]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
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 | ||
39279cc3 CM |
19 | #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> | |
39279cc3 CM |
25 | #include <linux/backing-dev.h> |
26 | #include <linux/mpage.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/writeback.h> | |
29 | #include <linux/statfs.h> | |
30 | #include <linux/compat.h> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
39279cc3 CM |
32 | #include "ctree.h" |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "btrfs_inode.h" | |
36 | #include "ioctl.h" | |
37 | #include "print-tree.h" | |
e02119d5 CM |
38 | #include "tree-log.h" |
39 | #include "locking.h" | |
12fa8ec6 | 40 | #include "compat.h" |
39279cc3 CM |
41 | |
42 | ||
d352ac68 CM |
43 | /* simple helper to fault in pages and copy. This should go away |
44 | * and be replaced with calls into generic code. | |
45 | */ | |
d397712b | 46 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
a1b32a59 CM |
47 | int write_bytes, |
48 | struct page **prepared_pages, | |
11c65dcc | 49 | struct iov_iter *i) |
39279cc3 | 50 | { |
914ee295 | 51 | size_t copied = 0; |
11c65dcc | 52 | int pg = 0; |
39279cc3 | 53 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
914ee295 | 54 | int total_copied = 0; |
39279cc3 | 55 | |
11c65dcc | 56 | while (write_bytes > 0) { |
39279cc3 CM |
57 | size_t count = min_t(size_t, |
58 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 59 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
60 | /* |
61 | * Copy data from userspace to the current page | |
62 | * | |
63 | * Disable pagefault to avoid recursive lock since | |
64 | * the pages are already locked | |
65 | */ | |
66 | pagefault_disable(); | |
67 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
68 | pagefault_enable(); | |
11c65dcc | 69 | |
39279cc3 CM |
70 | /* Flush processor's dcache for this page */ |
71 | flush_dcache_page(page); | |
11c65dcc JB |
72 | iov_iter_advance(i, copied); |
73 | write_bytes -= copied; | |
914ee295 | 74 | total_copied += copied; |
39279cc3 | 75 | |
914ee295 | 76 | /* Return to btrfs_file_aio_write to fault page */ |
11c65dcc | 77 | if (unlikely(copied == 0)) { |
914ee295 | 78 | break; |
11c65dcc JB |
79 | } |
80 | ||
81 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
82 | offset += copied; | |
83 | } else { | |
84 | pg++; | |
85 | offset = 0; | |
86 | } | |
39279cc3 | 87 | } |
914ee295 | 88 | return total_copied; |
39279cc3 CM |
89 | } |
90 | ||
d352ac68 CM |
91 | /* |
92 | * unlocks pages after btrfs_file_write is done with them | |
93 | */ | |
d397712b | 94 | static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
95 | { |
96 | size_t i; | |
97 | for (i = 0; i < num_pages; i++) { | |
98 | if (!pages[i]) | |
99 | break; | |
d352ac68 CM |
100 | /* page checked is some magic around finding pages that |
101 | * have been modified without going through btrfs_set_page_dirty | |
102 | * clear it here | |
103 | */ | |
4a096752 | 104 | ClearPageChecked(pages[i]); |
39279cc3 CM |
105 | unlock_page(pages[i]); |
106 | mark_page_accessed(pages[i]); | |
107 | page_cache_release(pages[i]); | |
108 | } | |
109 | } | |
110 | ||
d352ac68 CM |
111 | /* |
112 | * after copy_from_user, pages need to be dirtied and we need to make | |
113 | * sure holes are created between the current EOF and the start of | |
114 | * any next extents (if required). | |
115 | * | |
116 | * this also makes the decision about creating an inline extent vs | |
117 | * doing real data extents, marking pages dirty and delalloc as required. | |
118 | */ | |
d397712b | 119 | static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, |
39279cc3 CM |
120 | struct btrfs_root *root, |
121 | struct file *file, | |
122 | struct page **pages, | |
123 | size_t num_pages, | |
124 | loff_t pos, | |
125 | size_t write_bytes) | |
126 | { | |
39279cc3 | 127 | int err = 0; |
a52d9a80 | 128 | int i; |
6da6abae | 129 | struct inode *inode = fdentry(file)->d_inode; |
db94535d | 130 | u64 num_bytes; |
a52d9a80 CM |
131 | u64 start_pos; |
132 | u64 end_of_last_block; | |
133 | u64 end_pos = pos + write_bytes; | |
134 | loff_t isize = i_size_read(inode); | |
39279cc3 | 135 | |
5f39d397 | 136 | start_pos = pos & ~((u64)root->sectorsize - 1); |
db94535d CM |
137 | num_bytes = (write_bytes + pos - start_pos + |
138 | root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
39279cc3 | 139 | |
db94535d | 140 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 JB |
141 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
142 | NULL); | |
a22285a6 | 143 | BUG_ON(err); |
9ed74f2d | 144 | |
c8b97818 CM |
145 | for (i = 0; i < num_pages; i++) { |
146 | struct page *p = pages[i]; | |
147 | SetPageUptodate(p); | |
148 | ClearPageChecked(p); | |
149 | set_page_dirty(p); | |
a52d9a80 CM |
150 | } |
151 | if (end_pos > isize) { | |
152 | i_size_write(inode, end_pos); | |
f597bb19 CM |
153 | /* we've only changed i_size in ram, and we haven't updated |
154 | * the disk i_size. There is no need to log the inode | |
155 | * at this time. | |
156 | */ | |
39279cc3 | 157 | } |
a22285a6 | 158 | return 0; |
39279cc3 CM |
159 | } |
160 | ||
d352ac68 CM |
161 | /* |
162 | * this drops all the extents in the cache that intersect the range | |
163 | * [start, end]. Existing extents are split as required. | |
164 | */ | |
5b21f2ed ZY |
165 | int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
166 | int skip_pinned) | |
a52d9a80 CM |
167 | { |
168 | struct extent_map *em; | |
3b951516 CM |
169 | struct extent_map *split = NULL; |
170 | struct extent_map *split2 = NULL; | |
a52d9a80 | 171 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 172 | u64 len = end - start + 1; |
3b951516 CM |
173 | int ret; |
174 | int testend = 1; | |
5b21f2ed | 175 | unsigned long flags; |
c8b97818 | 176 | int compressed = 0; |
a52d9a80 | 177 | |
e6dcd2dc | 178 | WARN_ON(end < start); |
3b951516 | 179 | if (end == (u64)-1) { |
39b5637f | 180 | len = (u64)-1; |
3b951516 CM |
181 | testend = 0; |
182 | } | |
d397712b | 183 | while (1) { |
3b951516 CM |
184 | if (!split) |
185 | split = alloc_extent_map(GFP_NOFS); | |
186 | if (!split2) | |
187 | split2 = alloc_extent_map(GFP_NOFS); | |
c26a9203 | 188 | BUG_ON(!split || !split2); |
3b951516 | 189 | |
890871be | 190 | write_lock(&em_tree->lock); |
39b5637f | 191 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 192 | if (!em) { |
890871be | 193 | write_unlock(&em_tree->lock); |
a52d9a80 | 194 | break; |
d1310b2e | 195 | } |
5b21f2ed ZY |
196 | flags = em->flags; |
197 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { | |
55ef6899 | 198 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 199 | free_extent_map(em); |
a1ed835e | 200 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
201 | break; |
202 | } | |
55ef6899 YZ |
203 | start = em->start + em->len; |
204 | if (testend) | |
5b21f2ed | 205 | len = start + len - (em->start + em->len); |
5b21f2ed | 206 | free_extent_map(em); |
a1ed835e | 207 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
208 | continue; |
209 | } | |
c8b97818 | 210 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 211 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
a52d9a80 | 212 | remove_extent_mapping(em_tree, em); |
3b951516 CM |
213 | |
214 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
215 | em->start < start) { | |
216 | split->start = em->start; | |
217 | split->len = start - em->start; | |
ff5b7ee3 | 218 | split->orig_start = em->orig_start; |
3b951516 | 219 | split->block_start = em->block_start; |
c8b97818 CM |
220 | |
221 | if (compressed) | |
222 | split->block_len = em->block_len; | |
223 | else | |
224 | split->block_len = split->len; | |
225 | ||
3b951516 | 226 | split->bdev = em->bdev; |
5b21f2ed | 227 | split->flags = flags; |
261507a0 | 228 | split->compress_type = em->compress_type; |
3b951516 CM |
229 | ret = add_extent_mapping(em_tree, split); |
230 | BUG_ON(ret); | |
231 | free_extent_map(split); | |
232 | split = split2; | |
233 | split2 = NULL; | |
234 | } | |
235 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
236 | testend && em->start + em->len > start + len) { | |
237 | u64 diff = start + len - em->start; | |
238 | ||
239 | split->start = start + len; | |
240 | split->len = em->start + em->len - (start + len); | |
241 | split->bdev = em->bdev; | |
5b21f2ed | 242 | split->flags = flags; |
261507a0 | 243 | split->compress_type = em->compress_type; |
3b951516 | 244 | |
c8b97818 CM |
245 | if (compressed) { |
246 | split->block_len = em->block_len; | |
247 | split->block_start = em->block_start; | |
445a6944 | 248 | split->orig_start = em->orig_start; |
c8b97818 CM |
249 | } else { |
250 | split->block_len = split->len; | |
251 | split->block_start = em->block_start + diff; | |
445a6944 | 252 | split->orig_start = split->start; |
c8b97818 | 253 | } |
3b951516 CM |
254 | |
255 | ret = add_extent_mapping(em_tree, split); | |
256 | BUG_ON(ret); | |
257 | free_extent_map(split); | |
258 | split = NULL; | |
259 | } | |
890871be | 260 | write_unlock(&em_tree->lock); |
d1310b2e | 261 | |
a52d9a80 CM |
262 | /* once for us */ |
263 | free_extent_map(em); | |
264 | /* once for the tree*/ | |
265 | free_extent_map(em); | |
266 | } | |
3b951516 CM |
267 | if (split) |
268 | free_extent_map(split); | |
269 | if (split2) | |
270 | free_extent_map(split2); | |
a52d9a80 CM |
271 | return 0; |
272 | } | |
273 | ||
39279cc3 CM |
274 | /* |
275 | * this is very complex, but the basic idea is to drop all extents | |
276 | * in the range start - end. hint_block is filled in with a block number | |
277 | * that would be a good hint to the block allocator for this file. | |
278 | * | |
279 | * If an extent intersects the range but is not entirely inside the range | |
280 | * it is either truncated or split. Anything entirely inside the range | |
281 | * is deleted from the tree. | |
282 | */ | |
920bbbfb YZ |
283 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode, |
284 | u64 start, u64 end, u64 *hint_byte, int drop_cache) | |
39279cc3 | 285 | { |
920bbbfb | 286 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5f39d397 | 287 | struct extent_buffer *leaf; |
920bbbfb | 288 | struct btrfs_file_extent_item *fi; |
39279cc3 | 289 | struct btrfs_path *path; |
00f5c795 | 290 | struct btrfs_key key; |
920bbbfb YZ |
291 | struct btrfs_key new_key; |
292 | u64 search_start = start; | |
293 | u64 disk_bytenr = 0; | |
294 | u64 num_bytes = 0; | |
295 | u64 extent_offset = 0; | |
296 | u64 extent_end = 0; | |
297 | int del_nr = 0; | |
298 | int del_slot = 0; | |
299 | int extent_type; | |
ccd467d6 | 300 | int recow; |
00f5c795 | 301 | int ret; |
39279cc3 | 302 | |
a1ed835e CM |
303 | if (drop_cache) |
304 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 305 | |
39279cc3 CM |
306 | path = btrfs_alloc_path(); |
307 | if (!path) | |
308 | return -ENOMEM; | |
920bbbfb | 309 | |
d397712b | 310 | while (1) { |
ccd467d6 | 311 | recow = 0; |
39279cc3 CM |
312 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, |
313 | search_start, -1); | |
314 | if (ret < 0) | |
920bbbfb YZ |
315 | break; |
316 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
317 | leaf = path->nodes[0]; | |
318 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
319 | if (key.objectid == inode->i_ino && | |
320 | key.type == BTRFS_EXTENT_DATA_KEY) | |
321 | path->slots[0]--; | |
39279cc3 | 322 | } |
920bbbfb | 323 | ret = 0; |
8c2383c3 | 324 | next_slot: |
5f39d397 | 325 | leaf = path->nodes[0]; |
920bbbfb YZ |
326 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
327 | BUG_ON(del_nr > 0); | |
328 | ret = btrfs_next_leaf(root, path); | |
329 | if (ret < 0) | |
330 | break; | |
331 | if (ret > 0) { | |
332 | ret = 0; | |
333 | break; | |
8c2383c3 | 334 | } |
920bbbfb YZ |
335 | leaf = path->nodes[0]; |
336 | recow = 1; | |
337 | } | |
338 | ||
339 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
340 | if (key.objectid > inode->i_ino || | |
341 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
342 | break; | |
343 | ||
344 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
345 | struct btrfs_file_extent_item); | |
346 | extent_type = btrfs_file_extent_type(leaf, fi); | |
347 | ||
348 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
349 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
350 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
351 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
352 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
353 | extent_end = key.offset + | |
354 | btrfs_file_extent_num_bytes(leaf, fi); | |
355 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
356 | extent_end = key.offset + | |
357 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 358 | } else { |
920bbbfb | 359 | WARN_ON(1); |
8c2383c3 | 360 | extent_end = search_start; |
39279cc3 CM |
361 | } |
362 | ||
920bbbfb YZ |
363 | if (extent_end <= search_start) { |
364 | path->slots[0]++; | |
8c2383c3 | 365 | goto next_slot; |
39279cc3 CM |
366 | } |
367 | ||
920bbbfb YZ |
368 | search_start = max(key.offset, start); |
369 | if (recow) { | |
370 | btrfs_release_path(root, path); | |
371 | continue; | |
39279cc3 | 372 | } |
6643558d | 373 | |
920bbbfb YZ |
374 | /* |
375 | * | - range to drop - | | |
376 | * | -------- extent -------- | | |
377 | */ | |
378 | if (start > key.offset && end < extent_end) { | |
379 | BUG_ON(del_nr > 0); | |
380 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
381 | ||
382 | memcpy(&new_key, &key, sizeof(new_key)); | |
383 | new_key.offset = start; | |
384 | ret = btrfs_duplicate_item(trans, root, path, | |
385 | &new_key); | |
386 | if (ret == -EAGAIN) { | |
387 | btrfs_release_path(root, path); | |
388 | continue; | |
6643558d | 389 | } |
920bbbfb YZ |
390 | if (ret < 0) |
391 | break; | |
392 | ||
393 | leaf = path->nodes[0]; | |
394 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
395 | struct btrfs_file_extent_item); | |
396 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
397 | start - key.offset); | |
398 | ||
399 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
400 | struct btrfs_file_extent_item); | |
401 | ||
402 | extent_offset += start - key.offset; | |
403 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
404 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
405 | extent_end - start); | |
406 | btrfs_mark_buffer_dirty(leaf); | |
407 | ||
408 | if (disk_bytenr > 0) { | |
771ed689 | 409 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
410 | disk_bytenr, num_bytes, 0, |
411 | root->root_key.objectid, | |
412 | new_key.objectid, | |
413 | start - extent_offset); | |
771ed689 | 414 | BUG_ON(ret); |
920bbbfb | 415 | *hint_byte = disk_bytenr; |
771ed689 | 416 | } |
920bbbfb | 417 | key.offset = start; |
6643558d | 418 | } |
920bbbfb YZ |
419 | /* |
420 | * | ---- range to drop ----- | | |
421 | * | -------- extent -------- | | |
422 | */ | |
423 | if (start <= key.offset && end < extent_end) { | |
424 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 425 | |
920bbbfb YZ |
426 | memcpy(&new_key, &key, sizeof(new_key)); |
427 | new_key.offset = end; | |
428 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
6643558d | 429 | |
920bbbfb YZ |
430 | extent_offset += end - key.offset; |
431 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
432 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
433 | extent_end - end); | |
434 | btrfs_mark_buffer_dirty(leaf); | |
435 | if (disk_bytenr > 0) { | |
436 | inode_sub_bytes(inode, end - key.offset); | |
437 | *hint_byte = disk_bytenr; | |
39279cc3 | 438 | } |
920bbbfb | 439 | break; |
39279cc3 | 440 | } |
771ed689 | 441 | |
920bbbfb YZ |
442 | search_start = extent_end; |
443 | /* | |
444 | * | ---- range to drop ----- | | |
445 | * | -------- extent -------- | | |
446 | */ | |
447 | if (start > key.offset && end >= extent_end) { | |
448 | BUG_ON(del_nr > 0); | |
449 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 450 | |
920bbbfb YZ |
451 | btrfs_set_file_extent_num_bytes(leaf, fi, |
452 | start - key.offset); | |
453 | btrfs_mark_buffer_dirty(leaf); | |
454 | if (disk_bytenr > 0) { | |
455 | inode_sub_bytes(inode, extent_end - start); | |
456 | *hint_byte = disk_bytenr; | |
457 | } | |
458 | if (end == extent_end) | |
459 | break; | |
c8b97818 | 460 | |
920bbbfb YZ |
461 | path->slots[0]++; |
462 | goto next_slot; | |
31840ae1 ZY |
463 | } |
464 | ||
920bbbfb YZ |
465 | /* |
466 | * | ---- range to drop ----- | | |
467 | * | ------ extent ------ | | |
468 | */ | |
469 | if (start <= key.offset && end >= extent_end) { | |
470 | if (del_nr == 0) { | |
471 | del_slot = path->slots[0]; | |
472 | del_nr = 1; | |
473 | } else { | |
474 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
475 | del_nr++; | |
476 | } | |
31840ae1 | 477 | |
920bbbfb | 478 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
a76a3cd4 | 479 | inode_sub_bytes(inode, |
920bbbfb YZ |
480 | extent_end - key.offset); |
481 | extent_end = ALIGN(extent_end, | |
482 | root->sectorsize); | |
483 | } else if (disk_bytenr > 0) { | |
31840ae1 | 484 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
485 | disk_bytenr, num_bytes, 0, |
486 | root->root_key.objectid, | |
5d4f98a2 | 487 | key.objectid, key.offset - |
920bbbfb | 488 | extent_offset); |
31840ae1 | 489 | BUG_ON(ret); |
920bbbfb YZ |
490 | inode_sub_bytes(inode, |
491 | extent_end - key.offset); | |
492 | *hint_byte = disk_bytenr; | |
31840ae1 | 493 | } |
31840ae1 | 494 | |
920bbbfb YZ |
495 | if (end == extent_end) |
496 | break; | |
497 | ||
498 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
499 | path->slots[0]++; | |
500 | goto next_slot; | |
501 | } | |
502 | ||
503 | ret = btrfs_del_items(trans, root, path, del_slot, | |
504 | del_nr); | |
505 | BUG_ON(ret); | |
506 | ||
507 | del_nr = 0; | |
508 | del_slot = 0; | |
509 | ||
510 | btrfs_release_path(root, path); | |
511 | continue; | |
39279cc3 | 512 | } |
920bbbfb YZ |
513 | |
514 | BUG_ON(1); | |
39279cc3 | 515 | } |
920bbbfb YZ |
516 | |
517 | if (del_nr > 0) { | |
518 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); | |
519 | BUG_ON(ret); | |
6643558d | 520 | } |
920bbbfb YZ |
521 | |
522 | btrfs_free_path(path); | |
39279cc3 CM |
523 | return ret; |
524 | } | |
525 | ||
d899e052 | 526 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
527 | u64 objectid, u64 bytenr, u64 orig_offset, |
528 | u64 *start, u64 *end) | |
d899e052 YZ |
529 | { |
530 | struct btrfs_file_extent_item *fi; | |
531 | struct btrfs_key key; | |
532 | u64 extent_end; | |
533 | ||
534 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
535 | return 0; | |
536 | ||
537 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
538 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
539 | return 0; | |
540 | ||
541 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
542 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
543 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 544 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
545 | btrfs_file_extent_compression(leaf, fi) || |
546 | btrfs_file_extent_encryption(leaf, fi) || | |
547 | btrfs_file_extent_other_encoding(leaf, fi)) | |
548 | return 0; | |
549 | ||
550 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
551 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
552 | return 0; | |
553 | ||
554 | *start = key.offset; | |
555 | *end = extent_end; | |
556 | return 1; | |
557 | } | |
558 | ||
559 | /* | |
560 | * Mark extent in the range start - end as written. | |
561 | * | |
562 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
563 | * part of extent is marked as written, the extent will be split into | |
564 | * two or three. | |
565 | */ | |
566 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
567 | struct inode *inode, u64 start, u64 end) |
568 | { | |
920bbbfb | 569 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
570 | struct extent_buffer *leaf; |
571 | struct btrfs_path *path; | |
572 | struct btrfs_file_extent_item *fi; | |
573 | struct btrfs_key key; | |
920bbbfb | 574 | struct btrfs_key new_key; |
d899e052 YZ |
575 | u64 bytenr; |
576 | u64 num_bytes; | |
577 | u64 extent_end; | |
5d4f98a2 | 578 | u64 orig_offset; |
d899e052 YZ |
579 | u64 other_start; |
580 | u64 other_end; | |
920bbbfb YZ |
581 | u64 split; |
582 | int del_nr = 0; | |
583 | int del_slot = 0; | |
6c7d54ac | 584 | int recow; |
d899e052 YZ |
585 | int ret; |
586 | ||
587 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
588 | ||
589 | path = btrfs_alloc_path(); | |
590 | BUG_ON(!path); | |
591 | again: | |
6c7d54ac | 592 | recow = 0; |
920bbbfb | 593 | split = start; |
d899e052 YZ |
594 | key.objectid = inode->i_ino; |
595 | key.type = BTRFS_EXTENT_DATA_KEY; | |
920bbbfb | 596 | key.offset = split; |
d899e052 YZ |
597 | |
598 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
599 | if (ret > 0 && path->slots[0] > 0) | |
600 | path->slots[0]--; | |
601 | ||
602 | leaf = path->nodes[0]; | |
603 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
604 | BUG_ON(key.objectid != inode->i_ino || | |
605 | key.type != BTRFS_EXTENT_DATA_KEY); | |
606 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
607 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
608 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
609 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
610 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
611 | BUG_ON(key.offset > start || extent_end < end); | |
612 | ||
613 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
614 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 615 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
616 | memcpy(&new_key, &key, sizeof(new_key)); |
617 | ||
618 | if (start == key.offset && end < extent_end) { | |
619 | other_start = 0; | |
620 | other_end = start; | |
621 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
622 | inode->i_ino, bytenr, orig_offset, | |
623 | &other_start, &other_end)) { | |
624 | new_key.offset = end; | |
625 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
626 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
627 | struct btrfs_file_extent_item); | |
628 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
629 | extent_end - end); | |
630 | btrfs_set_file_extent_offset(leaf, fi, | |
631 | end - orig_offset); | |
632 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
633 | struct btrfs_file_extent_item); | |
634 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
635 | end - other_start); | |
636 | btrfs_mark_buffer_dirty(leaf); | |
637 | goto out; | |
638 | } | |
639 | } | |
640 | ||
641 | if (start > key.offset && end == extent_end) { | |
642 | other_start = end; | |
643 | other_end = 0; | |
644 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
645 | inode->i_ino, bytenr, orig_offset, | |
646 | &other_start, &other_end)) { | |
647 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
648 | struct btrfs_file_extent_item); | |
649 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
650 | start - key.offset); | |
651 | path->slots[0]++; | |
652 | new_key.offset = start; | |
653 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
654 | ||
655 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
656 | struct btrfs_file_extent_item); | |
657 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
658 | other_end - start); | |
659 | btrfs_set_file_extent_offset(leaf, fi, | |
660 | start - orig_offset); | |
661 | btrfs_mark_buffer_dirty(leaf); | |
662 | goto out; | |
663 | } | |
664 | } | |
d899e052 | 665 | |
920bbbfb YZ |
666 | while (start > key.offset || end < extent_end) { |
667 | if (key.offset == start) | |
668 | split = end; | |
669 | ||
920bbbfb YZ |
670 | new_key.offset = split; |
671 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
672 | if (ret == -EAGAIN) { | |
673 | btrfs_release_path(root, path); | |
674 | goto again; | |
d899e052 | 675 | } |
920bbbfb | 676 | BUG_ON(ret < 0); |
d899e052 | 677 | |
920bbbfb YZ |
678 | leaf = path->nodes[0]; |
679 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 680 | struct btrfs_file_extent_item); |
d899e052 | 681 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
682 | split - key.offset); |
683 | ||
684 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
685 | struct btrfs_file_extent_item); | |
686 | ||
687 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); | |
688 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
689 | extent_end - split); | |
d899e052 YZ |
690 | btrfs_mark_buffer_dirty(leaf); |
691 | ||
920bbbfb YZ |
692 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
693 | root->root_key.objectid, | |
694 | inode->i_ino, orig_offset); | |
d899e052 | 695 | BUG_ON(ret); |
d899e052 | 696 | |
920bbbfb YZ |
697 | if (split == start) { |
698 | key.offset = start; | |
699 | } else { | |
700 | BUG_ON(start != key.offset); | |
d899e052 | 701 | path->slots[0]--; |
920bbbfb | 702 | extent_end = end; |
d899e052 | 703 | } |
6c7d54ac | 704 | recow = 1; |
d899e052 YZ |
705 | } |
706 | ||
920bbbfb YZ |
707 | other_start = end; |
708 | other_end = 0; | |
6c7d54ac YZ |
709 | if (extent_mergeable(leaf, path->slots[0] + 1, |
710 | inode->i_ino, bytenr, orig_offset, | |
711 | &other_start, &other_end)) { | |
712 | if (recow) { | |
713 | btrfs_release_path(root, path); | |
714 | goto again; | |
715 | } | |
920bbbfb YZ |
716 | extent_end = other_end; |
717 | del_slot = path->slots[0] + 1; | |
718 | del_nr++; | |
719 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
720 | 0, root->root_key.objectid, | |
721 | inode->i_ino, orig_offset); | |
722 | BUG_ON(ret); | |
d899e052 | 723 | } |
920bbbfb YZ |
724 | other_start = 0; |
725 | other_end = start; | |
6c7d54ac YZ |
726 | if (extent_mergeable(leaf, path->slots[0] - 1, |
727 | inode->i_ino, bytenr, orig_offset, | |
728 | &other_start, &other_end)) { | |
729 | if (recow) { | |
730 | btrfs_release_path(root, path); | |
731 | goto again; | |
732 | } | |
920bbbfb YZ |
733 | key.offset = other_start; |
734 | del_slot = path->slots[0]; | |
735 | del_nr++; | |
736 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
737 | 0, root->root_key.objectid, | |
738 | inode->i_ino, orig_offset); | |
739 | BUG_ON(ret); | |
740 | } | |
741 | if (del_nr == 0) { | |
3f6fae95 SL |
742 | fi = btrfs_item_ptr(leaf, path->slots[0], |
743 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
744 | btrfs_set_file_extent_type(leaf, fi, |
745 | BTRFS_FILE_EXTENT_REG); | |
746 | btrfs_mark_buffer_dirty(leaf); | |
6c7d54ac | 747 | } else { |
3f6fae95 SL |
748 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
749 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
750 | btrfs_set_file_extent_type(leaf, fi, |
751 | BTRFS_FILE_EXTENT_REG); | |
752 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
753 | extent_end - key.offset); | |
754 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 755 | |
6c7d54ac YZ |
756 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
757 | BUG_ON(ret); | |
758 | } | |
920bbbfb | 759 | out: |
d899e052 YZ |
760 | btrfs_free_path(path); |
761 | return 0; | |
762 | } | |
763 | ||
39279cc3 | 764 | /* |
d352ac68 CM |
765 | * this gets pages into the page cache and locks them down, it also properly |
766 | * waits for data=ordered extents to finish before allowing the pages to be | |
767 | * modified. | |
39279cc3 | 768 | */ |
d397712b | 769 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
770 | struct page **pages, size_t num_pages, |
771 | loff_t pos, unsigned long first_index, | |
772 | unsigned long last_index, size_t write_bytes) | |
39279cc3 | 773 | { |
2ac55d41 | 774 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
775 | int i; |
776 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
6da6abae | 777 | struct inode *inode = fdentry(file)->d_inode; |
39279cc3 | 778 | int err = 0; |
8c2383c3 | 779 | u64 start_pos; |
e6dcd2dc | 780 | u64 last_pos; |
8c2383c3 | 781 | |
5f39d397 | 782 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 783 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 784 | |
9036c102 YZ |
785 | if (start_pos > inode->i_size) { |
786 | err = btrfs_cont_expand(inode, start_pos); | |
787 | if (err) | |
788 | return err; | |
789 | } | |
790 | ||
39279cc3 | 791 | memset(pages, 0, num_pages * sizeof(struct page *)); |
e6dcd2dc | 792 | again: |
39279cc3 CM |
793 | for (i = 0; i < num_pages; i++) { |
794 | pages[i] = grab_cache_page(inode->i_mapping, index + i); | |
795 | if (!pages[i]) { | |
7adf5dfb JB |
796 | int c; |
797 | for (c = i - 1; c >= 0; c--) { | |
798 | unlock_page(pages[c]); | |
799 | page_cache_release(pages[c]); | |
800 | } | |
801 | return -ENOMEM; | |
39279cc3 | 802 | } |
ccd467d6 | 803 | wait_on_page_writeback(pages[i]); |
39279cc3 | 804 | } |
0762704b | 805 | if (start_pos < inode->i_size) { |
e6dcd2dc | 806 | struct btrfs_ordered_extent *ordered; |
2ac55d41 JB |
807 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
808 | start_pos, last_pos - 1, 0, &cached_state, | |
809 | GFP_NOFS); | |
d397712b CM |
810 | ordered = btrfs_lookup_first_ordered_extent(inode, |
811 | last_pos - 1); | |
e6dcd2dc CM |
812 | if (ordered && |
813 | ordered->file_offset + ordered->len > start_pos && | |
814 | ordered->file_offset < last_pos) { | |
815 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
816 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
817 | start_pos, last_pos - 1, | |
818 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
819 | for (i = 0; i < num_pages; i++) { |
820 | unlock_page(pages[i]); | |
821 | page_cache_release(pages[i]); | |
822 | } | |
823 | btrfs_wait_ordered_range(inode, start_pos, | |
824 | last_pos - start_pos); | |
825 | goto again; | |
826 | } | |
827 | if (ordered) | |
828 | btrfs_put_ordered_extent(ordered); | |
829 | ||
2ac55d41 | 830 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
32c00aff | 831 | last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 832 | EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, |
0762704b | 833 | GFP_NOFS); |
2ac55d41 JB |
834 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
835 | start_pos, last_pos - 1, &cached_state, | |
836 | GFP_NOFS); | |
0762704b | 837 | } |
e6dcd2dc | 838 | for (i = 0; i < num_pages; i++) { |
f87f057b | 839 | clear_page_dirty_for_io(pages[i]); |
e6dcd2dc CM |
840 | set_page_extent_mapped(pages[i]); |
841 | WARN_ON(!PageLocked(pages[i])); | |
842 | } | |
39279cc3 | 843 | return 0; |
39279cc3 CM |
844 | } |
845 | ||
11c65dcc JB |
846 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
847 | const struct iovec *iov, | |
848 | unsigned long nr_segs, loff_t pos) | |
4b46fce2 | 849 | { |
11c65dcc JB |
850 | struct file *file = iocb->ki_filp; |
851 | struct inode *inode = fdentry(file)->d_inode; | |
852 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
853 | struct page *pinned[2]; | |
854 | struct page **pages = NULL; | |
855 | struct iov_iter i; | |
856 | loff_t *ppos = &iocb->ki_pos; | |
2ff3e9b6 CM |
857 | loff_t start_pos; |
858 | ssize_t num_written = 0; | |
859 | ssize_t err = 0; | |
11c65dcc JB |
860 | size_t count; |
861 | size_t ocount; | |
39279cc3 | 862 | int ret = 0; |
8c2383c3 | 863 | int nrptrs; |
39279cc3 CM |
864 | unsigned long first_index; |
865 | unsigned long last_index; | |
cb843a6f | 866 | int will_write; |
4b46fce2 | 867 | int buffered = 0; |
914ee295 XZ |
868 | int copied = 0; |
869 | int dirty_pages = 0; | |
cb843a6f | 870 | |
6b2f3d1f | 871 | will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) || |
cb843a6f | 872 | (file->f_flags & O_DIRECT)); |
8c2383c3 | 873 | |
39279cc3 CM |
874 | pinned[0] = NULL; |
875 | pinned[1] = NULL; | |
2ff3e9b6 | 876 | |
2ff3e9b6 CM |
877 | start_pos = pos; |
878 | ||
39279cc3 | 879 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
ab93dbec | 880 | |
ab93dbec CM |
881 | mutex_lock(&inode->i_mutex); |
882 | ||
11c65dcc JB |
883 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
884 | if (err) | |
885 | goto out; | |
886 | count = ocount; | |
887 | ||
39279cc3 CM |
888 | current->backing_dev_info = inode->i_mapping->backing_dev_info; |
889 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
890 | if (err) | |
ab93dbec CM |
891 | goto out; |
892 | ||
39279cc3 | 893 | if (count == 0) |
ab93dbec | 894 | goto out; |
2b1f55b0 | 895 | |
0ee0fda0 | 896 | err = file_remove_suid(file); |
39279cc3 | 897 | if (err) |
ab93dbec | 898 | goto out; |
9ed74f2d | 899 | |
acce952b | 900 | /* |
901 | * If BTRFS flips readonly due to some impossible error | |
902 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
903 | * although we have opened a file as writable, we have | |
904 | * to stop this write operation to ensure FS consistency. | |
905 | */ | |
906 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
907 | err = -EROFS; | |
908 | goto out; | |
909 | } | |
910 | ||
39279cc3 | 911 | file_update_time(file); |
4b46fce2 JB |
912 | BTRFS_I(inode)->sequence++; |
913 | ||
914 | if (unlikely(file->f_flags & O_DIRECT)) { | |
11c65dcc JB |
915 | num_written = generic_file_direct_write(iocb, iov, &nr_segs, |
916 | pos, ppos, count, | |
917 | ocount); | |
11c65dcc JB |
918 | /* |
919 | * the generic O_DIRECT will update in-memory i_size after the | |
920 | * DIOs are done. But our endio handlers that update the on | |
921 | * disk i_size never update past the in memory i_size. So we | |
922 | * need one more update here to catch any additions to the | |
923 | * file | |
924 | */ | |
925 | if (inode->i_size != BTRFS_I(inode)->disk_i_size) { | |
926 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
927 | mark_inode_dirty(inode); | |
928 | } | |
929 | ||
930 | if (num_written < 0) { | |
11c65dcc JB |
931 | ret = num_written; |
932 | num_written = 0; | |
4b46fce2 | 933 | goto out; |
11c65dcc JB |
934 | } else if (num_written == count) { |
935 | /* pick up pos changes done by the generic code */ | |
936 | pos = *ppos; | |
937 | goto out; | |
938 | } | |
4b46fce2 JB |
939 | /* |
940 | * We are going to do buffered for the rest of the range, so we | |
941 | * need to make sure to invalidate the buffered pages when we're | |
942 | * done. | |
943 | */ | |
944 | buffered = 1; | |
11c65dcc | 945 | pos += num_written; |
4b46fce2 JB |
946 | } |
947 | ||
11c65dcc JB |
948 | iov_iter_init(&i, iov, nr_segs, count, num_written); |
949 | nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) / | |
950 | PAGE_CACHE_SIZE, PAGE_CACHE_SIZE / | |
951 | (sizeof(struct page *))); | |
8c2383c3 | 952 | pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); |
2a29edc6 | 953 | if (!pages) { |
954 | ret = -ENOMEM; | |
955 | goto out; | |
956 | } | |
39279cc3 | 957 | |
ab93dbec CM |
958 | /* generic_write_checks can change our pos */ |
959 | start_pos = pos; | |
960 | ||
39279cc3 | 961 | first_index = pos >> PAGE_CACHE_SHIFT; |
11c65dcc | 962 | last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT; |
39279cc3 CM |
963 | |
964 | /* | |
965 | * there are lots of better ways to do this, but this code | |
966 | * makes sure the first and last page in the file range are | |
967 | * up to date and ready for cow | |
968 | */ | |
969 | if ((pos & (PAGE_CACHE_SIZE - 1))) { | |
970 | pinned[0] = grab_cache_page(inode->i_mapping, first_index); | |
971 | if (!PageUptodate(pinned[0])) { | |
9ebefb18 | 972 | ret = btrfs_readpage(NULL, pinned[0]); |
39279cc3 CM |
973 | BUG_ON(ret); |
974 | wait_on_page_locked(pinned[0]); | |
975 | } else { | |
976 | unlock_page(pinned[0]); | |
977 | } | |
978 | } | |
11c65dcc | 979 | if ((pos + iov_iter_count(&i)) & (PAGE_CACHE_SIZE - 1)) { |
39279cc3 CM |
980 | pinned[1] = grab_cache_page(inode->i_mapping, last_index); |
981 | if (!PageUptodate(pinned[1])) { | |
9ebefb18 | 982 | ret = btrfs_readpage(NULL, pinned[1]); |
39279cc3 CM |
983 | BUG_ON(ret); |
984 | wait_on_page_locked(pinned[1]); | |
985 | } else { | |
986 | unlock_page(pinned[1]); | |
987 | } | |
988 | } | |
989 | ||
11c65dcc | 990 | while (iov_iter_count(&i) > 0) { |
39279cc3 | 991 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
11c65dcc JB |
992 | size_t write_bytes = min(iov_iter_count(&i), |
993 | nrptrs * (size_t)PAGE_CACHE_SIZE - | |
8c2383c3 | 994 | offset); |
3a90983d YZ |
995 | size_t num_pages = (write_bytes + offset + |
996 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
39279cc3 | 997 | |
8c2383c3 | 998 | WARN_ON(num_pages > nrptrs); |
9aead435 | 999 | memset(pages, 0, sizeof(struct page *) * nrptrs); |
1832a6d5 | 1000 | |
914ee295 XZ |
1001 | /* |
1002 | * Fault pages before locking them in prepare_pages | |
1003 | * to avoid recursive lock | |
1004 | */ | |
1005 | if (unlikely(iov_iter_fault_in_readable(&i, write_bytes))) { | |
1006 | ret = -EFAULT; | |
1007 | goto out; | |
1008 | } | |
1009 | ||
1010 | ret = btrfs_delalloc_reserve_space(inode, | |
1011 | num_pages << PAGE_CACHE_SHIFT); | |
1832a6d5 CM |
1012 | if (ret) |
1013 | goto out; | |
1014 | ||
39279cc3 CM |
1015 | ret = prepare_pages(root, file, pages, num_pages, |
1016 | pos, first_index, last_index, | |
8c2383c3 | 1017 | write_bytes); |
6a63209f | 1018 | if (ret) { |
914ee295 XZ |
1019 | btrfs_delalloc_release_space(inode, |
1020 | num_pages << PAGE_CACHE_SHIFT); | |
54aa1f4d | 1021 | goto out; |
6a63209f | 1022 | } |
39279cc3 | 1023 | |
914ee295 | 1024 | copied = btrfs_copy_from_user(pos, num_pages, |
11c65dcc | 1025 | write_bytes, pages, &i); |
3a90983d YZ |
1026 | dirty_pages = (copied + offset + PAGE_CACHE_SIZE - 1) >> |
1027 | PAGE_CACHE_SHIFT; | |
914ee295 XZ |
1028 | |
1029 | if (num_pages > dirty_pages) { | |
1030 | if (copied > 0) | |
1031 | atomic_inc( | |
1032 | &BTRFS_I(inode)->outstanding_extents); | |
1033 | btrfs_delalloc_release_space(inode, | |
1034 | (num_pages - dirty_pages) << | |
1035 | PAGE_CACHE_SHIFT); | |
1036 | } | |
1037 | ||
1038 | if (copied > 0) { | |
0ca1f7ce | 1039 | dirty_and_release_pages(NULL, root, file, pages, |
914ee295 | 1040 | dirty_pages, pos, copied); |
54aa1f4d | 1041 | } |
39279cc3 | 1042 | |
39279cc3 CM |
1043 | btrfs_drop_pages(pages, num_pages); |
1044 | ||
914ee295 XZ |
1045 | if (copied > 0) { |
1046 | if (will_write) { | |
1047 | filemap_fdatawrite_range(inode->i_mapping, pos, | |
1048 | pos + copied - 1); | |
1049 | } else { | |
1050 | balance_dirty_pages_ratelimited_nr( | |
1051 | inode->i_mapping, | |
1052 | dirty_pages); | |
1053 | if (dirty_pages < | |
1054 | (root->leafsize >> PAGE_CACHE_SHIFT) + 1) | |
1055 | btrfs_btree_balance_dirty(root, 1); | |
1056 | btrfs_throttle(root); | |
1057 | } | |
cb843a6f CM |
1058 | } |
1059 | ||
914ee295 XZ |
1060 | pos += copied; |
1061 | num_written += copied; | |
39279cc3 | 1062 | |
39279cc3 CM |
1063 | cond_resched(); |
1064 | } | |
39279cc3 | 1065 | out: |
1832a6d5 | 1066 | mutex_unlock(&inode->i_mutex); |
6a63209f JB |
1067 | if (ret) |
1068 | err = ret; | |
5b92ee72 | 1069 | |
8c2383c3 | 1070 | kfree(pages); |
39279cc3 CM |
1071 | if (pinned[0]) |
1072 | page_cache_release(pinned[0]); | |
1073 | if (pinned[1]) | |
1074 | page_cache_release(pinned[1]); | |
1075 | *ppos = pos; | |
2ff3e9b6 | 1076 | |
5a3f23d5 CM |
1077 | /* |
1078 | * we want to make sure fsync finds this change | |
1079 | * but we haven't joined a transaction running right now. | |
1080 | * | |
1081 | * Later on, someone is sure to update the inode and get the | |
1082 | * real transid recorded. | |
1083 | * | |
1084 | * We set last_trans now to the fs_info generation + 1, | |
1085 | * this will either be one more than the running transaction | |
1086 | * or the generation used for the next transaction if there isn't | |
1087 | * one running right now. | |
1088 | */ | |
1089 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
1090 | ||
cb843a6f | 1091 | if (num_written > 0 && will_write) { |
e02119d5 CM |
1092 | struct btrfs_trans_handle *trans; |
1093 | ||
cb843a6f CM |
1094 | err = btrfs_wait_ordered_range(inode, start_pos, num_written); |
1095 | if (err) | |
2ff3e9b6 | 1096 | num_written = err; |
e02119d5 | 1097 | |
6b2f3d1f | 1098 | if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) { |
a22285a6 | 1099 | trans = btrfs_start_transaction(root, 0); |
495e8677 JB |
1100 | if (IS_ERR(trans)) { |
1101 | num_written = PTR_ERR(trans); | |
1102 | goto done; | |
1103 | } | |
1104 | mutex_lock(&inode->i_mutex); | |
cb843a6f CM |
1105 | ret = btrfs_log_dentry_safe(trans, root, |
1106 | file->f_dentry); | |
495e8677 | 1107 | mutex_unlock(&inode->i_mutex); |
cb843a6f | 1108 | if (ret == 0) { |
12fcfd22 CM |
1109 | ret = btrfs_sync_log(trans, root); |
1110 | if (ret == 0) | |
1111 | btrfs_end_transaction(trans, root); | |
1112 | else | |
1113 | btrfs_commit_transaction(trans, root); | |
257c62e1 | 1114 | } else if (ret != BTRFS_NO_LOG_SYNC) { |
cb843a6f | 1115 | btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1116 | } else { |
1117 | btrfs_end_transaction(trans, root); | |
cb843a6f CM |
1118 | } |
1119 | } | |
4b46fce2 | 1120 | if (file->f_flags & O_DIRECT && buffered) { |
cb843a6f CM |
1121 | invalidate_mapping_pages(inode->i_mapping, |
1122 | start_pos >> PAGE_CACHE_SHIFT, | |
1123 | (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT); | |
e02119d5 | 1124 | } |
2ff3e9b6 | 1125 | } |
495e8677 | 1126 | done: |
39279cc3 | 1127 | current->backing_dev_info = NULL; |
39279cc3 CM |
1128 | return num_written ? num_written : err; |
1129 | } | |
1130 | ||
d397712b | 1131 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1132 | { |
5a3f23d5 CM |
1133 | /* |
1134 | * ordered_data_close is set by settattr when we are about to truncate | |
1135 | * a file from a non-zero size to a zero size. This tries to | |
1136 | * flush down new bytes that may have been written if the | |
1137 | * application were using truncate to replace a file in place. | |
1138 | */ | |
1139 | if (BTRFS_I(inode)->ordered_data_close) { | |
1140 | BTRFS_I(inode)->ordered_data_close = 0; | |
1141 | btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); | |
1142 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) | |
1143 | filemap_flush(inode->i_mapping); | |
1144 | } | |
6bf13c0c SW |
1145 | if (filp->private_data) |
1146 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1147 | return 0; |
1148 | } | |
1149 | ||
d352ac68 CM |
1150 | /* |
1151 | * fsync call for both files and directories. This logs the inode into | |
1152 | * the tree log instead of forcing full commits whenever possible. | |
1153 | * | |
1154 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1155 | * in the metadata btree are up to date for copying to the log. | |
1156 | * | |
1157 | * It drops the inode mutex before doing the tree log commit. This is an | |
1158 | * important optimization for directories because holding the mutex prevents | |
1159 | * new operations on the dir while we write to disk. | |
1160 | */ | |
7ea80859 | 1161 | int btrfs_sync_file(struct file *file, int datasync) |
39279cc3 | 1162 | { |
7ea80859 | 1163 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1164 | struct inode *inode = dentry->d_inode; |
1165 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1166 | int ret = 0; |
39279cc3 CM |
1167 | struct btrfs_trans_handle *trans; |
1168 | ||
257c62e1 CM |
1169 | |
1170 | /* we wait first, since the writeback may change the inode */ | |
1171 | root->log_batch++; | |
1172 | /* the VFS called filemap_fdatawrite for us */ | |
1173 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
1174 | root->log_batch++; | |
1175 | ||
39279cc3 | 1176 | /* |
15ee9bc7 JB |
1177 | * check the transaction that last modified this inode |
1178 | * and see if its already been committed | |
39279cc3 | 1179 | */ |
15ee9bc7 JB |
1180 | if (!BTRFS_I(inode)->last_trans) |
1181 | goto out; | |
a2135011 | 1182 | |
257c62e1 CM |
1183 | /* |
1184 | * if the last transaction that changed this file was before | |
1185 | * the current transaction, we can bail out now without any | |
1186 | * syncing | |
1187 | */ | |
15ee9bc7 JB |
1188 | mutex_lock(&root->fs_info->trans_mutex); |
1189 | if (BTRFS_I(inode)->last_trans <= | |
1190 | root->fs_info->last_trans_committed) { | |
1191 | BTRFS_I(inode)->last_trans = 0; | |
1192 | mutex_unlock(&root->fs_info->trans_mutex); | |
1193 | goto out; | |
1194 | } | |
1195 | mutex_unlock(&root->fs_info->trans_mutex); | |
1196 | ||
1197 | /* | |
a52d9a80 CM |
1198 | * ok we haven't committed the transaction yet, lets do a commit |
1199 | */ | |
6f902af4 | 1200 | if (file->private_data) |
6bf13c0c SW |
1201 | btrfs_ioctl_trans_end(file); |
1202 | ||
a22285a6 YZ |
1203 | trans = btrfs_start_transaction(root, 0); |
1204 | if (IS_ERR(trans)) { | |
1205 | ret = PTR_ERR(trans); | |
39279cc3 CM |
1206 | goto out; |
1207 | } | |
e02119d5 | 1208 | |
2cfbd50b | 1209 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
d397712b | 1210 | if (ret < 0) |
e02119d5 | 1211 | goto out; |
49eb7e46 CM |
1212 | |
1213 | /* we've logged all the items and now have a consistent | |
1214 | * version of the file in the log. It is possible that | |
1215 | * someone will come in and modify the file, but that's | |
1216 | * fine because the log is consistent on disk, and we | |
1217 | * have references to all of the file's extents | |
1218 | * | |
1219 | * It is possible that someone will come in and log the | |
1220 | * file again, but that will end up using the synchronization | |
1221 | * inside btrfs_sync_log to keep things safe. | |
1222 | */ | |
2cfbd50b | 1223 | mutex_unlock(&dentry->d_inode->i_mutex); |
49eb7e46 | 1224 | |
257c62e1 CM |
1225 | if (ret != BTRFS_NO_LOG_SYNC) { |
1226 | if (ret > 0) { | |
12fcfd22 | 1227 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1228 | } else { |
1229 | ret = btrfs_sync_log(trans, root); | |
1230 | if (ret == 0) | |
1231 | ret = btrfs_end_transaction(trans, root); | |
1232 | else | |
1233 | ret = btrfs_commit_transaction(trans, root); | |
1234 | } | |
1235 | } else { | |
1236 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1237 | } |
2cfbd50b | 1238 | mutex_lock(&dentry->d_inode->i_mutex); |
39279cc3 | 1239 | out: |
014e4ac4 | 1240 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1241 | } |
1242 | ||
f0f37e2f | 1243 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1244 | .fault = filemap_fault, |
9ebefb18 CM |
1245 | .page_mkwrite = btrfs_page_mkwrite, |
1246 | }; | |
1247 | ||
1248 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1249 | { | |
058a457e MX |
1250 | struct address_space *mapping = filp->f_mapping; |
1251 | ||
1252 | if (!mapping->a_ops->readpage) | |
1253 | return -ENOEXEC; | |
1254 | ||
9ebefb18 | 1255 | file_accessed(filp); |
058a457e MX |
1256 | vma->vm_ops = &btrfs_file_vm_ops; |
1257 | vma->vm_flags |= VM_CAN_NONLINEAR; | |
1258 | ||
9ebefb18 CM |
1259 | return 0; |
1260 | } | |
1261 | ||
828c0950 | 1262 | const struct file_operations btrfs_file_operations = { |
39279cc3 CM |
1263 | .llseek = generic_file_llseek, |
1264 | .read = do_sync_read, | |
4a001071 | 1265 | .write = do_sync_write, |
9ebefb18 | 1266 | .aio_read = generic_file_aio_read, |
e9906a98 | 1267 | .splice_read = generic_file_splice_read, |
11c65dcc | 1268 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 1269 | .mmap = btrfs_file_mmap, |
39279cc3 | 1270 | .open = generic_file_open, |
e1b81e67 | 1271 | .release = btrfs_release_file, |
39279cc3 | 1272 | .fsync = btrfs_sync_file, |
34287aa3 | 1273 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 1274 | #ifdef CONFIG_COMPAT |
34287aa3 | 1275 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
1276 | #endif |
1277 | }; |