576};
Miao Xie9247f312012-11-26 09:24:43 +00002577
2578void btrfs_auto_defrag_exit(void)
2579{
2580 if (btrfs_inode_defrag_cachep)
2581 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2582}
2583
2584int btrfs_auto_defrag_init(void)
2585{
2586 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2587 sizeof(struct inode_defrag), 0,
2588 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2589 NULL);
2590 if (!btrfs_inode_defrag_cachep)
2591 return -ENOMEM;
2592
2593 return 0;
2594}