Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
39279cc3 CM |
6 | #include <linux/fs.h> |
7 | #include <linux/pagemap.h> | |
39279cc3 CM |
8 | #include <linux/time.h> |
9 | #include <linux/init.h> | |
10 | #include <linux/string.h> | |
39279cc3 | 11 | #include <linux/backing-dev.h> |
2fe17c10 | 12 | #include <linux/falloc.h> |
39279cc3 | 13 | #include <linux/writeback.h> |
39279cc3 | 14 | #include <linux/compat.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
55e301fd | 16 | #include <linux/btrfs.h> |
e2e40f2c | 17 | #include <linux/uio.h> |
ae5e165d | 18 | #include <linux/iversion.h> |
14605409 | 19 | #include <linux/fsverity.h> |
3ecb43cb | 20 | #include <linux/iomap.h> |
39279cc3 CM |
21 | #include "ctree.h" |
22 | #include "disk-io.h" | |
23 | #include "transaction.h" | |
24 | #include "btrfs_inode.h" | |
e02119d5 CM |
25 | #include "tree-log.h" |
26 | #include "locking.h" | |
fcebe456 | 27 | #include "qgroup.h" |
ebb8765b | 28 | #include "compression.h" |
86736342 | 29 | #include "delalloc-space.h" |
6a177381 | 30 | #include "reflink.h" |
f02a85d2 | 31 | #include "subpage.h" |
c7f13d42 | 32 | #include "fs.h" |
07e81dc9 | 33 | #include "accessors.h" |
a0231804 | 34 | #include "extent-tree.h" |
7c8ede16 | 35 | #include "file-item.h" |
7572dec8 | 36 | #include "ioctl.h" |
af142b6f | 37 | #include "file.h" |
7f0add25 | 38 | #include "super.h" |
39279cc3 | 39 | |
d352ac68 CM |
40 | /* simple helper to fault in pages and copy. This should go away |
41 | * and be replaced with calls into generic code. | |
42 | */ | |
ee22f0c4 | 43 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 44 | struct page **prepared_pages, |
11c65dcc | 45 | struct iov_iter *i) |
39279cc3 | 46 | { |
914ee295 | 47 | size_t copied = 0; |
d0215f3e | 48 | size_t total_copied = 0; |
11c65dcc | 49 | int pg = 0; |
7073017a | 50 | int offset = offset_in_page(pos); |
39279cc3 | 51 | |
11c65dcc | 52 | while (write_bytes > 0) { |
39279cc3 | 53 | size_t count = min_t(size_t, |
09cbfeaf | 54 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 55 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
56 | /* |
57 | * Copy data from userspace to the current page | |
914ee295 | 58 | */ |
f0b65f39 | 59 | copied = copy_page_from_iter_atomic(page, offset, count, i); |
11c65dcc | 60 | |
39279cc3 CM |
61 | /* Flush processor's dcache for this page */ |
62 | flush_dcache_page(page); | |
31339acd CM |
63 | |
64 | /* | |
65 | * if we get a partial write, we can end up with | |
66 | * partially up to date pages. These add | |
67 | * a lot of complexity, so make sure they don't | |
68 | * happen by forcing this copy to be retried. | |
69 | * | |
70 | * The rest of the btrfs_file_write code will fall | |
71 | * back to page at a time copies after we return 0. | |
72 | */ | |
f0b65f39 AV |
73 | if (unlikely(copied < count)) { |
74 | if (!PageUptodate(page)) { | |
75 | iov_iter_revert(i, copied); | |
76 | copied = 0; | |
77 | } | |
78 | if (!copied) | |
79 | break; | |
80 | } | |
31339acd | 81 | |
11c65dcc | 82 | write_bytes -= copied; |
914ee295 | 83 | total_copied += copied; |
f0b65f39 AV |
84 | offset += copied; |
85 | if (offset == PAGE_SIZE) { | |
11c65dcc JB |
86 | pg++; |
87 | offset = 0; | |
88 | } | |
39279cc3 | 89 | } |
914ee295 | 90 | return total_copied; |
39279cc3 CM |
91 | } |
92 | ||
d352ac68 CM |
93 | /* |
94 | * unlocks pages after btrfs_file_write is done with them | |
95 | */ | |
e4f94347 QW |
96 | static void btrfs_drop_pages(struct btrfs_fs_info *fs_info, |
97 | struct page **pages, size_t num_pages, | |
98 | u64 pos, u64 copied) | |
39279cc3 CM |
99 | { |
100 | size_t i; | |
e4f94347 QW |
101 | u64 block_start = round_down(pos, fs_info->sectorsize); |
102 | u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start; | |
103 | ||
104 | ASSERT(block_len <= U32_MAX); | |
39279cc3 | 105 | for (i = 0; i < num_pages; i++) { |
d352ac68 CM |
106 | /* page checked is some magic around finding pages that |
107 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
108 | * clear it here. There should be no need to mark the pages |
109 | * accessed as prepare_pages should have marked them accessed | |
110 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 111 | */ |
55151ea9 QW |
112 | btrfs_folio_clamp_clear_checked(fs_info, page_folio(pages[i]), |
113 | block_start, block_len); | |
39279cc3 | 114 | unlock_page(pages[i]); |
09cbfeaf | 115 | put_page(pages[i]); |
39279cc3 CM |
116 | } |
117 | } | |
118 | ||
d352ac68 | 119 | /* |
c0fab480 QW |
120 | * After btrfs_copy_from_user(), update the following things for delalloc: |
121 | * - Mark newly dirtied pages as DELALLOC in the io tree. | |
122 | * Used to advise which range is to be written back. | |
123 | * - Mark modified pages as Uptodate/Dirty and not needing COW fixup | |
124 | * - Update inode size for past EOF write | |
d352ac68 | 125 | */ |
088545f6 | 126 | int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages, |
2ff7e61e | 127 | size_t num_pages, loff_t pos, size_t write_bytes, |
aa8c1a41 | 128 | struct extent_state **cached, bool noreserve) |
39279cc3 | 129 | { |
088545f6 | 130 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
35cb2e90 | 131 | int ret = 0; |
a52d9a80 | 132 | int i; |
db94535d | 133 | u64 num_bytes; |
a52d9a80 CM |
134 | u64 start_pos; |
135 | u64 end_of_last_block; | |
136 | u64 end_pos = pos + write_bytes; | |
088545f6 | 137 | loff_t isize = i_size_read(&inode->vfs_inode); |
e3b8a485 | 138 | unsigned int extra_bits = 0; |
39279cc3 | 139 | |
aa8c1a41 GR |
140 | if (write_bytes == 0) |
141 | return 0; | |
142 | ||
143 | if (noreserve) | |
144 | extra_bits |= EXTENT_NORESERVE; | |
145 | ||
13f0dd8f | 146 | start_pos = round_down(pos, fs_info->sectorsize); |
da17066c | 147 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 148 | fs_info->sectorsize); |
f02a85d2 | 149 | ASSERT(num_bytes <= U32_MAX); |
39279cc3 | 150 | |
db94535d | 151 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 | 152 | |
7703bdd8 CM |
153 | /* |
154 | * The pages may have already been dirty, clear out old accounting so | |
155 | * we can set things up properly | |
156 | */ | |
088545f6 | 157 | clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block, |
e182163d | 158 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 159 | cached); |
7703bdd8 | 160 | |
35cb2e90 | 161 | ret = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
330a5827 | 162 | extra_bits, cached); |
35cb2e90 AJ |
163 | if (ret) |
164 | return ret; | |
9ed74f2d | 165 | |
c8b97818 CM |
166 | for (i = 0; i < num_pages; i++) { |
167 | struct page *p = pages[i]; | |
f02a85d2 | 168 | |
55151ea9 QW |
169 | btrfs_folio_clamp_set_uptodate(fs_info, page_folio(p), |
170 | start_pos, num_bytes); | |
171 | btrfs_folio_clamp_clear_checked(fs_info, page_folio(p), | |
172 | start_pos, num_bytes); | |
173 | btrfs_folio_clamp_set_dirty(fs_info, page_folio(p), | |
174 | start_pos, num_bytes); | |
a52d9a80 | 175 | } |
9f570b8d JB |
176 | |
177 | /* | |
178 | * we've only changed i_size in ram, and we haven't updated | |
179 | * the disk i_size. There is no need to log the inode | |
180 | * at this time. | |
181 | */ | |
182 | if (end_pos > isize) | |
088545f6 | 183 | i_size_write(&inode->vfs_inode, end_pos); |
a22285a6 | 184 | return 0; |
39279cc3 CM |
185 | } |
186 | ||
187 | /* | |
188 | * this is very complex, but the basic idea is to drop all extents | |
189 | * in the range start - end. hint_block is filled in with a block number | |
190 | * that would be a good hint to the block allocator for this file. | |
191 | * | |
192 | * If an extent intersects the range but is not entirely inside the range | |
193 | * it is either truncated or split. Anything entirely inside the range | |
194 | * is deleted from the tree. | |
2766ff61 FM |
195 | * |
196 | * Note: the VFS' inode number of bytes is not updated, it's up to the caller | |
197 | * to deal with that. We set the field 'bytes_found' of the arguments structure | |
198 | * with the number of allocated bytes found in the target range, so that the | |
199 | * caller can update the inode's number of bytes in an atomic way when | |
200 | * replacing extents in a range to avoid races with stat(2). | |
39279cc3 | 201 | */ |
5893dfb9 FM |
202 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, |
203 | struct btrfs_root *root, struct btrfs_inode *inode, | |
204 | struct btrfs_drop_extents_args *args) | |
39279cc3 | 205 | { |
0b246afa | 206 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 207 | struct extent_buffer *leaf; |
920bbbfb | 208 | struct btrfs_file_extent_item *fi; |
00f5c795 | 209 | struct btrfs_key key; |
920bbbfb | 210 | struct btrfs_key new_key; |
906c448c | 211 | u64 ino = btrfs_ino(inode); |
5893dfb9 | 212 | u64 search_start = args->start; |
920bbbfb YZ |
213 | u64 disk_bytenr = 0; |
214 | u64 num_bytes = 0; | |
215 | u64 extent_offset = 0; | |
216 | u64 extent_end = 0; | |
5893dfb9 | 217 | u64 last_end = args->start; |
920bbbfb YZ |
218 | int del_nr = 0; |
219 | int del_slot = 0; | |
220 | int extent_type; | |
ccd467d6 | 221 | int recow; |
00f5c795 | 222 | int ret; |
dc7fdde3 | 223 | int modify_tree = -1; |
27cdeb70 | 224 | int update_refs; |
c3308f84 | 225 | int found = 0; |
5893dfb9 FM |
226 | struct btrfs_path *path = args->path; |
227 | ||
2766ff61 | 228 | args->bytes_found = 0; |
5893dfb9 FM |
229 | args->extent_inserted = false; |
230 | ||
231 | /* Must always have a path if ->replace_extent is true */ | |
232 | ASSERT(!(args->replace_extent && !args->path)); | |
233 | ||
234 | if (!path) { | |
235 | path = btrfs_alloc_path(); | |
236 | if (!path) { | |
237 | ret = -ENOMEM; | |
238 | goto out; | |
239 | } | |
240 | } | |
39279cc3 | 241 | |
5893dfb9 | 242 | if (args->drop_cache) |
4c0c8cfc | 243 | btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false); |
a52d9a80 | 244 | |
5893dfb9 | 245 | if (args->start >= inode->disk_i_size && !args->replace_extent) |
dc7fdde3 CM |
246 | modify_tree = 0; |
247 | ||
e094f480 | 248 | update_refs = (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID); |
d397712b | 249 | while (1) { |
ccd467d6 | 250 | recow = 0; |
33345d01 | 251 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 252 | search_start, modify_tree); |
39279cc3 | 253 | if (ret < 0) |
920bbbfb | 254 | break; |
5893dfb9 | 255 | if (ret > 0 && path->slots[0] > 0 && search_start == args->start) { |
920bbbfb YZ |
256 | leaf = path->nodes[0]; |
257 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 258 | if (key.objectid == ino && |
920bbbfb YZ |
259 | key.type == BTRFS_EXTENT_DATA_KEY) |
260 | path->slots[0]--; | |
39279cc3 | 261 | } |
920bbbfb | 262 | ret = 0; |
8c2383c3 | 263 | next_slot: |
5f39d397 | 264 | leaf = path->nodes[0]; |
920bbbfb YZ |
265 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
266 | BUG_ON(del_nr > 0); | |
267 | ret = btrfs_next_leaf(root, path); | |
268 | if (ret < 0) | |
269 | break; | |
270 | if (ret > 0) { | |
271 | ret = 0; | |
272 | break; | |
8c2383c3 | 273 | } |
920bbbfb YZ |
274 | leaf = path->nodes[0]; |
275 | recow = 1; | |
276 | } | |
277 | ||
278 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
279 | |
280 | if (key.objectid > ino) | |
281 | break; | |
282 | if (WARN_ON_ONCE(key.objectid < ino) || | |
283 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
284 | ASSERT(del_nr == 0); | |
285 | path->slots[0]++; | |
286 | goto next_slot; | |
287 | } | |
5893dfb9 | 288 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end) |
920bbbfb YZ |
289 | break; |
290 | ||
291 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
292 | struct btrfs_file_extent_item); | |
293 | extent_type = btrfs_file_extent_type(leaf, fi); | |
294 | ||
295 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
296 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
297 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
298 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
299 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
300 | extent_end = key.offset + | |
301 | btrfs_file_extent_num_bytes(leaf, fi); | |
302 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
303 | extent_end = key.offset + | |
e41ca589 | 304 | btrfs_file_extent_ram_bytes(leaf, fi); |
8c2383c3 | 305 | } else { |
aeafbf84 FM |
306 | /* can't happen */ |
307 | BUG(); | |
39279cc3 CM |
308 | } |
309 | ||
fc19c5e7 FM |
310 | /* |
311 | * Don't skip extent items representing 0 byte lengths. They | |
312 | * used to be created (bug) if while punching holes we hit | |
313 | * -ENOSPC condition. So if we find one here, just ensure we | |
314 | * delete it, otherwise we would insert a new file extent item | |
315 | * with the same key (offset) as that 0 bytes length file | |
316 | * extent item in the call to setup_items_for_insert() later | |
317 | * in this function. | |
318 | */ | |
62fe51c1 JB |
319 | if (extent_end == key.offset && extent_end >= search_start) { |
320 | last_end = extent_end; | |
fc19c5e7 | 321 | goto delete_extent_item; |
62fe51c1 | 322 | } |
fc19c5e7 | 323 | |
920bbbfb YZ |
324 | if (extent_end <= search_start) { |
325 | path->slots[0]++; | |
8c2383c3 | 326 | goto next_slot; |
39279cc3 CM |
327 | } |
328 | ||
c3308f84 | 329 | found = 1; |
5893dfb9 | 330 | search_start = max(key.offset, args->start); |
dc7fdde3 CM |
331 | if (recow || !modify_tree) { |
332 | modify_tree = -1; | |
b3b4aa74 | 333 | btrfs_release_path(path); |
920bbbfb | 334 | continue; |
39279cc3 | 335 | } |
6643558d | 336 | |
920bbbfb YZ |
337 | /* |
338 | * | - range to drop - | | |
339 | * | -------- extent -------- | | |
340 | */ | |
5893dfb9 | 341 | if (args->start > key.offset && args->end < extent_end) { |
920bbbfb | 342 | BUG_ON(del_nr > 0); |
00fdf13a | 343 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 344 | ret = -EOPNOTSUPP; |
00fdf13a LB |
345 | break; |
346 | } | |
920bbbfb YZ |
347 | |
348 | memcpy(&new_key, &key, sizeof(new_key)); | |
5893dfb9 | 349 | new_key.offset = args->start; |
920bbbfb YZ |
350 | ret = btrfs_duplicate_item(trans, root, path, |
351 | &new_key); | |
352 | if (ret == -EAGAIN) { | |
b3b4aa74 | 353 | btrfs_release_path(path); |
920bbbfb | 354 | continue; |
6643558d | 355 | } |
920bbbfb YZ |
356 | if (ret < 0) |
357 | break; | |
358 | ||
359 | leaf = path->nodes[0]; | |
360 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
361 | struct btrfs_file_extent_item); | |
362 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 363 | args->start - key.offset); |
920bbbfb YZ |
364 | |
365 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
366 | struct btrfs_file_extent_item); | |
367 | ||
5893dfb9 | 368 | extent_offset += args->start - key.offset; |
920bbbfb YZ |
369 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
370 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 371 | extent_end - args->start); |
50564b65 | 372 | btrfs_mark_buffer_dirty(trans, leaf); |
920bbbfb | 373 | |
5dc562c5 | 374 | if (update_refs && disk_bytenr > 0) { |
4d09b4e9 JB |
375 | struct btrfs_ref ref = { |
376 | .action = BTRFS_ADD_DELAYED_REF, | |
377 | .bytenr = disk_bytenr, | |
12390e42 | 378 | .num_bytes = num_bytes, |
4d09b4e9 | 379 | .parent = 0, |
e094f480 JB |
380 | .owning_root = btrfs_root_id(root), |
381 | .ref_root = btrfs_root_id(root), | |
4d09b4e9 | 382 | }; |
f2e69a77 JB |
383 | btrfs_init_data_ref(&ref, new_key.objectid, |
384 | args->start - extent_offset, | |
385 | 0, false); | |
82fa113f | 386 | ret = btrfs_inc_extent_ref(trans, &ref); |
162d053e FM |
387 | if (ret) { |
388 | btrfs_abort_transaction(trans, ret); | |
389 | break; | |
390 | } | |
771ed689 | 391 | } |
5893dfb9 | 392 | key.offset = args->start; |
6643558d | 393 | } |
62fe51c1 JB |
394 | /* |
395 | * From here on out we will have actually dropped something, so | |
396 | * last_end can be updated. | |
397 | */ | |
398 | last_end = extent_end; | |
399 | ||
920bbbfb YZ |
400 | /* |
401 | * | ---- range to drop ----- | | |
402 | * | -------- extent -------- | | |
403 | */ | |
5893dfb9 | 404 | if (args->start <= key.offset && args->end < extent_end) { |
00fdf13a | 405 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 406 | ret = -EOPNOTSUPP; |
00fdf13a LB |
407 | break; |
408 | } | |
6643558d | 409 | |
920bbbfb | 410 | memcpy(&new_key, &key, sizeof(new_key)); |
5893dfb9 | 411 | new_key.offset = args->end; |
50564b65 | 412 | btrfs_set_item_key_safe(trans, path, &new_key); |
6643558d | 413 | |
5893dfb9 | 414 | extent_offset += args->end - key.offset; |
920bbbfb YZ |
415 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
416 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 417 | extent_end - args->end); |
50564b65 | 418 | btrfs_mark_buffer_dirty(trans, leaf); |
2671485d | 419 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 420 | args->bytes_found += args->end - key.offset; |
920bbbfb | 421 | break; |
39279cc3 | 422 | } |
771ed689 | 423 | |
920bbbfb YZ |
424 | search_start = extent_end; |
425 | /* | |
426 | * | ---- range to drop ----- | | |
427 | * | -------- extent -------- | | |
428 | */ | |
5893dfb9 | 429 | if (args->start > key.offset && args->end >= extent_end) { |
920bbbfb | 430 | BUG_ON(del_nr > 0); |
00fdf13a | 431 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 432 | ret = -EOPNOTSUPP; |
00fdf13a LB |
433 | break; |
434 | } | |
8c2383c3 | 435 | |
920bbbfb | 436 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5893dfb9 | 437 | args->start - key.offset); |
50564b65 | 438 | btrfs_mark_buffer_dirty(trans, leaf); |
2671485d | 439 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 440 | args->bytes_found += extent_end - args->start; |
5893dfb9 | 441 | if (args->end == extent_end) |
920bbbfb | 442 | break; |
c8b97818 | 443 | |
920bbbfb YZ |
444 | path->slots[0]++; |
445 | goto next_slot; | |
31840ae1 ZY |
446 | } |
447 | ||
920bbbfb YZ |
448 | /* |
449 | * | ---- range to drop ----- | | |
450 | * | ------ extent ------ | | |
451 | */ | |
5893dfb9 | 452 | if (args->start <= key.offset && args->end >= extent_end) { |
fc19c5e7 | 453 | delete_extent_item: |
920bbbfb YZ |
454 | if (del_nr == 0) { |
455 | del_slot = path->slots[0]; | |
456 | del_nr = 1; | |
457 | } else { | |
458 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
459 | del_nr++; | |
460 | } | |
31840ae1 | 461 | |
5dc562c5 JB |
462 | if (update_refs && |
463 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
2766ff61 | 464 | args->bytes_found += extent_end - key.offset; |
920bbbfb | 465 | extent_end = ALIGN(extent_end, |
0b246afa | 466 | fs_info->sectorsize); |
5dc562c5 | 467 | } else if (update_refs && disk_bytenr > 0) { |
4d09b4e9 JB |
468 | struct btrfs_ref ref = { |
469 | .action = BTRFS_DROP_DELAYED_REF, | |
470 | .bytenr = disk_bytenr, | |
12390e42 | 471 | .num_bytes = num_bytes, |
4d09b4e9 | 472 | .parent = 0, |
e094f480 JB |
473 | .owning_root = btrfs_root_id(root), |
474 | .ref_root = btrfs_root_id(root), | |
4d09b4e9 | 475 | }; |
f2e69a77 JB |
476 | btrfs_init_data_ref(&ref, key.objectid, |
477 | key.offset - extent_offset, | |
478 | 0, false); | |
ffd4bb2a | 479 | ret = btrfs_free_extent(trans, &ref); |
162d053e FM |
480 | if (ret) { |
481 | btrfs_abort_transaction(trans, ret); | |
482 | break; | |
483 | } | |
2766ff61 | 484 | args->bytes_found += extent_end - key.offset; |
31840ae1 | 485 | } |
31840ae1 | 486 | |
5893dfb9 | 487 | if (args->end == extent_end) |
920bbbfb YZ |
488 | break; |
489 | ||
490 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
491 | path->slots[0]++; | |
492 | goto next_slot; | |
493 | } | |
494 | ||
495 | ret = btrfs_del_items(trans, root, path, del_slot, | |
496 | del_nr); | |
79787eaa | 497 | if (ret) { |
66642832 | 498 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 499 | break; |
79787eaa | 500 | } |
920bbbfb YZ |
501 | |
502 | del_nr = 0; | |
503 | del_slot = 0; | |
504 | ||
b3b4aa74 | 505 | btrfs_release_path(path); |
920bbbfb | 506 | continue; |
39279cc3 | 507 | } |
920bbbfb | 508 | |
290342f6 | 509 | BUG(); |
39279cc3 | 510 | } |
920bbbfb | 511 | |
79787eaa | 512 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
513 | /* |
514 | * Set path->slots[0] to first slot, so that after the delete | |
515 | * if items are move off from our leaf to its immediate left or | |
516 | * right neighbor leafs, we end up with a correct and adjusted | |
5893dfb9 | 517 | * path->slots[0] for our insertion (if args->replace_extent). |
1acae57b FDBM |
518 | */ |
519 | path->slots[0] = del_slot; | |
920bbbfb | 520 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 521 | if (ret) |
66642832 | 522 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 523 | } |
1acae57b | 524 | |
d5f37527 FDBM |
525 | leaf = path->nodes[0]; |
526 | /* | |
527 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
528 | * which case it unlocked our path, so check path->locks[0] matches a | |
529 | * write lock. | |
530 | */ | |
7ecb4c31 | 531 | if (!ret && args->replace_extent && |
ac5887c8 | 532 | path->locks[0] == BTRFS_WRITE_LOCK && |
e902baac | 533 | btrfs_leaf_free_space(leaf) >= |
5893dfb9 | 534 | sizeof(struct btrfs_item) + args->extent_item_size) { |
d5f37527 FDBM |
535 | |
536 | key.objectid = ino; | |
537 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5893dfb9 | 538 | key.offset = args->start; |
d5f37527 FDBM |
539 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
540 | struct btrfs_key slot_key; | |
541 | ||
542 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
543 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
544 | path->slots[0]++; | |
1acae57b | 545 | } |
50564b65 FM |
546 | btrfs_setup_item_for_insert(trans, root, path, &key, |
547 | args->extent_item_size); | |
5893dfb9 | 548 | args->extent_inserted = true; |
6643558d | 549 | } |
920bbbfb | 550 | |
5893dfb9 FM |
551 | if (!args->path) |
552 | btrfs_free_path(path); | |
553 | else if (!args->extent_inserted) | |
1acae57b | 554 | btrfs_release_path(path); |
5893dfb9 FM |
555 | out: |
556 | args->drop_end = found ? min(args->end, last_end) : args->end; | |
5dc562c5 | 557 | |
39279cc3 CM |
558 | return ret; |
559 | } | |
560 | ||
d899e052 | 561 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
562 | u64 objectid, u64 bytenr, u64 orig_offset, |
563 | u64 *start, u64 *end) | |
d899e052 YZ |
564 | { |
565 | struct btrfs_file_extent_item *fi; | |
566 | struct btrfs_key key; | |
567 | u64 extent_end; | |
568 | ||
569 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
570 | return 0; | |
571 | ||
572 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
573 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
574 | return 0; | |
575 | ||
576 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
577 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
578 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 579 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
580 | btrfs_file_extent_compression(leaf, fi) || |
581 | btrfs_file_extent_encryption(leaf, fi) || | |
582 | btrfs_file_extent_other_encoding(leaf, fi)) | |
583 | return 0; | |
584 | ||
585 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
586 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
587 | return 0; | |
588 | ||
589 | *start = key.offset; | |
590 | *end = extent_end; | |
591 | return 1; | |
592 | } | |
593 | ||
594 | /* | |
595 | * Mark extent in the range start - end as written. | |
596 | * | |
597 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
598 | * part of extent is marked as written, the extent will be split into | |
599 | * two or three. | |
600 | */ | |
601 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 602 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 603 | { |
7a6d7067 | 604 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
605 | struct extent_buffer *leaf; |
606 | struct btrfs_path *path; | |
607 | struct btrfs_file_extent_item *fi; | |
82fa113f | 608 | struct btrfs_ref ref = { 0 }; |
d899e052 | 609 | struct btrfs_key key; |
920bbbfb | 610 | struct btrfs_key new_key; |
d899e052 YZ |
611 | u64 bytenr; |
612 | u64 num_bytes; | |
613 | u64 extent_end; | |
5d4f98a2 | 614 | u64 orig_offset; |
d899e052 YZ |
615 | u64 other_start; |
616 | u64 other_end; | |
920bbbfb YZ |
617 | u64 split; |
618 | int del_nr = 0; | |
619 | int del_slot = 0; | |
6c7d54ac | 620 | int recow; |
e7b2ec3d | 621 | int ret = 0; |
7a6d7067 | 622 | u64 ino = btrfs_ino(inode); |
d899e052 | 623 | |
d899e052 | 624 | path = btrfs_alloc_path(); |
d8926bb3 MF |
625 | if (!path) |
626 | return -ENOMEM; | |
d899e052 | 627 | again: |
6c7d54ac | 628 | recow = 0; |
920bbbfb | 629 | split = start; |
33345d01 | 630 | key.objectid = ino; |
d899e052 | 631 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 632 | key.offset = split; |
d899e052 YZ |
633 | |
634 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
635 | if (ret < 0) |
636 | goto out; | |
d899e052 YZ |
637 | if (ret > 0 && path->slots[0] > 0) |
638 | path->slots[0]--; | |
639 | ||
640 | leaf = path->nodes[0]; | |
641 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
642 | if (key.objectid != ino || |
643 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
644 | ret = -EINVAL; | |
645 | btrfs_abort_transaction(trans, ret); | |
646 | goto out; | |
647 | } | |
d899e052 YZ |
648 | fi = btrfs_item_ptr(leaf, path->slots[0], |
649 | struct btrfs_file_extent_item); | |
9c8e63db JB |
650 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
651 | ret = -EINVAL; | |
652 | btrfs_abort_transaction(trans, ret); | |
653 | goto out; | |
654 | } | |
d899e052 | 655 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
656 | if (key.offset > start || extent_end < end) { |
657 | ret = -EINVAL; | |
658 | btrfs_abort_transaction(trans, ret); | |
659 | goto out; | |
660 | } | |
d899e052 YZ |
661 | |
662 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
663 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 664 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
665 | memcpy(&new_key, &key, sizeof(new_key)); |
666 | ||
667 | if (start == key.offset && end < extent_end) { | |
668 | other_start = 0; | |
669 | other_end = start; | |
670 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 671 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
672 | &other_start, &other_end)) { |
673 | new_key.offset = end; | |
50564b65 | 674 | btrfs_set_item_key_safe(trans, path, &new_key); |
6c7d54ac YZ |
675 | fi = btrfs_item_ptr(leaf, path->slots[0], |
676 | struct btrfs_file_extent_item); | |
224ecce5 JB |
677 | btrfs_set_file_extent_generation(leaf, fi, |
678 | trans->transid); | |
6c7d54ac YZ |
679 | btrfs_set_file_extent_num_bytes(leaf, fi, |
680 | extent_end - end); | |
681 | btrfs_set_file_extent_offset(leaf, fi, | |
682 | end - orig_offset); | |
683 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
684 | struct btrfs_file_extent_item); | |
224ecce5 JB |
685 | btrfs_set_file_extent_generation(leaf, fi, |
686 | trans->transid); | |
6c7d54ac YZ |
687 | btrfs_set_file_extent_num_bytes(leaf, fi, |
688 | end - other_start); | |
50564b65 | 689 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac YZ |
690 | goto out; |
691 | } | |
692 | } | |
693 | ||
694 | if (start > key.offset && end == extent_end) { | |
695 | other_start = end; | |
696 | other_end = 0; | |
697 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 698 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
699 | &other_start, &other_end)) { |
700 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
701 | struct btrfs_file_extent_item); | |
702 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
703 | start - key.offset); | |
224ecce5 JB |
704 | btrfs_set_file_extent_generation(leaf, fi, |
705 | trans->transid); | |
6c7d54ac YZ |
706 | path->slots[0]++; |
707 | new_key.offset = start; | |
50564b65 | 708 | btrfs_set_item_key_safe(trans, path, &new_key); |
6c7d54ac YZ |
709 | |
710 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
711 | struct btrfs_file_extent_item); | |
224ecce5 JB |
712 | btrfs_set_file_extent_generation(leaf, fi, |
713 | trans->transid); | |
6c7d54ac YZ |
714 | btrfs_set_file_extent_num_bytes(leaf, fi, |
715 | other_end - start); | |
716 | btrfs_set_file_extent_offset(leaf, fi, | |
717 | start - orig_offset); | |
50564b65 | 718 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac YZ |
719 | goto out; |
720 | } | |
721 | } | |
d899e052 | 722 | |
920bbbfb YZ |
723 | while (start > key.offset || end < extent_end) { |
724 | if (key.offset == start) | |
725 | split = end; | |
726 | ||
920bbbfb YZ |
727 | new_key.offset = split; |
728 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
729 | if (ret == -EAGAIN) { | |
b3b4aa74 | 730 | btrfs_release_path(path); |
920bbbfb | 731 | goto again; |
d899e052 | 732 | } |
79787eaa | 733 | if (ret < 0) { |
66642832 | 734 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
735 | goto out; |
736 | } | |
d899e052 | 737 | |
920bbbfb YZ |
738 | leaf = path->nodes[0]; |
739 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 740 | struct btrfs_file_extent_item); |
224ecce5 | 741 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 742 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
743 | split - key.offset); |
744 | ||
745 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
746 | struct btrfs_file_extent_item); | |
747 | ||
224ecce5 | 748 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
749 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
750 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
751 | extent_end - split); | |
50564b65 | 752 | btrfs_mark_buffer_dirty(trans, leaf); |
d899e052 | 753 | |
4d09b4e9 JB |
754 | ref.action = BTRFS_ADD_DELAYED_REF; |
755 | ref.bytenr = bytenr; | |
12390e42 | 756 | ref.num_bytes = num_bytes; |
4d09b4e9 | 757 | ref.parent = 0; |
e094f480 JB |
758 | ref.owning_root = btrfs_root_id(root); |
759 | ref.ref_root = btrfs_root_id(root); | |
f2e69a77 | 760 | btrfs_init_data_ref(&ref, ino, orig_offset, 0, false); |
82fa113f | 761 | ret = btrfs_inc_extent_ref(trans, &ref); |
9c8e63db JB |
762 | if (ret) { |
763 | btrfs_abort_transaction(trans, ret); | |
764 | goto out; | |
765 | } | |
d899e052 | 766 | |
920bbbfb YZ |
767 | if (split == start) { |
768 | key.offset = start; | |
769 | } else { | |
9c8e63db JB |
770 | if (start != key.offset) { |
771 | ret = -EINVAL; | |
772 | btrfs_abort_transaction(trans, ret); | |
773 | goto out; | |
774 | } | |
d899e052 | 775 | path->slots[0]--; |
920bbbfb | 776 | extent_end = end; |
d899e052 | 777 | } |
6c7d54ac | 778 | recow = 1; |
d899e052 YZ |
779 | } |
780 | ||
920bbbfb YZ |
781 | other_start = end; |
782 | other_end = 0; | |
4d09b4e9 JB |
783 | |
784 | ref.action = BTRFS_DROP_DELAYED_REF; | |
785 | ref.bytenr = bytenr; | |
12390e42 | 786 | ref.num_bytes = num_bytes; |
4d09b4e9 | 787 | ref.parent = 0; |
e094f480 JB |
788 | ref.owning_root = btrfs_root_id(root); |
789 | ref.ref_root = btrfs_root_id(root); | |
f2e69a77 | 790 | btrfs_init_data_ref(&ref, ino, orig_offset, 0, false); |
6c7d54ac | 791 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 792 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
793 | &other_start, &other_end)) { |
794 | if (recow) { | |
b3b4aa74 | 795 | btrfs_release_path(path); |
6c7d54ac YZ |
796 | goto again; |
797 | } | |
920bbbfb YZ |
798 | extent_end = other_end; |
799 | del_slot = path->slots[0] + 1; | |
800 | del_nr++; | |
ffd4bb2a | 801 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
802 | if (ret) { |
803 | btrfs_abort_transaction(trans, ret); | |
804 | goto out; | |
805 | } | |
d899e052 | 806 | } |
920bbbfb YZ |
807 | other_start = 0; |
808 | other_end = start; | |
6c7d54ac | 809 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 810 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
811 | &other_start, &other_end)) { |
812 | if (recow) { | |
b3b4aa74 | 813 | btrfs_release_path(path); |
6c7d54ac YZ |
814 | goto again; |
815 | } | |
920bbbfb YZ |
816 | key.offset = other_start; |
817 | del_slot = path->slots[0]; | |
818 | del_nr++; | |
ffd4bb2a | 819 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
820 | if (ret) { |
821 | btrfs_abort_transaction(trans, ret); | |
822 | goto out; | |
823 | } | |
920bbbfb YZ |
824 | } |
825 | if (del_nr == 0) { | |
3f6fae95 SL |
826 | fi = btrfs_item_ptr(leaf, path->slots[0], |
827 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
828 | btrfs_set_file_extent_type(leaf, fi, |
829 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 830 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 831 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac | 832 | } else { |
3f6fae95 SL |
833 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
834 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
835 | btrfs_set_file_extent_type(leaf, fi, |
836 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 837 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
838 | btrfs_set_file_extent_num_bytes(leaf, fi, |
839 | extent_end - key.offset); | |
50564b65 | 840 | btrfs_mark_buffer_dirty(trans, leaf); |
920bbbfb | 841 | |
6c7d54ac | 842 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 843 | if (ret < 0) { |
66642832 | 844 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
845 | goto out; |
846 | } | |
6c7d54ac | 847 | } |
920bbbfb | 848 | out: |
d899e052 | 849 | btrfs_free_path(path); |
e7b2ec3d | 850 | return ret; |
d899e052 YZ |
851 | } |
852 | ||
b1bf862e CM |
853 | /* |
854 | * on error we return an unlocked page and the error value | |
855 | * on success we return a locked page and 0 | |
856 | */ | |
bb1591b4 CM |
857 | static int prepare_uptodate_page(struct inode *inode, |
858 | struct page *page, u64 pos, | |
b6316429 | 859 | bool force_uptodate) |
b1bf862e | 860 | { |
fb12489b | 861 | struct folio *folio = page_folio(page); |
b1bf862e CM |
862 | int ret = 0; |
863 | ||
09cbfeaf | 864 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 865 | !PageUptodate(page)) { |
fb12489b | 866 | ret = btrfs_read_folio(NULL, folio); |
b1bf862e CM |
867 | if (ret) |
868 | return ret; | |
869 | lock_page(page); | |
870 | if (!PageUptodate(page)) { | |
871 | unlock_page(page); | |
872 | return -EIO; | |
873 | } | |
e0467866 QW |
874 | |
875 | /* | |
fb12489b | 876 | * Since btrfs_read_folio() will unlock the folio before it |
f913cff3 | 877 | * returns, there is a window where btrfs_release_folio() can be |
7c11d0ae QW |
878 | * called to release the page. Here we check both inode |
879 | * mapping and PagePrivate() to make sure the page was not | |
880 | * released. | |
e0467866 QW |
881 | * |
882 | * The private flag check is essential for subpage as we need | |
cfbf07e2 | 883 | * to store extra bitmap using folio private. |
e0467866 | 884 | */ |
cfbf07e2 | 885 | if (page->mapping != inode->i_mapping || !folio_test_private(folio)) { |
bb1591b4 CM |
886 | unlock_page(page); |
887 | return -EAGAIN; | |
888 | } | |
b1bf862e CM |
889 | } |
890 | return 0; | |
891 | } | |
892 | ||
ffc143db | 893 | static fgf_t get_prepare_fgp_flags(bool nowait) |
fc226000 | 894 | { |
ffc143db | 895 | fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT; |
fc226000 SR |
896 | |
897 | if (nowait) | |
898 | fgp_flags |= FGP_NOWAIT; | |
899 | ||
900 | return fgp_flags; | |
901 | } | |
902 | ||
903 | static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait) | |
904 | { | |
905 | gfp_t gfp; | |
906 | ||
907 | gfp = btrfs_alloc_write_mask(inode->i_mapping); | |
908 | if (nowait) { | |
909 | gfp &= ~__GFP_DIRECT_RECLAIM; | |
910 | gfp |= GFP_NOWAIT; | |
911 | } | |
912 | ||
913 | return gfp; | |
914 | } | |
915 | ||
39279cc3 | 916 | /* |
376cc685 | 917 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 918 | */ |
b37392ea MX |
919 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
920 | size_t num_pages, loff_t pos, | |
fc226000 SR |
921 | size_t write_bytes, bool force_uptodate, |
922 | bool nowait) | |
39279cc3 CM |
923 | { |
924 | int i; | |
09cbfeaf | 925 | unsigned long index = pos >> PAGE_SHIFT; |
fc226000 | 926 | gfp_t mask = get_prepare_gfp_flags(inode, nowait); |
ffc143db | 927 | fgf_t fgp_flags = get_prepare_fgp_flags(nowait); |
aefee7f1 | 928 | int ret = 0; |
376cc685 | 929 | int faili; |
8c2383c3 | 930 | |
39279cc3 | 931 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 932 | again: |
fc226000 SR |
933 | pages[i] = pagecache_get_page(inode->i_mapping, index + i, |
934 | fgp_flags, mask | __GFP_WRITE); | |
39279cc3 | 935 | if (!pages[i]) { |
b1bf862e | 936 | faili = i - 1; |
fc226000 | 937 | if (nowait) |
aefee7f1 | 938 | ret = -EAGAIN; |
fc226000 | 939 | else |
aefee7f1 | 940 | ret = -ENOMEM; |
b1bf862e CM |
941 | goto fail; |
942 | } | |
943 | ||
aefee7f1 AJ |
944 | ret = set_page_extent_mapped(pages[i]); |
945 | if (ret < 0) { | |
32443de3 QW |
946 | faili = i; |
947 | goto fail; | |
948 | } | |
949 | ||
b1bf862e | 950 | if (i == 0) |
aefee7f1 | 951 | ret = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 952 | force_uptodate); |
aefee7f1 AJ |
953 | if (!ret && i == num_pages - 1) |
954 | ret = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 955 | pos + write_bytes, false); |
aefee7f1 | 956 | if (ret) { |
09cbfeaf | 957 | put_page(pages[i]); |
aefee7f1 AJ |
958 | if (!nowait && ret == -EAGAIN) { |
959 | ret = 0; | |
bb1591b4 CM |
960 | goto again; |
961 | } | |
b1bf862e CM |
962 | faili = i - 1; |
963 | goto fail; | |
39279cc3 | 964 | } |
ccd467d6 | 965 | wait_on_page_writeback(pages[i]); |
39279cc3 | 966 | } |
376cc685 MX |
967 | |
968 | return 0; | |
969 | fail: | |
970 | while (faili >= 0) { | |
971 | unlock_page(pages[faili]); | |
09cbfeaf | 972 | put_page(pages[faili]); |
376cc685 MX |
973 | faili--; |
974 | } | |
aefee7f1 | 975 | return ret; |
376cc685 MX |
976 | |
977 | } | |
978 | ||
979 | /* | |
980 | * This function locks the extent and properly waits for data=ordered extents | |
981 | * to finish before allowing the pages to be modified if need. | |
982 | * | |
983 | * The return value: | |
984 | * 1 - the extent is locked | |
985 | * 0 - the extent is not locked, and everything is OK | |
986 | * -EAGAIN - need re-prepare the pages | |
987 | * the other < 0 number - Something wrong happens | |
988 | */ | |
989 | static noinline int | |
2cff578c | 990 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 991 | size_t num_pages, loff_t pos, |
2e78c927 | 992 | size_t write_bytes, |
2fcab928 | 993 | u64 *lockstart, u64 *lockend, bool nowait, |
376cc685 MX |
994 | struct extent_state **cached_state) |
995 | { | |
3ffbd68c | 996 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
997 | u64 start_pos; |
998 | u64 last_pos; | |
999 | int i; | |
1000 | int ret = 0; | |
1001 | ||
0b246afa | 1002 | start_pos = round_down(pos, fs_info->sectorsize); |
e21139c6 | 1003 | last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1; |
376cc685 | 1004 | |
e3b8a485 | 1005 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 1006 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 1007 | |
2fcab928 | 1008 | if (nowait) { |
83ae4133 JB |
1009 | if (!try_lock_extent(&inode->io_tree, start_pos, last_pos, |
1010 | cached_state)) { | |
2fcab928 SR |
1011 | for (i = 0; i < num_pages; i++) { |
1012 | unlock_page(pages[i]); | |
1013 | put_page(pages[i]); | |
1014 | pages[i] = NULL; | |
1015 | } | |
1016 | ||
1017 | return -EAGAIN; | |
1018 | } | |
1019 | } else { | |
1020 | lock_extent(&inode->io_tree, start_pos, last_pos, cached_state); | |
1021 | } | |
1022 | ||
b88935bf MX |
1023 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1024 | last_pos - start_pos + 1); | |
e6dcd2dc | 1025 | if (ordered && |
bffe633e | 1026 | ordered->file_offset + ordered->num_bytes > start_pos && |
376cc685 | 1027 | ordered->file_offset <= last_pos) { |
570eb97b JB |
1028 | unlock_extent(&inode->io_tree, start_pos, last_pos, |
1029 | cached_state); | |
e6dcd2dc CM |
1030 | for (i = 0; i < num_pages; i++) { |
1031 | unlock_page(pages[i]); | |
09cbfeaf | 1032 | put_page(pages[i]); |
e6dcd2dc | 1033 | } |
36d45567 | 1034 | btrfs_start_ordered_extent(ordered); |
b88935bf MX |
1035 | btrfs_put_ordered_extent(ordered); |
1036 | return -EAGAIN; | |
e6dcd2dc CM |
1037 | } |
1038 | if (ordered) | |
1039 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1040 | |
376cc685 MX |
1041 | *lockstart = start_pos; |
1042 | *lockend = last_pos; | |
1043 | ret = 1; | |
0762704b | 1044 | } |
376cc685 | 1045 | |
7703bdd8 | 1046 | /* |
32443de3 QW |
1047 | * We should be called after prepare_pages() which should have locked |
1048 | * all pages in the range. | |
7703bdd8 | 1049 | */ |
32443de3 | 1050 | for (i = 0; i < num_pages; i++) |
e6dcd2dc | 1051 | WARN_ON(!PageLocked(pages[i])); |
b1bf862e | 1052 | |
376cc685 | 1053 | return ret; |
39279cc3 CM |
1054 | } |
1055 | ||
d7a8ab4e FM |
1056 | /* |
1057 | * Check if we can do nocow write into the range [@pos, @pos + @write_bytes) | |
1058 | * | |
1059 | * @pos: File offset. | |
1060 | * @write_bytes: The length to write, will be updated to the nocow writeable | |
1061 | * range. | |
1062 | * | |
1063 | * This function will flush ordered extents in the range to ensure proper | |
1064 | * nocow checks. | |
1065 | * | |
1066 | * Return: | |
1067 | * > 0 If we can nocow, and updates @write_bytes. | |
1068 | * 0 If we can't do a nocow write. | |
1069 | * -EAGAIN If we can't do a nocow write because snapshoting of the inode's | |
1070 | * root is in progress. | |
1071 | * < 0 If an error happened. | |
1072 | * | |
1073 | * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. | |
1074 | */ | |
1075 | int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, | |
80f9d241 | 1076 | size_t *write_bytes, bool nowait) |
7ee9e440 | 1077 | { |
3ffbd68c | 1078 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1079 | struct btrfs_root *root = inode->root; |
632ddfa2 | 1080 | struct extent_state *cached_state = NULL; |
7ee9e440 JB |
1081 | u64 lockstart, lockend; |
1082 | u64 num_bytes; | |
1083 | int ret; | |
1084 | ||
38d37aa9 QW |
1085 | if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) |
1086 | return 0; | |
1087 | ||
d7a8ab4e | 1088 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) |
5f791ec3 | 1089 | return -EAGAIN; |
8257b2dc | 1090 | |
0b246afa | 1091 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1092 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1093 | fs_info->sectorsize) - 1; |
5dbb75ed | 1094 | num_bytes = lockend - lockstart + 1; |
7ee9e440 | 1095 | |
80f9d241 | 1096 | if (nowait) { |
632ddfa2 JB |
1097 | if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend, |
1098 | &cached_state)) { | |
80f9d241 JB |
1099 | btrfs_drew_write_unlock(&root->snapshot_lock); |
1100 | return -EAGAIN; | |
1101 | } | |
1102 | } else { | |
632ddfa2 JB |
1103 | btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, |
1104 | &cached_state); | |
80f9d241 | 1105 | } |
85b7ab67 | 1106 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
80f9d241 JB |
1107 | NULL, NULL, NULL, nowait, false); |
1108 | if (ret <= 0) | |
d7a8ab4e | 1109 | btrfs_drew_write_unlock(&root->snapshot_lock); |
80f9d241 | 1110 | else |
c933956d MX |
1111 | *write_bytes = min_t(size_t, *write_bytes , |
1112 | num_bytes - pos + lockstart); | |
632ddfa2 | 1113 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
7ee9e440 JB |
1114 | |
1115 | return ret; | |
1116 | } | |
1117 | ||
38d37aa9 QW |
1118 | void btrfs_check_nocow_unlock(struct btrfs_inode *inode) |
1119 | { | |
1120 | btrfs_drew_write_unlock(&inode->root->snapshot_lock); | |
1121 | } | |
1122 | ||
efd34f03 CB |
1123 | static void update_time_for_write(struct inode *inode) |
1124 | { | |
b1c38a13 | 1125 | struct timespec64 now, ts; |
efd34f03 CB |
1126 | |
1127 | if (IS_NOCMTIME(inode)) | |
1128 | return; | |
1129 | ||
1130 | now = current_time(inode); | |
b1c38a13 JL |
1131 | ts = inode_get_mtime(inode); |
1132 | if (!timespec64_equal(&ts, &now)) | |
1133 | inode_set_mtime_to_ts(inode, now); | |
efd34f03 | 1134 | |
b1c38a13 JL |
1135 | ts = inode_get_ctime(inode); |
1136 | if (!timespec64_equal(&ts, &now)) | |
efd34f03 CB |
1137 | inode_set_ctime_to_ts(inode, now); |
1138 | ||
1139 | if (IS_I_VERSION(inode)) | |
1140 | inode_inc_iversion(inode); | |
1141 | } | |
1142 | ||
b8d8e1fd GR |
1143 | static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from, |
1144 | size_t count) | |
1145 | { | |
1146 | struct file *file = iocb->ki_filp; | |
1147 | struct inode *inode = file_inode(file); | |
41044b41 | 1148 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
b8d8e1fd GR |
1149 | loff_t pos = iocb->ki_pos; |
1150 | int ret; | |
1151 | loff_t oldsize; | |
1152 | loff_t start_pos; | |
1153 | ||
d7a8ab4e FM |
1154 | /* |
1155 | * Quickly bail out on NOWAIT writes if we don't have the nodatacow or | |
1156 | * prealloc flags, as without those flags we always have to COW. We will | |
1157 | * later check if we can really COW into the target range (using | |
1158 | * can_nocow_extent() at btrfs_get_blocks_direct_write()). | |
1159 | */ | |
1160 | if ((iocb->ki_flags & IOCB_NOWAIT) && | |
1161 | !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
1162 | return -EAGAIN; | |
b8d8e1fd | 1163 | |
b8d8e1fd GR |
1164 | ret = file_remove_privs(file); |
1165 | if (ret) | |
1166 | return ret; | |
1167 | ||
1168 | /* | |
1169 | * We reserve space for updating the inode when we reserve space for the | |
1170 | * extent we are going to write, so we will enospc out there. We don't | |
1171 | * need to start yet another transaction to update the inode as we will | |
1172 | * update the inode when we finish writing whatever data we write. | |
1173 | */ | |
efd34f03 | 1174 | update_time_for_write(inode); |
b8d8e1fd GR |
1175 | |
1176 | start_pos = round_down(pos, fs_info->sectorsize); | |
1177 | oldsize = i_size_read(inode); | |
1178 | if (start_pos > oldsize) { | |
1179 | /* Expand hole size to cover write data, preventing empty gap */ | |
1180 | loff_t end_pos = round_up(pos + count, fs_info->sectorsize); | |
1181 | ||
b06359a3 | 1182 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos); |
0d625446 | 1183 | if (ret) |
b8d8e1fd | 1184 | return ret; |
b8d8e1fd GR |
1185 | } |
1186 | ||
1187 | return 0; | |
1188 | } | |
1189 | ||
e4af400a GR |
1190 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1191 | struct iov_iter *i) | |
4b46fce2 | 1192 | { |
e4af400a | 1193 | struct file *file = iocb->ki_filp; |
c3523706 | 1194 | loff_t pos; |
496ad9aa | 1195 | struct inode *inode = file_inode(file); |
41044b41 | 1196 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
11c65dcc | 1197 | struct page **pages = NULL; |
364ecf36 | 1198 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1199 | u64 release_bytes = 0; |
376cc685 MX |
1200 | u64 lockstart; |
1201 | u64 lockend; | |
d0215f3e JB |
1202 | size_t num_written = 0; |
1203 | int nrptrs; | |
c3523706 | 1204 | ssize_t ret; |
7ee9e440 | 1205 | bool only_release_metadata = false; |
b6316429 | 1206 | bool force_page_uptodate = false; |
5e8b9ef3 | 1207 | loff_t old_isize = i_size_read(inode); |
c3523706 | 1208 | unsigned int ilock_flags = 0; |
304e45ac | 1209 | const bool nowait = (iocb->ki_flags & IOCB_NOWAIT); |
965f47ae | 1210 | unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0); |
c3523706 | 1211 | |
304e45ac | 1212 | if (nowait) |
c3523706 GR |
1213 | ilock_flags |= BTRFS_ILOCK_TRY; |
1214 | ||
29b6352b | 1215 | ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1216 | if (ret < 0) |
1217 | return ret; | |
4b46fce2 | 1218 | |
c3523706 GR |
1219 | ret = generic_write_checks(iocb, i); |
1220 | if (ret <= 0) | |
1221 | goto out; | |
1222 | ||
1223 | ret = btrfs_write_check(iocb, i, ret); | |
1224 | if (ret < 0) | |
1225 | goto out; | |
1226 | ||
1227 | pos = iocb->ki_pos; | |
09cbfeaf KS |
1228 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1229 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1230 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1231 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1232 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
c3523706 GR |
1233 | if (!pages) { |
1234 | ret = -ENOMEM; | |
1235 | goto out; | |
1236 | } | |
ab93dbec | 1237 | |
d0215f3e | 1238 | while (iov_iter_count(i) > 0) { |
c67d970f | 1239 | struct extent_state *cached_state = NULL; |
7073017a | 1240 | size_t offset = offset_in_page(pos); |
2e78c927 | 1241 | size_t sector_offset; |
d0215f3e | 1242 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1243 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1244 | offset); |
eefa45f5 | 1245 | size_t num_pages; |
7ee9e440 | 1246 | size_t reserve_bytes; |
d0215f3e JB |
1247 | size_t dirty_pages; |
1248 | size_t copied; | |
2e78c927 CR |
1249 | size_t dirty_sectors; |
1250 | size_t num_sectors; | |
79f015f2 | 1251 | int extents_locked; |
39279cc3 | 1252 | |
914ee295 XZ |
1253 | /* |
1254 | * Fault pages before locking them in prepare_pages | |
1255 | * to avoid recursive lock | |
1256 | */ | |
a6294593 | 1257 | if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) { |
914ee295 | 1258 | ret = -EFAULT; |
d0215f3e | 1259 | break; |
914ee295 XZ |
1260 | } |
1261 | ||
a0e248bb | 1262 | only_release_metadata = false; |
da17066c | 1263 | sector_offset = pos & (fs_info->sectorsize - 1); |
d9d8b2a5 | 1264 | |
364ecf36 | 1265 | extent_changeset_release(data_reserved); |
36ea6f3e NB |
1266 | ret = btrfs_check_data_free_space(BTRFS_I(inode), |
1267 | &data_reserved, pos, | |
304e45ac | 1268 | write_bytes, nowait); |
c6887cd1 | 1269 | if (ret < 0) { |
80f9d241 JB |
1270 | int can_nocow; |
1271 | ||
304e45ac SR |
1272 | if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) { |
1273 | ret = -EAGAIN; | |
1274 | break; | |
1275 | } | |
1276 | ||
eefa45f5 GR |
1277 | /* |
1278 | * If we don't have to COW at the offset, reserve | |
1279 | * metadata only. write_bytes may get smaller than | |
1280 | * requested here. | |
1281 | */ | |
80f9d241 | 1282 | can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos, |
304e45ac | 1283 | &write_bytes, nowait); |
80f9d241 JB |
1284 | if (can_nocow < 0) |
1285 | ret = can_nocow; | |
1286 | if (can_nocow > 0) | |
1287 | ret = 0; | |
1288 | if (ret) | |
c6887cd1 | 1289 | break; |
80f9d241 | 1290 | only_release_metadata = true; |
c6887cd1 | 1291 | } |
1832a6d5 | 1292 | |
eefa45f5 GR |
1293 | num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE); |
1294 | WARN_ON(num_pages > nrptrs); | |
1295 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1296 | fs_info->sectorsize); | |
8b62f87b | 1297 | WARN_ON(reserve_bytes == 0); |
9f3db423 | 1298 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
28c9b1e7 | 1299 | reserve_bytes, |
304e45ac | 1300 | reserve_bytes, nowait); |
7ee9e440 JB |
1301 | if (ret) { |
1302 | if (!only_release_metadata) | |
25ce28ca | 1303 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
bc42bda2 QW |
1304 | data_reserved, pos, |
1305 | write_bytes); | |
8257b2dc | 1306 | else |
38d37aa9 | 1307 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
a348c8d4 FM |
1308 | |
1309 | if (nowait && ret == -ENOSPC) | |
1310 | ret = -EAGAIN; | |
7ee9e440 JB |
1311 | break; |
1312 | } | |
1313 | ||
1314 | release_bytes = reserve_bytes; | |
376cc685 | 1315 | again: |
965f47ae | 1316 | ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags); |
eb81b682 FM |
1317 | if (ret) { |
1318 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); | |
965f47ae | 1319 | break; |
eb81b682 | 1320 | } |
965f47ae | 1321 | |
4a64001f JB |
1322 | /* |
1323 | * This is going to setup the pages array with the number of | |
1324 | * pages we want, so we don't really need to worry about the | |
1325 | * contents of pages from loop to loop | |
1326 | */ | |
b37392ea | 1327 | ret = prepare_pages(inode, pages, num_pages, |
fc226000 | 1328 | pos, write_bytes, force_page_uptodate, false); |
8b62f87b JB |
1329 | if (ret) { |
1330 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
8702ba93 | 1331 | reserve_bytes); |
d0215f3e | 1332 | break; |
8b62f87b | 1333 | } |
39279cc3 | 1334 | |
79f015f2 GR |
1335 | extents_locked = lock_and_cleanup_extent_if_need( |
1336 | BTRFS_I(inode), pages, | |
2cff578c | 1337 | num_pages, pos, write_bytes, &lockstart, |
304e45ac | 1338 | &lockend, nowait, &cached_state); |
79f015f2 | 1339 | if (extents_locked < 0) { |
304e45ac | 1340 | if (!nowait && extents_locked == -EAGAIN) |
376cc685 | 1341 | goto again; |
304e45ac | 1342 | |
8b62f87b | 1343 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
8702ba93 | 1344 | reserve_bytes); |
79f015f2 | 1345 | ret = extents_locked; |
376cc685 | 1346 | break; |
376cc685 MX |
1347 | } |
1348 | ||
ee22f0c4 | 1349 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1350 | |
0b246afa | 1351 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1352 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1353 | fs_info->sectorsize); |
1354 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1355 | |
b1bf862e CM |
1356 | /* |
1357 | * if we have trouble faulting in the pages, fall | |
1358 | * back to one page at a time | |
1359 | */ | |
1360 | if (copied < write_bytes) | |
1361 | nrptrs = 1; | |
1362 | ||
b6316429 JB |
1363 | if (copied == 0) { |
1364 | force_page_uptodate = true; | |
56244ef1 | 1365 | dirty_sectors = 0; |
b1bf862e | 1366 | dirty_pages = 0; |
b6316429 JB |
1367 | } else { |
1368 | force_page_uptodate = false; | |
ed6078f7 | 1369 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1370 | PAGE_SIZE); |
b6316429 | 1371 | } |
914ee295 | 1372 | |
2e78c927 | 1373 | if (num_sectors > dirty_sectors) { |
8b8b08cb | 1374 | /* release everything except the sectors we dirtied */ |
265fdfa6 | 1375 | release_bytes -= dirty_sectors << fs_info->sectorsize_bits; |
485290a7 | 1376 | if (only_release_metadata) { |
691fa059 | 1377 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1378 | release_bytes, true); |
485290a7 QW |
1379 | } else { |
1380 | u64 __pos; | |
1381 | ||
da17066c | 1382 | __pos = round_down(pos, |
0b246afa | 1383 | fs_info->sectorsize) + |
09cbfeaf | 1384 | (dirty_pages << PAGE_SHIFT); |
86d52921 | 1385 | btrfs_delalloc_release_space(BTRFS_I(inode), |
bc42bda2 | 1386 | data_reserved, __pos, |
43b18595 | 1387 | release_bytes, true); |
485290a7 | 1388 | } |
914ee295 XZ |
1389 | } |
1390 | ||
2e78c927 | 1391 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1392 | fs_info->sectorsize); |
376cc685 | 1393 | |
aa8c1a41 GR |
1394 | ret = btrfs_dirty_pages(BTRFS_I(inode), pages, |
1395 | dirty_pages, pos, copied, | |
1396 | &cached_state, only_release_metadata); | |
c67d970f FM |
1397 | |
1398 | /* | |
1399 | * If we have not locked the extent range, because the range's | |
1400 | * start offset is >= i_size, we might still have a non-NULL | |
1401 | * cached extent state, acquired while marking the extent range | |
1402 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1403 | * possible cached extent state to avoid a memory leak. | |
1404 | */ | |
79f015f2 | 1405 | if (extents_locked) |
570eb97b JB |
1406 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
1407 | lockend, &cached_state); | |
c67d970f FM |
1408 | else |
1409 | free_extent_state(cached_state); | |
1410 | ||
8702ba93 | 1411 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 | 1412 | if (ret) { |
e4f94347 | 1413 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
376cc685 | 1414 | break; |
f1de9683 | 1415 | } |
39279cc3 | 1416 | |
376cc685 | 1417 | release_bytes = 0; |
8257b2dc | 1418 | if (only_release_metadata) |
38d37aa9 | 1419 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
8257b2dc | 1420 | |
e4f94347 | 1421 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
f1de9683 | 1422 | |
d0215f3e JB |
1423 | cond_resched(); |
1424 | ||
914ee295 XZ |
1425 | pos += copied; |
1426 | num_written += copied; | |
d0215f3e | 1427 | } |
39279cc3 | 1428 | |
d0215f3e JB |
1429 | kfree(pages); |
1430 | ||
7ee9e440 | 1431 | if (release_bytes) { |
8257b2dc | 1432 | if (only_release_metadata) { |
38d37aa9 | 1433 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
691fa059 | 1434 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1435 | release_bytes, true); |
8257b2dc | 1436 | } else { |
86d52921 NB |
1437 | btrfs_delalloc_release_space(BTRFS_I(inode), |
1438 | data_reserved, | |
bc42bda2 | 1439 | round_down(pos, fs_info->sectorsize), |
43b18595 | 1440 | release_bytes, true); |
8257b2dc | 1441 | } |
7ee9e440 JB |
1442 | } |
1443 | ||
364ecf36 | 1444 | extent_changeset_free(data_reserved); |
5e8b9ef3 GR |
1445 | if (num_written > 0) { |
1446 | pagecache_isize_extended(inode, old_isize, iocb->ki_pos); | |
1447 | iocb->ki_pos += num_written; | |
1448 | } | |
c3523706 | 1449 | out: |
e5d4d75b | 1450 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
d0215f3e JB |
1451 | return num_written ? num_written : ret; |
1452 | } | |
1453 | ||
4e4cabec GR |
1454 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
1455 | const struct iov_iter *iter, loff_t offset) | |
1456 | { | |
1457 | const u32 blocksize_mask = fs_info->sectorsize - 1; | |
1458 | ||
1459 | if (offset & blocksize_mask) | |
1460 | return -EINVAL; | |
1461 | ||
1462 | if (iov_iter_alignment(iter) & blocksize_mask) | |
1463 | return -EINVAL; | |
1464 | ||
1465 | return 0; | |
1466 | } | |
1467 | ||
1468 | static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
d0215f3e JB |
1469 | { |
1470 | struct file *file = iocb->ki_filp; | |
728404da | 1471 | struct inode *inode = file_inode(file); |
41044b41 | 1472 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
c3523706 | 1473 | loff_t pos; |
4e4cabec | 1474 | ssize_t written = 0; |
d0215f3e | 1475 | ssize_t written_buffered; |
51bd9563 | 1476 | size_t prev_left = 0; |
d0215f3e | 1477 | loff_t endbyte; |
70f1e5b6 | 1478 | ssize_t ret; |
c3523706 | 1479 | unsigned int ilock_flags = 0; |
8184620a | 1480 | struct iomap_dio *dio; |
c3523706 GR |
1481 | |
1482 | if (iocb->ki_flags & IOCB_NOWAIT) | |
1483 | ilock_flags |= BTRFS_ILOCK_TRY; | |
1484 | ||
9af86694 BS |
1485 | /* |
1486 | * If the write DIO is within EOF, use a shared lock and also only if | |
1487 | * security bits will likely not be dropped by file_remove_privs() called | |
1488 | * from btrfs_write_check(). Either will need to be rechecked after the | |
1489 | * lock was acquired. | |
1490 | */ | |
1491 | if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode)) | |
e9adabb9 GR |
1492 | ilock_flags |= BTRFS_ILOCK_SHARED; |
1493 | ||
1494 | relock: | |
70f1e5b6 AJ |
1495 | ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
1496 | if (ret < 0) | |
1497 | return ret; | |
c3523706 | 1498 | |
9af86694 BS |
1499 | /* Shared lock cannot be used with security bits set. */ |
1500 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) { | |
1501 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); | |
1502 | ilock_flags &= ~BTRFS_ILOCK_SHARED; | |
1503 | goto relock; | |
1504 | } | |
1505 | ||
70f1e5b6 AJ |
1506 | ret = generic_write_checks(iocb, from); |
1507 | if (ret <= 0) { | |
e5d4d75b | 1508 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
70f1e5b6 | 1509 | return ret; |
c3523706 | 1510 | } |
d0215f3e | 1511 | |
70f1e5b6 AJ |
1512 | ret = btrfs_write_check(iocb, from, ret); |
1513 | if (ret < 0) { | |
e5d4d75b | 1514 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1515 | goto out; |
1516 | } | |
1517 | ||
1518 | pos = iocb->ki_pos; | |
e9adabb9 GR |
1519 | /* |
1520 | * Re-check since file size may have changed just before taking the | |
1521 | * lock or pos may have changed because of O_APPEND in generic_write_check() | |
1522 | */ | |
1523 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && | |
1524 | pos + iov_iter_count(from) > i_size_read(inode)) { | |
e5d4d75b | 1525 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
e9adabb9 GR |
1526 | ilock_flags &= ~BTRFS_ILOCK_SHARED; |
1527 | goto relock; | |
1528 | } | |
c3523706 GR |
1529 | |
1530 | if (check_direct_IO(fs_info, from, pos)) { | |
e5d4d75b | 1531 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
4e4cabec | 1532 | goto buffered; |
c3523706 | 1533 | } |
4e4cabec | 1534 | |
51bd9563 FM |
1535 | /* |
1536 | * The iov_iter can be mapped to the same file range we are writing to. | |
1537 | * If that's the case, then we will deadlock in the iomap code, because | |
1538 | * it first calls our callback btrfs_dio_iomap_begin(), which will create | |
1539 | * an ordered extent, and after that it will fault in the pages that the | |
1540 | * iov_iter refers to. During the fault in we end up in the readahead | |
1541 | * pages code (starting at btrfs_readahead()), which will lock the range, | |
1542 | * find that ordered extent and then wait for it to complete (at | |
1543 | * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since | |
1544 | * obviously the ordered extent can never complete as we didn't submit | |
1545 | * yet the respective bio(s). This always happens when the buffer is | |
1546 | * memory mapped to the same file range, since the iomap DIO code always | |
1547 | * invalidates pages in the target file range (after starting and waiting | |
1548 | * for any writeback). | |
1549 | * | |
1550 | * So here we disable page faults in the iov_iter and then retry if we | |
1551 | * got -EFAULT, faulting in the pages before the retry. | |
1552 | */ | |
51bd9563 | 1553 | from->nofault = true; |
8184620a | 1554 | dio = btrfs_dio_write(iocb, from, written); |
51bd9563 | 1555 | from->nofault = false; |
d0215f3e | 1556 | |
8184620a FM |
1557 | /* |
1558 | * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync | |
1559 | * iocb, and that needs to lock the inode. So unlock it before calling | |
1560 | * iomap_dio_complete() to avoid a deadlock. | |
1561 | */ | |
e5d4d75b | 1562 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
8184620a FM |
1563 | |
1564 | if (IS_ERR_OR_NULL(dio)) | |
70f1e5b6 | 1565 | ret = PTR_ERR_OR_ZERO(dio); |
8184620a | 1566 | else |
70f1e5b6 | 1567 | ret = iomap_dio_complete(dio); |
8184620a | 1568 | |
51bd9563 | 1569 | /* No increment (+=) because iomap returns a cumulative value. */ |
70f1e5b6 AJ |
1570 | if (ret > 0) |
1571 | written = ret; | |
51bd9563 | 1572 | |
70f1e5b6 | 1573 | if (iov_iter_count(from) > 0 && (ret == -EFAULT || ret > 0)) { |
51bd9563 FM |
1574 | const size_t left = iov_iter_count(from); |
1575 | /* | |
1576 | * We have more data left to write. Try to fault in as many as | |
1577 | * possible of the remainder pages and retry. We do this without | |
1578 | * releasing and locking again the inode, to prevent races with | |
1579 | * truncate. | |
1580 | * | |
1581 | * Also, in case the iov refers to pages in the file range of the | |
1582 | * file we want to write to (due to a mmap), we could enter an | |
1583 | * infinite loop if we retry after faulting the pages in, since | |
1584 | * iomap will invalidate any pages in the range early on, before | |
1585 | * it tries to fault in the pages of the iov. So we keep track of | |
1586 | * how much was left of iov in the previous EFAULT and fallback | |
1587 | * to buffered IO in case we haven't made any progress. | |
1588 | */ | |
1589 | if (left == prev_left) { | |
70f1e5b6 | 1590 | ret = -ENOTBLK; |
51bd9563 FM |
1591 | } else { |
1592 | fault_in_iov_iter_readable(from, left); | |
1593 | prev_left = left; | |
8184620a | 1594 | goto relock; |
51bd9563 | 1595 | } |
a42fa643 GR |
1596 | } |
1597 | ||
ac5e6669 | 1598 | /* |
70f1e5b6 | 1599 | * If 'ret' is -ENOTBLK or we have not written all data, then it means |
ac5e6669 FM |
1600 | * we must fallback to buffered IO. |
1601 | */ | |
70f1e5b6 | 1602 | if ((ret < 0 && ret != -ENOTBLK) || !iov_iter_count(from)) |
c3523706 | 1603 | goto out; |
d0215f3e | 1604 | |
4e4cabec | 1605 | buffered: |
ac5e6669 FM |
1606 | /* |
1607 | * If we are in a NOWAIT context, then return -EAGAIN to signal the caller | |
1608 | * it must retry the operation in a context where blocking is acceptable, | |
20af93d9 FM |
1609 | * because even if we end up not blocking during the buffered IO attempt |
1610 | * below, we will block when flushing and waiting for the IO. | |
ac5e6669 FM |
1611 | */ |
1612 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
70f1e5b6 | 1613 | ret = -EAGAIN; |
ac5e6669 FM |
1614 | goto out; |
1615 | } | |
1616 | ||
e4af400a GR |
1617 | pos = iocb->ki_pos; |
1618 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e | 1619 | if (written_buffered < 0) { |
70f1e5b6 | 1620 | ret = written_buffered; |
d0215f3e | 1621 | goto out; |
39279cc3 | 1622 | } |
075bdbdb FM |
1623 | /* |
1624 | * Ensure all data is persisted. We want the next direct IO read to be | |
1625 | * able to read what was just written. | |
1626 | */ | |
d0215f3e | 1627 | endbyte = pos + written_buffered - 1; |
70f1e5b6 AJ |
1628 | ret = btrfs_fdatawrite_range(inode, pos, endbyte); |
1629 | if (ret) | |
075bdbdb | 1630 | goto out; |
70f1e5b6 AJ |
1631 | ret = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
1632 | if (ret) | |
d0215f3e JB |
1633 | goto out; |
1634 | written += written_buffered; | |
867c4f93 | 1635 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1636 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1637 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1638 | out: |
70f1e5b6 | 1639 | return ret < 0 ? ret : written; |
d0215f3e | 1640 | } |
5b92ee72 | 1641 | |
7c0c7269 OS |
1642 | static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
1643 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
1644 | { | |
1645 | struct file *file = iocb->ki_filp; | |
1646 | struct inode *inode = file_inode(file); | |
1647 | loff_t count; | |
1648 | ssize_t ret; | |
1649 | ||
29b6352b | 1650 | btrfs_inode_lock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1651 | count = encoded->len; |
1652 | ret = generic_write_checks_count(iocb, &count); | |
1653 | if (ret == 0 && count != encoded->len) { | |
1654 | /* | |
1655 | * The write got truncated by generic_write_checks_count(). We | |
1656 | * can't do a partial encoded write. | |
1657 | */ | |
1658 | ret = -EFBIG; | |
1659 | } | |
1660 | if (ret || encoded->len == 0) | |
1661 | goto out; | |
1662 | ||
1663 | ret = btrfs_write_check(iocb, from, encoded->len); | |
1664 | if (ret < 0) | |
1665 | goto out; | |
1666 | ||
1667 | ret = btrfs_do_encoded_write(iocb, from, encoded); | |
1668 | out: | |
e5d4d75b | 1669 | btrfs_inode_unlock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1670 | return ret; |
1671 | } | |
1672 | ||
1673 | ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from, | |
1674 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
d0215f3e JB |
1675 | { |
1676 | struct file *file = iocb->ki_filp; | |
14971657 | 1677 | struct btrfs_inode *inode = BTRFS_I(file_inode(file)); |
7c0c7269 | 1678 | ssize_t num_written, num_sync; |
d0215f3e | 1679 | |
c86537a4 GR |
1680 | /* |
1681 | * If the fs flips readonly due to some impossible error, although we | |
1682 | * have opened a file as writable, we have to stop this write operation | |
1683 | * to ensure consistency. | |
1684 | */ | |
84961539 | 1685 | if (BTRFS_FS_ERROR(inode->root->fs_info)) |
c86537a4 GR |
1686 | return -EROFS; |
1687 | ||
926078b2 | 1688 | if (encoded && (iocb->ki_flags & IOCB_NOWAIT)) |
91f9943e CH |
1689 | return -EOPNOTSUPP; |
1690 | ||
7c0c7269 OS |
1691 | if (encoded) { |
1692 | num_written = btrfs_encoded_write(iocb, from, encoded); | |
1693 | num_sync = encoded->len; | |
1694 | } else if (iocb->ki_flags & IOCB_DIRECT) { | |
c1867eb3 DS |
1695 | num_written = btrfs_direct_write(iocb, from); |
1696 | num_sync = num_written; | |
7c0c7269 | 1697 | } else { |
c1867eb3 DS |
1698 | num_written = btrfs_buffered_write(iocb, from); |
1699 | num_sync = num_written; | |
7c0c7269 | 1700 | } |
d0215f3e | 1701 | |
bc0939fc FM |
1702 | btrfs_set_inode_last_sub_trans(inode); |
1703 | ||
7c0c7269 OS |
1704 | if (num_sync > 0) { |
1705 | num_sync = generic_write_sync(iocb, num_sync); | |
1706 | if (num_sync < 0) | |
1707 | num_written = num_sync; | |
1708 | } | |
0a3404dc | 1709 | |
c3523706 | 1710 | return num_written; |
39279cc3 CM |
1711 | } |
1712 | ||
7c0c7269 OS |
1713 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1714 | { | |
1715 | return btrfs_do_write_iter(iocb, from, NULL); | |
1716 | } | |
1717 | ||
d397712b | 1718 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1719 | { |
23b5ec74 JB |
1720 | struct btrfs_file_private *private = filp->private_data; |
1721 | ||
3c32c721 | 1722 | if (private) { |
23b5ec74 | 1723 | kfree(private->filldir_buf); |
3c32c721 FM |
1724 | free_extent_state(private->llseek_cached_state); |
1725 | kfree(private); | |
1726 | filp->private_data = NULL; | |
1727 | } | |
23b5ec74 | 1728 | |
f6dc45c7 | 1729 | /* |
1fd4033d NB |
1730 | * Set by setattr when we are about to truncate a file from a non-zero |
1731 | * size to a zero size. This tries to flush down new bytes that may | |
1732 | * have been written if the application were using truncate to replace | |
1733 | * a file in place. | |
f6dc45c7 | 1734 | */ |
1fd4033d | 1735 | if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
f6dc45c7 CM |
1736 | &BTRFS_I(inode)->runtime_flags)) |
1737 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
1738 | return 0; |
1739 | } | |
1740 | ||
669249ee FM |
1741 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
1742 | { | |
1743 | int ret; | |
343e4fc1 | 1744 | struct blk_plug plug; |
669249ee | 1745 | |
343e4fc1 LB |
1746 | /* |
1747 | * This is only called in fsync, which would do synchronous writes, so | |
1748 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
1749 | * multiple disks using raid profile, a large IO can be split to | |
1750 | * several segments of stripe length (currently 64K). | |
1751 | */ | |
1752 | blk_start_plug(&plug); | |
728404da | 1753 | ret = btrfs_fdatawrite_range(inode, start, end); |
343e4fc1 | 1754 | blk_finish_plug(&plug); |
669249ee FM |
1755 | |
1756 | return ret; | |
1757 | } | |
1758 | ||
626e9f41 FM |
1759 | static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx) |
1760 | { | |
1761 | struct btrfs_inode *inode = BTRFS_I(ctx->inode); | |
1762 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1763 | ||
4a4f8fe2 | 1764 | if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) && |
626e9f41 FM |
1765 | list_empty(&ctx->ordered_extents)) |
1766 | return true; | |
1767 | ||
1768 | /* | |
1769 | * If we are doing a fast fsync we can not bail out if the inode's | |
1770 | * last_trans is <= then the last committed transaction, because we only | |
1771 | * update the last_trans of the inode during ordered extent completion, | |
1772 | * and for a fast fsync we don't wait for that, we only wait for the | |
1773 | * writeback to complete. | |
1774 | */ | |
0124855f | 1775 | if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) && |
626e9f41 FM |
1776 | (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) || |
1777 | list_empty(&ctx->ordered_extents))) | |
1778 | return true; | |
1779 | ||
1780 | return false; | |
1781 | } | |
1782 | ||
d352ac68 CM |
1783 | /* |
1784 | * fsync call for both files and directories. This logs the inode into | |
1785 | * the tree log instead of forcing full commits whenever possible. | |
1786 | * | |
1787 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1788 | * in the metadata btree are up to date for copying to the log. | |
1789 | * | |
1790 | * It drops the inode mutex before doing the tree log commit. This is an | |
1791 | * important optimization for directories because holding the mutex prevents | |
1792 | * new operations on the dir while we write to disk. | |
1793 | */ | |
02c24a82 | 1794 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1795 | { |
de17e793 | 1796 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 1797 | struct inode *inode = d_inode(dentry); |
41044b41 | 1798 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
39279cc3 | 1799 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 1800 | struct btrfs_trans_handle *trans; |
8b050d35 | 1801 | struct btrfs_log_ctx ctx; |
333427a5 | 1802 | int ret = 0, err; |
48778179 FM |
1803 | u64 len; |
1804 | bool full_sync; | |
39279cc3 | 1805 | |
1abe9b8a | 1806 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1807 | |
ebb70442 LB |
1808 | btrfs_init_log_ctx(&ctx, inode); |
1809 | ||
95418ed1 | 1810 | /* |
48778179 FM |
1811 | * Always set the range to a full range, otherwise we can get into |
1812 | * several problems, from missing file extent items to represent holes | |
1813 | * when not using the NO_HOLES feature, to log tree corruption due to | |
1814 | * races between hole detection during logging and completion of ordered | |
1815 | * extents outside the range, to missing checksums due to ordered extents | |
1816 | * for which we flushed only a subset of their pages. | |
95418ed1 | 1817 | */ |
48778179 FM |
1818 | start = 0; |
1819 | end = LLONG_MAX; | |
1820 | len = (u64)LLONG_MAX + 1; | |
95418ed1 | 1821 | |
90abccf2 MX |
1822 | /* |
1823 | * We write the dirty pages in the range and wait until they complete | |
1824 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1825 | * multi-task, and make the performance up. See |
1826 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1827 | */ |
669249ee | 1828 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 1829 | if (ret) |
333427a5 | 1830 | goto out; |
90abccf2 | 1831 | |
29b6352b | 1832 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
c495144b | 1833 | |
2ecb7923 | 1834 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 1835 | |
aab15e8e | 1836 | /* |
885f46d8 FM |
1837 | * Before we acquired the inode's lock and the mmap lock, someone may |
1838 | * have dirtied more pages in the target range. We need to make sure | |
1839 | * that writeback for any such pages does not start while we are logging | |
1840 | * the inode, because if it does, any of the following might happen when | |
1841 | * we are not doing a full inode sync: | |
aab15e8e FM |
1842 | * |
1843 | * 1) We log an extent after its writeback finishes but before its | |
1844 | * checksums are added to the csum tree, leading to -EIO errors | |
1845 | * when attempting to read the extent after a log replay. | |
1846 | * | |
1847 | * 2) We can end up logging an extent before its writeback finishes. | |
1848 | * Therefore after the log replay we will have a file extent item | |
1849 | * pointing to an unwritten extent (and no data checksums as well). | |
1850 | * | |
1851 | * So trigger writeback for any eventual new dirty pages and then we | |
1852 | * wait for all ordered extents to complete below. | |
1853 | */ | |
1854 | ret = start_ordered_ops(inode, start, end); | |
1855 | if (ret) { | |
e5d4d75b | 1856 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
aab15e8e FM |
1857 | goto out; |
1858 | } | |
1859 | ||
cef7820d FM |
1860 | /* |
1861 | * Always check for the full sync flag while holding the inode's lock, | |
1862 | * to avoid races with other tasks. The flag must be either set all the | |
1863 | * time during logging or always off all the time while logging. | |
1864 | * We check the flag here after starting delalloc above, because when | |
1865 | * running delalloc the full sync flag may be set if we need to drop | |
1866 | * extra extent map ranges due to temporary memory allocation failures. | |
1867 | */ | |
1868 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1869 | &BTRFS_I(inode)->runtime_flags); | |
1870 | ||
669249ee | 1871 | /* |
b5e6c3e1 | 1872 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 1873 | * IO of a lower priority task while holding a transaction open. |
ba0b084a | 1874 | * |
48778179 FM |
1875 | * For a full fsync we wait for the ordered extents to complete while |
1876 | * for a fast fsync we wait just for writeback to complete, and then | |
1877 | * attach the ordered extents to the transaction so that a transaction | |
1878 | * commit waits for their completion, to avoid data loss if we fsync, | |
1879 | * the current transaction commits before the ordered extents complete | |
1880 | * and a power failure happens right after that. | |
d8e3fb10 NA |
1881 | * |
1882 | * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the | |
1883 | * logical address recorded in the ordered extent may change. We need | |
1884 | * to wait for the IO to stabilize the logical address. | |
669249ee | 1885 | */ |
d8e3fb10 | 1886 | if (full_sync || btrfs_is_zoned(fs_info)) { |
48778179 FM |
1887 | ret = btrfs_wait_ordered_range(inode, start, len); |
1888 | } else { | |
1889 | /* | |
1890 | * Get our ordered extents as soon as possible to avoid doing | |
1891 | * checksum lookups in the csum tree, and use instead the | |
1892 | * checksums attached to the ordered extents. | |
1893 | */ | |
1894 | btrfs_get_ordered_extents_for_logging(BTRFS_I(inode), | |
1895 | &ctx.ordered_extents); | |
1896 | ret = filemap_fdatawait_range(inode->i_mapping, start, end); | |
0ef8b726 | 1897 | } |
48778179 FM |
1898 | |
1899 | if (ret) | |
1900 | goto out_release_extents; | |
1901 | ||
2ecb7923 | 1902 | atomic_inc(&root->log_batch); |
257c62e1 | 1903 | |
626e9f41 | 1904 | if (skip_inode_logging(&ctx)) { |
5dc562c5 | 1905 | /* |
01327610 | 1906 | * We've had everything committed since the last time we were |
5dc562c5 JB |
1907 | * modified so clear this flag in case it was set for whatever |
1908 | * reason, it's no longer relevant. | |
1909 | */ | |
1910 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1911 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
1912 | /* |
1913 | * An ordered extent might have started before and completed | |
1914 | * already with io errors, in which case the inode was not | |
1915 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
1916 | * for any errors that might have happened since we last |
1917 | * checked called fsync. | |
0596a904 | 1918 | */ |
333427a5 | 1919 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
48778179 | 1920 | goto out_release_extents; |
15ee9bc7 | 1921 | } |
15ee9bc7 | 1922 | |
e383e158 FM |
1923 | btrfs_init_log_ctx_scratch_eb(&ctx); |
1924 | ||
5039eddc JB |
1925 | /* |
1926 | * We use start here because we will need to wait on the IO to complete | |
1927 | * in btrfs_sync_log, which could require joining a transaction (for | |
1928 | * example checking cross references in the nocow path). If we use join | |
1929 | * here we could get into a situation where we're waiting on IO to | |
1930 | * happen that is blocked on a transaction trying to commit. With start | |
1931 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 1932 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
1933 | * from thinking they are super smart and changing this to |
1934 | * btrfs_join_transaction *cough*Josef*cough*. | |
1935 | */ | |
a22285a6 YZ |
1936 | trans = btrfs_start_transaction(root, 0); |
1937 | if (IS_ERR(trans)) { | |
1938 | ret = PTR_ERR(trans); | |
48778179 | 1939 | goto out_release_extents; |
39279cc3 | 1940 | } |
d0c2f4fa | 1941 | trans->in_fsync = true; |
e02119d5 | 1942 | |
48778179 | 1943 | ret = btrfs_log_dentry_safe(trans, dentry, &ctx); |
e383e158 FM |
1944 | /* |
1945 | * Scratch eb no longer needed, release before syncing log or commit | |
1946 | * transaction, to avoid holding unnecessary memory during such long | |
1947 | * operations. | |
1948 | */ | |
1949 | if (ctx.scratch_eb) { | |
1950 | free_extent_buffer(ctx.scratch_eb); | |
1951 | ctx.scratch_eb = NULL; | |
1952 | } | |
48778179 | 1953 | btrfs_release_log_ctx_extents(&ctx); |
02c24a82 | 1954 | if (ret < 0) { |
a0634be5 | 1955 | /* Fallthrough and commit/free transaction. */ |
f31f09f6 | 1956 | ret = BTRFS_LOG_FORCE_COMMIT; |
02c24a82 | 1957 | } |
49eb7e46 CM |
1958 | |
1959 | /* we've logged all the items and now have a consistent | |
1960 | * version of the file in the log. It is possible that | |
1961 | * someone will come in and modify the file, but that's | |
1962 | * fine because the log is consistent on disk, and we | |
1963 | * have references to all of the file's extents | |
1964 | * | |
1965 | * It is possible that someone will come in and log the | |
1966 | * file again, but that will end up using the synchronization | |
1967 | * inside btrfs_sync_log to keep things safe. | |
1968 | */ | |
e5d4d75b | 1969 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
49eb7e46 | 1970 | |
bf7ba8ee JB |
1971 | if (ret == BTRFS_NO_LOG_SYNC) { |
1972 | ret = btrfs_end_transaction(trans); | |
1973 | goto out; | |
1974 | } | |
1975 | ||
1976 | /* We successfully logged the inode, attempt to sync the log. */ | |
1977 | if (!ret) { | |
1978 | ret = btrfs_sync_log(trans, root, &ctx); | |
0ef8b726 | 1979 | if (!ret) { |
bf7ba8ee JB |
1980 | ret = btrfs_end_transaction(trans); |
1981 | goto out; | |
48778179 | 1982 | } |
bf7ba8ee JB |
1983 | } |
1984 | ||
1985 | /* | |
1986 | * At this point we need to commit the transaction because we had | |
1987 | * btrfs_need_log_full_commit() or some other error. | |
1988 | * | |
1989 | * If we didn't do a full sync we have to stop the trans handle, wait on | |
1990 | * the ordered extents, start it again and commit the transaction. If | |
1991 | * we attempt to wait on the ordered extents here we could deadlock with | |
1992 | * something like fallocate() that is holding the extent lock trying to | |
1993 | * start a transaction while some other thread is trying to commit the | |
1994 | * transaction while we (fsync) are currently holding the transaction | |
1995 | * open. | |
1996 | */ | |
1997 | if (!full_sync) { | |
3a45bb20 | 1998 | ret = btrfs_end_transaction(trans); |
bf7ba8ee JB |
1999 | if (ret) |
2000 | goto out; | |
2001 | ret = btrfs_wait_ordered_range(inode, start, len); | |
2002 | if (ret) | |
2003 | goto out; | |
2004 | ||
2005 | /* | |
2006 | * This is safe to use here because we're only interested in | |
2007 | * making sure the transaction that had the ordered extents is | |
2008 | * committed. We aren't waiting on anything past this point, | |
2009 | * we're purely getting the transaction and committing it. | |
2010 | */ | |
2011 | trans = btrfs_attach_transaction_barrier(root); | |
2012 | if (IS_ERR(trans)) { | |
2013 | ret = PTR_ERR(trans); | |
2014 | ||
2015 | /* | |
2016 | * We committed the transaction and there's no currently | |
2017 | * running transaction, this means everything we care | |
2018 | * about made it to disk and we are done. | |
2019 | */ | |
2020 | if (ret == -ENOENT) | |
2021 | ret = 0; | |
2022 | goto out; | |
2023 | } | |
e02119d5 | 2024 | } |
bf7ba8ee JB |
2025 | |
2026 | ret = btrfs_commit_transaction(trans); | |
39279cc3 | 2027 | out: |
e383e158 | 2028 | free_extent_buffer(ctx.scratch_eb); |
ebb70442 | 2029 | ASSERT(list_empty(&ctx.list)); |
e09d94c9 | 2030 | ASSERT(list_empty(&ctx.conflict_inodes)); |
333427a5 JL |
2031 | err = file_check_and_advance_wb_err(file); |
2032 | if (!ret) | |
2033 | ret = err; | |
014e4ac4 | 2034 | return ret > 0 ? -EIO : ret; |
48778179 FM |
2035 | |
2036 | out_release_extents: | |
2037 | btrfs_release_log_ctx_extents(&ctx); | |
e5d4d75b | 2038 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
48778179 | 2039 | goto out; |
39279cc3 CM |
2040 | } |
2041 | ||
0ddefc2a FM |
2042 | /* |
2043 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
2044 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2045 | * be careful to check for EOF conditions here. We set the page up correctly | |
2046 | * for a written page which means we get ENOSPC checking when writing into | |
2047 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2048 | * support these features. | |
2049 | * | |
2050 | * We are not allowed to take the i_mutex here so we have to play games to | |
2051 | * protect against truncate races as the page could now be beyond EOF. Because | |
2052 | * truncate_setsize() writes the inode size before removing pages, once we have | |
2053 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
2054 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2055 | * unlock the page. | |
2056 | */ | |
2057 | static vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) | |
2058 | { | |
2059 | struct page *page = vmf->page; | |
2060 | struct folio *folio = page_folio(page); | |
2061 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
2062 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); | |
2063 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2064 | struct btrfs_ordered_extent *ordered; | |
2065 | struct extent_state *cached_state = NULL; | |
2066 | struct extent_changeset *data_reserved = NULL; | |
2067 | unsigned long zero_start; | |
2068 | loff_t size; | |
2069 | vm_fault_t ret; | |
2070 | int ret2; | |
2071 | int reserved = 0; | |
2072 | u64 reserved_space; | |
2073 | u64 page_start; | |
2074 | u64 page_end; | |
2075 | u64 end; | |
2076 | ||
2077 | ASSERT(folio_order(folio) == 0); | |
2078 | ||
2079 | reserved_space = PAGE_SIZE; | |
2080 | ||
2081 | sb_start_pagefault(inode->i_sb); | |
2082 | page_start = page_offset(page); | |
2083 | page_end = page_start + PAGE_SIZE - 1; | |
2084 | end = page_end; | |
2085 | ||
2086 | /* | |
2087 | * Reserving delalloc space after obtaining the page lock can lead to | |
2088 | * deadlock. For example, if a dirty page is locked by this function | |
2089 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
2090 | * dirty page write out, then the btrfs_writepages() function could | |
2091 | * end up waiting indefinitely to get a lock on the page currently | |
2092 | * being processed by btrfs_page_mkwrite() function. | |
2093 | */ | |
2094 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, | |
2095 | page_start, reserved_space); | |
2096 | if (!ret2) { | |
2097 | ret2 = file_update_time(vmf->vma->vm_file); | |
2098 | reserved = 1; | |
2099 | } | |
2100 | if (ret2) { | |
2101 | ret = vmf_error(ret2); | |
2102 | if (reserved) | |
2103 | goto out; | |
2104 | goto out_noreserve; | |
2105 | } | |
2106 | ||
2107 | /* Make the VM retry the fault. */ | |
2108 | ret = VM_FAULT_NOPAGE; | |
2109 | again: | |
2110 | down_read(&BTRFS_I(inode)->i_mmap_lock); | |
2111 | lock_page(page); | |
2112 | size = i_size_read(inode); | |
2113 | ||
2114 | if ((page->mapping != inode->i_mapping) || | |
2115 | (page_start >= size)) { | |
2116 | /* Page got truncated out from underneath us. */ | |
2117 | goto out_unlock; | |
2118 | } | |
2119 | wait_on_page_writeback(page); | |
2120 | ||
2121 | lock_extent(io_tree, page_start, page_end, &cached_state); | |
2122 | ret2 = set_page_extent_mapped(page); | |
2123 | if (ret2 < 0) { | |
2124 | ret = vmf_error(ret2); | |
2125 | unlock_extent(io_tree, page_start, page_end, &cached_state); | |
2126 | goto out_unlock; | |
2127 | } | |
2128 | ||
2129 | /* | |
2130 | * We can't set the delalloc bits if there are pending ordered | |
2131 | * extents. Drop our locks and wait for them to finish. | |
2132 | */ | |
2133 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, PAGE_SIZE); | |
2134 | if (ordered) { | |
2135 | unlock_extent(io_tree, page_start, page_end, &cached_state); | |
2136 | unlock_page(page); | |
2137 | up_read(&BTRFS_I(inode)->i_mmap_lock); | |
2138 | btrfs_start_ordered_extent(ordered); | |
2139 | btrfs_put_ordered_extent(ordered); | |
2140 | goto again; | |
2141 | } | |
2142 | ||
2143 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { | |
2144 | reserved_space = round_up(size - page_start, fs_info->sectorsize); | |
2145 | if (reserved_space < PAGE_SIZE) { | |
2146 | end = page_start + reserved_space - 1; | |
2147 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
2148 | data_reserved, page_start, | |
2149 | PAGE_SIZE - reserved_space, true); | |
2150 | } | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * page_mkwrite gets called when the page is firstly dirtied after it's | |
2155 | * faulted in, but write(2) could also dirty a page and set delalloc | |
2156 | * bits, thus in this case for space account reason, we still need to | |
2157 | * clear any delalloc bits within this page range since we have to | |
2158 | * reserve data&meta space before lock_page() (see above comments). | |
2159 | */ | |
2160 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, | |
2161 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
2162 | EXTENT_DEFRAG, &cached_state); | |
2163 | ||
2164 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, | |
2165 | &cached_state); | |
2166 | if (ret2) { | |
2167 | unlock_extent(io_tree, page_start, page_end, &cached_state); | |
2168 | ret = VM_FAULT_SIGBUS; | |
2169 | goto out_unlock; | |
2170 | } | |
2171 | ||
2172 | /* Page is wholly or partially inside EOF. */ | |
2173 | if (page_start + PAGE_SIZE > size) | |
2174 | zero_start = offset_in_page(size); | |
2175 | else | |
2176 | zero_start = PAGE_SIZE; | |
2177 | ||
2178 | if (zero_start != PAGE_SIZE) | |
2179 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); | |
2180 | ||
2181 | btrfs_folio_clear_checked(fs_info, folio, page_start, PAGE_SIZE); | |
2182 | btrfs_folio_set_dirty(fs_info, folio, page_start, end + 1 - page_start); | |
2183 | btrfs_folio_set_uptodate(fs_info, folio, page_start, end + 1 - page_start); | |
2184 | ||
2185 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); | |
2186 | ||
2187 | unlock_extent(io_tree, page_start, page_end, &cached_state); | |
2188 | up_read(&BTRFS_I(inode)->i_mmap_lock); | |
2189 | ||
2190 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); | |
2191 | sb_end_pagefault(inode->i_sb); | |
2192 | extent_changeset_free(data_reserved); | |
2193 | return VM_FAULT_LOCKED; | |
2194 | ||
2195 | out_unlock: | |
2196 | unlock_page(page); | |
2197 | up_read(&BTRFS_I(inode)->i_mmap_lock); | |
2198 | out: | |
2199 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); | |
2200 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, | |
2201 | reserved_space, (ret != 0)); | |
2202 | out_noreserve: | |
2203 | sb_end_pagefault(inode->i_sb); | |
2204 | extent_changeset_free(data_reserved); | |
2205 | return ret; | |
2206 | } | |
2207 | ||
f0f37e2f | 2208 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2209 | .fault = filemap_fault, |
f1820361 | 2210 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2211 | .page_mkwrite = btrfs_page_mkwrite, |
2212 | }; | |
2213 | ||
2214 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2215 | { | |
058a457e MX |
2216 | struct address_space *mapping = filp->f_mapping; |
2217 | ||
7e0a1265 | 2218 | if (!mapping->a_ops->read_folio) |
058a457e MX |
2219 | return -ENOEXEC; |
2220 | ||
9ebefb18 | 2221 | file_accessed(filp); |
058a457e | 2222 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2223 | |
9ebefb18 CM |
2224 | return 0; |
2225 | } | |
2226 | ||
35339c24 | 2227 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2228 | int slot, u64 start, u64 end) |
2229 | { | |
2230 | struct btrfs_file_extent_item *fi; | |
2231 | struct btrfs_key key; | |
2232 | ||
2233 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2234 | return 0; | |
2235 | ||
2236 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2237 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2238 | key.type != BTRFS_EXTENT_DATA_KEY) |
2239 | return 0; | |
2240 | ||
2241 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2242 | ||
2243 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2244 | return 0; | |
2245 | ||
2246 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2247 | return 0; | |
2248 | ||
2249 | if (key.offset == end) | |
2250 | return 1; | |
2251 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2252 | return 1; | |
2253 | return 0; | |
2254 | } | |
2255 | ||
a012a74e NB |
2256 | static int fill_holes(struct btrfs_trans_handle *trans, |
2257 | struct btrfs_inode *inode, | |
2258 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2259 | { |
3ffbd68c | 2260 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2261 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2262 | struct extent_buffer *leaf; |
2263 | struct btrfs_file_extent_item *fi; | |
2264 | struct extent_map *hole_em; | |
2aaa6655 JB |
2265 | struct btrfs_key key; |
2266 | int ret; | |
2267 | ||
0b246afa | 2268 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2269 | goto out; |
2270 | ||
a012a74e | 2271 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2272 | key.type = BTRFS_EXTENT_DATA_KEY; |
2273 | key.offset = offset; | |
2274 | ||
2aaa6655 | 2275 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2276 | if (ret <= 0) { |
2277 | /* | |
2278 | * We should have dropped this offset, so if we find it then | |
2279 | * something has gone horribly wrong. | |
2280 | */ | |
2281 | if (ret == 0) | |
2282 | ret = -EINVAL; | |
2aaa6655 | 2283 | return ret; |
f94480bd | 2284 | } |
2aaa6655 JB |
2285 | |
2286 | leaf = path->nodes[0]; | |
a012a74e | 2287 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2288 | u64 num_bytes; |
2289 | ||
2290 | path->slots[0]--; | |
2291 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2292 | struct btrfs_file_extent_item); | |
2293 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2294 | end - offset; | |
2295 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2296 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2297 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2298 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 2299 | btrfs_mark_buffer_dirty(trans, leaf); |
2aaa6655 JB |
2300 | goto out; |
2301 | } | |
2302 | ||
1707e26d | 2303 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2304 | u64 num_bytes; |
2305 | ||
2aaa6655 | 2306 | key.offset = offset; |
50564b65 | 2307 | btrfs_set_item_key_safe(trans, path, &key); |
2aaa6655 JB |
2308 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2309 | struct btrfs_file_extent_item); | |
2310 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2311 | offset; | |
2312 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2313 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2314 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2315 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 2316 | btrfs_mark_buffer_dirty(trans, leaf); |
2aaa6655 JB |
2317 | goto out; |
2318 | } | |
2319 | btrfs_release_path(path); | |
2320 | ||
d1f68ba0 OS |
2321 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, |
2322 | end - offset); | |
2aaa6655 JB |
2323 | if (ret) |
2324 | return ret; | |
2325 | ||
2326 | out: | |
2327 | btrfs_release_path(path); | |
2328 | ||
2329 | hole_em = alloc_extent_map(); | |
2330 | if (!hole_em) { | |
4c0c8cfc | 2331 | btrfs_drop_extent_map_range(inode, offset, end - 1, false); |
23e3337f | 2332 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2333 | } else { |
2334 | hole_em->start = offset; | |
2335 | hole_em->len = end - offset; | |
cc95bef6 | 2336 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2337 | hole_em->orig_start = offset; |
2338 | ||
2339 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2340 | hole_em->block_len = 0; | |
b4939680 | 2341 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2342 | hole_em->generation = trans->transid; |
2343 | ||
a1ba4c08 | 2344 | ret = btrfs_replace_extent_map_range(inode, hole_em, true); |
2aaa6655 JB |
2345 | free_extent_map(hole_em); |
2346 | if (ret) | |
23e3337f | 2347 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2348 | } |
2349 | ||
2350 | return 0; | |
2351 | } | |
2352 | ||
d7781546 QW |
2353 | /* |
2354 | * Find a hole extent on given inode and change start/len to the end of hole | |
2355 | * extent.(hole/vacuum extent whose em->start <= start && | |
2356 | * em->start + em->len > start) | |
2357 | * When a hole extent is found, return 1 and modify start/len. | |
2358 | */ | |
dea46d84 | 2359 | static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len) |
d7781546 | 2360 | { |
dea46d84 | 2361 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d7781546 QW |
2362 | struct extent_map *em; |
2363 | int ret = 0; | |
2364 | ||
8bab0a30 | 2365 | em = btrfs_get_extent(inode, NULL, |
609805d8 | 2366 | round_down(*start, fs_info->sectorsize), |
39b07b5d | 2367 | round_up(*len, fs_info->sectorsize)); |
9986277e DC |
2368 | if (IS_ERR(em)) |
2369 | return PTR_ERR(em); | |
d7781546 QW |
2370 | |
2371 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2372 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2373 | ret = 1; | |
2374 | *len = em->start + em->len > *start + *len ? | |
2375 | 0 : *start + *len - em->start - em->len; | |
2376 | *start = em->start + em->len; | |
2377 | } | |
2378 | free_extent_map(em); | |
2379 | return ret; | |
2380 | } | |
2381 | ||
55961c8a FM |
2382 | static void btrfs_punch_hole_lock_range(struct inode *inode, |
2383 | const u64 lockstart, | |
2384 | const u64 lockend, | |
2385 | struct extent_state **cached_state) | |
f27451f2 | 2386 | { |
0528476b QW |
2387 | /* |
2388 | * For subpage case, if the range is not at page boundary, we could | |
2389 | * have pages at the leading/tailing part of the range. | |
2390 | * This could lead to dead loop since filemap_range_has_page() | |
2391 | * will always return true. | |
2392 | * So here we need to do extra page alignment for | |
2393 | * filemap_range_has_page(). | |
2394 | */ | |
2395 | const u64 page_lockstart = round_up(lockstart, PAGE_SIZE); | |
2396 | const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1; | |
2397 | ||
f27451f2 | 2398 | while (1) { |
f27451f2 FM |
2399 | truncate_pagecache_range(inode, lockstart, lockend); |
2400 | ||
570eb97b JB |
2401 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2402 | cached_state); | |
f27451f2 | 2403 | /* |
55961c8a FM |
2404 | * We can't have ordered extents in the range, nor dirty/writeback |
2405 | * pages, because we have locked the inode's VFS lock in exclusive | |
2406 | * mode, we have locked the inode's i_mmap_lock in exclusive mode, | |
2407 | * we have flushed all delalloc in the range and we have waited | |
2408 | * for any ordered extents in the range to complete. | |
2409 | * We can race with anyone reading pages from this range, so after | |
2410 | * locking the range check if we have pages in the range, and if | |
2411 | * we do, unlock the range and retry. | |
f27451f2 | 2412 | */ |
55961c8a FM |
2413 | if (!filemap_range_has_page(inode->i_mapping, page_lockstart, |
2414 | page_lockend)) | |
f27451f2 | 2415 | break; |
55961c8a | 2416 | |
570eb97b JB |
2417 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2418 | cached_state); | |
f27451f2 | 2419 | } |
63c34cb4 FM |
2420 | |
2421 | btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend); | |
f27451f2 FM |
2422 | } |
2423 | ||
0cbb5bdf | 2424 | static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans, |
03fcb1ab | 2425 | struct btrfs_inode *inode, |
690a5dbf | 2426 | struct btrfs_path *path, |
bf385648 | 2427 | struct btrfs_replace_extent_info *extent_info, |
2766ff61 FM |
2428 | const u64 replace_len, |
2429 | const u64 bytes_to_drop) | |
690a5dbf | 2430 | { |
03fcb1ab NB |
2431 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2432 | struct btrfs_root *root = inode->root; | |
690a5dbf FM |
2433 | struct btrfs_file_extent_item *extent; |
2434 | struct extent_buffer *leaf; | |
2435 | struct btrfs_key key; | |
2436 | int slot; | |
690a5dbf FM |
2437 | int ret; |
2438 | ||
bf385648 | 2439 | if (replace_len == 0) |
690a5dbf FM |
2440 | return 0; |
2441 | ||
bf385648 | 2442 | if (extent_info->disk_offset == 0 && |
2766ff61 | 2443 | btrfs_fs_incompat(fs_info, NO_HOLES)) { |
03fcb1ab | 2444 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2445 | return 0; |
2766ff61 | 2446 | } |
690a5dbf | 2447 | |
03fcb1ab | 2448 | key.objectid = btrfs_ino(inode); |
690a5dbf | 2449 | key.type = BTRFS_EXTENT_DATA_KEY; |
bf385648 | 2450 | key.offset = extent_info->file_offset; |
690a5dbf | 2451 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
fb870f6c | 2452 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf FM |
2453 | if (ret) |
2454 | return ret; | |
2455 | leaf = path->nodes[0]; | |
2456 | slot = path->slots[0]; | |
bf385648 | 2457 | write_extent_buffer(leaf, extent_info->extent_buf, |
690a5dbf | 2458 | btrfs_item_ptr_offset(leaf, slot), |
fb870f6c | 2459 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf | 2460 | extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
fb870f6c | 2461 | ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE); |
bf385648 FM |
2462 | btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset); |
2463 | btrfs_set_file_extent_num_bytes(leaf, extent, replace_len); | |
2464 | if (extent_info->is_new_extent) | |
8fccebfa | 2465 | btrfs_set_file_extent_generation(leaf, extent, trans->transid); |
50564b65 | 2466 | btrfs_mark_buffer_dirty(trans, leaf); |
690a5dbf FM |
2467 | btrfs_release_path(path); |
2468 | ||
03fcb1ab NB |
2469 | ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset, |
2470 | replace_len); | |
9ddc959e JB |
2471 | if (ret) |
2472 | return ret; | |
2473 | ||
690a5dbf | 2474 | /* If it's a hole, nothing more needs to be done. */ |
2766ff61 | 2475 | if (extent_info->disk_offset == 0) { |
03fcb1ab | 2476 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2477 | return 0; |
2766ff61 | 2478 | } |
690a5dbf | 2479 | |
03fcb1ab | 2480 | btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop); |
8fccebfa | 2481 | |
bf385648 FM |
2482 | if (extent_info->is_new_extent && extent_info->insertions == 0) { |
2483 | key.objectid = extent_info->disk_offset; | |
8fccebfa | 2484 | key.type = BTRFS_EXTENT_ITEM_KEY; |
bf385648 | 2485 | key.offset = extent_info->disk_len; |
8fccebfa | 2486 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
03fcb1ab | 2487 | btrfs_ino(inode), |
bf385648 FM |
2488 | extent_info->file_offset, |
2489 | extent_info->qgroup_reserved, | |
8fccebfa FM |
2490 | &key); |
2491 | } else { | |
4d09b4e9 JB |
2492 | struct btrfs_ref ref = { |
2493 | .action = BTRFS_ADD_DELAYED_REF, | |
2494 | .bytenr = extent_info->disk_offset, | |
12390e42 | 2495 | .num_bytes = extent_info->disk_len, |
e094f480 JB |
2496 | .owning_root = btrfs_root_id(root), |
2497 | .ref_root = btrfs_root_id(root), | |
4d09b4e9 | 2498 | }; |
8fccebfa FM |
2499 | u64 ref_offset; |
2500 | ||
bf385648 | 2501 | ref_offset = extent_info->file_offset - extent_info->data_offset; |
f2e69a77 | 2502 | btrfs_init_data_ref(&ref, btrfs_ino(inode), ref_offset, 0, false); |
8fccebfa FM |
2503 | ret = btrfs_inc_extent_ref(trans, &ref); |
2504 | } | |
2505 | ||
bf385648 | 2506 | extent_info->insertions++; |
690a5dbf FM |
2507 | |
2508 | return ret; | |
2509 | } | |
2510 | ||
9cba40a6 FM |
2511 | /* |
2512 | * The respective range must have been previously locked, as well as the inode. | |
2513 | * The end offset is inclusive (last byte of the range). | |
bf385648 FM |
2514 | * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing |
2515 | * the file range with an extent. | |
2516 | * When not punching a hole, we don't want to end up in a state where we dropped | |
2517 | * extents without inserting a new one, so we must abort the transaction to avoid | |
2518 | * a corruption. | |
9cba40a6 | 2519 | */ |
bfc78479 NB |
2520 | int btrfs_replace_file_extents(struct btrfs_inode *inode, |
2521 | struct btrfs_path *path, const u64 start, | |
2522 | const u64 end, | |
2523 | struct btrfs_replace_extent_info *extent_info, | |
2524 | struct btrfs_trans_handle **trans_out) | |
9cba40a6 | 2525 | { |
5893dfb9 | 2526 | struct btrfs_drop_extents_args drop_args = { 0 }; |
bfc78479 NB |
2527 | struct btrfs_root *root = inode->root; |
2528 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2bd36e7b | 2529 | u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1); |
bfc78479 | 2530 | u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize); |
9cba40a6 FM |
2531 | struct btrfs_trans_handle *trans = NULL; |
2532 | struct btrfs_block_rsv *rsv; | |
2533 | unsigned int rsv_count; | |
2534 | u64 cur_offset; | |
9cba40a6 FM |
2535 | u64 len = end - start; |
2536 | int ret = 0; | |
2537 | ||
2538 | if (end <= start) | |
2539 | return -EINVAL; | |
2540 | ||
2541 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); | |
2542 | if (!rsv) { | |
2543 | ret = -ENOMEM; | |
2544 | goto out; | |
2545 | } | |
2bd36e7b | 2546 | rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1); |
710d5921 | 2547 | rsv->failfast = true; |
9cba40a6 FM |
2548 | |
2549 | /* | |
2550 | * 1 - update the inode | |
2551 | * 1 - removing the extents in the range | |
bf385648 FM |
2552 | * 1 - adding the hole extent if no_holes isn't set or if we are |
2553 | * replacing the range with a new extent | |
9cba40a6 | 2554 | */ |
bf385648 | 2555 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info) |
690a5dbf FM |
2556 | rsv_count = 3; |
2557 | else | |
2558 | rsv_count = 2; | |
2559 | ||
9cba40a6 FM |
2560 | trans = btrfs_start_transaction(root, rsv_count); |
2561 | if (IS_ERR(trans)) { | |
2562 | ret = PTR_ERR(trans); | |
2563 | trans = NULL; | |
2564 | goto out_free; | |
2565 | } | |
2566 | ||
2567 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, | |
2568 | min_size, false); | |
650c9cab FM |
2569 | if (WARN_ON(ret)) |
2570 | goto out_trans; | |
9cba40a6 FM |
2571 | trans->block_rsv = rsv; |
2572 | ||
2573 | cur_offset = start; | |
5893dfb9 FM |
2574 | drop_args.path = path; |
2575 | drop_args.end = end + 1; | |
2576 | drop_args.drop_cache = true; | |
9cba40a6 | 2577 | while (cur_offset < end) { |
5893dfb9 | 2578 | drop_args.start = cur_offset; |
bfc78479 | 2579 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
2766ff61 FM |
2580 | /* If we are punching a hole decrement the inode's byte count */ |
2581 | if (!extent_info) | |
bfc78479 | 2582 | btrfs_update_inode_bytes(inode, 0, |
2766ff61 | 2583 | drop_args.bytes_found); |
690a5dbf FM |
2584 | if (ret != -ENOSPC) { |
2585 | /* | |
4afb912f JB |
2586 | * The only time we don't want to abort is if we are |
2587 | * attempting to clone a partial inline extent, in which | |
2588 | * case we'll get EOPNOTSUPP. However if we aren't | |
2589 | * clone we need to abort no matter what, because if we | |
2590 | * got EOPNOTSUPP via prealloc then we messed up and | |
2591 | * need to abort. | |
690a5dbf | 2592 | */ |
4afb912f JB |
2593 | if (ret && |
2594 | (ret != -EOPNOTSUPP || | |
2595 | (extent_info && extent_info->is_new_extent))) | |
690a5dbf | 2596 | btrfs_abort_transaction(trans, ret); |
9cba40a6 | 2597 | break; |
690a5dbf | 2598 | } |
9cba40a6 FM |
2599 | |
2600 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2601 | ||
5893dfb9 | 2602 | if (!extent_info && cur_offset < drop_args.drop_end && |
690a5dbf | 2603 | cur_offset < ino_size) { |
bfc78479 NB |
2604 | ret = fill_holes(trans, inode, path, cur_offset, |
2605 | drop_args.drop_end); | |
9cba40a6 FM |
2606 | if (ret) { |
2607 | /* | |
2608 | * If we failed then we didn't insert our hole | |
2609 | * entries for the area we dropped, so now the | |
2610 | * fs is corrupted, so we must abort the | |
2611 | * transaction. | |
2612 | */ | |
2613 | btrfs_abort_transaction(trans, ret); | |
2614 | break; | |
2615 | } | |
5893dfb9 | 2616 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e JB |
2617 | /* |
2618 | * We are past the i_size here, but since we didn't | |
2619 | * insert holes we need to clear the mapped area so we | |
2620 | * know to not set disk_i_size in this area until a new | |
2621 | * file extent is inserted here. | |
2622 | */ | |
bfc78479 | 2623 | ret = btrfs_inode_clear_file_extent_range(inode, |
5893dfb9 FM |
2624 | cur_offset, |
2625 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2626 | if (ret) { |
2627 | /* | |
2628 | * We couldn't clear our area, so we could | |
2629 | * presumably adjust up and corrupt the fs, so | |
2630 | * we need to abort. | |
2631 | */ | |
2632 | btrfs_abort_transaction(trans, ret); | |
2633 | break; | |
2634 | } | |
9cba40a6 FM |
2635 | } |
2636 | ||
5893dfb9 FM |
2637 | if (extent_info && |
2638 | drop_args.drop_end > extent_info->file_offset) { | |
2639 | u64 replace_len = drop_args.drop_end - | |
2640 | extent_info->file_offset; | |
690a5dbf | 2641 | |
bfc78479 NB |
2642 | ret = btrfs_insert_replace_extent(trans, inode, path, |
2643 | extent_info, replace_len, | |
03fcb1ab | 2644 | drop_args.bytes_found); |
690a5dbf FM |
2645 | if (ret) { |
2646 | btrfs_abort_transaction(trans, ret); | |
2647 | break; | |
2648 | } | |
bf385648 FM |
2649 | extent_info->data_len -= replace_len; |
2650 | extent_info->data_offset += replace_len; | |
2651 | extent_info->file_offset += replace_len; | |
690a5dbf FM |
2652 | } |
2653 | ||
983d8209 FM |
2654 | /* |
2655 | * We are releasing our handle on the transaction, balance the | |
2656 | * dirty pages of the btree inode and flush delayed items, and | |
2657 | * then get a new transaction handle, which may now point to a | |
2658 | * new transaction in case someone else may have committed the | |
2659 | * transaction we used to replace/drop file extent items. So | |
2660 | * bump the inode's iversion and update mtime and ctime except | |
2661 | * if we are called from a dedupe context. This is because a | |
2662 | * power failure/crash may happen after the transaction is | |
2663 | * committed and before we finish replacing/dropping all the | |
2664 | * file extent items we need. | |
2665 | */ | |
2666 | inode_inc_iversion(&inode->vfs_inode); | |
2667 | ||
2a9462de | 2668 | if (!extent_info || extent_info->update_times) |
b1c38a13 JL |
2669 | inode_set_mtime_to_ts(&inode->vfs_inode, |
2670 | inode_set_ctime_current(&inode->vfs_inode)); | |
983d8209 | 2671 | |
8b9d0322 | 2672 | ret = btrfs_update_inode(trans, inode); |
9cba40a6 FM |
2673 | if (ret) |
2674 | break; | |
2675 | ||
2676 | btrfs_end_transaction(trans); | |
2677 | btrfs_btree_balance_dirty(fs_info); | |
2678 | ||
2679 | trans = btrfs_start_transaction(root, rsv_count); | |
2680 | if (IS_ERR(trans)) { | |
2681 | ret = PTR_ERR(trans); | |
2682 | trans = NULL; | |
2683 | break; | |
2684 | } | |
2685 | ||
2686 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, | |
2687 | rsv, min_size, false); | |
650c9cab FM |
2688 | if (WARN_ON(ret)) |
2689 | break; | |
9cba40a6 FM |
2690 | trans->block_rsv = rsv; |
2691 | ||
3227788c BC |
2692 | cur_offset = drop_args.drop_end; |
2693 | len = end - cur_offset; | |
2694 | if (!extent_info && len) { | |
bfc78479 | 2695 | ret = find_first_non_hole(inode, &cur_offset, &len); |
690a5dbf FM |
2696 | if (unlikely(ret < 0)) |
2697 | break; | |
2698 | if (ret && !len) { | |
2699 | ret = 0; | |
2700 | break; | |
2701 | } | |
9cba40a6 FM |
2702 | } |
2703 | } | |
2704 | ||
690a5dbf FM |
2705 | /* |
2706 | * If we were cloning, force the next fsync to be a full one since we | |
2707 | * we replaced (or just dropped in the case of cloning holes when | |
e2b84217 FM |
2708 | * NO_HOLES is enabled) file extent items and did not setup new extent |
2709 | * maps for the replacement extents (or holes). | |
690a5dbf | 2710 | */ |
bf385648 | 2711 | if (extent_info && !extent_info->is_new_extent) |
23e3337f | 2712 | btrfs_set_inode_full_sync(inode); |
690a5dbf | 2713 | |
9cba40a6 FM |
2714 | if (ret) |
2715 | goto out_trans; | |
2716 | ||
2717 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2718 | /* | |
2719 | * If we are using the NO_HOLES feature we might have had already an | |
2720 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2721 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2722 | * represent holes, drop_end can be less than lockend and so we must | |
2723 | * make sure we have an extent map representing the existing hole (the | |
2724 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2725 | * map representing the existing hole), otherwise the fast fsync path | |
2726 | * will not record the existence of the hole region | |
2727 | * [existing_hole_start, lockend]. | |
2728 | */ | |
5893dfb9 FM |
2729 | if (drop_args.drop_end <= end) |
2730 | drop_args.drop_end = end + 1; | |
9cba40a6 FM |
2731 | /* |
2732 | * Don't insert file hole extent item if it's for a range beyond eof | |
2733 | * (because it's useless) or if it represents a 0 bytes range (when | |
2734 | * cur_offset == drop_end). | |
2735 | */ | |
5893dfb9 FM |
2736 | if (!extent_info && cur_offset < ino_size && |
2737 | cur_offset < drop_args.drop_end) { | |
bfc78479 NB |
2738 | ret = fill_holes(trans, inode, path, cur_offset, |
2739 | drop_args.drop_end); | |
9cba40a6 FM |
2740 | if (ret) { |
2741 | /* Same comment as above. */ | |
2742 | btrfs_abort_transaction(trans, ret); | |
2743 | goto out_trans; | |
2744 | } | |
5893dfb9 | 2745 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e | 2746 | /* See the comment in the loop above for the reasoning here. */ |
bfc78479 NB |
2747 | ret = btrfs_inode_clear_file_extent_range(inode, cur_offset, |
2748 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2749 | if (ret) { |
2750 | btrfs_abort_transaction(trans, ret); | |
2751 | goto out_trans; | |
2752 | } | |
2753 | ||
9cba40a6 | 2754 | } |
bf385648 | 2755 | if (extent_info) { |
bfc78479 | 2756 | ret = btrfs_insert_replace_extent(trans, inode, path, |
03fcb1ab NB |
2757 | extent_info, extent_info->data_len, |
2758 | drop_args.bytes_found); | |
690a5dbf FM |
2759 | if (ret) { |
2760 | btrfs_abort_transaction(trans, ret); | |
2761 | goto out_trans; | |
2762 | } | |
2763 | } | |
9cba40a6 FM |
2764 | |
2765 | out_trans: | |
2766 | if (!trans) | |
2767 | goto out_free; | |
2768 | ||
2769 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2770 | if (ret) | |
2771 | btrfs_end_transaction(trans); | |
2772 | else | |
2773 | *trans_out = trans; | |
2774 | out_free: | |
2775 | btrfs_free_block_rsv(fs_info, rsv); | |
2776 | out: | |
2777 | return ret; | |
2778 | } | |
2779 | ||
05fd9564 | 2780 | static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len) |
2aaa6655 | 2781 | { |
05fd9564 | 2782 | struct inode *inode = file_inode(file); |
41044b41 | 2783 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
2aaa6655 JB |
2784 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2785 | struct extent_state *cached_state = NULL; | |
2786 | struct btrfs_path *path; | |
9cba40a6 | 2787 | struct btrfs_trans_handle *trans = NULL; |
d7781546 QW |
2788 | u64 lockstart; |
2789 | u64 lockend; | |
2790 | u64 tail_start; | |
2791 | u64 tail_len; | |
2792 | u64 orig_start = offset; | |
2aaa6655 | 2793 | int ret = 0; |
9703fefe | 2794 | bool same_block; |
a1a50f60 | 2795 | u64 ino_size; |
9703fefe | 2796 | bool truncated_block = false; |
e8c1c76e | 2797 | bool updated_inode = false; |
2aaa6655 | 2798 | |
29b6352b | 2799 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
bd6526d0 | 2800 | |
0ef8b726 JB |
2801 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2802 | if (ret) | |
bd6526d0 | 2803 | goto out_only_mutex; |
2aaa6655 | 2804 | |
0b246afa | 2805 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
dea46d84 | 2806 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2807 | if (ret < 0) |
2808 | goto out_only_mutex; | |
2809 | if (ret && !len) { | |
2810 | /* Already in a large hole */ | |
2811 | ret = 0; | |
2812 | goto out_only_mutex; | |
2813 | } | |
2814 | ||
05fd9564 DW |
2815 | ret = file_modified(file); |
2816 | if (ret) | |
2817 | goto out_only_mutex; | |
2818 | ||
ee8ba05c JB |
2819 | lockstart = round_up(offset, fs_info->sectorsize); |
2820 | lockend = round_down(offset + len, fs_info->sectorsize) - 1; | |
0b246afa JM |
2821 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2822 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2823 | /* |
9703fefe | 2824 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2825 | * because we are sure there is no data there. |
2826 | */ | |
2aaa6655 | 2827 | /* |
9703fefe CR |
2828 | * Only do this if we are in the same block and we aren't doing the |
2829 | * entire block. | |
2aaa6655 | 2830 | */ |
0b246afa | 2831 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2832 | if (offset < ino_size) { |
9703fefe | 2833 | truncated_block = true; |
217f42eb NB |
2834 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2835 | 0); | |
e8c1c76e FM |
2836 | } else { |
2837 | ret = 0; | |
2838 | } | |
d7781546 | 2839 | goto out_only_mutex; |
2aaa6655 JB |
2840 | } |
2841 | ||
9703fefe | 2842 | /* zero back part of the first block */ |
12870f1c | 2843 | if (offset < ino_size) { |
9703fefe | 2844 | truncated_block = true; |
217f42eb | 2845 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
7426cc04 | 2846 | if (ret) { |
e5d4d75b | 2847 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
7426cc04 MX |
2848 | return ret; |
2849 | } | |
2aaa6655 JB |
2850 | } |
2851 | ||
d7781546 QW |
2852 | /* Check the aligned pages after the first unaligned page, |
2853 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2854 | * including several following pages are already in holes, |
d7781546 QW |
2855 | * the extra check can be skipped */ |
2856 | if (offset == orig_start) { | |
2857 | /* after truncate page, check hole again */ | |
2858 | len = offset + len - lockstart; | |
2859 | offset = lockstart; | |
dea46d84 | 2860 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2861 | if (ret < 0) |
2862 | goto out_only_mutex; | |
2863 | if (ret && !len) { | |
2864 | ret = 0; | |
2865 | goto out_only_mutex; | |
2866 | } | |
2867 | lockstart = offset; | |
2868 | } | |
2869 | ||
2870 | /* Check the tail unaligned part is in a hole */ | |
2871 | tail_start = lockend + 1; | |
2872 | tail_len = offset + len - tail_start; | |
2873 | if (tail_len) { | |
dea46d84 | 2874 | ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len); |
d7781546 QW |
2875 | if (unlikely(ret < 0)) |
2876 | goto out_only_mutex; | |
2877 | if (!ret) { | |
2878 | /* zero the front end of the last page */ | |
2879 | if (tail_start + tail_len < ino_size) { | |
9703fefe | 2880 | truncated_block = true; |
217f42eb | 2881 | ret = btrfs_truncate_block(BTRFS_I(inode), |
9703fefe CR |
2882 | tail_start + tail_len, |
2883 | 0, 1); | |
d7781546 QW |
2884 | if (ret) |
2885 | goto out_only_mutex; | |
51f395ad | 2886 | } |
0061280d | 2887 | } |
2aaa6655 JB |
2888 | } |
2889 | ||
2890 | if (lockend < lockstart) { | |
e8c1c76e FM |
2891 | ret = 0; |
2892 | goto out_only_mutex; | |
2aaa6655 JB |
2893 | } |
2894 | ||
55961c8a | 2895 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state); |
2aaa6655 JB |
2896 | |
2897 | path = btrfs_alloc_path(); | |
2898 | if (!path) { | |
2899 | ret = -ENOMEM; | |
2900 | goto out; | |
2901 | } | |
2902 | ||
bfc78479 NB |
2903 | ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart, |
2904 | lockend, NULL, &trans); | |
9cba40a6 FM |
2905 | btrfs_free_path(path); |
2906 | if (ret) | |
2907 | goto out; | |
2aaa6655 | 2908 | |
9cba40a6 | 2909 | ASSERT(trans != NULL); |
e1f5790e | 2910 | inode_inc_iversion(inode); |
b1c38a13 | 2911 | inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); |
8b9d0322 | 2912 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
e8c1c76e | 2913 | updated_inode = true; |
3a45bb20 | 2914 | btrfs_end_transaction(trans); |
2ff7e61e | 2915 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2916 | out: |
570eb97b JB |
2917 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2918 | &cached_state); | |
d7781546 | 2919 | out_only_mutex: |
9cba40a6 | 2920 | if (!updated_inode && truncated_block && !ret) { |
e8c1c76e FM |
2921 | /* |
2922 | * If we only end up zeroing part of a page, we still need to | |
2923 | * update the inode item, so that all the time fields are | |
2924 | * updated as well as the necessary btrfs inode in memory fields | |
2925 | * for detecting, at fsync time, if the inode isn't yet in the | |
2926 | * log tree or it's there but not up to date. | |
2927 | */ | |
2a9462de | 2928 | struct timespec64 now = inode_set_ctime_current(inode); |
17900668 FM |
2929 | |
2930 | inode_inc_iversion(inode); | |
b1c38a13 | 2931 | inode_set_mtime_to_ts(inode, now); |
e8c1c76e FM |
2932 | trans = btrfs_start_transaction(root, 1); |
2933 | if (IS_ERR(trans)) { | |
9cba40a6 | 2934 | ret = PTR_ERR(trans); |
e8c1c76e | 2935 | } else { |
9cba40a6 FM |
2936 | int ret2; |
2937 | ||
8b9d0322 | 2938 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
9cba40a6 FM |
2939 | ret2 = btrfs_end_transaction(trans); |
2940 | if (!ret) | |
2941 | ret = ret2; | |
e8c1c76e FM |
2942 | } |
2943 | } | |
e5d4d75b | 2944 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
9cba40a6 | 2945 | return ret; |
2aaa6655 JB |
2946 | } |
2947 | ||
14524a84 QW |
2948 | /* Helper structure to record which range is already reserved */ |
2949 | struct falloc_range { | |
2950 | struct list_head list; | |
2951 | u64 start; | |
2952 | u64 len; | |
2953 | }; | |
2954 | ||
2955 | /* | |
2956 | * Helper function to add falloc range | |
2957 | * | |
2958 | * Caller should have locked the larger range of extent containing | |
2959 | * [start, len) | |
2960 | */ | |
2961 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2962 | { | |
14524a84 QW |
2963 | struct falloc_range *range = NULL; |
2964 | ||
77d25534 NB |
2965 | if (!list_empty(head)) { |
2966 | /* | |
2967 | * As fallocate iterates by bytenr order, we only need to check | |
2968 | * the last range. | |
2969 | */ | |
2970 | range = list_last_entry(head, struct falloc_range, list); | |
2971 | if (range->start + range->len == start) { | |
2972 | range->len += len; | |
2973 | return 0; | |
2974 | } | |
14524a84 | 2975 | } |
77d25534 | 2976 | |
32fc932e | 2977 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2978 | if (!range) |
2979 | return -ENOMEM; | |
2980 | range->start = start; | |
2981 | range->len = len; | |
2982 | list_add_tail(&range->list, head); | |
2983 | return 0; | |
2984 | } | |
2985 | ||
f27451f2 FM |
2986 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2987 | const u64 end, | |
2988 | const int mode) | |
2989 | { | |
2990 | struct btrfs_trans_handle *trans; | |
2991 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2992 | int ret; | |
2993 | int ret2; | |
2994 | ||
2995 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2996 | return 0; | |
2997 | ||
2998 | trans = btrfs_start_transaction(root, 1); | |
2999 | if (IS_ERR(trans)) | |
3000 | return PTR_ERR(trans); | |
3001 | ||
2a9462de | 3002 | inode_set_ctime_current(inode); |
f27451f2 | 3003 | i_size_write(inode, end); |
76aea537 | 3004 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
8b9d0322 | 3005 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
f27451f2 FM |
3006 | ret2 = btrfs_end_transaction(trans); |
3007 | ||
3008 | return ret ? ret : ret2; | |
3009 | } | |
3010 | ||
81fdf638 | 3011 | enum { |
f262fa8d DS |
3012 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
3013 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
3014 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
3015 | }; |
3016 | ||
948dfeb8 | 3017 | static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode, |
f27451f2 FM |
3018 | u64 offset) |
3019 | { | |
ee8ba05c | 3020 | const u64 sectorsize = inode->root->fs_info->sectorsize; |
f27451f2 | 3021 | struct extent_map *em; |
81fdf638 | 3022 | int ret; |
f27451f2 FM |
3023 | |
3024 | offset = round_down(offset, sectorsize); | |
8bab0a30 | 3025 | em = btrfs_get_extent(inode, NULL, offset, sectorsize); |
f27451f2 FM |
3026 | if (IS_ERR(em)) |
3027 | return PTR_ERR(em); | |
3028 | ||
3029 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 | 3030 | ret = RANGE_BOUNDARY_HOLE; |
f86f7a75 | 3031 | else if (em->flags & EXTENT_FLAG_PREALLOC) |
81fdf638 FM |
3032 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; |
3033 | else | |
3034 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
3035 | |
3036 | free_extent_map(em); | |
3037 | return ret; | |
3038 | } | |
3039 | ||
3040 | static int btrfs_zero_range(struct inode *inode, | |
3041 | loff_t offset, | |
3042 | loff_t len, | |
3043 | const int mode) | |
3044 | { | |
3045 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
3046 | struct extent_map *em; | |
3047 | struct extent_changeset *data_reserved = NULL; | |
3048 | int ret; | |
3049 | u64 alloc_hint = 0; | |
ee8ba05c | 3050 | const u64 sectorsize = fs_info->sectorsize; |
f27451f2 FM |
3051 | u64 alloc_start = round_down(offset, sectorsize); |
3052 | u64 alloc_end = round_up(offset + len, sectorsize); | |
3053 | u64 bytes_to_reserve = 0; | |
3054 | bool space_reserved = false; | |
3055 | ||
8bab0a30 | 3056 | em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start, |
39b07b5d | 3057 | alloc_end - alloc_start); |
f27451f2 FM |
3058 | if (IS_ERR(em)) { |
3059 | ret = PTR_ERR(em); | |
3060 | goto out; | |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * Avoid hole punching and extent allocation for some cases. More cases | |
3065 | * could be considered, but these are unlikely common and we keep things | |
3066 | * as simple as possible for now. Also, intentionally, if the target | |
3067 | * range contains one or more prealloc extents together with regular | |
3068 | * extents and holes, we drop all the existing extents and allocate a | |
3069 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
3070 | */ | |
f86f7a75 | 3071 | if (em->start <= alloc_start && (em->flags & EXTENT_FLAG_PREALLOC)) { |
f27451f2 FM |
3072 | const u64 em_end = em->start + em->len; |
3073 | ||
3074 | if (em_end >= offset + len) { | |
3075 | /* | |
3076 | * The whole range is already a prealloc extent, | |
3077 | * do nothing except updating the inode's i_size if | |
3078 | * needed. | |
3079 | */ | |
3080 | free_extent_map(em); | |
3081 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
3082 | mode); | |
3083 | goto out; | |
3084 | } | |
3085 | /* | |
3086 | * Part of the range is already a prealloc extent, so operate | |
3087 | * only on the remaining part of the range. | |
3088 | */ | |
3089 | alloc_start = em_end; | |
3090 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
3091 | len = offset + len - alloc_start; | |
3092 | offset = alloc_start; | |
3093 | alloc_hint = em->block_start + em->len; | |
3094 | } | |
3095 | free_extent_map(em); | |
3096 | ||
3097 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
3098 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
8bab0a30 | 3099 | em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start, sectorsize); |
f27451f2 FM |
3100 | if (IS_ERR(em)) { |
3101 | ret = PTR_ERR(em); | |
3102 | goto out; | |
3103 | } | |
3104 | ||
f86f7a75 | 3105 | if (em->flags & EXTENT_FLAG_PREALLOC) { |
f27451f2 FM |
3106 | free_extent_map(em); |
3107 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
3108 | mode); | |
3109 | goto out; | |
3110 | } | |
3111 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
3112 | free_extent_map(em); | |
217f42eb NB |
3113 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
3114 | 0); | |
f27451f2 FM |
3115 | if (!ret) |
3116 | ret = btrfs_fallocate_update_isize(inode, | |
3117 | offset + len, | |
3118 | mode); | |
3119 | return ret; | |
3120 | } | |
3121 | free_extent_map(em); | |
3122 | alloc_start = round_down(offset, sectorsize); | |
3123 | alloc_end = alloc_start + sectorsize; | |
3124 | goto reserve_space; | |
3125 | } | |
3126 | ||
3127 | alloc_start = round_up(offset, sectorsize); | |
3128 | alloc_end = round_down(offset + len, sectorsize); | |
3129 | ||
3130 | /* | |
3131 | * For unaligned ranges, check the pages at the boundaries, they might | |
3132 | * map to an extent, in which case we need to partially zero them, or | |
3133 | * they might map to a hole, in which case we need our allocation range | |
3134 | * to cover them. | |
3135 | */ | |
3136 | if (!IS_ALIGNED(offset, sectorsize)) { | |
948dfeb8 NB |
3137 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
3138 | offset); | |
f27451f2 FM |
3139 | if (ret < 0) |
3140 | goto out; | |
81fdf638 | 3141 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
3142 | alloc_start = round_down(offset, sectorsize); |
3143 | ret = 0; | |
81fdf638 | 3144 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb | 3145 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
f27451f2 FM |
3146 | if (ret) |
3147 | goto out; | |
81fdf638 FM |
3148 | } else { |
3149 | ret = 0; | |
f27451f2 FM |
3150 | } |
3151 | } | |
3152 | ||
3153 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
948dfeb8 | 3154 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
f27451f2 FM |
3155 | offset + len); |
3156 | if (ret < 0) | |
3157 | goto out; | |
81fdf638 | 3158 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
3159 | alloc_end = round_up(offset + len, sectorsize); |
3160 | ret = 0; | |
81fdf638 | 3161 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb NB |
3162 | ret = btrfs_truncate_block(BTRFS_I(inode), offset + len, |
3163 | 0, 1); | |
f27451f2 FM |
3164 | if (ret) |
3165 | goto out; | |
81fdf638 FM |
3166 | } else { |
3167 | ret = 0; | |
f27451f2 FM |
3168 | } |
3169 | } | |
3170 | ||
3171 | reserve_space: | |
3172 | if (alloc_start < alloc_end) { | |
3173 | struct extent_state *cached_state = NULL; | |
3174 | const u64 lockstart = alloc_start; | |
3175 | const u64 lockend = alloc_end - 1; | |
3176 | ||
3177 | bytes_to_reserve = alloc_end - alloc_start; | |
3178 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3179 | bytes_to_reserve); | |
3180 | if (ret < 0) | |
3181 | goto out; | |
3182 | space_reserved = true; | |
55961c8a FM |
3183 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
3184 | &cached_state); | |
7661a3e0 | 3185 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved, |
a7f8b1c2 | 3186 | alloc_start, bytes_to_reserve); |
4f6a49de | 3187 | if (ret) { |
570eb97b JB |
3188 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
3189 | lockend, &cached_state); | |
a7f8b1c2 | 3190 | goto out; |
4f6a49de | 3191 | } |
f27451f2 FM |
3192 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, |
3193 | alloc_end - alloc_start, | |
dc527961 | 3194 | fs_info->sectorsize, |
f27451f2 | 3195 | offset + len, &alloc_hint); |
570eb97b JB |
3196 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3197 | &cached_state); | |
f27451f2 | 3198 | /* btrfs_prealloc_file_range releases reserved space on error */ |
9f13ce74 | 3199 | if (ret) { |
f27451f2 | 3200 | space_reserved = false; |
9f13ce74 FM |
3201 | goto out; |
3202 | } | |
f27451f2 | 3203 | } |
9f13ce74 | 3204 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3205 | out: |
3206 | if (ret && space_reserved) | |
25ce28ca | 3207 | btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved, |
f27451f2 FM |
3208 | alloc_start, bytes_to_reserve); |
3209 | extent_changeset_free(data_reserved); | |
3210 | ||
3211 | return ret; | |
3212 | } | |
3213 | ||
2fe17c10 CH |
3214 | static long btrfs_fallocate(struct file *file, int mode, |
3215 | loff_t offset, loff_t len) | |
3216 | { | |
496ad9aa | 3217 | struct inode *inode = file_inode(file); |
2fe17c10 | 3218 | struct extent_state *cached_state = NULL; |
364ecf36 | 3219 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3220 | struct falloc_range *range; |
3221 | struct falloc_range *tmp; | |
84af994b | 3222 | LIST_HEAD(reserve_list); |
2fe17c10 CH |
3223 | u64 cur_offset; |
3224 | u64 last_byte; | |
3225 | u64 alloc_start; | |
3226 | u64 alloc_end; | |
3227 | u64 alloc_hint = 0; | |
3228 | u64 locked_end; | |
14524a84 | 3229 | u64 actual_end = 0; |
47e1d1c7 FM |
3230 | u64 data_space_needed = 0; |
3231 | u64 data_space_reserved = 0; | |
3232 | u64 qgroup_reserved = 0; | |
2fe17c10 | 3233 | struct extent_map *em; |
ee8ba05c | 3234 | int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize; |
2fe17c10 CH |
3235 | int ret; |
3236 | ||
f1569c4c | 3237 | /* Do not allow fallocate in ZONED mode */ |
41044b41 | 3238 | if (btrfs_is_zoned(inode_to_fs_info(inode))) |
f1569c4c NA |
3239 | return -EOPNOTSUPP; |
3240 | ||
797f4277 MX |
3241 | alloc_start = round_down(offset, blocksize); |
3242 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3243 | cur_offset = alloc_start; |
2fe17c10 | 3244 | |
2aaa6655 | 3245 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3246 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3247 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3248 | return -EOPNOTSUPP; |
3249 | ||
2aaa6655 | 3250 | if (mode & FALLOC_FL_PUNCH_HOLE) |
05fd9564 | 3251 | return btrfs_punch_hole(file, offset, len); |
2aaa6655 | 3252 | |
29b6352b | 3253 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
2a162ce9 DI |
3254 | |
3255 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3256 | ret = inode_newsize_ok(inode, offset + len); | |
3257 | if (ret) | |
3258 | goto out; | |
3259 | } | |
2fe17c10 | 3260 | |
05fd9564 DW |
3261 | ret = file_modified(file); |
3262 | if (ret) | |
3263 | goto out; | |
3264 | ||
14524a84 QW |
3265 | /* |
3266 | * TODO: Move these two operations after we have checked | |
3267 | * accurate reserved space, or fallocate can still fail but | |
3268 | * with page truncated or size expanded. | |
3269 | * | |
3270 | * But that's a minor problem and won't do much harm BTW. | |
3271 | */ | |
2fe17c10 | 3272 | if (alloc_start > inode->i_size) { |
b06359a3 | 3273 | ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode), |
a41ad394 | 3274 | alloc_start); |
2fe17c10 CH |
3275 | if (ret) |
3276 | goto out; | |
0f6925fa | 3277 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3278 | /* |
3279 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3280 | * need to zero out the end of the block if i_size lands in the |
3281 | * middle of a block. | |
a71754fc | 3282 | */ |
217f42eb | 3283 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
a71754fc JB |
3284 | if (ret) |
3285 | goto out; | |
2fe17c10 CH |
3286 | } |
3287 | ||
a71754fc | 3288 | /* |
ffa8fc60 FM |
3289 | * We have locked the inode at the VFS level (in exclusive mode) and we |
3290 | * have locked the i_mmap_lock lock (in exclusive mode). Now before | |
3291 | * locking the file range, flush all dealloc in the range and wait for | |
3292 | * all ordered extents in the range to complete. After this we can lock | |
3293 | * the file range and, due to the previous locking we did, we know there | |
3294 | * can't be more delalloc or ordered extents in the range. | |
a71754fc | 3295 | */ |
0ef8b726 JB |
3296 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3297 | alloc_end - alloc_start); | |
3298 | if (ret) | |
3299 | goto out; | |
a71754fc | 3300 | |
f27451f2 FM |
3301 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3302 | ret = btrfs_zero_range(inode, offset, len, mode); | |
e5d4d75b | 3303 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
f27451f2 FM |
3304 | return ret; |
3305 | } | |
3306 | ||
2fe17c10 | 3307 | locked_end = alloc_end - 1; |
570eb97b JB |
3308 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3309 | &cached_state); | |
2fe17c10 | 3310 | |
63c34cb4 FM |
3311 | btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end); |
3312 | ||
14524a84 | 3313 | /* First, check if we exceed the qgroup limit */ |
6b7d6e93 | 3314 | while (cur_offset < alloc_end) { |
8bab0a30 | 3315 | em = btrfs_get_extent(BTRFS_I(inode), NULL, cur_offset, |
39b07b5d | 3316 | alloc_end - cur_offset); |
9986277e DC |
3317 | if (IS_ERR(em)) { |
3318 | ret = PTR_ERR(em); | |
79787eaa JM |
3319 | break; |
3320 | } | |
2fe17c10 | 3321 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3322 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3323 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3324 | if (em->block_start == EXTENT_MAP_HOLE || |
3325 | (cur_offset >= inode->i_size && | |
f86f7a75 | 3326 | !(em->flags & EXTENT_FLAG_PREALLOC))) { |
47e1d1c7 FM |
3327 | const u64 range_len = last_byte - cur_offset; |
3328 | ||
3329 | ret = add_falloc_range(&reserve_list, cur_offset, range_len); | |
14524a84 QW |
3330 | if (ret < 0) { |
3331 | free_extent_map(em); | |
3332 | break; | |
3d850dd4 | 3333 | } |
7661a3e0 | 3334 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), |
47e1d1c7 | 3335 | &data_reserved, cur_offset, range_len); |
be2d253c FM |
3336 | if (ret < 0) { |
3337 | free_extent_map(em); | |
14524a84 | 3338 | break; |
be2d253c | 3339 | } |
47e1d1c7 FM |
3340 | qgroup_reserved += range_len; |
3341 | data_space_needed += range_len; | |
2fe17c10 CH |
3342 | } |
3343 | free_extent_map(em); | |
2fe17c10 | 3344 | cur_offset = last_byte; |
14524a84 QW |
3345 | } |
3346 | ||
47e1d1c7 FM |
3347 | if (!ret && data_space_needed > 0) { |
3348 | /* | |
3349 | * We are safe to reserve space here as we can't have delalloc | |
3350 | * in the range, see above. | |
3351 | */ | |
3352 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3353 | data_space_needed); | |
3354 | if (!ret) | |
3355 | data_space_reserved = data_space_needed; | |
3356 | } | |
3357 | ||
14524a84 QW |
3358 | /* |
3359 | * If ret is still 0, means we're OK to fallocate. | |
3360 | * Or just cleanup the list and exit. | |
3361 | */ | |
3362 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
47e1d1c7 | 3363 | if (!ret) { |
14524a84 QW |
3364 | ret = btrfs_prealloc_file_range(inode, mode, |
3365 | range->start, | |
dc527961 | 3366 | range->len, blocksize, |
14524a84 | 3367 | offset + len, &alloc_hint); |
47e1d1c7 FM |
3368 | /* |
3369 | * btrfs_prealloc_file_range() releases space even | |
3370 | * if it returns an error. | |
3371 | */ | |
3372 | data_space_reserved -= range->len; | |
3373 | qgroup_reserved -= range->len; | |
3374 | } else if (data_space_reserved > 0) { | |
25ce28ca | 3375 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
47e1d1c7 FM |
3376 | data_reserved, range->start, |
3377 | range->len); | |
3378 | data_space_reserved -= range->len; | |
3379 | qgroup_reserved -= range->len; | |
3380 | } else if (qgroup_reserved > 0) { | |
3381 | btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved, | |
9e65bfca | 3382 | range->start, range->len, NULL); |
47e1d1c7 FM |
3383 | qgroup_reserved -= range->len; |
3384 | } | |
14524a84 QW |
3385 | list_del(&range->list); |
3386 | kfree(range); | |
3387 | } | |
3388 | if (ret < 0) | |
3389 | goto out_unlock; | |
3390 | ||
f27451f2 FM |
3391 | /* |
3392 | * We didn't need to allocate any more space, but we still extended the | |
3393 | * size of the file so we need to update i_size and the inode item. | |
3394 | */ | |
3395 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3396 | out_unlock: |
570eb97b JB |
3397 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3398 | &cached_state); | |
2fe17c10 | 3399 | out: |
e5d4d75b | 3400 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
364ecf36 | 3401 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3402 | return ret; |
3403 | } | |
3404 | ||
b6e83356 | 3405 | /* |
ac3c0d36 FM |
3406 | * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range |
3407 | * that has unflushed and/or flushing delalloc. There might be other adjacent | |
3408 | * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps | |
3409 | * looping while it gets adjacent subranges, and merging them together. | |
b6e83356 FM |
3410 | */ |
3411 | static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end, | |
b3e744fe | 3412 | struct extent_state **cached_state, |
af979fd6 | 3413 | bool *search_io_tree, |
b6e83356 FM |
3414 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
3415 | { | |
40daf3e0 | 3416 | u64 len = end + 1 - start; |
8ddc8274 FM |
3417 | u64 delalloc_len = 0; |
3418 | struct btrfs_ordered_extent *oe; | |
3419 | u64 oe_start; | |
3420 | u64 oe_end; | |
b6e83356 FM |
3421 | |
3422 | /* | |
3423 | * Search the io tree first for EXTENT_DELALLOC. If we find any, it | |
3424 | * means we have delalloc (dirty pages) for which writeback has not | |
3425 | * started yet. | |
3426 | */ | |
8ddc8274 FM |
3427 | if (*search_io_tree) { |
3428 | spin_lock(&inode->lock); | |
3429 | if (inode->delalloc_bytes > 0) { | |
3430 | spin_unlock(&inode->lock); | |
3431 | *delalloc_start_ret = start; | |
3432 | delalloc_len = count_range_bits(&inode->io_tree, | |
3433 | delalloc_start_ret, end, | |
8c6e53a7 | 3434 | len, EXTENT_DELALLOC, 1, |
b3e744fe | 3435 | cached_state); |
8ddc8274 FM |
3436 | } else { |
3437 | spin_unlock(&inode->lock); | |
3438 | } | |
a2853ffc FM |
3439 | } |
3440 | ||
40daf3e0 FM |
3441 | if (delalloc_len > 0) { |
3442 | /* | |
3443 | * If delalloc was found then *delalloc_start_ret has a sector size | |
3444 | * aligned value (rounded down). | |
3445 | */ | |
b6e83356 FM |
3446 | *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1; |
3447 | ||
40daf3e0 FM |
3448 | if (*delalloc_start_ret == start) { |
3449 | /* Delalloc for the whole range, nothing more to do. */ | |
3450 | if (*delalloc_end_ret == end) | |
3451 | return true; | |
8ddc8274 | 3452 | /* Else trim our search range for ordered extents. */ |
40daf3e0 FM |
3453 | start = *delalloc_end_ret + 1; |
3454 | len = end + 1 - start; | |
3455 | } | |
af979fd6 FM |
3456 | } else { |
3457 | /* No delalloc, future calls don't need to search again. */ | |
3458 | *search_io_tree = false; | |
40daf3e0 FM |
3459 | } |
3460 | ||
a2853ffc | 3461 | /* |
8ddc8274 FM |
3462 | * Now also check if there's any ordered extent in the range. |
3463 | * We do this because: | |
b6e83356 FM |
3464 | * |
3465 | * 1) When delalloc is flushed, the file range is locked, we clear the | |
8ddc8274 FM |
3466 | * EXTENT_DELALLOC bit from the io tree and create an extent map and |
3467 | * an ordered extent for the write. So we might just have been called | |
3468 | * after delalloc is flushed and before the ordered extent completes | |
3469 | * and inserts the new file extent item in the subvolume's btree; | |
b6e83356 | 3470 | * |
8ddc8274 | 3471 | * 2) We may have an ordered extent created by flushing delalloc for a |
b6e83356 FM |
3472 | * subrange that starts before the subrange we found marked with |
3473 | * EXTENT_DELALLOC in the io tree. | |
8ddc8274 FM |
3474 | * |
3475 | * We could also use the extent map tree to find such delalloc that is | |
3476 | * being flushed, but using the ordered extents tree is more efficient | |
3477 | * because it's usually much smaller as ordered extents are removed from | |
3478 | * the tree once they complete. With the extent maps, we mau have them | |
3479 | * in the extent map tree for a very long time, and they were either | |
3480 | * created by previous writes or loaded by read operations. | |
b6e83356 | 3481 | */ |
8ddc8274 FM |
3482 | oe = btrfs_lookup_first_ordered_range(inode, start, len); |
3483 | if (!oe) | |
d47704bd | 3484 | return (delalloc_len > 0); |
d47704bd | 3485 | |
8ddc8274 FM |
3486 | /* The ordered extent may span beyond our search range. */ |
3487 | oe_start = max(oe->file_offset, start); | |
3488 | oe_end = min(oe->file_offset + oe->num_bytes - 1, end); | |
b6e83356 | 3489 | |
8ddc8274 | 3490 | btrfs_put_ordered_extent(oe); |
b6e83356 | 3491 | |
8ddc8274 | 3492 | /* Don't have unflushed delalloc, return the ordered extent range. */ |
b6e83356 | 3493 | if (delalloc_len == 0) { |
8ddc8274 FM |
3494 | *delalloc_start_ret = oe_start; |
3495 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3496 | return true; |
3497 | } | |
3498 | ||
3499 | /* | |
8ddc8274 FM |
3500 | * We have both unflushed delalloc (io_tree) and an ordered extent. |
3501 | * If the ranges are adjacent returned a combined range, otherwise | |
3502 | * return the leftmost range. | |
b6e83356 | 3503 | */ |
8ddc8274 FM |
3504 | if (oe_start < *delalloc_start_ret) { |
3505 | if (oe_end < *delalloc_start_ret) | |
3506 | *delalloc_end_ret = oe_end; | |
3507 | *delalloc_start_ret = oe_start; | |
3508 | } else if (*delalloc_end_ret + 1 == oe_start) { | |
3509 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3510 | } |
3511 | ||
b6e83356 FM |
3512 | return true; |
3513 | } | |
3514 | ||
3515 | /* | |
3516 | * Check if there's delalloc in a given range. | |
3517 | * | |
3518 | * @inode: The inode. | |
3519 | * @start: The start offset of the range. It does not need to be | |
3520 | * sector size aligned. | |
3521 | * @end: The end offset (inclusive value) of the search range. | |
3522 | * It does not need to be sector size aligned. | |
b3e744fe FM |
3523 | * @cached_state: Extent state record used for speeding up delalloc |
3524 | * searches in the inode's io_tree. Can be NULL. | |
b6e83356 FM |
3525 | * @delalloc_start_ret: Output argument, set to the start offset of the |
3526 | * subrange found with delalloc (may not be sector size | |
3527 | * aligned). | |
3528 | * @delalloc_end_ret: Output argument, set to he end offset (inclusive value) | |
3529 | * of the subrange found with delalloc. | |
3530 | * | |
3531 | * Returns true if a subrange with delalloc is found within the given range, and | |
3532 | * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and | |
3533 | * end offsets of the subrange. | |
3534 | */ | |
ac3c0d36 | 3535 | bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end, |
b3e744fe | 3536 | struct extent_state **cached_state, |
ac3c0d36 | 3537 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
b6e83356 FM |
3538 | { |
3539 | u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize); | |
3540 | u64 prev_delalloc_end = 0; | |
af979fd6 | 3541 | bool search_io_tree = true; |
b6e83356 FM |
3542 | bool ret = false; |
3543 | ||
2f2e84ca | 3544 | while (cur_offset <= end) { |
b6e83356 FM |
3545 | u64 delalloc_start; |
3546 | u64 delalloc_end; | |
3547 | bool delalloc; | |
3548 | ||
3549 | delalloc = find_delalloc_subrange(inode, cur_offset, end, | |
b3e744fe | 3550 | cached_state, &search_io_tree, |
b6e83356 FM |
3551 | &delalloc_start, |
3552 | &delalloc_end); | |
3553 | if (!delalloc) | |
3554 | break; | |
3555 | ||
3556 | if (prev_delalloc_end == 0) { | |
3557 | /* First subrange found. */ | |
3558 | *delalloc_start_ret = max(delalloc_start, start); | |
3559 | *delalloc_end_ret = delalloc_end; | |
3560 | ret = true; | |
3561 | } else if (delalloc_start == prev_delalloc_end + 1) { | |
3562 | /* Subrange adjacent to the previous one, merge them. */ | |
3563 | *delalloc_end_ret = delalloc_end; | |
3564 | } else { | |
3565 | /* Subrange not adjacent to the previous one, exit. */ | |
3566 | break; | |
3567 | } | |
3568 | ||
3569 | prev_delalloc_end = delalloc_end; | |
3570 | cur_offset = delalloc_end + 1; | |
3571 | cond_resched(); | |
3572 | } | |
3573 | ||
3574 | return ret; | |
3575 | } | |
3576 | ||
3577 | /* | |
3578 | * Check if there's a hole or delalloc range in a range representing a hole (or | |
3579 | * prealloc extent) found in the inode's subvolume btree. | |
3580 | * | |
3581 | * @inode: The inode. | |
3582 | * @whence: Seek mode (SEEK_DATA or SEEK_HOLE). | |
3583 | * @start: Start offset of the hole region. It does not need to be sector | |
3584 | * size aligned. | |
3585 | * @end: End offset (inclusive value) of the hole region. It does not | |
3586 | * need to be sector size aligned. | |
3587 | * @start_ret: Return parameter, used to set the start of the subrange in the | |
3588 | * hole that matches the search criteria (seek mode), if such | |
3589 | * subrange is found (return value of the function is true). | |
3590 | * The value returned here may not be sector size aligned. | |
3591 | * | |
3592 | * Returns true if a subrange matching the given seek mode is found, and if one | |
3593 | * is found, it updates @start_ret with the start of the subrange. | |
3594 | */ | |
3595 | static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence, | |
3c32c721 | 3596 | struct extent_state **cached_state, |
b6e83356 FM |
3597 | u64 start, u64 end, u64 *start_ret) |
3598 | { | |
3599 | u64 delalloc_start; | |
3600 | u64 delalloc_end; | |
3601 | bool delalloc; | |
3602 | ||
3c32c721 | 3603 | delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state, |
ac3c0d36 | 3604 | &delalloc_start, &delalloc_end); |
b6e83356 FM |
3605 | if (delalloc && whence == SEEK_DATA) { |
3606 | *start_ret = delalloc_start; | |
3607 | return true; | |
3608 | } | |
3609 | ||
3610 | if (delalloc && whence == SEEK_HOLE) { | |
3611 | /* | |
3612 | * We found delalloc but it starts after out start offset. So we | |
3613 | * have a hole between our start offset and the delalloc start. | |
3614 | */ | |
3615 | if (start < delalloc_start) { | |
3616 | *start_ret = start; | |
3617 | return true; | |
3618 | } | |
3619 | /* | |
3620 | * Delalloc range starts at our start offset. | |
3621 | * If the delalloc range's length is smaller than our range, | |
3622 | * then it means we have a hole that starts where the delalloc | |
3623 | * subrange ends. | |
3624 | */ | |
3625 | if (delalloc_end < end) { | |
3626 | *start_ret = delalloc_end + 1; | |
3627 | return true; | |
3628 | } | |
3629 | ||
3630 | /* There's delalloc for the whole range. */ | |
3631 | return false; | |
3632 | } | |
3633 | ||
3634 | if (!delalloc && whence == SEEK_HOLE) { | |
3635 | *start_ret = start; | |
3636 | return true; | |
3637 | } | |
3638 | ||
3639 | /* | |
3640 | * No delalloc in the range and we are seeking for data. The caller has | |
3641 | * to iterate to the next extent item in the subvolume btree. | |
3642 | */ | |
3643 | return false; | |
3644 | } | |
3645 | ||
3c32c721 | 3646 | static loff_t find_desired_extent(struct file *file, loff_t offset, int whence) |
b2675157 | 3647 | { |
3c32c721 FM |
3648 | struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host); |
3649 | struct btrfs_file_private *private = file->private_data; | |
cca5de97 | 3650 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
b2675157 | 3651 | struct extent_state *cached_state = NULL; |
3c32c721 | 3652 | struct extent_state **delalloc_cached_state; |
b6e83356 FM |
3653 | const loff_t i_size = i_size_read(&inode->vfs_inode); |
3654 | const u64 ino = btrfs_ino(inode); | |
3655 | struct btrfs_root *root = inode->root; | |
3656 | struct btrfs_path *path; | |
3657 | struct btrfs_key key; | |
3658 | u64 last_extent_end; | |
4d1a40c6 LB |
3659 | u64 lockstart; |
3660 | u64 lockend; | |
3661 | u64 start; | |
b6e83356 FM |
3662 | int ret; |
3663 | bool found = false; | |
b2675157 | 3664 | |
bc80230e | 3665 | if (i_size == 0 || offset >= i_size) |
4d1a40c6 LB |
3666 | return -ENXIO; |
3667 | ||
b6e83356 FM |
3668 | /* |
3669 | * Quick path. If the inode has no prealloc extents and its number of | |
3670 | * bytes used matches its i_size, then it can not have holes. | |
3671 | */ | |
3672 | if (whence == SEEK_HOLE && | |
3673 | !(inode->flags & BTRFS_INODE_PREALLOC) && | |
3674 | inode_get_bytes(&inode->vfs_inode) == i_size) | |
3675 | return i_size; | |
3676 | ||
3c32c721 FM |
3677 | if (!private) { |
3678 | private = kzalloc(sizeof(*private), GFP_KERNEL); | |
3679 | /* | |
3680 | * No worries if memory allocation failed. | |
3681 | * The private structure is used only for speeding up multiple | |
3682 | * lseek SEEK_HOLE/DATA calls to a file when there's delalloc, | |
3683 | * so everything will still be correct. | |
3684 | */ | |
3685 | file->private_data = private; | |
3686 | } | |
3687 | ||
3688 | if (private) | |
3689 | delalloc_cached_state = &private->llseek_cached_state; | |
3690 | else | |
3691 | delalloc_cached_state = NULL; | |
3692 | ||
4d1a40c6 | 3693 | /* |
bc80230e | 3694 | * offset can be negative, in this case we start finding DATA/HOLE from |
4d1a40c6 LB |
3695 | * the very start of the file. |
3696 | */ | |
bc80230e | 3697 | start = max_t(loff_t, 0, offset); |
4d1a40c6 | 3698 | |
0b246afa | 3699 | lockstart = round_down(start, fs_info->sectorsize); |
d79b7c26 | 3700 | lockend = round_up(i_size, fs_info->sectorsize); |
b2675157 | 3701 | if (lockend <= lockstart) |
0b246afa | 3702 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3703 | lockend--; |
b6e83356 FM |
3704 | |
3705 | path = btrfs_alloc_path(); | |
3706 | if (!path) | |
3707 | return -ENOMEM; | |
3708 | path->reada = READA_FORWARD; | |
3709 | ||
3710 | key.objectid = ino; | |
3711 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3712 | key.offset = start; | |
3713 | ||
3714 | last_extent_end = lockstart; | |
b2675157 | 3715 | |
570eb97b | 3716 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b2675157 | 3717 | |
b6e83356 FM |
3718 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3719 | if (ret < 0) { | |
3720 | goto out; | |
3721 | } else if (ret > 0 && path->slots[0] > 0) { | |
3722 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3723 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3724 | path->slots[0]--; | |
3725 | } | |
3726 | ||
d79b7c26 | 3727 | while (start < i_size) { |
b6e83356 FM |
3728 | struct extent_buffer *leaf = path->nodes[0]; |
3729 | struct btrfs_file_extent_item *extent; | |
3730 | u64 extent_end; | |
1f55ee6d | 3731 | u8 type; |
b6e83356 FM |
3732 | |
3733 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3734 | ret = btrfs_next_leaf(root, path); | |
3735 | if (ret < 0) | |
3736 | goto out; | |
3737 | else if (ret > 0) | |
3738 | break; | |
3739 | ||
3740 | leaf = path->nodes[0]; | |
b2675157 JB |
3741 | } |
3742 | ||
b6e83356 FM |
3743 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3744 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
7f4ca37c | 3745 | break; |
b2675157 | 3746 | |
b6e83356 FM |
3747 | extent_end = btrfs_file_extent_end(path); |
3748 | ||
3749 | /* | |
3750 | * In the first iteration we may have a slot that points to an | |
3751 | * extent that ends before our start offset, so skip it. | |
3752 | */ | |
3753 | if (extent_end <= start) { | |
3754 | path->slots[0]++; | |
3755 | continue; | |
3756 | } | |
3757 | ||
3758 | /* We have an implicit hole, NO_HOLES feature is likely set. */ | |
3759 | if (last_extent_end < key.offset) { | |
3760 | u64 search_start = last_extent_end; | |
3761 | u64 found_start; | |
3762 | ||
3763 | /* | |
3764 | * First iteration, @start matches @offset and it's | |
3765 | * within the hole. | |
3766 | */ | |
3767 | if (start == offset) | |
3768 | search_start = offset; | |
3769 | ||
3770 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3771 | delalloc_cached_state, |
b6e83356 FM |
3772 | search_start, |
3773 | key.offset - 1, | |
3774 | &found_start); | |
3775 | if (found) { | |
3776 | start = found_start; | |
3777 | break; | |
3778 | } | |
3779 | /* | |
3780 | * Didn't find data or a hole (due to delalloc) in the | |
3781 | * implicit hole range, so need to analyze the extent. | |
3782 | */ | |
3783 | } | |
3784 | ||
3785 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
3786 | struct btrfs_file_extent_item); | |
1f55ee6d | 3787 | type = btrfs_file_extent_type(leaf, extent); |
b6e83356 | 3788 | |
1f55ee6d FM |
3789 | /* |
3790 | * Can't access the extent's disk_bytenr field if this is an | |
3791 | * inline extent, since at that offset, it's where the extent | |
3792 | * data starts. | |
3793 | */ | |
3794 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
3795 | (type == BTRFS_FILE_EXTENT_REG && | |
3796 | btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) { | |
b6e83356 FM |
3797 | /* |
3798 | * Explicit hole or prealloc extent, search for delalloc. | |
3799 | * A prealloc extent is treated like a hole. | |
3800 | */ | |
3801 | u64 search_start = key.offset; | |
3802 | u64 found_start; | |
3803 | ||
3804 | /* | |
3805 | * First iteration, @start matches @offset and it's | |
3806 | * within the hole. | |
3807 | */ | |
3808 | if (start == offset) | |
3809 | search_start = offset; | |
3810 | ||
3811 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3812 | delalloc_cached_state, |
b6e83356 FM |
3813 | search_start, |
3814 | extent_end - 1, | |
3815 | &found_start); | |
3816 | if (found) { | |
3817 | start = found_start; | |
3818 | break; | |
3819 | } | |
3820 | /* | |
3821 | * Didn't find data or a hole (due to delalloc) in the | |
3822 | * implicit hole range, so need to analyze the next | |
3823 | * extent item. | |
3824 | */ | |
3825 | } else { | |
3826 | /* | |
3827 | * Found a regular or inline extent. | |
3828 | * If we are seeking for data, adjust the start offset | |
3829 | * and stop, we're done. | |
3830 | */ | |
3831 | if (whence == SEEK_DATA) { | |
3832 | start = max_t(u64, key.offset, offset); | |
3833 | found = true; | |
3834 | break; | |
3835 | } | |
3836 | /* | |
3837 | * Else, we are seeking for a hole, check the next file | |
3838 | * extent item. | |
3839 | */ | |
3840 | } | |
3841 | ||
3842 | start = extent_end; | |
3843 | last_extent_end = extent_end; | |
3844 | path->slots[0]++; | |
aed0ca18 FM |
3845 | if (fatal_signal_pending(current)) { |
3846 | ret = -EINTR; | |
b6e83356 | 3847 | goto out; |
aed0ca18 | 3848 | } |
b2675157 JB |
3849 | cond_resched(); |
3850 | } | |
b6e83356 FM |
3851 | |
3852 | /* We have an implicit hole from the last extent found up to i_size. */ | |
3853 | if (!found && start < i_size) { | |
3c32c721 FM |
3854 | found = find_desired_extent_in_hole(inode, whence, |
3855 | delalloc_cached_state, start, | |
b6e83356 FM |
3856 | i_size - 1, &start); |
3857 | if (!found) | |
3858 | start = i_size; | |
3859 | } | |
3860 | ||
3861 | out: | |
570eb97b | 3862 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b6e83356 FM |
3863 | btrfs_free_path(path); |
3864 | ||
3865 | if (ret < 0) | |
3866 | return ret; | |
3867 | ||
3868 | if (whence == SEEK_DATA && start >= i_size) | |
3869 | return -ENXIO; | |
bc80230e | 3870 | |
b6e83356 | 3871 | return min_t(loff_t, start, i_size); |
b2675157 JB |
3872 | } |
3873 | ||
965c8e59 | 3874 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3875 | { |
3876 | struct inode *inode = file->f_mapping->host; | |
b2675157 | 3877 | |
965c8e59 | 3878 | switch (whence) { |
2034f3b4 NB |
3879 | default: |
3880 | return generic_file_llseek(file, offset, whence); | |
b2675157 JB |
3881 | case SEEK_DATA: |
3882 | case SEEK_HOLE: | |
29b6352b | 3883 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
3c32c721 | 3884 | offset = find_desired_extent(file, offset, whence); |
e5d4d75b | 3885 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
bc80230e | 3886 | break; |
b2675157 JB |
3887 | } |
3888 | ||
bc80230e NB |
3889 | if (offset < 0) |
3890 | return offset; | |
3891 | ||
2034f3b4 | 3892 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
3893 | } |
3894 | ||
edf064e7 GR |
3895 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3896 | { | |
14605409 BB |
3897 | int ret; |
3898 | ||
210a03c9 | 3899 | filp->f_mode |= FMODE_NOWAIT | FMODE_CAN_ODIRECT; |
14605409 BB |
3900 | |
3901 | ret = fsverity_file_open(inode, filp); | |
3902 | if (ret) | |
3903 | return ret; | |
edf064e7 GR |
3904 | return generic_file_open(inode, filp); |
3905 | } | |
3906 | ||
4e4cabec GR |
3907 | static int check_direct_read(struct btrfs_fs_info *fs_info, |
3908 | const struct iov_iter *iter, loff_t offset) | |
3909 | { | |
3910 | int ret; | |
3911 | int i, seg; | |
3912 | ||
3913 | ret = check_direct_IO(fs_info, iter, offset); | |
3914 | if (ret < 0) | |
3915 | return ret; | |
3916 | ||
3917 | if (!iter_is_iovec(iter)) | |
3918 | return 0; | |
3919 | ||
de4f5fed JA |
3920 | for (seg = 0; seg < iter->nr_segs; seg++) { |
3921 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
3922 | const struct iovec *iov1 = iter_iov(iter) + seg; | |
3923 | const struct iovec *iov2 = iter_iov(iter) + i; | |
3924 | ||
3925 | if (iov1->iov_base == iov2->iov_base) | |
4e4cabec | 3926 | return -EINVAL; |
de4f5fed JA |
3927 | } |
3928 | } | |
4e4cabec GR |
3929 | return 0; |
3930 | } | |
3931 | ||
3932 | static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to) | |
3933 | { | |
3934 | struct inode *inode = file_inode(iocb->ki_filp); | |
51bd9563 FM |
3935 | size_t prev_left = 0; |
3936 | ssize_t read = 0; | |
4e4cabec GR |
3937 | ssize_t ret; |
3938 | ||
14605409 BB |
3939 | if (fsverity_active(inode)) |
3940 | return 0; | |
3941 | ||
41044b41 | 3942 | if (check_direct_read(inode_to_fs_info(inode), to, iocb->ki_pos)) |
4e4cabec GR |
3943 | return 0; |
3944 | ||
29b6352b | 3945 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 FM |
3946 | again: |
3947 | /* | |
3948 | * This is similar to what we do for direct IO writes, see the comment | |
3949 | * at btrfs_direct_write(), but we also disable page faults in addition | |
3950 | * to disabling them only at the iov_iter level. This is because when | |
3951 | * reading from a hole or prealloc extent, iomap calls iov_iter_zero(), | |
3952 | * which can still trigger page fault ins despite having set ->nofault | |
3953 | * to true of our 'to' iov_iter. | |
3954 | * | |
3955 | * The difference to direct IO writes is that we deadlock when trying | |
3956 | * to lock the extent range in the inode's tree during he page reads | |
3957 | * triggered by the fault in (while for writes it is due to waiting for | |
3958 | * our own ordered extent). This is because for direct IO reads, | |
3959 | * btrfs_dio_iomap_begin() returns with the extent range locked, which | |
3960 | * is only unlocked in the endio callback (end_bio_extent_readpage()). | |
3961 | */ | |
3962 | pagefault_disable(); | |
3963 | to->nofault = true; | |
8184620a | 3964 | ret = btrfs_dio_read(iocb, to, read); |
51bd9563 FM |
3965 | to->nofault = false; |
3966 | pagefault_enable(); | |
3967 | ||
3968 | /* No increment (+=) because iomap returns a cumulative value. */ | |
3969 | if (ret > 0) | |
3970 | read = ret; | |
3971 | ||
3972 | if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) { | |
3973 | const size_t left = iov_iter_count(to); | |
3974 | ||
3975 | if (left == prev_left) { | |
3976 | /* | |
3977 | * We didn't make any progress since the last attempt, | |
3978 | * fallback to a buffered read for the remainder of the | |
3979 | * range. This is just to avoid any possibility of looping | |
3980 | * for too long. | |
3981 | */ | |
3982 | ret = read; | |
3983 | } else { | |
3984 | /* | |
3985 | * We made some progress since the last retry or this is | |
3986 | * the first time we are retrying. Fault in as many pages | |
3987 | * as possible and retry. | |
3988 | */ | |
3989 | fault_in_iov_iter_writeable(to, left); | |
3990 | prev_left = left; | |
3991 | goto again; | |
3992 | } | |
3993 | } | |
e5d4d75b | 3994 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 | 3995 | return ret < 0 ? ret : read; |
4e4cabec GR |
3996 | } |
3997 | ||
f85781fb GR |
3998 | static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
3999 | { | |
4000 | ssize_t ret = 0; | |
4001 | ||
4002 | if (iocb->ki_flags & IOCB_DIRECT) { | |
4e4cabec | 4003 | ret = btrfs_direct_read(iocb, to); |
0425e7ba JT |
4004 | if (ret < 0 || !iov_iter_count(to) || |
4005 | iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp))) | |
f85781fb GR |
4006 | return ret; |
4007 | } | |
4008 | ||
87fa0f3e | 4009 | return filemap_read(iocb, to, ret); |
f85781fb GR |
4010 | } |
4011 | ||
828c0950 | 4012 | const struct file_operations btrfs_file_operations = { |
b2675157 | 4013 | .llseek = btrfs_file_llseek, |
f85781fb | 4014 | .read_iter = btrfs_file_read_iter, |
2cb1e089 | 4015 | .splice_read = filemap_splice_read, |
b30ac0fc | 4016 | .write_iter = btrfs_file_write_iter, |
d7776591 | 4017 | .splice_write = iter_file_splice_write, |
9ebefb18 | 4018 | .mmap = btrfs_file_mmap, |
edf064e7 | 4019 | .open = btrfs_file_open, |
e1b81e67 | 4020 | .release = btrfs_release_file, |
b0c58223 | 4021 | .get_unmapped_area = thp_get_unmapped_area, |
39279cc3 | 4022 | .fsync = btrfs_sync_file, |
2fe17c10 | 4023 | .fallocate = btrfs_fallocate, |
34287aa3 | 4024 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 4025 | #ifdef CONFIG_COMPAT |
4c63c245 | 4026 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 4027 | #endif |
2e5dfc99 | 4028 | .remap_file_range = btrfs_remap_file_range, |
210a03c9 | 4029 | .fop_flags = FOP_BUFFER_RASYNC | FOP_BUFFER_WASYNC, |
39279cc3 | 4030 | }; |
9247f317 | 4031 | |
728404da FM |
4032 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) |
4033 | { | |
4034 | int ret; | |
4035 | ||
4036 | /* | |
4037 | * So with compression we will find and lock a dirty page and clear the | |
4038 | * first one as dirty, setup an async extent, and immediately return | |
4039 | * with the entire range locked but with nobody actually marked with | |
4040 | * writeback. So we can't just filemap_write_and_wait_range() and | |
4041 | * expect it to work since it will just kick off a thread to do the | |
4042 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
4043 | * since it will wait on the page lock, which won't be unlocked until | |
4044 | * after the pages have been marked as writeback and so we're good to go | |
4045 | * from there. We have to do this otherwise we'll miss the ordered | |
4046 | * extents and that results in badness. Please Josef, do not think you | |
4047 | * know better and pull this out at some point in the future, it is | |
4048 | * right and you are wrong. | |
4049 | */ | |
4050 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
4051 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
4052 | &BTRFS_I(inode)->runtime_flags)) | |
4053 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
4054 | ||
4055 | return ret; | |
4056 | } |