]>
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> | |
2fe17c10 | 27 | #include <linux/falloc.h> |
39279cc3 CM |
28 | #include <linux/swap.h> |
29 | #include <linux/writeback.h> | |
30 | #include <linux/statfs.h> | |
31 | #include <linux/compat.h> | |
5a0e3ad6 | 32 | #include <linux/slab.h> |
39279cc3 CM |
33 | #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" | |
e02119d5 CM |
39 | #include "tree-log.h" |
40 | #include "locking.h" | |
12fa8ec6 | 41 | #include "compat.h" |
39279cc3 CM |
42 | |
43 | ||
d352ac68 CM |
44 | /* simple helper to fault in pages and copy. This should go away |
45 | * and be replaced with calls into generic code. | |
46 | */ | |
d397712b | 47 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
d0215f3e | 48 | size_t write_bytes, |
a1b32a59 | 49 | struct page **prepared_pages, |
11c65dcc | 50 | struct iov_iter *i) |
39279cc3 | 51 | { |
914ee295 | 52 | size_t copied = 0; |
d0215f3e | 53 | size_t total_copied = 0; |
11c65dcc | 54 | int pg = 0; |
39279cc3 CM |
55 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
56 | ||
11c65dcc | 57 | while (write_bytes > 0) { |
39279cc3 CM |
58 | size_t count = min_t(size_t, |
59 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 60 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
61 | /* |
62 | * Copy data from userspace to the current page | |
63 | * | |
64 | * Disable pagefault to avoid recursive lock since | |
65 | * the pages are already locked | |
66 | */ | |
67 | pagefault_disable(); | |
68 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
69 | pagefault_enable(); | |
11c65dcc | 70 | |
39279cc3 CM |
71 | /* Flush processor's dcache for this page */ |
72 | flush_dcache_page(page); | |
31339acd CM |
73 | |
74 | /* | |
75 | * if we get a partial write, we can end up with | |
76 | * partially up to date pages. These add | |
77 | * a lot of complexity, so make sure they don't | |
78 | * happen by forcing this copy to be retried. | |
79 | * | |
80 | * The rest of the btrfs_file_write code will fall | |
81 | * back to page at a time copies after we return 0. | |
82 | */ | |
83 | if (!PageUptodate(page) && copied < count) | |
84 | copied = 0; | |
85 | ||
11c65dcc JB |
86 | iov_iter_advance(i, copied); |
87 | write_bytes -= copied; | |
914ee295 | 88 | total_copied += copied; |
39279cc3 | 89 | |
914ee295 | 90 | /* Return to btrfs_file_aio_write to fault page */ |
9f570b8d | 91 | if (unlikely(copied == 0)) |
914ee295 | 92 | break; |
11c65dcc JB |
93 | |
94 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
95 | offset += copied; | |
96 | } else { | |
97 | pg++; | |
98 | offset = 0; | |
99 | } | |
39279cc3 | 100 | } |
914ee295 | 101 | return total_copied; |
39279cc3 CM |
102 | } |
103 | ||
d352ac68 CM |
104 | /* |
105 | * unlocks pages after btrfs_file_write is done with them | |
106 | */ | |
d397712b | 107 | static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
108 | { |
109 | size_t i; | |
110 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
111 | /* page checked is some magic around finding pages that |
112 | * have been modified without going through btrfs_set_page_dirty | |
113 | * clear it here | |
114 | */ | |
4a096752 | 115 | ClearPageChecked(pages[i]); |
39279cc3 CM |
116 | unlock_page(pages[i]); |
117 | mark_page_accessed(pages[i]); | |
118 | page_cache_release(pages[i]); | |
119 | } | |
120 | } | |
121 | ||
d352ac68 CM |
122 | /* |
123 | * after copy_from_user, pages need to be dirtied and we need to make | |
124 | * sure holes are created between the current EOF and the start of | |
125 | * any next extents (if required). | |
126 | * | |
127 | * this also makes the decision about creating an inline extent vs | |
128 | * doing real data extents, marking pages dirty and delalloc as required. | |
129 | */ | |
d0215f3e JB |
130 | static noinline int dirty_and_release_pages(struct btrfs_root *root, |
131 | struct file *file, | |
132 | struct page **pages, | |
133 | size_t num_pages, | |
134 | loff_t pos, | |
135 | size_t write_bytes) | |
39279cc3 | 136 | { |
39279cc3 | 137 | int err = 0; |
a52d9a80 | 138 | int i; |
6da6abae | 139 | struct inode *inode = fdentry(file)->d_inode; |
db94535d | 140 | u64 num_bytes; |
a52d9a80 CM |
141 | u64 start_pos; |
142 | u64 end_of_last_block; | |
143 | u64 end_pos = pos + write_bytes; | |
144 | loff_t isize = i_size_read(inode); | |
39279cc3 | 145 | |
5f39d397 | 146 | start_pos = pos & ~((u64)root->sectorsize - 1); |
db94535d CM |
147 | num_bytes = (write_bytes + pos - start_pos + |
148 | root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
39279cc3 | 149 | |
db94535d | 150 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 JB |
151 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
152 | NULL); | |
d0215f3e JB |
153 | if (err) |
154 | return err; | |
9ed74f2d | 155 | |
c8b97818 CM |
156 | for (i = 0; i < num_pages; i++) { |
157 | struct page *p = pages[i]; | |
158 | SetPageUptodate(p); | |
159 | ClearPageChecked(p); | |
160 | set_page_dirty(p); | |
a52d9a80 | 161 | } |
9f570b8d JB |
162 | |
163 | /* | |
164 | * we've only changed i_size in ram, and we haven't updated | |
165 | * the disk i_size. There is no need to log the inode | |
166 | * at this time. | |
167 | */ | |
168 | if (end_pos > isize) | |
a52d9a80 | 169 | i_size_write(inode, end_pos); |
a22285a6 | 170 | return 0; |
39279cc3 CM |
171 | } |
172 | ||
d352ac68 CM |
173 | /* |
174 | * this drops all the extents in the cache that intersect the range | |
175 | * [start, end]. Existing extents are split as required. | |
176 | */ | |
5b21f2ed ZY |
177 | int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
178 | int skip_pinned) | |
a52d9a80 CM |
179 | { |
180 | struct extent_map *em; | |
3b951516 CM |
181 | struct extent_map *split = NULL; |
182 | struct extent_map *split2 = NULL; | |
a52d9a80 | 183 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 184 | u64 len = end - start + 1; |
3b951516 CM |
185 | int ret; |
186 | int testend = 1; | |
5b21f2ed | 187 | unsigned long flags; |
c8b97818 | 188 | int compressed = 0; |
a52d9a80 | 189 | |
e6dcd2dc | 190 | WARN_ON(end < start); |
3b951516 | 191 | if (end == (u64)-1) { |
39b5637f | 192 | len = (u64)-1; |
3b951516 CM |
193 | testend = 0; |
194 | } | |
d397712b | 195 | while (1) { |
3b951516 CM |
196 | if (!split) |
197 | split = alloc_extent_map(GFP_NOFS); | |
198 | if (!split2) | |
199 | split2 = alloc_extent_map(GFP_NOFS); | |
c26a9203 | 200 | BUG_ON(!split || !split2); |
3b951516 | 201 | |
890871be | 202 | write_lock(&em_tree->lock); |
39b5637f | 203 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 204 | if (!em) { |
890871be | 205 | write_unlock(&em_tree->lock); |
a52d9a80 | 206 | break; |
d1310b2e | 207 | } |
5b21f2ed ZY |
208 | flags = em->flags; |
209 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { | |
55ef6899 | 210 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 211 | free_extent_map(em); |
a1ed835e | 212 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
213 | break; |
214 | } | |
55ef6899 YZ |
215 | start = em->start + em->len; |
216 | if (testend) | |
5b21f2ed | 217 | len = start + len - (em->start + em->len); |
5b21f2ed | 218 | free_extent_map(em); |
a1ed835e | 219 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
220 | continue; |
221 | } | |
c8b97818 | 222 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 223 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
a52d9a80 | 224 | remove_extent_mapping(em_tree, em); |
3b951516 CM |
225 | |
226 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
227 | em->start < start) { | |
228 | split->start = em->start; | |
229 | split->len = start - em->start; | |
ff5b7ee3 | 230 | split->orig_start = em->orig_start; |
3b951516 | 231 | split->block_start = em->block_start; |
c8b97818 CM |
232 | |
233 | if (compressed) | |
234 | split->block_len = em->block_len; | |
235 | else | |
236 | split->block_len = split->len; | |
237 | ||
3b951516 | 238 | split->bdev = em->bdev; |
5b21f2ed | 239 | split->flags = flags; |
261507a0 | 240 | split->compress_type = em->compress_type; |
3b951516 CM |
241 | ret = add_extent_mapping(em_tree, split); |
242 | BUG_ON(ret); | |
243 | free_extent_map(split); | |
244 | split = split2; | |
245 | split2 = NULL; | |
246 | } | |
247 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
248 | testend && em->start + em->len > start + len) { | |
249 | u64 diff = start + len - em->start; | |
250 | ||
251 | split->start = start + len; | |
252 | split->len = em->start + em->len - (start + len); | |
253 | split->bdev = em->bdev; | |
5b21f2ed | 254 | split->flags = flags; |
261507a0 | 255 | split->compress_type = em->compress_type; |
3b951516 | 256 | |
c8b97818 CM |
257 | if (compressed) { |
258 | split->block_len = em->block_len; | |
259 | split->block_start = em->block_start; | |
445a6944 | 260 | split->orig_start = em->orig_start; |
c8b97818 CM |
261 | } else { |
262 | split->block_len = split->len; | |
263 | split->block_start = em->block_start + diff; | |
445a6944 | 264 | split->orig_start = split->start; |
c8b97818 | 265 | } |
3b951516 CM |
266 | |
267 | ret = add_extent_mapping(em_tree, split); | |
268 | BUG_ON(ret); | |
269 | free_extent_map(split); | |
270 | split = NULL; | |
271 | } | |
890871be | 272 | write_unlock(&em_tree->lock); |
d1310b2e | 273 | |
a52d9a80 CM |
274 | /* once for us */ |
275 | free_extent_map(em); | |
276 | /* once for the tree*/ | |
277 | free_extent_map(em); | |
278 | } | |
3b951516 CM |
279 | if (split) |
280 | free_extent_map(split); | |
281 | if (split2) | |
282 | free_extent_map(split2); | |
a52d9a80 CM |
283 | return 0; |
284 | } | |
285 | ||
39279cc3 CM |
286 | /* |
287 | * this is very complex, but the basic idea is to drop all extents | |
288 | * in the range start - end. hint_block is filled in with a block number | |
289 | * that would be a good hint to the block allocator for this file. | |
290 | * | |
291 | * If an extent intersects the range but is not entirely inside the range | |
292 | * it is either truncated or split. Anything entirely inside the range | |
293 | * is deleted from the tree. | |
294 | */ | |
920bbbfb YZ |
295 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode, |
296 | u64 start, u64 end, u64 *hint_byte, int drop_cache) | |
39279cc3 | 297 | { |
920bbbfb | 298 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5f39d397 | 299 | struct extent_buffer *leaf; |
920bbbfb | 300 | struct btrfs_file_extent_item *fi; |
39279cc3 | 301 | struct btrfs_path *path; |
00f5c795 | 302 | struct btrfs_key key; |
920bbbfb YZ |
303 | struct btrfs_key new_key; |
304 | u64 search_start = start; | |
305 | u64 disk_bytenr = 0; | |
306 | u64 num_bytes = 0; | |
307 | u64 extent_offset = 0; | |
308 | u64 extent_end = 0; | |
309 | int del_nr = 0; | |
310 | int del_slot = 0; | |
311 | int extent_type; | |
ccd467d6 | 312 | int recow; |
00f5c795 | 313 | int ret; |
39279cc3 | 314 | |
a1ed835e CM |
315 | if (drop_cache) |
316 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 317 | |
39279cc3 CM |
318 | path = btrfs_alloc_path(); |
319 | if (!path) | |
320 | return -ENOMEM; | |
920bbbfb | 321 | |
d397712b | 322 | while (1) { |
ccd467d6 | 323 | recow = 0; |
39279cc3 CM |
324 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, |
325 | search_start, -1); | |
326 | if (ret < 0) | |
920bbbfb YZ |
327 | break; |
328 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
329 | leaf = path->nodes[0]; | |
330 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
331 | if (key.objectid == inode->i_ino && | |
332 | key.type == BTRFS_EXTENT_DATA_KEY) | |
333 | path->slots[0]--; | |
39279cc3 | 334 | } |
920bbbfb | 335 | ret = 0; |
8c2383c3 | 336 | next_slot: |
5f39d397 | 337 | leaf = path->nodes[0]; |
920bbbfb YZ |
338 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
339 | BUG_ON(del_nr > 0); | |
340 | ret = btrfs_next_leaf(root, path); | |
341 | if (ret < 0) | |
342 | break; | |
343 | if (ret > 0) { | |
344 | ret = 0; | |
345 | break; | |
8c2383c3 | 346 | } |
920bbbfb YZ |
347 | leaf = path->nodes[0]; |
348 | recow = 1; | |
349 | } | |
350 | ||
351 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
352 | if (key.objectid > inode->i_ino || | |
353 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
354 | break; | |
355 | ||
356 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
357 | struct btrfs_file_extent_item); | |
358 | extent_type = btrfs_file_extent_type(leaf, fi); | |
359 | ||
360 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
361 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
362 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
363 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
364 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
365 | extent_end = key.offset + | |
366 | btrfs_file_extent_num_bytes(leaf, fi); | |
367 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
368 | extent_end = key.offset + | |
369 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 370 | } else { |
920bbbfb | 371 | WARN_ON(1); |
8c2383c3 | 372 | extent_end = search_start; |
39279cc3 CM |
373 | } |
374 | ||
920bbbfb YZ |
375 | if (extent_end <= search_start) { |
376 | path->slots[0]++; | |
8c2383c3 | 377 | goto next_slot; |
39279cc3 CM |
378 | } |
379 | ||
920bbbfb YZ |
380 | search_start = max(key.offset, start); |
381 | if (recow) { | |
382 | btrfs_release_path(root, path); | |
383 | continue; | |
39279cc3 | 384 | } |
6643558d | 385 | |
920bbbfb YZ |
386 | /* |
387 | * | - range to drop - | | |
388 | * | -------- extent -------- | | |
389 | */ | |
390 | if (start > key.offset && end < extent_end) { | |
391 | BUG_ON(del_nr > 0); | |
392 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
393 | ||
394 | memcpy(&new_key, &key, sizeof(new_key)); | |
395 | new_key.offset = start; | |
396 | ret = btrfs_duplicate_item(trans, root, path, | |
397 | &new_key); | |
398 | if (ret == -EAGAIN) { | |
399 | btrfs_release_path(root, path); | |
400 | continue; | |
6643558d | 401 | } |
920bbbfb YZ |
402 | if (ret < 0) |
403 | break; | |
404 | ||
405 | leaf = path->nodes[0]; | |
406 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
407 | struct btrfs_file_extent_item); | |
408 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
409 | start - key.offset); | |
410 | ||
411 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
412 | struct btrfs_file_extent_item); | |
413 | ||
414 | extent_offset += start - key.offset; | |
415 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
416 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
417 | extent_end - start); | |
418 | btrfs_mark_buffer_dirty(leaf); | |
419 | ||
420 | if (disk_bytenr > 0) { | |
771ed689 | 421 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
422 | disk_bytenr, num_bytes, 0, |
423 | root->root_key.objectid, | |
424 | new_key.objectid, | |
425 | start - extent_offset); | |
771ed689 | 426 | BUG_ON(ret); |
920bbbfb | 427 | *hint_byte = disk_bytenr; |
771ed689 | 428 | } |
920bbbfb | 429 | key.offset = start; |
6643558d | 430 | } |
920bbbfb YZ |
431 | /* |
432 | * | ---- range to drop ----- | | |
433 | * | -------- extent -------- | | |
434 | */ | |
435 | if (start <= key.offset && end < extent_end) { | |
436 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 437 | |
920bbbfb YZ |
438 | memcpy(&new_key, &key, sizeof(new_key)); |
439 | new_key.offset = end; | |
440 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
6643558d | 441 | |
920bbbfb YZ |
442 | extent_offset += end - key.offset; |
443 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
444 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
445 | extent_end - end); | |
446 | btrfs_mark_buffer_dirty(leaf); | |
447 | if (disk_bytenr > 0) { | |
448 | inode_sub_bytes(inode, end - key.offset); | |
449 | *hint_byte = disk_bytenr; | |
39279cc3 | 450 | } |
920bbbfb | 451 | break; |
39279cc3 | 452 | } |
771ed689 | 453 | |
920bbbfb YZ |
454 | search_start = extent_end; |
455 | /* | |
456 | * | ---- range to drop ----- | | |
457 | * | -------- extent -------- | | |
458 | */ | |
459 | if (start > key.offset && end >= extent_end) { | |
460 | BUG_ON(del_nr > 0); | |
461 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 462 | |
920bbbfb YZ |
463 | btrfs_set_file_extent_num_bytes(leaf, fi, |
464 | start - key.offset); | |
465 | btrfs_mark_buffer_dirty(leaf); | |
466 | if (disk_bytenr > 0) { | |
467 | inode_sub_bytes(inode, extent_end - start); | |
468 | *hint_byte = disk_bytenr; | |
469 | } | |
470 | if (end == extent_end) | |
471 | break; | |
c8b97818 | 472 | |
920bbbfb YZ |
473 | path->slots[0]++; |
474 | goto next_slot; | |
31840ae1 ZY |
475 | } |
476 | ||
920bbbfb YZ |
477 | /* |
478 | * | ---- range to drop ----- | | |
479 | * | ------ extent ------ | | |
480 | */ | |
481 | if (start <= key.offset && end >= extent_end) { | |
482 | if (del_nr == 0) { | |
483 | del_slot = path->slots[0]; | |
484 | del_nr = 1; | |
485 | } else { | |
486 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
487 | del_nr++; | |
488 | } | |
31840ae1 | 489 | |
920bbbfb | 490 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
a76a3cd4 | 491 | inode_sub_bytes(inode, |
920bbbfb YZ |
492 | extent_end - key.offset); |
493 | extent_end = ALIGN(extent_end, | |
494 | root->sectorsize); | |
495 | } else if (disk_bytenr > 0) { | |
31840ae1 | 496 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
497 | disk_bytenr, num_bytes, 0, |
498 | root->root_key.objectid, | |
5d4f98a2 | 499 | key.objectid, key.offset - |
920bbbfb | 500 | extent_offset); |
31840ae1 | 501 | BUG_ON(ret); |
920bbbfb YZ |
502 | inode_sub_bytes(inode, |
503 | extent_end - key.offset); | |
504 | *hint_byte = disk_bytenr; | |
31840ae1 | 505 | } |
31840ae1 | 506 | |
920bbbfb YZ |
507 | if (end == extent_end) |
508 | break; | |
509 | ||
510 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
511 | path->slots[0]++; | |
512 | goto next_slot; | |
513 | } | |
514 | ||
515 | ret = btrfs_del_items(trans, root, path, del_slot, | |
516 | del_nr); | |
517 | BUG_ON(ret); | |
518 | ||
519 | del_nr = 0; | |
520 | del_slot = 0; | |
521 | ||
522 | btrfs_release_path(root, path); | |
523 | continue; | |
39279cc3 | 524 | } |
920bbbfb YZ |
525 | |
526 | BUG_ON(1); | |
39279cc3 | 527 | } |
920bbbfb YZ |
528 | |
529 | if (del_nr > 0) { | |
530 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); | |
531 | BUG_ON(ret); | |
6643558d | 532 | } |
920bbbfb YZ |
533 | |
534 | btrfs_free_path(path); | |
39279cc3 CM |
535 | return ret; |
536 | } | |
537 | ||
d899e052 | 538 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
539 | u64 objectid, u64 bytenr, u64 orig_offset, |
540 | u64 *start, u64 *end) | |
d899e052 YZ |
541 | { |
542 | struct btrfs_file_extent_item *fi; | |
543 | struct btrfs_key key; | |
544 | u64 extent_end; | |
545 | ||
546 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
547 | return 0; | |
548 | ||
549 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
550 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
551 | return 0; | |
552 | ||
553 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
554 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
555 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 556 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
557 | btrfs_file_extent_compression(leaf, fi) || |
558 | btrfs_file_extent_encryption(leaf, fi) || | |
559 | btrfs_file_extent_other_encoding(leaf, fi)) | |
560 | return 0; | |
561 | ||
562 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
563 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
564 | return 0; | |
565 | ||
566 | *start = key.offset; | |
567 | *end = extent_end; | |
568 | return 1; | |
569 | } | |
570 | ||
571 | /* | |
572 | * Mark extent in the range start - end as written. | |
573 | * | |
574 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
575 | * part of extent is marked as written, the extent will be split into | |
576 | * two or three. | |
577 | */ | |
578 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
579 | struct inode *inode, u64 start, u64 end) |
580 | { | |
920bbbfb | 581 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
582 | struct extent_buffer *leaf; |
583 | struct btrfs_path *path; | |
584 | struct btrfs_file_extent_item *fi; | |
585 | struct btrfs_key key; | |
920bbbfb | 586 | struct btrfs_key new_key; |
d899e052 YZ |
587 | u64 bytenr; |
588 | u64 num_bytes; | |
589 | u64 extent_end; | |
5d4f98a2 | 590 | u64 orig_offset; |
d899e052 YZ |
591 | u64 other_start; |
592 | u64 other_end; | |
920bbbfb YZ |
593 | u64 split; |
594 | int del_nr = 0; | |
595 | int del_slot = 0; | |
6c7d54ac | 596 | int recow; |
d899e052 YZ |
597 | int ret; |
598 | ||
599 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
600 | ||
601 | path = btrfs_alloc_path(); | |
602 | BUG_ON(!path); | |
603 | again: | |
6c7d54ac | 604 | recow = 0; |
920bbbfb | 605 | split = start; |
d899e052 YZ |
606 | key.objectid = inode->i_ino; |
607 | key.type = BTRFS_EXTENT_DATA_KEY; | |
920bbbfb | 608 | key.offset = split; |
d899e052 YZ |
609 | |
610 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
611 | if (ret < 0) |
612 | goto out; | |
d899e052 YZ |
613 | if (ret > 0 && path->slots[0] > 0) |
614 | path->slots[0]--; | |
615 | ||
616 | leaf = path->nodes[0]; | |
617 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
618 | BUG_ON(key.objectid != inode->i_ino || | |
619 | key.type != BTRFS_EXTENT_DATA_KEY); | |
620 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
621 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
622 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
623 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
624 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
625 | BUG_ON(key.offset > start || extent_end < end); | |
626 | ||
627 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
628 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 629 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
630 | memcpy(&new_key, &key, sizeof(new_key)); |
631 | ||
632 | if (start == key.offset && end < extent_end) { | |
633 | other_start = 0; | |
634 | other_end = start; | |
635 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
636 | inode->i_ino, bytenr, orig_offset, | |
637 | &other_start, &other_end)) { | |
638 | new_key.offset = end; | |
639 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
640 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
641 | struct btrfs_file_extent_item); | |
642 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
643 | extent_end - end); | |
644 | btrfs_set_file_extent_offset(leaf, fi, | |
645 | end - orig_offset); | |
646 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
647 | struct btrfs_file_extent_item); | |
648 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
649 | end - other_start); | |
650 | btrfs_mark_buffer_dirty(leaf); | |
651 | goto out; | |
652 | } | |
653 | } | |
654 | ||
655 | if (start > key.offset && end == extent_end) { | |
656 | other_start = end; | |
657 | other_end = 0; | |
658 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
659 | inode->i_ino, bytenr, orig_offset, | |
660 | &other_start, &other_end)) { | |
661 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
662 | struct btrfs_file_extent_item); | |
663 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
664 | start - key.offset); | |
665 | path->slots[0]++; | |
666 | new_key.offset = start; | |
667 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
668 | ||
669 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
670 | struct btrfs_file_extent_item); | |
671 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
672 | other_end - start); | |
673 | btrfs_set_file_extent_offset(leaf, fi, | |
674 | start - orig_offset); | |
675 | btrfs_mark_buffer_dirty(leaf); | |
676 | goto out; | |
677 | } | |
678 | } | |
d899e052 | 679 | |
920bbbfb YZ |
680 | while (start > key.offset || end < extent_end) { |
681 | if (key.offset == start) | |
682 | split = end; | |
683 | ||
920bbbfb YZ |
684 | new_key.offset = split; |
685 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
686 | if (ret == -EAGAIN) { | |
687 | btrfs_release_path(root, path); | |
688 | goto again; | |
d899e052 | 689 | } |
920bbbfb | 690 | BUG_ON(ret < 0); |
d899e052 | 691 | |
920bbbfb YZ |
692 | leaf = path->nodes[0]; |
693 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 694 | struct btrfs_file_extent_item); |
d899e052 | 695 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
696 | split - key.offset); |
697 | ||
698 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
699 | struct btrfs_file_extent_item); | |
700 | ||
701 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); | |
702 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
703 | extent_end - split); | |
d899e052 YZ |
704 | btrfs_mark_buffer_dirty(leaf); |
705 | ||
920bbbfb YZ |
706 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
707 | root->root_key.objectid, | |
708 | inode->i_ino, orig_offset); | |
d899e052 | 709 | BUG_ON(ret); |
d899e052 | 710 | |
920bbbfb YZ |
711 | if (split == start) { |
712 | key.offset = start; | |
713 | } else { | |
714 | BUG_ON(start != key.offset); | |
d899e052 | 715 | path->slots[0]--; |
920bbbfb | 716 | extent_end = end; |
d899e052 | 717 | } |
6c7d54ac | 718 | recow = 1; |
d899e052 YZ |
719 | } |
720 | ||
920bbbfb YZ |
721 | other_start = end; |
722 | other_end = 0; | |
6c7d54ac YZ |
723 | if (extent_mergeable(leaf, path->slots[0] + 1, |
724 | inode->i_ino, bytenr, orig_offset, | |
725 | &other_start, &other_end)) { | |
726 | if (recow) { | |
727 | btrfs_release_path(root, path); | |
728 | goto again; | |
729 | } | |
920bbbfb YZ |
730 | extent_end = other_end; |
731 | del_slot = path->slots[0] + 1; | |
732 | del_nr++; | |
733 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
734 | 0, root->root_key.objectid, | |
735 | inode->i_ino, orig_offset); | |
736 | BUG_ON(ret); | |
d899e052 | 737 | } |
920bbbfb YZ |
738 | other_start = 0; |
739 | other_end = start; | |
6c7d54ac YZ |
740 | if (extent_mergeable(leaf, path->slots[0] - 1, |
741 | inode->i_ino, bytenr, orig_offset, | |
742 | &other_start, &other_end)) { | |
743 | if (recow) { | |
744 | btrfs_release_path(root, path); | |
745 | goto again; | |
746 | } | |
920bbbfb YZ |
747 | key.offset = other_start; |
748 | del_slot = path->slots[0]; | |
749 | del_nr++; | |
750 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
751 | 0, root->root_key.objectid, | |
752 | inode->i_ino, orig_offset); | |
753 | BUG_ON(ret); | |
754 | } | |
755 | if (del_nr == 0) { | |
3f6fae95 SL |
756 | fi = btrfs_item_ptr(leaf, path->slots[0], |
757 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
758 | btrfs_set_file_extent_type(leaf, fi, |
759 | BTRFS_FILE_EXTENT_REG); | |
760 | btrfs_mark_buffer_dirty(leaf); | |
6c7d54ac | 761 | } else { |
3f6fae95 SL |
762 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
763 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
764 | btrfs_set_file_extent_type(leaf, fi, |
765 | BTRFS_FILE_EXTENT_REG); | |
766 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
767 | extent_end - key.offset); | |
768 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 769 | |
6c7d54ac YZ |
770 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
771 | BUG_ON(ret); | |
772 | } | |
920bbbfb | 773 | out: |
d899e052 YZ |
774 | btrfs_free_path(path); |
775 | return 0; | |
776 | } | |
777 | ||
b1bf862e CM |
778 | /* |
779 | * on error we return an unlocked page and the error value | |
780 | * on success we return a locked page and 0 | |
781 | */ | |
782 | static int prepare_uptodate_page(struct page *page, u64 pos) | |
783 | { | |
784 | int ret = 0; | |
785 | ||
786 | if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) { | |
787 | ret = btrfs_readpage(NULL, page); | |
788 | if (ret) | |
789 | return ret; | |
790 | lock_page(page); | |
791 | if (!PageUptodate(page)) { | |
792 | unlock_page(page); | |
793 | return -EIO; | |
794 | } | |
795 | } | |
796 | return 0; | |
797 | } | |
798 | ||
39279cc3 | 799 | /* |
d352ac68 CM |
800 | * this gets pages into the page cache and locks them down, it also properly |
801 | * waits for data=ordered extents to finish before allowing the pages to be | |
802 | * modified. | |
39279cc3 | 803 | */ |
d397712b | 804 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
805 | struct page **pages, size_t num_pages, |
806 | loff_t pos, unsigned long first_index, | |
807 | unsigned long last_index, size_t write_bytes) | |
39279cc3 | 808 | { |
2ac55d41 | 809 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
810 | int i; |
811 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
6da6abae | 812 | struct inode *inode = fdentry(file)->d_inode; |
39279cc3 | 813 | int err = 0; |
b1bf862e | 814 | int faili = 0; |
8c2383c3 | 815 | u64 start_pos; |
e6dcd2dc | 816 | u64 last_pos; |
8c2383c3 | 817 | |
5f39d397 | 818 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 819 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 820 | |
9036c102 | 821 | if (start_pos > inode->i_size) { |
a41ad394 | 822 | err = btrfs_cont_expand(inode, i_size_read(inode), start_pos); |
9036c102 YZ |
823 | if (err) |
824 | return err; | |
825 | } | |
826 | ||
e6dcd2dc | 827 | again: |
39279cc3 CM |
828 | for (i = 0; i < num_pages; i++) { |
829 | pages[i] = grab_cache_page(inode->i_mapping, index + i); | |
830 | if (!pages[i]) { | |
b1bf862e CM |
831 | faili = i - 1; |
832 | err = -ENOMEM; | |
833 | goto fail; | |
834 | } | |
835 | ||
836 | if (i == 0) | |
837 | err = prepare_uptodate_page(pages[i], pos); | |
838 | if (i == num_pages - 1) | |
839 | err = prepare_uptodate_page(pages[i], | |
840 | pos + write_bytes); | |
841 | if (err) { | |
842 | page_cache_release(pages[i]); | |
843 | faili = i - 1; | |
844 | goto fail; | |
39279cc3 | 845 | } |
ccd467d6 | 846 | wait_on_page_writeback(pages[i]); |
39279cc3 | 847 | } |
b1bf862e | 848 | err = 0; |
0762704b | 849 | if (start_pos < inode->i_size) { |
e6dcd2dc | 850 | struct btrfs_ordered_extent *ordered; |
2ac55d41 JB |
851 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
852 | start_pos, last_pos - 1, 0, &cached_state, | |
853 | GFP_NOFS); | |
d397712b CM |
854 | ordered = btrfs_lookup_first_ordered_extent(inode, |
855 | last_pos - 1); | |
e6dcd2dc CM |
856 | if (ordered && |
857 | ordered->file_offset + ordered->len > start_pos && | |
858 | ordered->file_offset < last_pos) { | |
859 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
860 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
861 | start_pos, last_pos - 1, | |
862 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
863 | for (i = 0; i < num_pages; i++) { |
864 | unlock_page(pages[i]); | |
865 | page_cache_release(pages[i]); | |
866 | } | |
867 | btrfs_wait_ordered_range(inode, start_pos, | |
868 | last_pos - start_pos); | |
869 | goto again; | |
870 | } | |
871 | if (ordered) | |
872 | btrfs_put_ordered_extent(ordered); | |
873 | ||
2ac55d41 | 874 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
32c00aff | 875 | last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 876 | EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, |
0762704b | 877 | GFP_NOFS); |
2ac55d41 JB |
878 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
879 | start_pos, last_pos - 1, &cached_state, | |
880 | GFP_NOFS); | |
0762704b | 881 | } |
e6dcd2dc | 882 | for (i = 0; i < num_pages; i++) { |
f87f057b | 883 | clear_page_dirty_for_io(pages[i]); |
e6dcd2dc CM |
884 | set_page_extent_mapped(pages[i]); |
885 | WARN_ON(!PageLocked(pages[i])); | |
886 | } | |
39279cc3 | 887 | return 0; |
b1bf862e CM |
888 | fail: |
889 | while (faili >= 0) { | |
890 | unlock_page(pages[faili]); | |
891 | page_cache_release(pages[faili]); | |
892 | faili--; | |
893 | } | |
894 | return err; | |
895 | ||
39279cc3 CM |
896 | } |
897 | ||
d0215f3e JB |
898 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
899 | struct iov_iter *i, | |
900 | loff_t pos) | |
4b46fce2 | 901 | { |
11c65dcc JB |
902 | struct inode *inode = fdentry(file)->d_inode; |
903 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
11c65dcc | 904 | struct page **pages = NULL; |
39279cc3 CM |
905 | unsigned long first_index; |
906 | unsigned long last_index; | |
d0215f3e JB |
907 | size_t num_written = 0; |
908 | int nrptrs; | |
909 | int ret; | |
4b46fce2 | 910 | |
d0215f3e | 911 | nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) / |
11c65dcc JB |
912 | PAGE_CACHE_SIZE, PAGE_CACHE_SIZE / |
913 | (sizeof(struct page *))); | |
8c2383c3 | 914 | pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
915 | if (!pages) |
916 | return -ENOMEM; | |
ab93dbec | 917 | |
39279cc3 | 918 | first_index = pos >> PAGE_CACHE_SHIFT; |
d0215f3e | 919 | last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT; |
39279cc3 | 920 | |
d0215f3e | 921 | while (iov_iter_count(i) > 0) { |
39279cc3 | 922 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
d0215f3e | 923 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 924 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 925 | offset); |
3a90983d YZ |
926 | size_t num_pages = (write_bytes + offset + |
927 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0215f3e JB |
928 | size_t dirty_pages; |
929 | size_t copied; | |
39279cc3 | 930 | |
8c2383c3 | 931 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 932 | |
914ee295 XZ |
933 | /* |
934 | * Fault pages before locking them in prepare_pages | |
935 | * to avoid recursive lock | |
936 | */ | |
d0215f3e | 937 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 938 | ret = -EFAULT; |
d0215f3e | 939 | break; |
914ee295 XZ |
940 | } |
941 | ||
942 | ret = btrfs_delalloc_reserve_space(inode, | |
943 | num_pages << PAGE_CACHE_SHIFT); | |
1832a6d5 | 944 | if (ret) |
d0215f3e | 945 | break; |
1832a6d5 | 946 | |
4a64001f JB |
947 | /* |
948 | * This is going to setup the pages array with the number of | |
949 | * pages we want, so we don't really need to worry about the | |
950 | * contents of pages from loop to loop | |
951 | */ | |
39279cc3 CM |
952 | ret = prepare_pages(root, file, pages, num_pages, |
953 | pos, first_index, last_index, | |
8c2383c3 | 954 | write_bytes); |
6a63209f | 955 | if (ret) { |
914ee295 XZ |
956 | btrfs_delalloc_release_space(inode, |
957 | num_pages << PAGE_CACHE_SHIFT); | |
d0215f3e | 958 | break; |
6a63209f | 959 | } |
39279cc3 | 960 | |
914ee295 | 961 | copied = btrfs_copy_from_user(pos, num_pages, |
d0215f3e | 962 | write_bytes, pages, i); |
b1bf862e CM |
963 | |
964 | /* | |
965 | * if we have trouble faulting in the pages, fall | |
966 | * back to one page at a time | |
967 | */ | |
968 | if (copied < write_bytes) | |
969 | nrptrs = 1; | |
970 | ||
971 | if (copied == 0) | |
972 | dirty_pages = 0; | |
973 | else | |
974 | dirty_pages = (copied + offset + | |
975 | PAGE_CACHE_SIZE - 1) >> | |
976 | PAGE_CACHE_SHIFT; | |
914ee295 | 977 | |
d0215f3e JB |
978 | /* |
979 | * If we had a short copy we need to release the excess delaloc | |
980 | * bytes we reserved. We need to increment outstanding_extents | |
981 | * because btrfs_delalloc_release_space will decrement it, but | |
982 | * we still have an outstanding extent for the chunk we actually | |
983 | * managed to copy. | |
984 | */ | |
914ee295 XZ |
985 | if (num_pages > dirty_pages) { |
986 | if (copied > 0) | |
987 | atomic_inc( | |
988 | &BTRFS_I(inode)->outstanding_extents); | |
989 | btrfs_delalloc_release_space(inode, | |
990 | (num_pages - dirty_pages) << | |
991 | PAGE_CACHE_SHIFT); | |
992 | } | |
993 | ||
994 | if (copied > 0) { | |
d0215f3e JB |
995 | ret = dirty_and_release_pages(root, file, pages, |
996 | dirty_pages, pos, | |
997 | copied); | |
998 | if (ret) { | |
999 | btrfs_delalloc_release_space(inode, | |
1000 | dirty_pages << PAGE_CACHE_SHIFT); | |
1001 | btrfs_drop_pages(pages, num_pages); | |
1002 | break; | |
1003 | } | |
54aa1f4d | 1004 | } |
39279cc3 | 1005 | |
39279cc3 CM |
1006 | btrfs_drop_pages(pages, num_pages); |
1007 | ||
d0215f3e JB |
1008 | cond_resched(); |
1009 | ||
1010 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
1011 | dirty_pages); | |
1012 | if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1) | |
1013 | btrfs_btree_balance_dirty(root, 1); | |
1014 | btrfs_throttle(root); | |
cb843a6f | 1015 | |
914ee295 XZ |
1016 | pos += copied; |
1017 | num_written += copied; | |
d0215f3e | 1018 | } |
39279cc3 | 1019 | |
d0215f3e JB |
1020 | kfree(pages); |
1021 | ||
1022 | return num_written ? num_written : ret; | |
1023 | } | |
1024 | ||
1025 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
1026 | const struct iovec *iov, | |
1027 | unsigned long nr_segs, loff_t pos, | |
1028 | loff_t *ppos, size_t count, size_t ocount) | |
1029 | { | |
1030 | struct file *file = iocb->ki_filp; | |
1031 | struct inode *inode = fdentry(file)->d_inode; | |
1032 | struct iov_iter i; | |
1033 | ssize_t written; | |
1034 | ssize_t written_buffered; | |
1035 | loff_t endbyte; | |
1036 | int err; | |
1037 | ||
1038 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos, | |
1039 | count, ocount); | |
1040 | ||
1041 | /* | |
1042 | * the generic O_DIRECT will update in-memory i_size after the | |
1043 | * DIOs are done. But our endio handlers that update the on | |
1044 | * disk i_size never update past the in memory i_size. So we | |
1045 | * need one more update here to catch any additions to the | |
1046 | * file | |
1047 | */ | |
1048 | if (inode->i_size != BTRFS_I(inode)->disk_i_size) { | |
1049 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
1050 | mark_inode_dirty(inode); | |
1051 | } | |
1052 | ||
1053 | if (written < 0 || written == count) | |
1054 | return written; | |
1055 | ||
1056 | pos += written; | |
1057 | count -= written; | |
1058 | iov_iter_init(&i, iov, nr_segs, count, written); | |
1059 | written_buffered = __btrfs_buffered_write(file, &i, pos); | |
1060 | if (written_buffered < 0) { | |
1061 | err = written_buffered; | |
1062 | goto out; | |
39279cc3 | 1063 | } |
d0215f3e JB |
1064 | endbyte = pos + written_buffered - 1; |
1065 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); | |
1066 | if (err) | |
1067 | goto out; | |
1068 | written += written_buffered; | |
1069 | *ppos = pos + written_buffered; | |
1070 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, | |
1071 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1072 | out: |
d0215f3e JB |
1073 | return written ? written : err; |
1074 | } | |
5b92ee72 | 1075 | |
d0215f3e JB |
1076 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
1077 | const struct iovec *iov, | |
1078 | unsigned long nr_segs, loff_t pos) | |
1079 | { | |
1080 | struct file *file = iocb->ki_filp; | |
1081 | struct inode *inode = fdentry(file)->d_inode; | |
1082 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1083 | loff_t *ppos = &iocb->ki_pos; | |
1084 | ssize_t num_written = 0; | |
1085 | ssize_t err = 0; | |
1086 | size_t count, ocount; | |
1087 | ||
1088 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
1089 | ||
1090 | mutex_lock(&inode->i_mutex); | |
1091 | ||
1092 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
1093 | if (err) { | |
1094 | mutex_unlock(&inode->i_mutex); | |
1095 | goto out; | |
1096 | } | |
1097 | count = ocount; | |
1098 | ||
1099 | current->backing_dev_info = inode->i_mapping->backing_dev_info; | |
1100 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1101 | if (err) { | |
1102 | mutex_unlock(&inode->i_mutex); | |
1103 | goto out; | |
1104 | } | |
1105 | ||
1106 | if (count == 0) { | |
1107 | mutex_unlock(&inode->i_mutex); | |
1108 | goto out; | |
1109 | } | |
1110 | ||
1111 | err = file_remove_suid(file); | |
1112 | if (err) { | |
1113 | mutex_unlock(&inode->i_mutex); | |
1114 | goto out; | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * If BTRFS flips readonly due to some impossible error | |
1119 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1120 | * although we have opened a file as writable, we have | |
1121 | * to stop this write operation to ensure FS consistency. | |
1122 | */ | |
1123 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
1124 | mutex_unlock(&inode->i_mutex); | |
1125 | err = -EROFS; | |
1126 | goto out; | |
1127 | } | |
1128 | ||
1129 | file_update_time(file); | |
1130 | BTRFS_I(inode)->sequence++; | |
1131 | ||
1132 | if (unlikely(file->f_flags & O_DIRECT)) { | |
1133 | num_written = __btrfs_direct_write(iocb, iov, nr_segs, | |
1134 | pos, ppos, count, ocount); | |
1135 | } else { | |
1136 | struct iov_iter i; | |
1137 | ||
1138 | iov_iter_init(&i, iov, nr_segs, count, num_written); | |
1139 | ||
1140 | num_written = __btrfs_buffered_write(file, &i, pos); | |
1141 | if (num_written > 0) | |
1142 | *ppos = pos + num_written; | |
1143 | } | |
1144 | ||
1145 | mutex_unlock(&inode->i_mutex); | |
2ff3e9b6 | 1146 | |
5a3f23d5 CM |
1147 | /* |
1148 | * we want to make sure fsync finds this change | |
1149 | * but we haven't joined a transaction running right now. | |
1150 | * | |
1151 | * Later on, someone is sure to update the inode and get the | |
1152 | * real transid recorded. | |
1153 | * | |
1154 | * We set last_trans now to the fs_info generation + 1, | |
1155 | * this will either be one more than the running transaction | |
1156 | * or the generation used for the next transaction if there isn't | |
1157 | * one running right now. | |
1158 | */ | |
1159 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
d0215f3e JB |
1160 | if (num_written > 0 || num_written == -EIOCBQUEUED) { |
1161 | err = generic_write_sync(file, pos, num_written); | |
1162 | if (err < 0 && num_written > 0) | |
2ff3e9b6 CM |
1163 | num_written = err; |
1164 | } | |
d0215f3e | 1165 | out: |
39279cc3 | 1166 | current->backing_dev_info = NULL; |
39279cc3 CM |
1167 | return num_written ? num_written : err; |
1168 | } | |
1169 | ||
d397712b | 1170 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1171 | { |
5a3f23d5 CM |
1172 | /* |
1173 | * ordered_data_close is set by settattr when we are about to truncate | |
1174 | * a file from a non-zero size to a zero size. This tries to | |
1175 | * flush down new bytes that may have been written if the | |
1176 | * application were using truncate to replace a file in place. | |
1177 | */ | |
1178 | if (BTRFS_I(inode)->ordered_data_close) { | |
1179 | BTRFS_I(inode)->ordered_data_close = 0; | |
1180 | btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); | |
1181 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) | |
1182 | filemap_flush(inode->i_mapping); | |
1183 | } | |
6bf13c0c SW |
1184 | if (filp->private_data) |
1185 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1186 | return 0; |
1187 | } | |
1188 | ||
d352ac68 CM |
1189 | /* |
1190 | * fsync call for both files and directories. This logs the inode into | |
1191 | * the tree log instead of forcing full commits whenever possible. | |
1192 | * | |
1193 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1194 | * in the metadata btree are up to date for copying to the log. | |
1195 | * | |
1196 | * It drops the inode mutex before doing the tree log commit. This is an | |
1197 | * important optimization for directories because holding the mutex prevents | |
1198 | * new operations on the dir while we write to disk. | |
1199 | */ | |
7ea80859 | 1200 | int btrfs_sync_file(struct file *file, int datasync) |
39279cc3 | 1201 | { |
7ea80859 | 1202 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1203 | struct inode *inode = dentry->d_inode; |
1204 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1205 | int ret = 0; |
39279cc3 CM |
1206 | struct btrfs_trans_handle *trans; |
1207 | ||
257c62e1 CM |
1208 | |
1209 | /* we wait first, since the writeback may change the inode */ | |
1210 | root->log_batch++; | |
1211 | /* the VFS called filemap_fdatawrite for us */ | |
1212 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
1213 | root->log_batch++; | |
1214 | ||
39279cc3 | 1215 | /* |
15ee9bc7 JB |
1216 | * check the transaction that last modified this inode |
1217 | * and see if its already been committed | |
39279cc3 | 1218 | */ |
15ee9bc7 JB |
1219 | if (!BTRFS_I(inode)->last_trans) |
1220 | goto out; | |
a2135011 | 1221 | |
257c62e1 CM |
1222 | /* |
1223 | * if the last transaction that changed this file was before | |
1224 | * the current transaction, we can bail out now without any | |
1225 | * syncing | |
1226 | */ | |
15ee9bc7 JB |
1227 | mutex_lock(&root->fs_info->trans_mutex); |
1228 | if (BTRFS_I(inode)->last_trans <= | |
1229 | root->fs_info->last_trans_committed) { | |
1230 | BTRFS_I(inode)->last_trans = 0; | |
1231 | mutex_unlock(&root->fs_info->trans_mutex); | |
1232 | goto out; | |
1233 | } | |
1234 | mutex_unlock(&root->fs_info->trans_mutex); | |
1235 | ||
1236 | /* | |
a52d9a80 CM |
1237 | * ok we haven't committed the transaction yet, lets do a commit |
1238 | */ | |
6f902af4 | 1239 | if (file->private_data) |
6bf13c0c SW |
1240 | btrfs_ioctl_trans_end(file); |
1241 | ||
a22285a6 YZ |
1242 | trans = btrfs_start_transaction(root, 0); |
1243 | if (IS_ERR(trans)) { | |
1244 | ret = PTR_ERR(trans); | |
39279cc3 CM |
1245 | goto out; |
1246 | } | |
e02119d5 | 1247 | |
2cfbd50b | 1248 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
d397712b | 1249 | if (ret < 0) |
e02119d5 | 1250 | goto out; |
49eb7e46 CM |
1251 | |
1252 | /* we've logged all the items and now have a consistent | |
1253 | * version of the file in the log. It is possible that | |
1254 | * someone will come in and modify the file, but that's | |
1255 | * fine because the log is consistent on disk, and we | |
1256 | * have references to all of the file's extents | |
1257 | * | |
1258 | * It is possible that someone will come in and log the | |
1259 | * file again, but that will end up using the synchronization | |
1260 | * inside btrfs_sync_log to keep things safe. | |
1261 | */ | |
2cfbd50b | 1262 | mutex_unlock(&dentry->d_inode->i_mutex); |
49eb7e46 | 1263 | |
257c62e1 CM |
1264 | if (ret != BTRFS_NO_LOG_SYNC) { |
1265 | if (ret > 0) { | |
12fcfd22 | 1266 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1267 | } else { |
1268 | ret = btrfs_sync_log(trans, root); | |
1269 | if (ret == 0) | |
1270 | ret = btrfs_end_transaction(trans, root); | |
1271 | else | |
1272 | ret = btrfs_commit_transaction(trans, root); | |
1273 | } | |
1274 | } else { | |
1275 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1276 | } |
2cfbd50b | 1277 | mutex_lock(&dentry->d_inode->i_mutex); |
39279cc3 | 1278 | out: |
014e4ac4 | 1279 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1280 | } |
1281 | ||
f0f37e2f | 1282 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1283 | .fault = filemap_fault, |
9ebefb18 CM |
1284 | .page_mkwrite = btrfs_page_mkwrite, |
1285 | }; | |
1286 | ||
1287 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1288 | { | |
058a457e MX |
1289 | struct address_space *mapping = filp->f_mapping; |
1290 | ||
1291 | if (!mapping->a_ops->readpage) | |
1292 | return -ENOEXEC; | |
1293 | ||
9ebefb18 | 1294 | file_accessed(filp); |
058a457e MX |
1295 | vma->vm_ops = &btrfs_file_vm_ops; |
1296 | vma->vm_flags |= VM_CAN_NONLINEAR; | |
1297 | ||
9ebefb18 CM |
1298 | return 0; |
1299 | } | |
1300 | ||
2fe17c10 CH |
1301 | static long btrfs_fallocate(struct file *file, int mode, |
1302 | loff_t offset, loff_t len) | |
1303 | { | |
1304 | struct inode *inode = file->f_path.dentry->d_inode; | |
1305 | struct extent_state *cached_state = NULL; | |
1306 | u64 cur_offset; | |
1307 | u64 last_byte; | |
1308 | u64 alloc_start; | |
1309 | u64 alloc_end; | |
1310 | u64 alloc_hint = 0; | |
1311 | u64 locked_end; | |
1312 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
1313 | struct extent_map *em; | |
1314 | int ret; | |
1315 | ||
1316 | alloc_start = offset & ~mask; | |
1317 | alloc_end = (offset + len + mask) & ~mask; | |
1318 | ||
1319 | /* We only support the FALLOC_FL_KEEP_SIZE mode */ | |
1320 | if (mode & ~FALLOC_FL_KEEP_SIZE) | |
1321 | return -EOPNOTSUPP; | |
1322 | ||
1323 | /* | |
1324 | * wait for ordered IO before we have any locks. We'll loop again | |
1325 | * below with the locks held. | |
1326 | */ | |
1327 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
1328 | ||
1329 | mutex_lock(&inode->i_mutex); | |
1330 | ret = inode_newsize_ok(inode, alloc_end); | |
1331 | if (ret) | |
1332 | goto out; | |
1333 | ||
1334 | if (alloc_start > inode->i_size) { | |
a41ad394 JB |
1335 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
1336 | alloc_start); | |
2fe17c10 CH |
1337 | if (ret) |
1338 | goto out; | |
1339 | } | |
1340 | ||
1341 | ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start); | |
1342 | if (ret) | |
1343 | goto out; | |
1344 | ||
1345 | locked_end = alloc_end - 1; | |
1346 | while (1) { | |
1347 | struct btrfs_ordered_extent *ordered; | |
1348 | ||
1349 | /* the extent lock is ordered inside the running | |
1350 | * transaction | |
1351 | */ | |
1352 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
1353 | locked_end, 0, &cached_state, GFP_NOFS); | |
1354 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
1355 | alloc_end - 1); | |
1356 | if (ordered && | |
1357 | ordered->file_offset + ordered->len > alloc_start && | |
1358 | ordered->file_offset < alloc_end) { | |
1359 | btrfs_put_ordered_extent(ordered); | |
1360 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
1361 | alloc_start, locked_end, | |
1362 | &cached_state, GFP_NOFS); | |
1363 | /* | |
1364 | * we can't wait on the range with the transaction | |
1365 | * running or with the extent lock held | |
1366 | */ | |
1367 | btrfs_wait_ordered_range(inode, alloc_start, | |
1368 | alloc_end - alloc_start); | |
1369 | } else { | |
1370 | if (ordered) | |
1371 | btrfs_put_ordered_extent(ordered); | |
1372 | break; | |
1373 | } | |
1374 | } | |
1375 | ||
1376 | cur_offset = alloc_start; | |
1377 | while (1) { | |
1378 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
1379 | alloc_end - cur_offset, 0); | |
1380 | BUG_ON(IS_ERR(em) || !em); | |
1381 | last_byte = min(extent_map_end(em), alloc_end); | |
1382 | last_byte = (last_byte + mask) & ~mask; | |
1383 | if (em->block_start == EXTENT_MAP_HOLE || | |
1384 | (cur_offset >= inode->i_size && | |
1385 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
1386 | ret = btrfs_prealloc_file_range(inode, mode, cur_offset, | |
1387 | last_byte - cur_offset, | |
1388 | 1 << inode->i_blkbits, | |
1389 | offset + len, | |
1390 | &alloc_hint); | |
1391 | if (ret < 0) { | |
1392 | free_extent_map(em); | |
1393 | break; | |
1394 | } | |
1395 | } | |
1396 | free_extent_map(em); | |
1397 | ||
1398 | cur_offset = last_byte; | |
1399 | if (cur_offset >= alloc_end) { | |
1400 | ret = 0; | |
1401 | break; | |
1402 | } | |
1403 | } | |
1404 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, | |
1405 | &cached_state, GFP_NOFS); | |
1406 | ||
1407 | btrfs_free_reserved_data_space(inode, alloc_end - alloc_start); | |
1408 | out: | |
1409 | mutex_unlock(&inode->i_mutex); | |
1410 | return ret; | |
1411 | } | |
1412 | ||
828c0950 | 1413 | const struct file_operations btrfs_file_operations = { |
39279cc3 CM |
1414 | .llseek = generic_file_llseek, |
1415 | .read = do_sync_read, | |
4a001071 | 1416 | .write = do_sync_write, |
9ebefb18 | 1417 | .aio_read = generic_file_aio_read, |
e9906a98 | 1418 | .splice_read = generic_file_splice_read, |
11c65dcc | 1419 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 1420 | .mmap = btrfs_file_mmap, |
39279cc3 | 1421 | .open = generic_file_open, |
e1b81e67 | 1422 | .release = btrfs_release_file, |
39279cc3 | 1423 | .fsync = btrfs_sync_file, |
2fe17c10 | 1424 | .fallocate = btrfs_fallocate, |
34287aa3 | 1425 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 1426 | #ifdef CONFIG_COMPAT |
34287aa3 | 1427 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
1428 | #endif |
1429 | }; |