blob: 70dc8ca73e257bc3a1e7a96ea48009bff093af9a [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason39279cc2007-06-12 06:35:45 -040019#include <linux/fs.h>
20#include <linux/pagemap.h>
21#include <linux/highmem.h>
22#include <linux/time.h>
23#include <linux/init.h>
24#include <linux/string.h>
Chris Mason39279cc2007-06-12 06:35:45 -040025#include <linux/backing-dev.h>
26#include <linux/mpage.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010027#include <linux/falloc.h>
Chris Mason39279cc2007-06-12 06:35:45 -040028#include <linux/swap.h>
29#include <linux/writeback.h>
30#include <linux/statfs.h>
31#include <linux/compat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/slab.h>
Chris Mason39279cc2007-06-12 06:35:45 -040033#include "ctree.h"
34#include "disk-io.h"
35#include "transaction.h"
36#include "btrfs_inode.h"
37#include "ioctl.h"
38#include "print-tree.h"
Chris Masone02119d2008-09-05 16:13:11 -040039#include "tree-log.h"
40#include "locking.h"
Jeff Mahoney12fa8ec2008-05-02 15:03:58 -040041#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040042
Chris Mason4cb53002011-05-24 15:35:30 -040043/*
44 * when auto defrag is enabled we
45 * queue up these defrag structs to remember which
46 * inodes need defragging passes
47 */
48struct inode_defrag {
49 struct rb_node rb_node;
50 /* objectid */
51 u64 ino;
52 /*
53 * transid where the defrag was added, we search for
54 * extents newer than this
55 */
56 u64 transid;
57
58 /* root objectid */
59 u64 root;
60
61 /* last offset we were able to defrag */
62 u64 last_offset;
63
64 /* if we've wrapped around back to zero once already */
65 int cycled;
66};
67
Miao Xie762f2262012-05-24 18:58:27 +080068static int __compare_inode_defrag(struct inode_defrag *defrag1,
69 struct inode_defrag *defrag2)
70{
71 if (defrag1->root > defrag2->root)
72 return 1;
73 else if (defrag1->root < defrag2->root)
74 return -1;
75 else if (defrag1->ino > defrag2->ino)
76 return 1;
77 else if (defrag1->ino < defrag2->ino)
78 return -1;
79 else
80 return 0;
81}
82
Chris Mason4cb53002011-05-24 15:35:30 -040083/* pop a record for an inode into the defrag tree. The lock
84 * must be held already
85 *
86 * If you're inserting a record for an older transid than an
87 * existing record, the transid already in the tree is lowered
88 *
89 * If an existing record is found the defrag item you
90 * pass in is freed
91 */
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +000092static void __btrfs_add_inode_defrag(struct inode *inode,
Chris Mason4cb53002011-05-24 15:35:30 -040093 struct inode_defrag *defrag)
94{
95 struct btrfs_root *root = BTRFS_I(inode)->root;
96 struct inode_defrag *entry;
97 struct rb_node **p;
98 struct rb_node *parent = NULL;
Miao Xie762f2262012-05-24 18:58:27 +080099 int ret;
Chris Mason4cb53002011-05-24 15:35:30 -0400100
101 p = &root->fs_info->defrag_inodes.rb_node;
102 while (*p) {
103 parent = *p;
104 entry = rb_entry(parent, struct inode_defrag, rb_node);
105
Miao Xie762f2262012-05-24 18:58:27 +0800106 ret = __compare_inode_defrag(defrag, entry);
107 if (ret < 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400108 p = &parent->rb_left;
Miao Xie762f2262012-05-24 18:58:27 +0800109 else if (ret > 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400110 p = &parent->rb_right;
111 else {
112 /* if we're reinserting an entry for
113 * an old defrag run, make sure to
114 * lower the transid of our existing record
115 */
116 if (defrag->transid < entry->transid)
117 entry->transid = defrag->transid;
118 if (defrag->last_offset > entry->last_offset)
119 entry->last_offset = defrag->last_offset;
120 goto exists;
121 }
122 }
Josef Bacik72ac3c02012-05-23 14:13:11 -0400123 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
Chris Mason4cb53002011-05-24 15:35:30 -0400124 rb_link_node(&defrag->rb_node, parent, p);
125 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000126 return;
Chris Mason4cb53002011-05-24 15:35:30 -0400127
128exists:
129 kfree(defrag);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000130 return;
Chris Mason4cb53002011-05-24 15:35:30 -0400131
132}
133
134/*
135 * insert a defrag record for this inode if auto defrag is
136 * enabled
137 */
138int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
139 struct inode *inode)
140{
141 struct btrfs_root *root = BTRFS_I(inode)->root;
142 struct inode_defrag *defrag;
Chris Mason4cb53002011-05-24 15:35:30 -0400143 u64 transid;
144
145 if (!btrfs_test_opt(root, AUTO_DEFRAG))
146 return 0;
147
David Sterba7841cb22011-05-31 18:07:27 +0200148 if (btrfs_fs_closing(root->fs_info))
Chris Mason4cb53002011-05-24 15:35:30 -0400149 return 0;
150
Josef Bacik72ac3c02012-05-23 14:13:11 -0400151 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
Chris Mason4cb53002011-05-24 15:35:30 -0400152 return 0;
153
154 if (trans)
155 transid = trans->transid;
156 else
157 transid = BTRFS_I(inode)->root->last_trans;
158
159 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
160 if (!defrag)
161 return -ENOMEM;
162
David Sterbaa4689d22011-05-31 17:08:14 +0000163 defrag->ino = btrfs_ino(inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400164 defrag->transid = transid;
165 defrag->root = root->root_key.objectid;
166
167 spin_lock(&root->fs_info->defrag_inodes_lock);
Josef Bacik72ac3c02012-05-23 14:13:11 -0400168 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000169 __btrfs_add_inode_defrag(inode, defrag);
Dan Carpenterf4ac9042011-08-05 14:19:00 +0000170 else
171 kfree(defrag);
Chris Mason4cb53002011-05-24 15:35:30 -0400172 spin_unlock(&root->fs_info->defrag_inodes_lock);
Wanlong Gaoa0f98dd2011-07-18 12:19:35 +0000173 return 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400174}
175
176/*
177 * must be called with the defrag_inodes lock held
178 */
Miao Xie762f2262012-05-24 18:58:27 +0800179struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info,
180 u64 root, u64 ino,
Chris Mason4cb53002011-05-24 15:35:30 -0400181 struct rb_node **next)
182{
183 struct inode_defrag *entry = NULL;
Miao Xie762f2262012-05-24 18:58:27 +0800184 struct inode_defrag tmp;
Chris Mason4cb53002011-05-24 15:35:30 -0400185 struct rb_node *p;
186 struct rb_node *parent = NULL;
Miao Xie762f2262012-05-24 18:58:27 +0800187 int ret;
188
189 tmp.ino = ino;
190 tmp.root = root;
Chris Mason4cb53002011-05-24 15:35:30 -0400191
192 p = info->defrag_inodes.rb_node;
193 while (p) {
194 parent = p;
195 entry = rb_entry(parent, struct inode_defrag, rb_node);
196
Miao Xie762f2262012-05-24 18:58:27 +0800197 ret = __compare_inode_defrag(&tmp, entry);
198 if (ret < 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400199 p = parent->rb_left;
Miao Xie762f2262012-05-24 18:58:27 +0800200 else if (ret > 0)
Chris Mason4cb53002011-05-24 15:35:30 -0400201 p = parent->rb_right;
202 else
203 return entry;
204 }
205
206 if (next) {
Miao Xie762f2262012-05-24 18:58:27 +0800207 while (parent && __compare_inode_defrag(&tmp, entry) > 0) {
Chris Mason4cb53002011-05-24 15:35:30 -0400208 parent = rb_next(parent);
209 entry = rb_entry(parent, struct inode_defrag, rb_node);
210 }
211 *next = parent;
212 }
213 return NULL;
214}
215
216/*
217 * run through the list of inodes in the FS that need
218 * defragging
219 */
220int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
221{
222 struct inode_defrag *defrag;
223 struct btrfs_root *inode_root;
224 struct inode *inode;
225 struct rb_node *n;
226 struct btrfs_key key;
227 struct btrfs_ioctl_defrag_range_args range;
228 u64 first_ino = 0;
Miao Xie762f2262012-05-24 18:58:27 +0800229 u64 root_objectid = 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400230 int num_defrag;
231 int defrag_batch = 1024;
232
233 memset(&range, 0, sizeof(range));
234 range.len = (u64)-1;
235
236 atomic_inc(&fs_info->defrag_running);
237 spin_lock(&fs_info->defrag_inodes_lock);
238 while(1) {
239 n = NULL;
240
241 /* find an inode to defrag */
Miao Xie762f2262012-05-24 18:58:27 +0800242 defrag = btrfs_find_defrag_inode(fs_info, root_objectid,
243 first_ino, &n);
Chris Mason4cb53002011-05-24 15:35:30 -0400244 if (!defrag) {
Miao Xie762f2262012-05-24 18:58:27 +0800245 if (n) {
246 defrag = rb_entry(n, struct inode_defrag,
247 rb_node);
248 } else if (root_objectid || first_ino) {
249 root_objectid = 0;
Chris Mason4cb53002011-05-24 15:35:30 -0400250 first_ino = 0;
251 continue;
252 } else {
253 break;
254 }
255 }
256
257 /* remove it from the rbtree */
258 first_ino = defrag->ino + 1;
Miao Xie762f2262012-05-24 18:58:27 +0800259 root_objectid = defrag->root;
Chris Mason4cb53002011-05-24 15:35:30 -0400260 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
261
David Sterba7841cb22011-05-31 18:07:27 +0200262 if (btrfs_fs_closing(fs_info))
Chris Mason4cb53002011-05-24 15:35:30 -0400263 goto next_free;
264
265 spin_unlock(&fs_info->defrag_inodes_lock);
266
267 /* get the inode */
268 key.objectid = defrag->root;
269 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
270 key.offset = (u64)-1;
271 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
272 if (IS_ERR(inode_root))
273 goto next;
274
275 key.objectid = defrag->ino;
276 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
277 key.offset = 0;
278
279 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
280 if (IS_ERR(inode))
281 goto next;
282
283 /* do a chunk of defrag */
Josef Bacik72ac3c02012-05-23 14:13:11 -0400284 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
Chris Mason4cb53002011-05-24 15:35:30 -0400285 range.start = defrag->last_offset;
286 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
287 defrag_batch);
288 /*
289 * if we filled the whole defrag batch, there
290 * must be more work to do. Queue this defrag
291 * again
292 */
293 if (num_defrag == defrag_batch) {
294 defrag->last_offset = range.start;
295 __btrfs_add_inode_defrag(inode, defrag);
296 /*
297 * we don't want to kfree defrag, we added it back to
298 * the rbtree
299 */
300 defrag = NULL;
301 } else if (defrag->last_offset && !defrag->cycled) {
302 /*
303 * we didn't fill our defrag batch, but
304 * we didn't start at zero. Make sure we loop
305 * around to the start of the file.
306 */
307 defrag->last_offset = 0;
308 defrag->cycled = 1;
309 __btrfs_add_inode_defrag(inode, defrag);
310 defrag = NULL;
311 }
312
313 iput(inode);
314next:
315 spin_lock(&fs_info->defrag_inodes_lock);
316next_free:
317 kfree(defrag);
318 }
319 spin_unlock(&fs_info->defrag_inodes_lock);
320
321 atomic_dec(&fs_info->defrag_running);
322
323 /*
324 * during unmount, we use the transaction_wait queue to
325 * wait for the defragger to stop
326 */
327 wake_up(&fs_info->transaction_wait);
328 return 0;
329}
Chris Mason39279cc2007-06-12 06:35:45 -0400330
Chris Masond352ac62008-09-29 15:18:18 -0400331/* simple helper to fault in pages and copy. This should go away
332 * and be replaced with calls into generic code.
333 */
Chris Masond3977122009-01-05 21:25:51 -0500334static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -0500335 size_t write_bytes,
Chris Masona1b32a52008-09-05 16:09:51 -0400336 struct page **prepared_pages,
Josef Bacik11c65dc2010-05-23 11:07:21 -0400337 struct iov_iter *i)
Chris Mason39279cc2007-06-12 06:35:45 -0400338{
Xin Zhong914ee292010-12-09 09:30:14 +0000339 size_t copied = 0;
Josef Bacikd0215f32011-01-25 14:57:24 -0500340 size_t total_copied = 0;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400341 int pg = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400342 int offset = pos & (PAGE_CACHE_SIZE - 1);
343
Josef Bacik11c65dc2010-05-23 11:07:21 -0400344 while (write_bytes > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -0400345 size_t count = min_t(size_t,
346 PAGE_CACHE_SIZE - offset, write_bytes);
Josef Bacik11c65dc2010-05-23 11:07:21 -0400347 struct page *page = prepared_pages[pg];
Xin Zhong914ee292010-12-09 09:30:14 +0000348 /*
349 * Copy data from userspace to the current page
350 *
351 * Disable pagefault to avoid recursive lock since
352 * the pages are already locked
353 */
354 pagefault_disable();
355 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
356 pagefault_enable();
Josef Bacik11c65dc2010-05-23 11:07:21 -0400357
Chris Mason39279cc2007-06-12 06:35:45 -0400358 /* Flush processor's dcache for this page */
359 flush_dcache_page(page);
Chris Mason31339ac2011-03-07 11:10:24 -0500360
361 /*
362 * if we get a partial write, we can end up with
363 * partially up to date pages. These add
364 * a lot of complexity, so make sure they don't
365 * happen by forcing this copy to be retried.
366 *
367 * The rest of the btrfs_file_write code will fall
368 * back to page at a time copies after we return 0.
369 */
370 if (!PageUptodate(page) && copied < count)
371 copied = 0;
372
Josef Bacik11c65dc2010-05-23 11:07:21 -0400373 iov_iter_advance(i, copied);
374 write_bytes -= copied;
Xin Zhong914ee292010-12-09 09:30:14 +0000375 total_copied += copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400376
Xin Zhong914ee292010-12-09 09:30:14 +0000377 /* Return to btrfs_file_aio_write to fault page */
Josef Bacik9f570b82011-01-25 12:42:37 -0500378 if (unlikely(copied == 0))
Xin Zhong914ee292010-12-09 09:30:14 +0000379 break;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400380
381 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
382 offset += copied;
383 } else {
384 pg++;
385 offset = 0;
386 }
Chris Mason39279cc2007-06-12 06:35:45 -0400387 }
Xin Zhong914ee292010-12-09 09:30:14 +0000388 return total_copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400389}
390
Chris Masond352ac62008-09-29 15:18:18 -0400391/*
392 * unlocks pages after btrfs_file_write is done with them
393 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400394void btrfs_drop_pages(struct page **pages, size_t num_pages)
Chris Mason39279cc2007-06-12 06:35:45 -0400395{
396 size_t i;
397 for (i = 0; i < num_pages; i++) {
Chris Masond352ac62008-09-29 15:18:18 -0400398 /* page checked is some magic around finding pages that
399 * have been modified without going through btrfs_set_page_dirty
400 * clear it here
401 */
Chris Mason4a096752008-07-21 10:29:44 -0400402 ClearPageChecked(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400403 unlock_page(pages[i]);
404 mark_page_accessed(pages[i]);
405 page_cache_release(pages[i]);
406 }
407}
408
Chris Masond352ac62008-09-29 15:18:18 -0400409/*
410 * after copy_from_user, pages need to be dirtied and we need to make
411 * sure holes are created between the current EOF and the start of
412 * any next extents (if required).
413 *
414 * this also makes the decision about creating an inline extent vs
415 * doing real data extents, marking pages dirty and delalloc as required.
416 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400417int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
418 struct page **pages, size_t num_pages,
419 loff_t pos, size_t write_bytes,
420 struct extent_state **cached)
Chris Mason39279cc2007-06-12 06:35:45 -0400421{
Chris Mason39279cc2007-06-12 06:35:45 -0400422 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400423 int i;
Chris Masondb945352007-10-15 16:15:53 -0400424 u64 num_bytes;
Chris Masona52d9a82007-08-27 16:49:44 -0400425 u64 start_pos;
426 u64 end_of_last_block;
427 u64 end_pos = pos + write_bytes;
428 loff_t isize = i_size_read(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400429
Chris Mason5f39d392007-10-15 16:14:19 -0400430 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -0400431 num_bytes = (write_bytes + pos - start_pos +
432 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
Chris Mason39279cc2007-06-12 06:35:45 -0400433
Chris Masondb945352007-10-15 16:15:53 -0400434 end_of_last_block = start_pos + num_bytes - 1;
Josef Bacik2ac55d42010-02-03 19:33:23 +0000435 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400436 cached);
Josef Bacikd0215f32011-01-25 14:57:24 -0500437 if (err)
438 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -0400439
Chris Masonc8b97812008-10-29 14:49:59 -0400440 for (i = 0; i < num_pages; i++) {
441 struct page *p = pages[i];
442 SetPageUptodate(p);
443 ClearPageChecked(p);
444 set_page_dirty(p);
Chris Masona52d9a82007-08-27 16:49:44 -0400445 }
Josef Bacik9f570b82011-01-25 12:42:37 -0500446
447 /*
448 * we've only changed i_size in ram, and we haven't updated
449 * the disk i_size. There is no need to log the inode
450 * at this time.
451 */
452 if (end_pos > isize)
Chris Masona52d9a82007-08-27 16:49:44 -0400453 i_size_write(inode, end_pos);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400454 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400455}
456
Chris Masond352ac62008-09-29 15:18:18 -0400457/*
458 * this drops all the extents in the cache that intersect the range
459 * [start, end]. Existing extents are split as required.
460 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400461int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
462 int skip_pinned)
Chris Masona52d9a82007-08-27 16:49:44 -0400463{
464 struct extent_map *em;
Chris Mason3b951512008-04-17 11:29:12 -0400465 struct extent_map *split = NULL;
466 struct extent_map *split2 = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -0400467 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan39b56372008-02-15 10:40:50 -0500468 u64 len = end - start + 1;
Chris Mason3b951512008-04-17 11:29:12 -0400469 int ret;
470 int testend = 1;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400471 unsigned long flags;
Chris Masonc8b97812008-10-29 14:49:59 -0400472 int compressed = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400473
Chris Masone6dcd2d2008-07-17 12:53:50 -0400474 WARN_ON(end < start);
Chris Mason3b951512008-04-17 11:29:12 -0400475 if (end == (u64)-1) {
Yan39b56372008-02-15 10:40:50 -0500476 len = (u64)-1;
Chris Mason3b951512008-04-17 11:29:12 -0400477 testend = 0;
478 }
Chris Masond3977122009-01-05 21:25:51 -0500479 while (1) {
Chris Mason3b951512008-04-17 11:29:12 -0400480 if (!split)
David Sterba172ddd62011-04-21 00:48:27 +0200481 split = alloc_extent_map();
Chris Mason3b951512008-04-17 11:29:12 -0400482 if (!split2)
David Sterba172ddd62011-04-21 00:48:27 +0200483 split2 = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100484 BUG_ON(!split || !split2); /* -ENOMEM */
Chris Mason3b951512008-04-17 11:29:12 -0400485
Chris Mason890871b2009-09-02 16:24:52 -0400486 write_lock(&em_tree->lock);
Yan39b56372008-02-15 10:40:50 -0500487 em = lookup_extent_mapping(em_tree, start, len);
Chris Masond1310b22008-01-24 16:13:08 -0500488 if (!em) {
Chris Mason890871b2009-09-02 16:24:52 -0400489 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -0400490 break;
Chris Masond1310b22008-01-24 16:13:08 -0500491 }
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400492 flags = em->flags;
493 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
Yan, Zheng55ef6892009-11-12 09:36:44 +0000494 if (testend && em->start + em->len >= start + len) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400495 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400496 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400497 break;
498 }
Yan, Zheng55ef6892009-11-12 09:36:44 +0000499 start = em->start + em->len;
500 if (testend)
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400501 len = start + len - (em->start + em->len);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400502 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400503 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400504 continue;
505 }
Chris Masonc8b97812008-10-29 14:49:59 -0400506 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Chris Mason3ce7e672008-07-31 15:42:54 -0400507 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -0400508 remove_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -0400509
510 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
511 em->start < start) {
512 split->start = em->start;
513 split->len = start - em->start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500514 split->orig_start = em->orig_start;
Chris Mason3b951512008-04-17 11:29:12 -0400515 split->block_start = em->block_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400516
517 if (compressed)
518 split->block_len = em->block_len;
519 else
520 split->block_len = split->len;
521
Chris Mason3b951512008-04-17 11:29:12 -0400522 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400523 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800524 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400525 ret = add_extent_mapping(em_tree, split);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100526 BUG_ON(ret); /* Logic error */
Chris Mason3b951512008-04-17 11:29:12 -0400527 free_extent_map(split);
528 split = split2;
529 split2 = NULL;
530 }
531 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
532 testend && em->start + em->len > start + len) {
533 u64 diff = start + len - em->start;
534
535 split->start = start + len;
536 split->len = em->start + em->len - (start + len);
537 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400538 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800539 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400540
Chris Masonc8b97812008-10-29 14:49:59 -0400541 if (compressed) {
542 split->block_len = em->block_len;
543 split->block_start = em->block_start;
Chris Mason445a6942008-11-10 11:53:33 -0500544 split->orig_start = em->orig_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400545 } else {
546 split->block_len = split->len;
547 split->block_start = em->block_start + diff;
Chris Mason445a6942008-11-10 11:53:33 -0500548 split->orig_start = split->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400549 }
Chris Mason3b951512008-04-17 11:29:12 -0400550
551 ret = add_extent_mapping(em_tree, split);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100552 BUG_ON(ret); /* Logic error */
Chris Mason3b951512008-04-17 11:29:12 -0400553 free_extent_map(split);
554 split = NULL;
555 }
Chris Mason890871b2009-09-02 16:24:52 -0400556 write_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -0500557
Chris Masona52d9a82007-08-27 16:49:44 -0400558 /* once for us */
559 free_extent_map(em);
560 /* once for the tree*/
561 free_extent_map(em);
562 }
Chris Mason3b951512008-04-17 11:29:12 -0400563 if (split)
564 free_extent_map(split);
565 if (split2)
566 free_extent_map(split2);
Chris Masona52d9a82007-08-27 16:49:44 -0400567 return 0;
568}
569
Chris Mason39279cc2007-06-12 06:35:45 -0400570/*
571 * this is very complex, but the basic idea is to drop all extents
572 * in the range start - end. hint_block is filled in with a block number
573 * that would be a good hint to the block allocator for this file.
574 *
575 * If an extent intersects the range but is not entirely inside the range
576 * it is either truncated or split. Anything entirely inside the range
577 * is deleted from the tree.
578 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000579int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
580 u64 start, u64 end, u64 *hint_byte, int drop_cache)
Chris Mason39279cc2007-06-12 06:35:45 -0400581{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000582 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason00f5c792007-11-30 10:09:33 -0500583 struct extent_buffer *leaf;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000584 struct btrfs_file_extent_item *fi;
Chris Mason00f5c792007-11-30 10:09:33 -0500585 struct btrfs_path *path;
586 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000587 struct btrfs_key new_key;
Li Zefan33345d012011-04-20 10:31:50 +0800588 u64 ino = btrfs_ino(inode);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000589 u64 search_start = start;
590 u64 disk_bytenr = 0;
591 u64 num_bytes = 0;
592 u64 extent_offset = 0;
593 u64 extent_end = 0;
594 int del_nr = 0;
595 int del_slot = 0;
596 int extent_type;
Chris Masonccd467d2007-06-28 15:57:36 -0400597 int recow;
Chris Mason00f5c792007-11-30 10:09:33 -0500598 int ret;
Chris Masondc7fdde2012-04-27 14:31:29 -0400599 int modify_tree = -1;
Chris Mason39279cc2007-06-12 06:35:45 -0400600
Chris Masona1ed8352009-09-11 12:27:37 -0400601 if (drop_cache)
602 btrfs_drop_extent_cache(inode, start, end - 1, 0);
Chris Masona52d9a82007-08-27 16:49:44 -0400603
Chris Mason39279cc2007-06-12 06:35:45 -0400604 path = btrfs_alloc_path();
605 if (!path)
606 return -ENOMEM;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000607
Chris Masondc7fdde2012-04-27 14:31:29 -0400608 if (start >= BTRFS_I(inode)->disk_i_size)
609 modify_tree = 0;
610
Chris Masond3977122009-01-05 21:25:51 -0500611 while (1) {
Chris Masonccd467d2007-06-28 15:57:36 -0400612 recow = 0;
Li Zefan33345d012011-04-20 10:31:50 +0800613 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Chris Masondc7fdde2012-04-27 14:31:29 -0400614 search_start, modify_tree);
Chris Mason39279cc2007-06-12 06:35:45 -0400615 if (ret < 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000616 break;
617 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
618 leaf = path->nodes[0];
619 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +0800620 if (key.objectid == ino &&
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000621 key.type == BTRFS_EXTENT_DATA_KEY)
622 path->slots[0]--;
Chris Mason39279cc2007-06-12 06:35:45 -0400623 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400624 ret = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000625next_slot:
626 leaf = path->nodes[0];
627 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
628 BUG_ON(del_nr > 0);
629 ret = btrfs_next_leaf(root, path);
630 if (ret < 0)
631 break;
632 if (ret > 0) {
633 ret = 0;
634 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400635 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000636 leaf = path->nodes[0];
637 recow = 1;
638 }
639
640 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800641 if (key.objectid > ino ||
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000642 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
643 break;
644
645 fi = btrfs_item_ptr(leaf, path->slots[0],
646 struct btrfs_file_extent_item);
647 extent_type = btrfs_file_extent_type(leaf, fi);
648
649 if (extent_type == BTRFS_FILE_EXTENT_REG ||
650 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
651 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
652 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
653 extent_offset = btrfs_file_extent_offset(leaf, fi);
654 extent_end = key.offset +
655 btrfs_file_extent_num_bytes(leaf, fi);
656 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
657 extent_end = key.offset +
658 btrfs_file_extent_inline_len(leaf, fi);
Chris Mason8c2383c2007-06-18 09:57:58 -0400659 } else {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000660 WARN_ON(1);
Chris Mason8c2383c2007-06-18 09:57:58 -0400661 extent_end = search_start;
Chris Mason39279cc2007-06-12 06:35:45 -0400662 }
663
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000664 if (extent_end <= search_start) {
665 path->slots[0]++;
Chris Mason8c2383c2007-06-18 09:57:58 -0400666 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400667 }
668
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000669 search_start = max(key.offset, start);
Chris Masondc7fdde2012-04-27 14:31:29 -0400670 if (recow || !modify_tree) {
671 modify_tree = -1;
David Sterbab3b4aa72011-04-21 01:20:15 +0200672 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000673 continue;
Chris Mason39279cc2007-06-12 06:35:45 -0400674 }
Chris Mason771ed682008-11-06 22:02:51 -0500675
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000676 /*
677 * | - range to drop - |
678 * | -------- extent -------- |
679 */
680 if (start > key.offset && end < extent_end) {
681 BUG_ON(del_nr > 0);
682 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
683
684 memcpy(&new_key, &key, sizeof(new_key));
685 new_key.offset = start;
686 ret = btrfs_duplicate_item(trans, root, path,
687 &new_key);
688 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200689 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000690 continue;
691 }
692 if (ret < 0)
693 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400694
Chris Mason5f39d392007-10-15 16:14:19 -0400695 leaf = path->nodes[0];
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000696 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
697 struct btrfs_file_extent_item);
698 btrfs_set_file_extent_num_bytes(leaf, fi,
699 start - key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400700
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000701 fi = btrfs_item_ptr(leaf, path->slots[0],
702 struct btrfs_file_extent_item);
Chris Masonc8b97812008-10-29 14:49:59 -0400703
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000704 extent_offset += start - key.offset;
705 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
706 btrfs_set_file_extent_num_bytes(leaf, fi,
707 extent_end - start);
708 btrfs_mark_buffer_dirty(leaf);
Chris Masondb945352007-10-15 16:15:53 -0400709
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000710 if (disk_bytenr > 0) {
711 ret = btrfs_inc_extent_ref(trans, root,
712 disk_bytenr, num_bytes, 0,
713 root->root_key.objectid,
714 new_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200715 start - extent_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100716 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000717 *hint_byte = disk_bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -0400718 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000719 key.offset = start;
720 }
721 /*
722 * | ---- range to drop ----- |
723 * | -------- extent -------- |
724 */
725 if (start <= key.offset && end < extent_end) {
726 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
727
728 memcpy(&new_key, &key, sizeof(new_key));
729 new_key.offset = end;
730 btrfs_set_item_key_safe(trans, root, path, &new_key);
731
732 extent_offset += end - key.offset;
733 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
734 btrfs_set_file_extent_num_bytes(leaf, fi,
735 extent_end - end);
736 btrfs_mark_buffer_dirty(leaf);
737 if (disk_bytenr > 0) {
738 inode_sub_bytes(inode, end - key.offset);
739 *hint_byte = disk_bytenr;
740 }
741 break;
Zheng Yan31840ae2008-09-23 13:14:14 -0400742 }
743
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000744 search_start = extent_end;
745 /*
746 * | ---- range to drop ----- |
747 * | -------- extent -------- |
748 */
749 if (start > key.offset && end >= extent_end) {
750 BUG_ON(del_nr > 0);
751 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
752
753 btrfs_set_file_extent_num_bytes(leaf, fi,
754 start - key.offset);
755 btrfs_mark_buffer_dirty(leaf);
756 if (disk_bytenr > 0) {
757 inode_sub_bytes(inode, extent_end - start);
758 *hint_byte = disk_bytenr;
759 }
760 if (end == extent_end)
761 break;
762
763 path->slots[0]++;
764 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400765 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000766
767 /*
768 * | ---- range to drop ----- |
769 * | ------ extent ------ |
770 */
771 if (start <= key.offset && end >= extent_end) {
772 if (del_nr == 0) {
773 del_slot = path->slots[0];
774 del_nr = 1;
775 } else {
776 BUG_ON(del_slot + del_nr != path->slots[0]);
777 del_nr++;
778 }
779
780 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
781 inode_sub_bytes(inode,
782 extent_end - key.offset);
783 extent_end = ALIGN(extent_end,
784 root->sectorsize);
785 } else if (disk_bytenr > 0) {
786 ret = btrfs_free_extent(trans, root,
787 disk_bytenr, num_bytes, 0,
788 root->root_key.objectid,
789 key.objectid, key.offset -
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200790 extent_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100791 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000792 inode_sub_bytes(inode,
793 extent_end - key.offset);
794 *hint_byte = disk_bytenr;
795 }
796
797 if (end == extent_end)
798 break;
799
800 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
801 path->slots[0]++;
802 goto next_slot;
803 }
804
805 ret = btrfs_del_items(trans, root, path, del_slot,
806 del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100807 if (ret) {
808 btrfs_abort_transaction(trans, root, ret);
809 goto out;
810 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000811
812 del_nr = 0;
813 del_slot = 0;
814
David Sterbab3b4aa72011-04-21 01:20:15 +0200815 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000816 continue;
817 }
818
819 BUG_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -0400820 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000821
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100822 if (!ret && del_nr > 0) {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000823 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100824 if (ret)
825 btrfs_abort_transaction(trans, root, ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000826 }
827
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100828out:
Chris Mason39279cc2007-06-12 06:35:45 -0400829 btrfs_free_path(path);
830 return ret;
831}
832
Yan Zhengd899e052008-10-30 14:25:28 -0400833static int extent_mergeable(struct extent_buffer *leaf, int slot,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000834 u64 objectid, u64 bytenr, u64 orig_offset,
835 u64 *start, u64 *end)
Yan Zhengd899e052008-10-30 14:25:28 -0400836{
837 struct btrfs_file_extent_item *fi;
838 struct btrfs_key key;
839 u64 extent_end;
840
841 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
842 return 0;
843
844 btrfs_item_key_to_cpu(leaf, &key, slot);
845 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
846 return 0;
847
848 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
849 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
850 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000851 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Yan Zhengd899e052008-10-30 14:25:28 -0400852 btrfs_file_extent_compression(leaf, fi) ||
853 btrfs_file_extent_encryption(leaf, fi) ||
854 btrfs_file_extent_other_encoding(leaf, fi))
855 return 0;
856
857 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
858 if ((*start && *start != key.offset) || (*end && *end != extent_end))
859 return 0;
860
861 *start = key.offset;
862 *end = extent_end;
863 return 1;
864}
865
866/*
867 * Mark extent in the range start - end as written.
868 *
869 * This changes extent type from 'pre-allocated' to 'regular'. If only
870 * part of extent is marked as written, the extent will be split into
871 * two or three.
872 */
873int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
Yan Zhengd899e052008-10-30 14:25:28 -0400874 struct inode *inode, u64 start, u64 end)
875{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000876 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zhengd899e052008-10-30 14:25:28 -0400877 struct extent_buffer *leaf;
878 struct btrfs_path *path;
879 struct btrfs_file_extent_item *fi;
880 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000881 struct btrfs_key new_key;
Yan Zhengd899e052008-10-30 14:25:28 -0400882 u64 bytenr;
883 u64 num_bytes;
884 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400885 u64 orig_offset;
Yan Zhengd899e052008-10-30 14:25:28 -0400886 u64 other_start;
887 u64 other_end;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000888 u64 split;
889 int del_nr = 0;
890 int del_slot = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000891 int recow;
Yan Zhengd899e052008-10-30 14:25:28 -0400892 int ret;
Li Zefan33345d012011-04-20 10:31:50 +0800893 u64 ino = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -0400894
895 btrfs_drop_extent_cache(inode, start, end - 1, 0);
896
897 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -0700898 if (!path)
899 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -0400900again:
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000901 recow = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000902 split = start;
Li Zefan33345d012011-04-20 10:31:50 +0800903 key.objectid = ino;
Yan Zhengd899e052008-10-30 14:25:28 -0400904 key.type = BTRFS_EXTENT_DATA_KEY;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000905 key.offset = split;
Yan Zhengd899e052008-10-30 14:25:28 -0400906
907 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Josef Bacik41415732011-03-16 13:59:32 -0400908 if (ret < 0)
909 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -0400910 if (ret > 0 && path->slots[0] > 0)
911 path->slots[0]--;
912
913 leaf = path->nodes[0];
914 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800915 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
Yan Zhengd899e052008-10-30 14:25:28 -0400916 fi = btrfs_item_ptr(leaf, path->slots[0],
917 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000918 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
919 BTRFS_FILE_EXTENT_PREALLOC);
Yan Zhengd899e052008-10-30 14:25:28 -0400920 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
921 BUG_ON(key.offset > start || extent_end < end);
922
923 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
924 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400925 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000926 memcpy(&new_key, &key, sizeof(new_key));
927
928 if (start == key.offset && end < extent_end) {
929 other_start = 0;
930 other_end = start;
931 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +0800932 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000933 &other_start, &other_end)) {
934 new_key.offset = end;
935 btrfs_set_item_key_safe(trans, root, path, &new_key);
936 fi = btrfs_item_ptr(leaf, path->slots[0],
937 struct btrfs_file_extent_item);
938 btrfs_set_file_extent_num_bytes(leaf, fi,
939 extent_end - end);
940 btrfs_set_file_extent_offset(leaf, fi,
941 end - orig_offset);
942 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
943 struct btrfs_file_extent_item);
944 btrfs_set_file_extent_num_bytes(leaf, fi,
945 end - other_start);
946 btrfs_mark_buffer_dirty(leaf);
947 goto out;
948 }
949 }
950
951 if (start > key.offset && end == extent_end) {
952 other_start = end;
953 other_end = 0;
954 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +0800955 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000956 &other_start, &other_end)) {
957 fi = btrfs_item_ptr(leaf, path->slots[0],
958 struct btrfs_file_extent_item);
959 btrfs_set_file_extent_num_bytes(leaf, fi,
960 start - key.offset);
961 path->slots[0]++;
962 new_key.offset = start;
963 btrfs_set_item_key_safe(trans, root, path, &new_key);
964
965 fi = btrfs_item_ptr(leaf, path->slots[0],
966 struct btrfs_file_extent_item);
967 btrfs_set_file_extent_num_bytes(leaf, fi,
968 other_end - start);
969 btrfs_set_file_extent_offset(leaf, fi,
970 start - orig_offset);
971 btrfs_mark_buffer_dirty(leaf);
972 goto out;
973 }
974 }
Yan Zhengd899e052008-10-30 14:25:28 -0400975
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000976 while (start > key.offset || end < extent_end) {
977 if (key.offset == start)
978 split = end;
Yan Zhengd899e052008-10-30 14:25:28 -0400979
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000980 new_key.offset = split;
981 ret = btrfs_duplicate_item(trans, root, path, &new_key);
982 if (ret == -EAGAIN) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200983 btrfs_release_path(path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000984 goto again;
Yan Zhengd899e052008-10-30 14:25:28 -0400985 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100986 if (ret < 0) {
987 btrfs_abort_transaction(trans, root, ret);
988 goto out;
989 }
Yan Zhengd899e052008-10-30 14:25:28 -0400990
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000991 leaf = path->nodes[0];
992 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Yan Zhengd899e052008-10-30 14:25:28 -0400993 struct btrfs_file_extent_item);
Yan Zhengd899e052008-10-30 14:25:28 -0400994 btrfs_set_file_extent_num_bytes(leaf, fi,
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000995 split - key.offset);
996
997 fi = btrfs_item_ptr(leaf, path->slots[0],
998 struct btrfs_file_extent_item);
999
1000 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1001 btrfs_set_file_extent_num_bytes(leaf, fi,
1002 extent_end - split);
Yan Zhengd899e052008-10-30 14:25:28 -04001003 btrfs_mark_buffer_dirty(leaf);
1004
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001005 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1006 root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001007 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001008 BUG_ON(ret); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001009
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001010 if (split == start) {
1011 key.offset = start;
1012 } else {
1013 BUG_ON(start != key.offset);
Yan Zhengd899e052008-10-30 14:25:28 -04001014 path->slots[0]--;
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001015 extent_end = end;
Yan Zhengd899e052008-10-30 14:25:28 -04001016 }
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001017 recow = 1;
Yan Zhengd899e052008-10-30 14:25:28 -04001018 }
1019
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001020 other_start = end;
1021 other_end = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001022 if (extent_mergeable(leaf, path->slots[0] + 1,
Li Zefan33345d012011-04-20 10:31:50 +08001023 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001024 &other_start, &other_end)) {
1025 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001026 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001027 goto again;
1028 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001029 extent_end = other_end;
1030 del_slot = path->slots[0] + 1;
1031 del_nr++;
1032 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1033 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001034 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001035 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001036 }
1037 other_start = 0;
1038 other_end = start;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001039 if (extent_mergeable(leaf, path->slots[0] - 1,
Li Zefan33345d012011-04-20 10:31:50 +08001040 ino, bytenr, orig_offset,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001041 &other_start, &other_end)) {
1042 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001043 btrfs_release_path(path);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001044 goto again;
1045 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001046 key.offset = other_start;
1047 del_slot = path->slots[0];
1048 del_nr++;
1049 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1050 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001051 ino, orig_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001052 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001053 }
1054 if (del_nr == 0) {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001055 fi = btrfs_item_ptr(leaf, path->slots[0],
1056 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001057 btrfs_set_file_extent_type(leaf, fi,
1058 BTRFS_FILE_EXTENT_REG);
1059 btrfs_mark_buffer_dirty(leaf);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001060 } else {
Shaohua Li3f6fae92010-02-11 07:43:00 +00001061 fi = btrfs_item_ptr(leaf, del_slot - 1,
1062 struct btrfs_file_extent_item);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +00001063 btrfs_set_file_extent_type(leaf, fi,
1064 BTRFS_FILE_EXTENT_REG);
1065 btrfs_set_file_extent_num_bytes(leaf, fi,
1066 extent_end - key.offset);
1067 btrfs_mark_buffer_dirty(leaf);
1068
1069 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001070 if (ret < 0) {
1071 btrfs_abort_transaction(trans, root, ret);
1072 goto out;
1073 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001074 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001075out:
Yan Zhengd899e052008-10-30 14:25:28 -04001076 btrfs_free_path(path);
1077 return 0;
1078}
1079
Chris Mason39279cc2007-06-12 06:35:45 -04001080/*
Chris Masonb1bf8622011-02-28 09:52:08 -05001081 * on error we return an unlocked page and the error value
1082 * on success we return a locked page and 0
1083 */
Josef Bacikb63164292011-09-30 15:23:54 -04001084static int prepare_uptodate_page(struct page *page, u64 pos,
1085 bool force_uptodate)
Chris Masonb1bf8622011-02-28 09:52:08 -05001086{
1087 int ret = 0;
1088
Josef Bacikb63164292011-09-30 15:23:54 -04001089 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1090 !PageUptodate(page)) {
Chris Masonb1bf8622011-02-28 09:52:08 -05001091 ret = btrfs_readpage(NULL, page);
1092 if (ret)
1093 return ret;
1094 lock_page(page);
1095 if (!PageUptodate(page)) {
1096 unlock_page(page);
1097 return -EIO;
1098 }
1099 }
1100 return 0;
1101}
1102
1103/*
Chris Masond352ac62008-09-29 15:18:18 -04001104 * this gets pages into the page cache and locks them down, it also properly
1105 * waits for data=ordered extents to finish before allowing the pages to be
1106 * modified.
Chris Mason39279cc2007-06-12 06:35:45 -04001107 */
Chris Masond3977122009-01-05 21:25:51 -05001108static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
Chris Mason98ed5172008-01-03 10:01:48 -05001109 struct page **pages, size_t num_pages,
1110 loff_t pos, unsigned long first_index,
Josef Bacikb63164292011-09-30 15:23:54 -04001111 size_t write_bytes, bool force_uptodate)
Chris Mason39279cc2007-06-12 06:35:45 -04001112{
Josef Bacik2ac55d42010-02-03 19:33:23 +00001113 struct extent_state *cached_state = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001114 int i;
1115 unsigned long index = pos >> PAGE_CACHE_SHIFT;
Chris Mason6da6aba2007-12-18 16:15:09 -05001116 struct inode *inode = fdentry(file)->d_inode;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04001117 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
Chris Mason39279cc2007-06-12 06:35:45 -04001118 int err = 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001119 int faili = 0;
Chris Mason8c2383c2007-06-18 09:57:58 -04001120 u64 start_pos;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001121 u64 last_pos;
Chris Mason8c2383c2007-06-18 09:57:58 -04001122
Chris Mason5f39d392007-10-15 16:14:19 -04001123 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001124 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001125
Chris Masone6dcd2d2008-07-17 12:53:50 -04001126again:
Chris Mason39279cc2007-06-12 06:35:45 -04001127 for (i = 0; i < num_pages; i++) {
Josef Bacika94733d2011-07-11 10:47:06 -04001128 pages[i] = find_or_create_page(inode->i_mapping, index + i,
Johannes Weinere3a41a52012-01-10 15:07:55 -08001129 mask | __GFP_WRITE);
Chris Mason39279cc2007-06-12 06:35:45 -04001130 if (!pages[i]) {
Chris Masonb1bf8622011-02-28 09:52:08 -05001131 faili = i - 1;
1132 err = -ENOMEM;
1133 goto fail;
1134 }
1135
1136 if (i == 0)
Josef Bacikb63164292011-09-30 15:23:54 -04001137 err = prepare_uptodate_page(pages[i], pos,
1138 force_uptodate);
Chris Masonb1bf8622011-02-28 09:52:08 -05001139 if (i == num_pages - 1)
1140 err = prepare_uptodate_page(pages[i],
Josef Bacikb63164292011-09-30 15:23:54 -04001141 pos + write_bytes, false);
Chris Masonb1bf8622011-02-28 09:52:08 -05001142 if (err) {
1143 page_cache_release(pages[i]);
1144 faili = i - 1;
1145 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -04001146 }
Chris Masonccd467d2007-06-28 15:57:36 -04001147 wait_on_page_writeback(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -04001148 }
Chris Masonb1bf8622011-02-28 09:52:08 -05001149 err = 0;
Chris Mason07627042008-02-19 11:29:24 -05001150 if (start_pos < inode->i_size) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04001151 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001152 lock_extent_bits(&BTRFS_I(inode)->io_tree,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001153 start_pos, last_pos - 1, 0, &cached_state);
Chris Masond3977122009-01-05 21:25:51 -05001154 ordered = btrfs_lookup_first_ordered_extent(inode,
1155 last_pos - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001156 if (ordered &&
1157 ordered->file_offset + ordered->len > start_pos &&
1158 ordered->file_offset < last_pos) {
1159 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001160 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1161 start_pos, last_pos - 1,
1162 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001163 for (i = 0; i < num_pages; i++) {
1164 unlock_page(pages[i]);
1165 page_cache_release(pages[i]);
1166 }
1167 btrfs_wait_ordered_range(inode, start_pos,
1168 last_pos - start_pos);
1169 goto again;
1170 }
1171 if (ordered)
1172 btrfs_put_ordered_extent(ordered);
1173
Josef Bacik2ac55d42010-02-03 19:33:23 +00001174 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
Josef Bacik32c00af2009-10-08 13:34:05 -04001175 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00001176 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
Chris Mason07627042008-02-19 11:29:24 -05001177 GFP_NOFS);
Josef Bacik2ac55d42010-02-03 19:33:23 +00001178 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1179 start_pos, last_pos - 1, &cached_state,
1180 GFP_NOFS);
Chris Mason07627042008-02-19 11:29:24 -05001181 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001182 for (i = 0; i < num_pages; i++) {
Wu Fengguang32c7f202011-08-08 15:19:47 -06001183 if (clear_page_dirty_for_io(pages[i]))
1184 account_page_redirty(pages[i]);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001185 set_page_extent_mapped(pages[i]);
1186 WARN_ON(!PageLocked(pages[i]));
1187 }
Chris Mason39279cc2007-06-12 06:35:45 -04001188 return 0;
Chris Masonb1bf8622011-02-28 09:52:08 -05001189fail:
1190 while (faili >= 0) {
1191 unlock_page(pages[faili]);
1192 page_cache_release(pages[faili]);
1193 faili--;
1194 }
1195 return err;
1196
Chris Mason39279cc2007-06-12 06:35:45 -04001197}
1198
Josef Bacikd0215f32011-01-25 14:57:24 -05001199static noinline ssize_t __btrfs_buffered_write(struct file *file,
1200 struct iov_iter *i,
1201 loff_t pos)
Josef Bacik4b46fce2010-05-23 11:00:55 -04001202{
Josef Bacik11c65dc2010-05-23 11:07:21 -04001203 struct inode *inode = fdentry(file)->d_inode;
1204 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik11c65dc2010-05-23 11:07:21 -04001205 struct page **pages = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001206 unsigned long first_index;
Josef Bacikd0215f32011-01-25 14:57:24 -05001207 size_t num_written = 0;
1208 int nrptrs;
Tsutomu Itohc9149232011-03-30 00:57:23 +00001209 int ret = 0;
Josef Bacikb63164292011-09-30 15:23:54 -04001210 bool force_page_uptodate = false;
Chris Masoncb843a62008-10-03 12:30:02 -04001211
Josef Bacikd0215f32011-01-25 14:57:24 -05001212 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
Josef Bacik11c65dc2010-05-23 11:07:21 -04001213 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1214 (sizeof(struct page *)));
Wu Fengguang142349f2011-12-16 12:32:57 -05001215 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1216 nrptrs = max(nrptrs, 8);
Chris Mason8c2383c2007-06-18 09:57:58 -04001217 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001218 if (!pages)
1219 return -ENOMEM;
Chris Masonab93dbe2009-10-01 12:29:10 -04001220
Chris Mason39279cc2007-06-12 06:35:45 -04001221 first_index = pos >> PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -04001222
Josef Bacikd0215f32011-01-25 14:57:24 -05001223 while (iov_iter_count(i) > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -04001224 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
Josef Bacikd0215f32011-01-25 14:57:24 -05001225 size_t write_bytes = min(iov_iter_count(i),
Josef Bacik11c65dc2010-05-23 11:07:21 -04001226 nrptrs * (size_t)PAGE_CACHE_SIZE -
Chris Mason8c2383c2007-06-18 09:57:58 -04001227 offset);
Yan, Zheng3a909832011-01-18 13:34:40 +08001228 size_t num_pages = (write_bytes + offset +
1229 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Josef Bacikd0215f32011-01-25 14:57:24 -05001230 size_t dirty_pages;
1231 size_t copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001232
Chris Mason8c2383c2007-06-18 09:57:58 -04001233 WARN_ON(num_pages > nrptrs);
Chris Mason1832a6d2007-12-21 16:27:21 -05001234
Xin Zhong914ee292010-12-09 09:30:14 +00001235 /*
1236 * Fault pages before locking them in prepare_pages
1237 * to avoid recursive lock
1238 */
Josef Bacikd0215f32011-01-25 14:57:24 -05001239 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
Xin Zhong914ee292010-12-09 09:30:14 +00001240 ret = -EFAULT;
Josef Bacikd0215f32011-01-25 14:57:24 -05001241 break;
Xin Zhong914ee292010-12-09 09:30:14 +00001242 }
1243
1244 ret = btrfs_delalloc_reserve_space(inode,
1245 num_pages << PAGE_CACHE_SHIFT);
Chris Mason1832a6d2007-12-21 16:27:21 -05001246 if (ret)
Josef Bacikd0215f32011-01-25 14:57:24 -05001247 break;
Chris Mason1832a6d2007-12-21 16:27:21 -05001248
Josef Bacik4a640012011-01-25 15:10:08 -05001249 /*
1250 * This is going to setup the pages array with the number of
1251 * pages we want, so we don't really need to worry about the
1252 * contents of pages from loop to loop
1253 */
Chris Mason39279cc2007-06-12 06:35:45 -04001254 ret = prepare_pages(root, file, pages, num_pages,
Josef Bacikb63164292011-09-30 15:23:54 -04001255 pos, first_index, write_bytes,
1256 force_page_uptodate);
Josef Bacik6a632092009-02-20 11:00:09 -05001257 if (ret) {
Xin Zhong914ee292010-12-09 09:30:14 +00001258 btrfs_delalloc_release_space(inode,
1259 num_pages << PAGE_CACHE_SHIFT);
Josef Bacikd0215f32011-01-25 14:57:24 -05001260 break;
Josef Bacik6a632092009-02-20 11:00:09 -05001261 }
Chris Mason39279cc2007-06-12 06:35:45 -04001262
Xin Zhong914ee292010-12-09 09:30:14 +00001263 copied = btrfs_copy_from_user(pos, num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -05001264 write_bytes, pages, i);
Chris Masonb1bf8622011-02-28 09:52:08 -05001265
1266 /*
1267 * if we have trouble faulting in the pages, fall
1268 * back to one page at a time
1269 */
1270 if (copied < write_bytes)
1271 nrptrs = 1;
1272
Josef Bacikb63164292011-09-30 15:23:54 -04001273 if (copied == 0) {
1274 force_page_uptodate = true;
Chris Masonb1bf8622011-02-28 09:52:08 -05001275 dirty_pages = 0;
Josef Bacikb63164292011-09-30 15:23:54 -04001276 } else {
1277 force_page_uptodate = false;
Chris Masonb1bf8622011-02-28 09:52:08 -05001278 dirty_pages = (copied + offset +
1279 PAGE_CACHE_SIZE - 1) >>
1280 PAGE_CACHE_SHIFT;
Josef Bacikb63164292011-09-30 15:23:54 -04001281 }
Xin Zhong914ee292010-12-09 09:30:14 +00001282
Josef Bacikd0215f32011-01-25 14:57:24 -05001283 /*
1284 * If we had a short copy we need to release the excess delaloc
1285 * bytes we reserved. We need to increment outstanding_extents
1286 * because btrfs_delalloc_release_space will decrement it, but
1287 * we still have an outstanding extent for the chunk we actually
1288 * managed to copy.
1289 */
Xin Zhong914ee292010-12-09 09:30:14 +00001290 if (num_pages > dirty_pages) {
Josef Bacik9e0baf62011-07-15 15:16:44 +00001291 if (copied > 0) {
1292 spin_lock(&BTRFS_I(inode)->lock);
1293 BTRFS_I(inode)->outstanding_extents++;
1294 spin_unlock(&BTRFS_I(inode)->lock);
1295 }
Xin Zhong914ee292010-12-09 09:30:14 +00001296 btrfs_delalloc_release_space(inode,
1297 (num_pages - dirty_pages) <<
1298 PAGE_CACHE_SHIFT);
1299 }
1300
1301 if (copied > 0) {
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001302 ret = btrfs_dirty_pages(root, inode, pages,
1303 dirty_pages, pos, copied,
1304 NULL);
Josef Bacikd0215f32011-01-25 14:57:24 -05001305 if (ret) {
1306 btrfs_delalloc_release_space(inode,
1307 dirty_pages << PAGE_CACHE_SHIFT);
1308 btrfs_drop_pages(pages, num_pages);
1309 break;
1310 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001311 }
Chris Mason39279cc2007-06-12 06:35:45 -04001312
Chris Mason39279cc2007-06-12 06:35:45 -04001313 btrfs_drop_pages(pages, num_pages);
Xin Zhong914ee292010-12-09 09:30:14 +00001314
Josef Bacikd0215f32011-01-25 14:57:24 -05001315 cond_resched();
1316
1317 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1318 dirty_pages);
1319 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1320 btrfs_btree_balance_dirty(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04001321
Xin Zhong914ee292010-12-09 09:30:14 +00001322 pos += copied;
1323 num_written += copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001324 }
Chris Mason5b92ee72008-01-03 13:46:11 -05001325
Chris Mason8c2383c2007-06-18 09:57:58 -04001326 kfree(pages);
Josef Bacikd0215f32011-01-25 14:57:24 -05001327
1328 return num_written ? num_written : ret;
1329}
1330
1331static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1332 const struct iovec *iov,
1333 unsigned long nr_segs, loff_t pos,
1334 loff_t *ppos, size_t count, size_t ocount)
1335{
1336 struct file *file = iocb->ki_filp;
1337 struct inode *inode = fdentry(file)->d_inode;
1338 struct iov_iter i;
1339 ssize_t written;
1340 ssize_t written_buffered;
1341 loff_t endbyte;
1342 int err;
1343
1344 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1345 count, ocount);
1346
1347 /*
1348 * the generic O_DIRECT will update in-memory i_size after the
1349 * DIOs are done. But our endio handlers that update the on
1350 * disk i_size never update past the in memory i_size. So we
1351 * need one more update here to catch any additions to the
1352 * file
1353 */
1354 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1355 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1356 mark_inode_dirty(inode);
1357 }
1358
1359 if (written < 0 || written == count)
1360 return written;
1361
1362 pos += written;
1363 count -= written;
1364 iov_iter_init(&i, iov, nr_segs, count, written);
1365 written_buffered = __btrfs_buffered_write(file, &i, pos);
1366 if (written_buffered < 0) {
1367 err = written_buffered;
1368 goto out;
1369 }
1370 endbyte = pos + written_buffered - 1;
1371 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1372 if (err)
1373 goto out;
1374 written += written_buffered;
1375 *ppos = pos + written_buffered;
1376 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1377 endbyte >> PAGE_CACHE_SHIFT);
1378out:
1379 return written ? written : err;
1380}
1381
1382static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1383 const struct iovec *iov,
1384 unsigned long nr_segs, loff_t pos)
1385{
1386 struct file *file = iocb->ki_filp;
1387 struct inode *inode = fdentry(file)->d_inode;
1388 struct btrfs_root *root = BTRFS_I(inode)->root;
1389 loff_t *ppos = &iocb->ki_pos;
Miao Xie0c1a98c2011-09-11 10:52:24 -04001390 u64 start_pos;
Josef Bacikd0215f32011-01-25 14:57:24 -05001391 ssize_t num_written = 0;
1392 ssize_t err = 0;
1393 size_t count, ocount;
1394
1395 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1396
1397 mutex_lock(&inode->i_mutex);
1398
1399 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1400 if (err) {
1401 mutex_unlock(&inode->i_mutex);
1402 goto out;
1403 }
1404 count = ocount;
1405
1406 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1407 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1408 if (err) {
1409 mutex_unlock(&inode->i_mutex);
1410 goto out;
1411 }
1412
1413 if (count == 0) {
1414 mutex_unlock(&inode->i_mutex);
1415 goto out;
1416 }
1417
1418 err = file_remove_suid(file);
1419 if (err) {
1420 mutex_unlock(&inode->i_mutex);
1421 goto out;
1422 }
1423
1424 /*
1425 * If BTRFS flips readonly due to some impossible error
1426 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1427 * although we have opened a file as writable, we have
1428 * to stop this write operation to ensure FS consistency.
1429 */
1430 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1431 mutex_unlock(&inode->i_mutex);
1432 err = -EROFS;
1433 goto out;
1434 }
1435
Josef Bacike41f9412012-03-26 09:46:47 -04001436 err = file_update_time(file);
Josef Bacik22c44fe2011-11-30 10:45:38 -05001437 if (err) {
1438 mutex_unlock(&inode->i_mutex);
1439 goto out;
1440 }
Josef Bacikd0215f32011-01-25 14:57:24 -05001441
Miao Xie0c1a98c2011-09-11 10:52:24 -04001442 start_pos = round_down(pos, root->sectorsize);
1443 if (start_pos > i_size_read(inode)) {
1444 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1445 if (err) {
1446 mutex_unlock(&inode->i_mutex);
1447 goto out;
1448 }
1449 }
1450
Josef Bacikd0215f32011-01-25 14:57:24 -05001451 if (unlikely(file->f_flags & O_DIRECT)) {
1452 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1453 pos, ppos, count, ocount);
1454 } else {
1455 struct iov_iter i;
1456
1457 iov_iter_init(&i, iov, nr_segs, count, num_written);
1458
1459 num_written = __btrfs_buffered_write(file, &i, pos);
1460 if (num_written > 0)
1461 *ppos = pos + num_written;
1462 }
1463
1464 mutex_unlock(&inode->i_mutex);
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001465
Chris Mason5a3f23d2009-03-31 13:27:11 -04001466 /*
1467 * we want to make sure fsync finds this change
1468 * but we haven't joined a transaction running right now.
1469 *
1470 * Later on, someone is sure to update the inode and get the
1471 * real transid recorded.
1472 *
1473 * We set last_trans now to the fs_info generation + 1,
1474 * this will either be one more than the running transaction
1475 * or the generation used for the next transaction if there isn't
1476 * one running right now.
1477 */
1478 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
Josef Bacikd0215f32011-01-25 14:57:24 -05001479 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1480 err = generic_write_sync(file, pos, num_written);
1481 if (err < 0 && num_written > 0)
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001482 num_written = err;
1483 }
Josef Bacikd0215f32011-01-25 14:57:24 -05001484out:
Chris Mason39279cc2007-06-12 06:35:45 -04001485 current->backing_dev_info = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001486 return num_written ? num_written : err;
1487}
1488
Chris Masond3977122009-01-05 21:25:51 -05001489int btrfs_release_file(struct inode *inode, struct file *filp)
Mingminge1b81e62008-05-27 10:55:43 -04001490{
Chris Mason5a3f23d2009-03-31 13:27:11 -04001491 /*
1492 * ordered_data_close is set by settattr when we are about to truncate
1493 * a file from a non-zero size to a zero size. This tries to
1494 * flush down new bytes that may have been written if the
1495 * application were using truncate to replace a file in place.
1496 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04001497 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1498 &BTRFS_I(inode)->runtime_flags)) {
Chris Mason5a3f23d2009-03-31 13:27:11 -04001499 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1500 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1501 filemap_flush(inode->i_mapping);
1502 }
Sage Weil6bf13c02008-06-10 10:07:39 -04001503 if (filp->private_data)
1504 btrfs_ioctl_trans_end(filp);
Mingminge1b81e62008-05-27 10:55:43 -04001505 return 0;
1506}
1507
Chris Masond352ac62008-09-29 15:18:18 -04001508/*
1509 * fsync call for both files and directories. This logs the inode into
1510 * the tree log instead of forcing full commits whenever possible.
1511 *
1512 * It needs to call filemap_fdatawait so that all ordered extent updates are
1513 * in the metadata btree are up to date for copying to the log.
1514 *
1515 * It drops the inode mutex before doing the tree log commit. This is an
1516 * important optimization for directories because holding the mutex prevents
1517 * new operations on the dir while we write to disk.
1518 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001519int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
Chris Mason39279cc2007-06-12 06:35:45 -04001520{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001521 struct dentry *dentry = file->f_path.dentry;
Chris Mason39279cc2007-06-12 06:35:45 -04001522 struct inode *inode = dentry->d_inode;
1523 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001524 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001525 struct btrfs_trans_handle *trans;
1526
liubo1abe9b82011-03-24 11:18:59 +00001527 trace_btrfs_sync_file(file, datasync);
Chris Mason257c62e2009-10-13 13:21:08 -04001528
Josef Bacik02c24a82011-07-16 20:44:56 -04001529 mutex_lock(&inode->i_mutex);
1530
Josef Bacik0885ef52012-04-23 15:09:39 -04001531 /*
1532 * we wait first, since the writeback may change the inode, also wait
1533 * ordered range does a filemape_write_and_wait_range which is why we
1534 * don't do it above like other file systems.
1535 */
Chris Mason257c62e2009-10-13 13:21:08 -04001536 root->log_batch++;
Josef Bacik0885ef52012-04-23 15:09:39 -04001537 btrfs_wait_ordered_range(inode, start, end);
Chris Mason257c62e2009-10-13 13:21:08 -04001538 root->log_batch++;
1539
Chris Mason39279cc2007-06-12 06:35:45 -04001540 /*
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001541 * check the transaction that last modified this inode
1542 * and see if its already been committed
Chris Mason39279cc2007-06-12 06:35:45 -04001543 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001544 if (!BTRFS_I(inode)->last_trans) {
1545 mutex_unlock(&inode->i_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001546 goto out;
Josef Bacik02c24a82011-07-16 20:44:56 -04001547 }
Chris Masona2135012008-06-25 16:01:30 -04001548
Chris Mason257c62e2009-10-13 13:21:08 -04001549 /*
1550 * if the last transaction that changed this file was before
1551 * the current transaction, we can bail out now without any
1552 * syncing
1553 */
Josef Bacika4abeea2011-04-11 17:25:13 -04001554 smp_mb();
Josef Bacik22ee6982012-05-29 16:57:49 -04001555 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1556 BTRFS_I(inode)->last_trans <=
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001557 root->fs_info->last_trans_committed) {
1558 BTRFS_I(inode)->last_trans = 0;
Josef Bacik02c24a82011-07-16 20:44:56 -04001559 mutex_unlock(&inode->i_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001560 goto out;
1561 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001562
1563 /*
Chris Masona52d9a82007-08-27 16:49:44 -04001564 * ok we haven't committed the transaction yet, lets do a commit
1565 */
Dan Carpenter6f902af2010-05-29 09:49:07 +00001566 if (file->private_data)
Sage Weil6bf13c02008-06-10 10:07:39 -04001567 btrfs_ioctl_trans_end(file);
1568
Yan, Zhenga22285a2010-05-16 10:48:46 -04001569 trans = btrfs_start_transaction(root, 0);
1570 if (IS_ERR(trans)) {
1571 ret = PTR_ERR(trans);
Josef Bacik02c24a82011-07-16 20:44:56 -04001572 mutex_unlock(&inode->i_mutex);
Chris Mason39279cc2007-06-12 06:35:45 -04001573 goto out;
1574 }
Chris Masone02119d2008-09-05 16:13:11 -04001575
Chris Mason2cfbd502009-02-20 10:55:10 -05001576 ret = btrfs_log_dentry_safe(trans, root, dentry);
Josef Bacik02c24a82011-07-16 20:44:56 -04001577 if (ret < 0) {
1578 mutex_unlock(&inode->i_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04001579 goto out;
Josef Bacik02c24a82011-07-16 20:44:56 -04001580 }
Chris Mason49eb7e42008-09-11 15:53:12 -04001581
1582 /* we've logged all the items and now have a consistent
1583 * version of the file in the log. It is possible that
1584 * someone will come in and modify the file, but that's
1585 * fine because the log is consistent on disk, and we
1586 * have references to all of the file's extents
1587 *
1588 * It is possible that someone will come in and log the
1589 * file again, but that will end up using the synchronization
1590 * inside btrfs_sync_log to keep things safe.
1591 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001592 mutex_unlock(&inode->i_mutex);
Chris Mason49eb7e42008-09-11 15:53:12 -04001593
Chris Mason257c62e2009-10-13 13:21:08 -04001594 if (ret != BTRFS_NO_LOG_SYNC) {
1595 if (ret > 0) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001596 ret = btrfs_commit_transaction(trans, root);
Chris Mason257c62e2009-10-13 13:21:08 -04001597 } else {
1598 ret = btrfs_sync_log(trans, root);
1599 if (ret == 0)
1600 ret = btrfs_end_transaction(trans, root);
1601 else
1602 ret = btrfs_commit_transaction(trans, root);
1603 }
1604 } else {
1605 ret = btrfs_end_transaction(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001606 }
Chris Mason39279cc2007-06-12 06:35:45 -04001607out:
Roel Kluin014e4ac2010-01-29 10:42:11 +00001608 return ret > 0 ? -EIO : ret;
Chris Mason39279cc2007-06-12 06:35:45 -04001609}
1610
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001611static const struct vm_operations_struct btrfs_file_vm_ops = {
Chris Mason92fee662007-07-25 12:31:35 -04001612 .fault = filemap_fault,
Chris Mason9ebefb182007-06-15 13:50:00 -04001613 .page_mkwrite = btrfs_page_mkwrite,
1614};
1615
1616static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1617{
Miao Xie058a4572010-05-20 07:21:50 +00001618 struct address_space *mapping = filp->f_mapping;
1619
1620 if (!mapping->a_ops->readpage)
1621 return -ENOEXEC;
1622
Chris Mason9ebefb182007-06-15 13:50:00 -04001623 file_accessed(filp);
Miao Xie058a4572010-05-20 07:21:50 +00001624 vma->vm_ops = &btrfs_file_vm_ops;
1625 vma->vm_flags |= VM_CAN_NONLINEAR;
1626
Chris Mason9ebefb182007-06-15 13:50:00 -04001627 return 0;
1628}
1629
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001630static long btrfs_fallocate(struct file *file, int mode,
1631 loff_t offset, loff_t len)
1632{
1633 struct inode *inode = file->f_path.dentry->d_inode;
1634 struct extent_state *cached_state = NULL;
1635 u64 cur_offset;
1636 u64 last_byte;
1637 u64 alloc_start;
1638 u64 alloc_end;
1639 u64 alloc_hint = 0;
1640 u64 locked_end;
1641 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1642 struct extent_map *em;
1643 int ret;
1644
1645 alloc_start = offset & ~mask;
1646 alloc_end = (offset + len + mask) & ~mask;
1647
1648 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1649 if (mode & ~FALLOC_FL_KEEP_SIZE)
1650 return -EOPNOTSUPP;
1651
1652 /*
Chris Masond98456f2012-01-31 20:27:41 -05001653 * Make sure we have enough space before we do the
1654 * allocation.
1655 */
1656 ret = btrfs_check_data_free_space(inode, len);
1657 if (ret)
1658 return ret;
1659
1660 /*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001661 * wait for ordered IO before we have any locks. We'll loop again
1662 * below with the locks held.
1663 */
1664 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1665
1666 mutex_lock(&inode->i_mutex);
1667 ret = inode_newsize_ok(inode, alloc_end);
1668 if (ret)
1669 goto out;
1670
1671 if (alloc_start > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -05001672 ret = btrfs_cont_expand(inode, i_size_read(inode),
1673 alloc_start);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001674 if (ret)
1675 goto out;
1676 }
1677
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001678 locked_end = alloc_end - 1;
1679 while (1) {
1680 struct btrfs_ordered_extent *ordered;
1681
1682 /* the extent lock is ordered inside the running
1683 * transaction
1684 */
1685 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001686 locked_end, 0, &cached_state);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001687 ordered = btrfs_lookup_first_ordered_extent(inode,
1688 alloc_end - 1);
1689 if (ordered &&
1690 ordered->file_offset + ordered->len > alloc_start &&
1691 ordered->file_offset < alloc_end) {
1692 btrfs_put_ordered_extent(ordered);
1693 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1694 alloc_start, locked_end,
1695 &cached_state, GFP_NOFS);
1696 /*
1697 * we can't wait on the range with the transaction
1698 * running or with the extent lock held
1699 */
1700 btrfs_wait_ordered_range(inode, alloc_start,
1701 alloc_end - alloc_start);
1702 } else {
1703 if (ordered)
1704 btrfs_put_ordered_extent(ordered);
1705 break;
1706 }
1707 }
1708
1709 cur_offset = alloc_start;
1710 while (1) {
Josef Bacikf1e490a2011-08-18 10:36:39 -04001711 u64 actual_end;
1712
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001713 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1714 alloc_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001715 if (IS_ERR_OR_NULL(em)) {
1716 if (!em)
1717 ret = -ENOMEM;
1718 else
1719 ret = PTR_ERR(em);
1720 break;
1721 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001722 last_byte = min(extent_map_end(em), alloc_end);
Josef Bacikf1e490a2011-08-18 10:36:39 -04001723 actual_end = min_t(u64, extent_map_end(em), offset + len);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001724 last_byte = (last_byte + mask) & ~mask;
Josef Bacikf1e490a2011-08-18 10:36:39 -04001725
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001726 if (em->block_start == EXTENT_MAP_HOLE ||
1727 (cur_offset >= inode->i_size &&
1728 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1729 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1730 last_byte - cur_offset,
1731 1 << inode->i_blkbits,
1732 offset + len,
1733 &alloc_hint);
Josef Bacik1b9c3322011-08-17 10:19:52 -04001734
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001735 if (ret < 0) {
1736 free_extent_map(em);
1737 break;
1738 }
Josef Bacikf1e490a2011-08-18 10:36:39 -04001739 } else if (actual_end > inode->i_size &&
1740 !(mode & FALLOC_FL_KEEP_SIZE)) {
1741 /*
1742 * We didn't need to allocate any more space, but we
1743 * still extended the size of the file so we need to
1744 * update i_size.
1745 */
1746 inode->i_ctime = CURRENT_TIME;
1747 i_size_write(inode, actual_end);
1748 btrfs_ordered_update_i_size(inode, actual_end, NULL);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001749 }
1750 free_extent_map(em);
1751
1752 cur_offset = last_byte;
1753 if (cur_offset >= alloc_end) {
1754 ret = 0;
1755 break;
1756 }
1757 }
1758 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1759 &cached_state, GFP_NOFS);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001760out:
1761 mutex_unlock(&inode->i_mutex);
Chris Masond98456f2012-01-31 20:27:41 -05001762 /* Let go of our reservation. */
1763 btrfs_free_reserved_data_space(inode, len);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001764 return ret;
1765}
1766
Josef Bacikb2675152011-07-18 13:21:36 -04001767static int find_desired_extent(struct inode *inode, loff_t *offset, int origin)
1768{
1769 struct btrfs_root *root = BTRFS_I(inode)->root;
1770 struct extent_map *em;
1771 struct extent_state *cached_state = NULL;
1772 u64 lockstart = *offset;
1773 u64 lockend = i_size_read(inode);
1774 u64 start = *offset;
1775 u64 orig_start = *offset;
1776 u64 len = i_size_read(inode);
1777 u64 last_end = 0;
1778 int ret = 0;
1779
1780 lockend = max_t(u64, root->sectorsize, lockend);
1781 if (lockend <= lockstart)
1782 lockend = lockstart + root->sectorsize;
1783
1784 len = lockend - lockstart + 1;
1785
1786 len = max_t(u64, len, root->sectorsize);
1787 if (inode->i_size == 0)
1788 return -ENXIO;
1789
1790 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001791 &cached_state);
Josef Bacikb2675152011-07-18 13:21:36 -04001792
1793 /*
1794 * Delalloc is such a pain. If we have a hole and we have pending
1795 * delalloc for a portion of the hole we will get back a hole that
1796 * exists for the entire range since it hasn't been actually written
1797 * yet. So to take care of this case we need to look for an extent just
1798 * before the position we want in case there is outstanding delalloc
1799 * going on here.
1800 */
1801 if (origin == SEEK_HOLE && start != 0) {
1802 if (start <= root->sectorsize)
1803 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
1804 root->sectorsize, 0);
1805 else
1806 em = btrfs_get_extent_fiemap(inode, NULL, 0,
1807 start - root->sectorsize,
1808 root->sectorsize, 0);
1809 if (IS_ERR(em)) {
Jeff Liu6af021d2012-02-09 14:25:50 +08001810 ret = PTR_ERR(em);
Josef Bacikb2675152011-07-18 13:21:36 -04001811 goto out;
1812 }
1813 last_end = em->start + em->len;
1814 if (em->block_start == EXTENT_MAP_DELALLOC)
1815 last_end = min_t(u64, last_end, inode->i_size);
1816 free_extent_map(em);
1817 }
1818
1819 while (1) {
1820 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
1821 if (IS_ERR(em)) {
Jeff Liu6af021d2012-02-09 14:25:50 +08001822 ret = PTR_ERR(em);
Josef Bacikb2675152011-07-18 13:21:36 -04001823 break;
1824 }
1825
1826 if (em->block_start == EXTENT_MAP_HOLE) {
1827 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1828 if (last_end <= orig_start) {
1829 free_extent_map(em);
1830 ret = -ENXIO;
1831 break;
1832 }
1833 }
1834
1835 if (origin == SEEK_HOLE) {
1836 *offset = start;
1837 free_extent_map(em);
1838 break;
1839 }
1840 } else {
1841 if (origin == SEEK_DATA) {
1842 if (em->block_start == EXTENT_MAP_DELALLOC) {
1843 if (start >= inode->i_size) {
1844 free_extent_map(em);
1845 ret = -ENXIO;
1846 break;
1847 }
1848 }
1849
1850 *offset = start;
1851 free_extent_map(em);
1852 break;
1853 }
1854 }
1855
1856 start = em->start + em->len;
1857 last_end = em->start + em->len;
1858
1859 if (em->block_start == EXTENT_MAP_DELALLOC)
1860 last_end = min_t(u64, last_end, inode->i_size);
1861
1862 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1863 free_extent_map(em);
1864 ret = -ENXIO;
1865 break;
1866 }
1867 free_extent_map(em);
1868 cond_resched();
1869 }
1870 if (!ret)
1871 *offset = min(*offset, inode->i_size);
1872out:
1873 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1874 &cached_state, GFP_NOFS);
1875 return ret;
1876}
1877
1878static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin)
1879{
1880 struct inode *inode = file->f_mapping->host;
1881 int ret;
1882
1883 mutex_lock(&inode->i_mutex);
1884 switch (origin) {
1885 case SEEK_END:
1886 case SEEK_CUR:
Andi Kleenef3d0fd2011-09-15 16:06:48 -07001887 offset = generic_file_llseek(file, offset, origin);
Josef Bacikb2675152011-07-18 13:21:36 -04001888 goto out;
1889 case SEEK_DATA:
1890 case SEEK_HOLE:
Jeff Liu48802c82011-09-18 10:34:02 -04001891 if (offset >= i_size_read(inode)) {
1892 mutex_unlock(&inode->i_mutex);
1893 return -ENXIO;
1894 }
1895
Josef Bacikb2675152011-07-18 13:21:36 -04001896 ret = find_desired_extent(inode, &offset, origin);
1897 if (ret) {
1898 mutex_unlock(&inode->i_mutex);
1899 return ret;
1900 }
1901 }
1902
Dan Carpenter9a4327c2011-08-18 10:16:05 -04001903 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
Jeff Liu48802c82011-09-18 10:34:02 -04001904 offset = -EINVAL;
Dan Carpenter9a4327c2011-08-18 10:16:05 -04001905 goto out;
1906 }
1907 if (offset > inode->i_sb->s_maxbytes) {
Jeff Liu48802c82011-09-18 10:34:02 -04001908 offset = -EINVAL;
Dan Carpenter9a4327c2011-08-18 10:16:05 -04001909 goto out;
1910 }
Josef Bacikb2675152011-07-18 13:21:36 -04001911
1912 /* Special lock needed here? */
1913 if (offset != file->f_pos) {
1914 file->f_pos = offset;
1915 file->f_version = 0;
1916 }
1917out:
1918 mutex_unlock(&inode->i_mutex);
1919 return offset;
1920}
1921
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001922const struct file_operations btrfs_file_operations = {
Josef Bacikb2675152011-07-18 13:21:36 -04001923 .llseek = btrfs_file_llseek,
Chris Mason39279cc2007-06-12 06:35:45 -04001924 .read = do_sync_read,
Miao Xie4a0010712010-06-07 03:38:51 +00001925 .write = do_sync_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001926 .aio_read = generic_file_aio_read,
Chris Masone9906a92007-12-14 12:56:58 -05001927 .splice_read = generic_file_splice_read,
Josef Bacik11c65dc2010-05-23 11:07:21 -04001928 .aio_write = btrfs_file_aio_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001929 .mmap = btrfs_file_mmap,
Chris Mason39279cc2007-06-12 06:35:45 -04001930 .open = generic_file_open,
Mingminge1b81e62008-05-27 10:55:43 -04001931 .release = btrfs_release_file,
Chris Mason39279cc2007-06-12 06:35:45 -04001932 .fsync = btrfs_sync_file,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001933 .fallocate = btrfs_fallocate,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001934 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001935#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001936 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001937#endif
1938};