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> |
39279cc3 CM |
20 | #include "ctree.h" |
21 | #include "disk-io.h" | |
22 | #include "transaction.h" | |
23 | #include "btrfs_inode.h" | |
39279cc3 | 24 | #include "print-tree.h" |
e02119d5 CM |
25 | #include "tree-log.h" |
26 | #include "locking.h" | |
2aaa6655 | 27 | #include "volumes.h" |
fcebe456 | 28 | #include "qgroup.h" |
ebb8765b | 29 | #include "compression.h" |
86736342 | 30 | #include "delalloc-space.h" |
6a177381 | 31 | #include "reflink.h" |
f02a85d2 | 32 | #include "subpage.h" |
c7f13d42 | 33 | #include "fs.h" |
07e81dc9 | 34 | #include "accessors.h" |
a0231804 | 35 | #include "extent-tree.h" |
7c8ede16 | 36 | #include "file-item.h" |
7572dec8 | 37 | #include "ioctl.h" |
af142b6f | 38 | #include "file.h" |
7f0add25 | 39 | #include "super.h" |
39279cc3 | 40 | |
d352ac68 CM |
41 | /* simple helper to fault in pages and copy. This should go away |
42 | * and be replaced with calls into generic code. | |
43 | */ | |
ee22f0c4 | 44 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 45 | struct page **prepared_pages, |
11c65dcc | 46 | struct iov_iter *i) |
39279cc3 | 47 | { |
914ee295 | 48 | size_t copied = 0; |
d0215f3e | 49 | size_t total_copied = 0; |
11c65dcc | 50 | int pg = 0; |
7073017a | 51 | int offset = offset_in_page(pos); |
39279cc3 | 52 | |
11c65dcc | 53 | while (write_bytes > 0) { |
39279cc3 | 54 | size_t count = min_t(size_t, |
09cbfeaf | 55 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 56 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
57 | /* |
58 | * Copy data from userspace to the current page | |
914ee295 | 59 | */ |
f0b65f39 | 60 | copied = copy_page_from_iter_atomic(page, offset, count, i); |
11c65dcc | 61 | |
39279cc3 CM |
62 | /* Flush processor's dcache for this page */ |
63 | flush_dcache_page(page); | |
31339acd CM |
64 | |
65 | /* | |
66 | * if we get a partial write, we can end up with | |
67 | * partially up to date pages. These add | |
68 | * a lot of complexity, so make sure they don't | |
69 | * happen by forcing this copy to be retried. | |
70 | * | |
71 | * The rest of the btrfs_file_write code will fall | |
72 | * back to page at a time copies after we return 0. | |
73 | */ | |
f0b65f39 AV |
74 | if (unlikely(copied < count)) { |
75 | if (!PageUptodate(page)) { | |
76 | iov_iter_revert(i, copied); | |
77 | copied = 0; | |
78 | } | |
79 | if (!copied) | |
80 | break; | |
81 | } | |
31339acd | 82 | |
11c65dcc | 83 | write_bytes -= copied; |
914ee295 | 84 | total_copied += copied; |
f0b65f39 AV |
85 | offset += copied; |
86 | if (offset == PAGE_SIZE) { | |
11c65dcc JB |
87 | pg++; |
88 | offset = 0; | |
89 | } | |
39279cc3 | 90 | } |
914ee295 | 91 | return total_copied; |
39279cc3 CM |
92 | } |
93 | ||
d352ac68 CM |
94 | /* |
95 | * unlocks pages after btrfs_file_write is done with them | |
96 | */ | |
e4f94347 QW |
97 | static void btrfs_drop_pages(struct btrfs_fs_info *fs_info, |
98 | struct page **pages, size_t num_pages, | |
99 | u64 pos, u64 copied) | |
39279cc3 CM |
100 | { |
101 | size_t i; | |
e4f94347 QW |
102 | u64 block_start = round_down(pos, fs_info->sectorsize); |
103 | u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start; | |
104 | ||
105 | ASSERT(block_len <= U32_MAX); | |
39279cc3 | 106 | for (i = 0; i < num_pages; i++) { |
d352ac68 CM |
107 | /* page checked is some magic around finding pages that |
108 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
109 | * clear it here. There should be no need to mark the pages |
110 | * accessed as prepare_pages should have marked them accessed | |
111 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 112 | */ |
e4f94347 QW |
113 | btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start, |
114 | block_len); | |
39279cc3 | 115 | unlock_page(pages[i]); |
09cbfeaf | 116 | put_page(pages[i]); |
39279cc3 CM |
117 | } |
118 | } | |
119 | ||
d352ac68 | 120 | /* |
c0fab480 QW |
121 | * After btrfs_copy_from_user(), update the following things for delalloc: |
122 | * - Mark newly dirtied pages as DELALLOC in the io tree. | |
123 | * Used to advise which range is to be written back. | |
124 | * - Mark modified pages as Uptodate/Dirty and not needing COW fixup | |
125 | * - Update inode size for past EOF write | |
d352ac68 | 126 | */ |
088545f6 | 127 | int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages, |
2ff7e61e | 128 | size_t num_pages, loff_t pos, size_t write_bytes, |
aa8c1a41 | 129 | struct extent_state **cached, bool noreserve) |
39279cc3 | 130 | { |
088545f6 | 131 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
39279cc3 | 132 | int err = 0; |
a52d9a80 | 133 | int i; |
db94535d | 134 | u64 num_bytes; |
a52d9a80 CM |
135 | u64 start_pos; |
136 | u64 end_of_last_block; | |
137 | u64 end_pos = pos + write_bytes; | |
088545f6 | 138 | loff_t isize = i_size_read(&inode->vfs_inode); |
e3b8a485 | 139 | unsigned int extra_bits = 0; |
39279cc3 | 140 | |
aa8c1a41 GR |
141 | if (write_bytes == 0) |
142 | return 0; | |
143 | ||
144 | if (noreserve) | |
145 | extra_bits |= EXTENT_NORESERVE; | |
146 | ||
13f0dd8f | 147 | start_pos = round_down(pos, fs_info->sectorsize); |
da17066c | 148 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 149 | fs_info->sectorsize); |
f02a85d2 | 150 | ASSERT(num_bytes <= U32_MAX); |
39279cc3 | 151 | |
db94535d | 152 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 | 153 | |
7703bdd8 CM |
154 | /* |
155 | * The pages may have already been dirty, clear out old accounting so | |
156 | * we can set things up properly | |
157 | */ | |
088545f6 | 158 | clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block, |
e182163d | 159 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 160 | cached); |
7703bdd8 | 161 | |
088545f6 | 162 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
330a5827 | 163 | extra_bits, cached); |
d0215f3e JB |
164 | if (err) |
165 | return err; | |
9ed74f2d | 166 | |
c8b97818 CM |
167 | for (i = 0; i < num_pages; i++) { |
168 | struct page *p = pages[i]; | |
f02a85d2 QW |
169 | |
170 | btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes); | |
e4f94347 | 171 | btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes); |
f02a85d2 | 172 | btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes); |
a52d9a80 | 173 | } |
9f570b8d JB |
174 | |
175 | /* | |
176 | * we've only changed i_size in ram, and we haven't updated | |
177 | * the disk i_size. There is no need to log the inode | |
178 | * at this time. | |
179 | */ | |
180 | if (end_pos > isize) | |
088545f6 | 181 | i_size_write(&inode->vfs_inode, end_pos); |
a22285a6 | 182 | return 0; |
39279cc3 CM |
183 | } |
184 | ||
185 | /* | |
186 | * this is very complex, but the basic idea is to drop all extents | |
187 | * in the range start - end. hint_block is filled in with a block number | |
188 | * that would be a good hint to the block allocator for this file. | |
189 | * | |
190 | * If an extent intersects the range but is not entirely inside the range | |
191 | * it is either truncated or split. Anything entirely inside the range | |
192 | * is deleted from the tree. | |
2766ff61 FM |
193 | * |
194 | * Note: the VFS' inode number of bytes is not updated, it's up to the caller | |
195 | * to deal with that. We set the field 'bytes_found' of the arguments structure | |
196 | * with the number of allocated bytes found in the target range, so that the | |
197 | * caller can update the inode's number of bytes in an atomic way when | |
198 | * replacing extents in a range to avoid races with stat(2). | |
39279cc3 | 199 | */ |
5893dfb9 FM |
200 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, |
201 | struct btrfs_root *root, struct btrfs_inode *inode, | |
202 | struct btrfs_drop_extents_args *args) | |
39279cc3 | 203 | { |
0b246afa | 204 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 205 | struct extent_buffer *leaf; |
920bbbfb | 206 | struct btrfs_file_extent_item *fi; |
82fa113f | 207 | struct btrfs_ref ref = { 0 }; |
00f5c795 | 208 | struct btrfs_key key; |
920bbbfb | 209 | struct btrfs_key new_key; |
906c448c | 210 | u64 ino = btrfs_ino(inode); |
5893dfb9 | 211 | u64 search_start = args->start; |
920bbbfb YZ |
212 | u64 disk_bytenr = 0; |
213 | u64 num_bytes = 0; | |
214 | u64 extent_offset = 0; | |
215 | u64 extent_end = 0; | |
5893dfb9 | 216 | u64 last_end = args->start; |
920bbbfb YZ |
217 | int del_nr = 0; |
218 | int del_slot = 0; | |
219 | int extent_type; | |
ccd467d6 | 220 | int recow; |
00f5c795 | 221 | int ret; |
dc7fdde3 | 222 | int modify_tree = -1; |
27cdeb70 | 223 | int update_refs; |
c3308f84 | 224 | int found = 0; |
5893dfb9 FM |
225 | struct btrfs_path *path = args->path; |
226 | ||
2766ff61 | 227 | args->bytes_found = 0; |
5893dfb9 FM |
228 | args->extent_inserted = false; |
229 | ||
230 | /* Must always have a path if ->replace_extent is true */ | |
231 | ASSERT(!(args->replace_extent && !args->path)); | |
232 | ||
233 | if (!path) { | |
234 | path = btrfs_alloc_path(); | |
235 | if (!path) { | |
236 | ret = -ENOMEM; | |
237 | goto out; | |
238 | } | |
239 | } | |
39279cc3 | 240 | |
5893dfb9 | 241 | if (args->drop_cache) |
4c0c8cfc | 242 | btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false); |
a52d9a80 | 243 | |
5893dfb9 | 244 | if (args->start >= inode->disk_i_size && !args->replace_extent) |
dc7fdde3 CM |
245 | modify_tree = 0; |
246 | ||
d175209b | 247 | update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID); |
d397712b | 248 | while (1) { |
ccd467d6 | 249 | recow = 0; |
33345d01 | 250 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 251 | search_start, modify_tree); |
39279cc3 | 252 | if (ret < 0) |
920bbbfb | 253 | break; |
5893dfb9 | 254 | if (ret > 0 && path->slots[0] > 0 && search_start == args->start) { |
920bbbfb YZ |
255 | leaf = path->nodes[0]; |
256 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 257 | if (key.objectid == ino && |
920bbbfb YZ |
258 | key.type == BTRFS_EXTENT_DATA_KEY) |
259 | path->slots[0]--; | |
39279cc3 | 260 | } |
920bbbfb | 261 | ret = 0; |
8c2383c3 | 262 | next_slot: |
5f39d397 | 263 | leaf = path->nodes[0]; |
920bbbfb YZ |
264 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
265 | BUG_ON(del_nr > 0); | |
266 | ret = btrfs_next_leaf(root, path); | |
267 | if (ret < 0) | |
268 | break; | |
269 | if (ret > 0) { | |
270 | ret = 0; | |
271 | break; | |
8c2383c3 | 272 | } |
920bbbfb YZ |
273 | leaf = path->nodes[0]; |
274 | recow = 1; | |
275 | } | |
276 | ||
277 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
278 | |
279 | if (key.objectid > ino) | |
280 | break; | |
281 | if (WARN_ON_ONCE(key.objectid < ino) || | |
282 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
283 | ASSERT(del_nr == 0); | |
284 | path->slots[0]++; | |
285 | goto next_slot; | |
286 | } | |
5893dfb9 | 287 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end) |
920bbbfb YZ |
288 | break; |
289 | ||
290 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
291 | struct btrfs_file_extent_item); | |
292 | extent_type = btrfs_file_extent_type(leaf, fi); | |
293 | ||
294 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
295 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
296 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
297 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
298 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
299 | extent_end = key.offset + | |
300 | btrfs_file_extent_num_bytes(leaf, fi); | |
301 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
302 | extent_end = key.offset + | |
e41ca589 | 303 | btrfs_file_extent_ram_bytes(leaf, fi); |
8c2383c3 | 304 | } else { |
aeafbf84 FM |
305 | /* can't happen */ |
306 | BUG(); | |
39279cc3 CM |
307 | } |
308 | ||
fc19c5e7 FM |
309 | /* |
310 | * Don't skip extent items representing 0 byte lengths. They | |
311 | * used to be created (bug) if while punching holes we hit | |
312 | * -ENOSPC condition. So if we find one here, just ensure we | |
313 | * delete it, otherwise we would insert a new file extent item | |
314 | * with the same key (offset) as that 0 bytes length file | |
315 | * extent item in the call to setup_items_for_insert() later | |
316 | * in this function. | |
317 | */ | |
62fe51c1 JB |
318 | if (extent_end == key.offset && extent_end >= search_start) { |
319 | last_end = extent_end; | |
fc19c5e7 | 320 | goto delete_extent_item; |
62fe51c1 | 321 | } |
fc19c5e7 | 322 | |
920bbbfb YZ |
323 | if (extent_end <= search_start) { |
324 | path->slots[0]++; | |
8c2383c3 | 325 | goto next_slot; |
39279cc3 CM |
326 | } |
327 | ||
c3308f84 | 328 | found = 1; |
5893dfb9 | 329 | search_start = max(key.offset, args->start); |
dc7fdde3 CM |
330 | if (recow || !modify_tree) { |
331 | modify_tree = -1; | |
b3b4aa74 | 332 | btrfs_release_path(path); |
920bbbfb | 333 | continue; |
39279cc3 | 334 | } |
6643558d | 335 | |
920bbbfb YZ |
336 | /* |
337 | * | - range to drop - | | |
338 | * | -------- extent -------- | | |
339 | */ | |
5893dfb9 | 340 | if (args->start > key.offset && args->end < extent_end) { |
920bbbfb | 341 | BUG_ON(del_nr > 0); |
00fdf13a | 342 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 343 | ret = -EOPNOTSUPP; |
00fdf13a LB |
344 | break; |
345 | } | |
920bbbfb YZ |
346 | |
347 | memcpy(&new_key, &key, sizeof(new_key)); | |
5893dfb9 | 348 | new_key.offset = args->start; |
920bbbfb YZ |
349 | ret = btrfs_duplicate_item(trans, root, path, |
350 | &new_key); | |
351 | if (ret == -EAGAIN) { | |
b3b4aa74 | 352 | btrfs_release_path(path); |
920bbbfb | 353 | continue; |
6643558d | 354 | } |
920bbbfb YZ |
355 | if (ret < 0) |
356 | break; | |
357 | ||
358 | leaf = path->nodes[0]; | |
359 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
360 | struct btrfs_file_extent_item); | |
361 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 362 | args->start - key.offset); |
920bbbfb YZ |
363 | |
364 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
365 | struct btrfs_file_extent_item); | |
366 | ||
5893dfb9 | 367 | extent_offset += args->start - key.offset; |
920bbbfb YZ |
368 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
369 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 370 | extent_end - args->start); |
920bbbfb YZ |
371 | btrfs_mark_buffer_dirty(leaf); |
372 | ||
5dc562c5 | 373 | if (update_refs && disk_bytenr > 0) { |
82fa113f QW |
374 | btrfs_init_generic_ref(&ref, |
375 | BTRFS_ADD_DELAYED_REF, | |
376 | disk_bytenr, num_bytes, 0); | |
377 | btrfs_init_data_ref(&ref, | |
920bbbfb YZ |
378 | root->root_key.objectid, |
379 | new_key.objectid, | |
f42c5da6 NB |
380 | args->start - extent_offset, |
381 | 0, false); | |
82fa113f | 382 | ret = btrfs_inc_extent_ref(trans, &ref); |
79787eaa | 383 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 384 | } |
5893dfb9 | 385 | key.offset = args->start; |
6643558d | 386 | } |
62fe51c1 JB |
387 | /* |
388 | * From here on out we will have actually dropped something, so | |
389 | * last_end can be updated. | |
390 | */ | |
391 | last_end = extent_end; | |
392 | ||
920bbbfb YZ |
393 | /* |
394 | * | ---- range to drop ----- | | |
395 | * | -------- extent -------- | | |
396 | */ | |
5893dfb9 | 397 | if (args->start <= key.offset && args->end < extent_end) { |
00fdf13a | 398 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 399 | ret = -EOPNOTSUPP; |
00fdf13a LB |
400 | break; |
401 | } | |
6643558d | 402 | |
920bbbfb | 403 | memcpy(&new_key, &key, sizeof(new_key)); |
5893dfb9 | 404 | new_key.offset = args->end; |
0b246afa | 405 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 406 | |
5893dfb9 | 407 | extent_offset += args->end - key.offset; |
920bbbfb YZ |
408 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
409 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 410 | extent_end - args->end); |
920bbbfb | 411 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 412 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 413 | args->bytes_found += args->end - key.offset; |
920bbbfb | 414 | break; |
39279cc3 | 415 | } |
771ed689 | 416 | |
920bbbfb YZ |
417 | search_start = extent_end; |
418 | /* | |
419 | * | ---- range to drop ----- | | |
420 | * | -------- extent -------- | | |
421 | */ | |
5893dfb9 | 422 | if (args->start > key.offset && args->end >= extent_end) { |
920bbbfb | 423 | BUG_ON(del_nr > 0); |
00fdf13a | 424 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 425 | ret = -EOPNOTSUPP; |
00fdf13a LB |
426 | break; |
427 | } | |
8c2383c3 | 428 | |
920bbbfb | 429 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5893dfb9 | 430 | args->start - key.offset); |
920bbbfb | 431 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 432 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 433 | args->bytes_found += extent_end - args->start; |
5893dfb9 | 434 | if (args->end == extent_end) |
920bbbfb | 435 | break; |
c8b97818 | 436 | |
920bbbfb YZ |
437 | path->slots[0]++; |
438 | goto next_slot; | |
31840ae1 ZY |
439 | } |
440 | ||
920bbbfb YZ |
441 | /* |
442 | * | ---- range to drop ----- | | |
443 | * | ------ extent ------ | | |
444 | */ | |
5893dfb9 | 445 | if (args->start <= key.offset && args->end >= extent_end) { |
fc19c5e7 | 446 | delete_extent_item: |
920bbbfb YZ |
447 | if (del_nr == 0) { |
448 | del_slot = path->slots[0]; | |
449 | del_nr = 1; | |
450 | } else { | |
451 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
452 | del_nr++; | |
453 | } | |
31840ae1 | 454 | |
5dc562c5 JB |
455 | if (update_refs && |
456 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
2766ff61 | 457 | args->bytes_found += extent_end - key.offset; |
920bbbfb | 458 | extent_end = ALIGN(extent_end, |
0b246afa | 459 | fs_info->sectorsize); |
5dc562c5 | 460 | } else if (update_refs && disk_bytenr > 0) { |
ffd4bb2a QW |
461 | btrfs_init_generic_ref(&ref, |
462 | BTRFS_DROP_DELAYED_REF, | |
463 | disk_bytenr, num_bytes, 0); | |
464 | btrfs_init_data_ref(&ref, | |
920bbbfb | 465 | root->root_key.objectid, |
ffd4bb2a | 466 | key.objectid, |
f42c5da6 NB |
467 | key.offset - extent_offset, 0, |
468 | false); | |
ffd4bb2a | 469 | ret = btrfs_free_extent(trans, &ref); |
79787eaa | 470 | BUG_ON(ret); /* -ENOMEM */ |
2766ff61 | 471 | args->bytes_found += extent_end - key.offset; |
31840ae1 | 472 | } |
31840ae1 | 473 | |
5893dfb9 | 474 | if (args->end == extent_end) |
920bbbfb YZ |
475 | break; |
476 | ||
477 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
478 | path->slots[0]++; | |
479 | goto next_slot; | |
480 | } | |
481 | ||
482 | ret = btrfs_del_items(trans, root, path, del_slot, | |
483 | del_nr); | |
79787eaa | 484 | if (ret) { |
66642832 | 485 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 486 | break; |
79787eaa | 487 | } |
920bbbfb YZ |
488 | |
489 | del_nr = 0; | |
490 | del_slot = 0; | |
491 | ||
b3b4aa74 | 492 | btrfs_release_path(path); |
920bbbfb | 493 | continue; |
39279cc3 | 494 | } |
920bbbfb | 495 | |
290342f6 | 496 | BUG(); |
39279cc3 | 497 | } |
920bbbfb | 498 | |
79787eaa | 499 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
500 | /* |
501 | * Set path->slots[0] to first slot, so that after the delete | |
502 | * if items are move off from our leaf to its immediate left or | |
503 | * right neighbor leafs, we end up with a correct and adjusted | |
5893dfb9 | 504 | * path->slots[0] for our insertion (if args->replace_extent). |
1acae57b FDBM |
505 | */ |
506 | path->slots[0] = del_slot; | |
920bbbfb | 507 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 508 | if (ret) |
66642832 | 509 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 510 | } |
1acae57b | 511 | |
d5f37527 FDBM |
512 | leaf = path->nodes[0]; |
513 | /* | |
514 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
515 | * which case it unlocked our path, so check path->locks[0] matches a | |
516 | * write lock. | |
517 | */ | |
7ecb4c31 | 518 | if (!ret && args->replace_extent && |
ac5887c8 | 519 | path->locks[0] == BTRFS_WRITE_LOCK && |
e902baac | 520 | btrfs_leaf_free_space(leaf) >= |
5893dfb9 | 521 | sizeof(struct btrfs_item) + args->extent_item_size) { |
d5f37527 FDBM |
522 | |
523 | key.objectid = ino; | |
524 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5893dfb9 | 525 | key.offset = args->start; |
d5f37527 FDBM |
526 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
527 | struct btrfs_key slot_key; | |
528 | ||
529 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
530 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
531 | path->slots[0]++; | |
1acae57b | 532 | } |
f0641656 | 533 | btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size); |
5893dfb9 | 534 | args->extent_inserted = true; |
6643558d | 535 | } |
920bbbfb | 536 | |
5893dfb9 FM |
537 | if (!args->path) |
538 | btrfs_free_path(path); | |
539 | else if (!args->extent_inserted) | |
1acae57b | 540 | btrfs_release_path(path); |
5893dfb9 FM |
541 | out: |
542 | args->drop_end = found ? min(args->end, last_end) : args->end; | |
5dc562c5 | 543 | |
39279cc3 CM |
544 | return ret; |
545 | } | |
546 | ||
d899e052 | 547 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
548 | u64 objectid, u64 bytenr, u64 orig_offset, |
549 | u64 *start, u64 *end) | |
d899e052 YZ |
550 | { |
551 | struct btrfs_file_extent_item *fi; | |
552 | struct btrfs_key key; | |
553 | u64 extent_end; | |
554 | ||
555 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
556 | return 0; | |
557 | ||
558 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
559 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
560 | return 0; | |
561 | ||
562 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
563 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
564 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 565 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
566 | btrfs_file_extent_compression(leaf, fi) || |
567 | btrfs_file_extent_encryption(leaf, fi) || | |
568 | btrfs_file_extent_other_encoding(leaf, fi)) | |
569 | return 0; | |
570 | ||
571 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
572 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
573 | return 0; | |
574 | ||
575 | *start = key.offset; | |
576 | *end = extent_end; | |
577 | return 1; | |
578 | } | |
579 | ||
580 | /* | |
581 | * Mark extent in the range start - end as written. | |
582 | * | |
583 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
584 | * part of extent is marked as written, the extent will be split into | |
585 | * two or three. | |
586 | */ | |
587 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 588 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 589 | { |
3ffbd68c | 590 | struct btrfs_fs_info *fs_info = trans->fs_info; |
7a6d7067 | 591 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
592 | struct extent_buffer *leaf; |
593 | struct btrfs_path *path; | |
594 | struct btrfs_file_extent_item *fi; | |
82fa113f | 595 | struct btrfs_ref ref = { 0 }; |
d899e052 | 596 | struct btrfs_key key; |
920bbbfb | 597 | struct btrfs_key new_key; |
d899e052 YZ |
598 | u64 bytenr; |
599 | u64 num_bytes; | |
600 | u64 extent_end; | |
5d4f98a2 | 601 | u64 orig_offset; |
d899e052 YZ |
602 | u64 other_start; |
603 | u64 other_end; | |
920bbbfb YZ |
604 | u64 split; |
605 | int del_nr = 0; | |
606 | int del_slot = 0; | |
6c7d54ac | 607 | int recow; |
e7b2ec3d | 608 | int ret = 0; |
7a6d7067 | 609 | u64 ino = btrfs_ino(inode); |
d899e052 | 610 | |
d899e052 | 611 | path = btrfs_alloc_path(); |
d8926bb3 MF |
612 | if (!path) |
613 | return -ENOMEM; | |
d899e052 | 614 | again: |
6c7d54ac | 615 | recow = 0; |
920bbbfb | 616 | split = start; |
33345d01 | 617 | key.objectid = ino; |
d899e052 | 618 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 619 | key.offset = split; |
d899e052 YZ |
620 | |
621 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
622 | if (ret < 0) |
623 | goto out; | |
d899e052 YZ |
624 | if (ret > 0 && path->slots[0] > 0) |
625 | path->slots[0]--; | |
626 | ||
627 | leaf = path->nodes[0]; | |
628 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
629 | if (key.objectid != ino || |
630 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
631 | ret = -EINVAL; | |
632 | btrfs_abort_transaction(trans, ret); | |
633 | goto out; | |
634 | } | |
d899e052 YZ |
635 | fi = btrfs_item_ptr(leaf, path->slots[0], |
636 | struct btrfs_file_extent_item); | |
9c8e63db JB |
637 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
638 | ret = -EINVAL; | |
639 | btrfs_abort_transaction(trans, ret); | |
640 | goto out; | |
641 | } | |
d899e052 | 642 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
643 | if (key.offset > start || extent_end < end) { |
644 | ret = -EINVAL; | |
645 | btrfs_abort_transaction(trans, ret); | |
646 | goto out; | |
647 | } | |
d899e052 YZ |
648 | |
649 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
650 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 651 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
652 | memcpy(&new_key, &key, sizeof(new_key)); |
653 | ||
654 | if (start == key.offset && end < extent_end) { | |
655 | other_start = 0; | |
656 | other_end = start; | |
657 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 658 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
659 | &other_start, &other_end)) { |
660 | new_key.offset = end; | |
0b246afa | 661 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
662 | fi = btrfs_item_ptr(leaf, path->slots[0], |
663 | struct btrfs_file_extent_item); | |
224ecce5 JB |
664 | btrfs_set_file_extent_generation(leaf, fi, |
665 | trans->transid); | |
6c7d54ac YZ |
666 | btrfs_set_file_extent_num_bytes(leaf, fi, |
667 | extent_end - end); | |
668 | btrfs_set_file_extent_offset(leaf, fi, | |
669 | end - orig_offset); | |
670 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
671 | struct btrfs_file_extent_item); | |
224ecce5 JB |
672 | btrfs_set_file_extent_generation(leaf, fi, |
673 | trans->transid); | |
6c7d54ac YZ |
674 | btrfs_set_file_extent_num_bytes(leaf, fi, |
675 | end - other_start); | |
676 | btrfs_mark_buffer_dirty(leaf); | |
677 | goto out; | |
678 | } | |
679 | } | |
680 | ||
681 | if (start > key.offset && end == extent_end) { | |
682 | other_start = end; | |
683 | other_end = 0; | |
684 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 685 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
686 | &other_start, &other_end)) { |
687 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
688 | struct btrfs_file_extent_item); | |
689 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
690 | start - key.offset); | |
224ecce5 JB |
691 | btrfs_set_file_extent_generation(leaf, fi, |
692 | trans->transid); | |
6c7d54ac YZ |
693 | path->slots[0]++; |
694 | new_key.offset = start; | |
0b246afa | 695 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
696 | |
697 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
698 | struct btrfs_file_extent_item); | |
224ecce5 JB |
699 | btrfs_set_file_extent_generation(leaf, fi, |
700 | trans->transid); | |
6c7d54ac YZ |
701 | btrfs_set_file_extent_num_bytes(leaf, fi, |
702 | other_end - start); | |
703 | btrfs_set_file_extent_offset(leaf, fi, | |
704 | start - orig_offset); | |
705 | btrfs_mark_buffer_dirty(leaf); | |
706 | goto out; | |
707 | } | |
708 | } | |
d899e052 | 709 | |
920bbbfb YZ |
710 | while (start > key.offset || end < extent_end) { |
711 | if (key.offset == start) | |
712 | split = end; | |
713 | ||
920bbbfb YZ |
714 | new_key.offset = split; |
715 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
716 | if (ret == -EAGAIN) { | |
b3b4aa74 | 717 | btrfs_release_path(path); |
920bbbfb | 718 | goto again; |
d899e052 | 719 | } |
79787eaa | 720 | if (ret < 0) { |
66642832 | 721 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
722 | goto out; |
723 | } | |
d899e052 | 724 | |
920bbbfb YZ |
725 | leaf = path->nodes[0]; |
726 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 727 | struct btrfs_file_extent_item); |
224ecce5 | 728 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 729 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
730 | split - key.offset); |
731 | ||
732 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
733 | struct btrfs_file_extent_item); | |
734 | ||
224ecce5 | 735 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
736 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
737 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
738 | extent_end - split); | |
d899e052 YZ |
739 | btrfs_mark_buffer_dirty(leaf); |
740 | ||
82fa113f QW |
741 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr, |
742 | num_bytes, 0); | |
743 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, | |
f42c5da6 | 744 | orig_offset, 0, false); |
82fa113f | 745 | ret = btrfs_inc_extent_ref(trans, &ref); |
9c8e63db JB |
746 | if (ret) { |
747 | btrfs_abort_transaction(trans, ret); | |
748 | goto out; | |
749 | } | |
d899e052 | 750 | |
920bbbfb YZ |
751 | if (split == start) { |
752 | key.offset = start; | |
753 | } else { | |
9c8e63db JB |
754 | if (start != key.offset) { |
755 | ret = -EINVAL; | |
756 | btrfs_abort_transaction(trans, ret); | |
757 | goto out; | |
758 | } | |
d899e052 | 759 | path->slots[0]--; |
920bbbfb | 760 | extent_end = end; |
d899e052 | 761 | } |
6c7d54ac | 762 | recow = 1; |
d899e052 YZ |
763 | } |
764 | ||
920bbbfb YZ |
765 | other_start = end; |
766 | other_end = 0; | |
ffd4bb2a QW |
767 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
768 | num_bytes, 0); | |
f42c5da6 NB |
769 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset, |
770 | 0, false); | |
6c7d54ac | 771 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 772 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
773 | &other_start, &other_end)) { |
774 | if (recow) { | |
b3b4aa74 | 775 | btrfs_release_path(path); |
6c7d54ac YZ |
776 | goto again; |
777 | } | |
920bbbfb YZ |
778 | extent_end = other_end; |
779 | del_slot = path->slots[0] + 1; | |
780 | del_nr++; | |
ffd4bb2a | 781 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
782 | if (ret) { |
783 | btrfs_abort_transaction(trans, ret); | |
784 | goto out; | |
785 | } | |
d899e052 | 786 | } |
920bbbfb YZ |
787 | other_start = 0; |
788 | other_end = start; | |
6c7d54ac | 789 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 790 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
791 | &other_start, &other_end)) { |
792 | if (recow) { | |
b3b4aa74 | 793 | btrfs_release_path(path); |
6c7d54ac YZ |
794 | goto again; |
795 | } | |
920bbbfb YZ |
796 | key.offset = other_start; |
797 | del_slot = path->slots[0]; | |
798 | del_nr++; | |
ffd4bb2a | 799 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
800 | if (ret) { |
801 | btrfs_abort_transaction(trans, ret); | |
802 | goto out; | |
803 | } | |
920bbbfb YZ |
804 | } |
805 | if (del_nr == 0) { | |
3f6fae95 SL |
806 | fi = btrfs_item_ptr(leaf, path->slots[0], |
807 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
808 | btrfs_set_file_extent_type(leaf, fi, |
809 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 810 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 811 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 812 | } else { |
3f6fae95 SL |
813 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
814 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
815 | btrfs_set_file_extent_type(leaf, fi, |
816 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 817 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
818 | btrfs_set_file_extent_num_bytes(leaf, fi, |
819 | extent_end - key.offset); | |
820 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 821 | |
6c7d54ac | 822 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 823 | if (ret < 0) { |
66642832 | 824 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
825 | goto out; |
826 | } | |
6c7d54ac | 827 | } |
920bbbfb | 828 | out: |
d899e052 | 829 | btrfs_free_path(path); |
e7b2ec3d | 830 | return ret; |
d899e052 YZ |
831 | } |
832 | ||
b1bf862e CM |
833 | /* |
834 | * on error we return an unlocked page and the error value | |
835 | * on success we return a locked page and 0 | |
836 | */ | |
bb1591b4 CM |
837 | static int prepare_uptodate_page(struct inode *inode, |
838 | struct page *page, u64 pos, | |
b6316429 | 839 | bool force_uptodate) |
b1bf862e | 840 | { |
fb12489b | 841 | struct folio *folio = page_folio(page); |
b1bf862e CM |
842 | int ret = 0; |
843 | ||
09cbfeaf | 844 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 845 | !PageUptodate(page)) { |
fb12489b | 846 | ret = btrfs_read_folio(NULL, folio); |
b1bf862e CM |
847 | if (ret) |
848 | return ret; | |
849 | lock_page(page); | |
850 | if (!PageUptodate(page)) { | |
851 | unlock_page(page); | |
852 | return -EIO; | |
853 | } | |
e0467866 QW |
854 | |
855 | /* | |
fb12489b | 856 | * Since btrfs_read_folio() will unlock the folio before it |
f913cff3 | 857 | * returns, there is a window where btrfs_release_folio() can be |
7c11d0ae QW |
858 | * called to release the page. Here we check both inode |
859 | * mapping and PagePrivate() to make sure the page was not | |
860 | * released. | |
e0467866 QW |
861 | * |
862 | * The private flag check is essential for subpage as we need | |
863 | * to store extra bitmap using page->private. | |
864 | */ | |
865 | if (page->mapping != inode->i_mapping || !PagePrivate(page)) { | |
bb1591b4 CM |
866 | unlock_page(page); |
867 | return -EAGAIN; | |
868 | } | |
b1bf862e CM |
869 | } |
870 | return 0; | |
871 | } | |
872 | ||
fc226000 SR |
873 | static unsigned int get_prepare_fgp_flags(bool nowait) |
874 | { | |
875 | unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT; | |
876 | ||
877 | if (nowait) | |
878 | fgp_flags |= FGP_NOWAIT; | |
879 | ||
880 | return fgp_flags; | |
881 | } | |
882 | ||
883 | static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait) | |
884 | { | |
885 | gfp_t gfp; | |
886 | ||
887 | gfp = btrfs_alloc_write_mask(inode->i_mapping); | |
888 | if (nowait) { | |
889 | gfp &= ~__GFP_DIRECT_RECLAIM; | |
890 | gfp |= GFP_NOWAIT; | |
891 | } | |
892 | ||
893 | return gfp; | |
894 | } | |
895 | ||
39279cc3 | 896 | /* |
376cc685 | 897 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 898 | */ |
b37392ea MX |
899 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
900 | size_t num_pages, loff_t pos, | |
fc226000 SR |
901 | size_t write_bytes, bool force_uptodate, |
902 | bool nowait) | |
39279cc3 CM |
903 | { |
904 | int i; | |
09cbfeaf | 905 | unsigned long index = pos >> PAGE_SHIFT; |
fc226000 SR |
906 | gfp_t mask = get_prepare_gfp_flags(inode, nowait); |
907 | unsigned int fgp_flags = get_prepare_fgp_flags(nowait); | |
fc28b62d | 908 | int err = 0; |
376cc685 | 909 | int faili; |
8c2383c3 | 910 | |
39279cc3 | 911 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 912 | again: |
fc226000 SR |
913 | pages[i] = pagecache_get_page(inode->i_mapping, index + i, |
914 | fgp_flags, mask | __GFP_WRITE); | |
39279cc3 | 915 | if (!pages[i]) { |
b1bf862e | 916 | faili = i - 1; |
fc226000 SR |
917 | if (nowait) |
918 | err = -EAGAIN; | |
919 | else | |
920 | err = -ENOMEM; | |
b1bf862e CM |
921 | goto fail; |
922 | } | |
923 | ||
32443de3 QW |
924 | err = set_page_extent_mapped(pages[i]); |
925 | if (err < 0) { | |
926 | faili = i; | |
927 | goto fail; | |
928 | } | |
929 | ||
b1bf862e | 930 | if (i == 0) |
bb1591b4 | 931 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 932 | force_uptodate); |
bb1591b4 CM |
933 | if (!err && i == num_pages - 1) |
934 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 935 | pos + write_bytes, false); |
b1bf862e | 936 | if (err) { |
09cbfeaf | 937 | put_page(pages[i]); |
fc226000 | 938 | if (!nowait && err == -EAGAIN) { |
bb1591b4 CM |
939 | err = 0; |
940 | goto again; | |
941 | } | |
b1bf862e CM |
942 | faili = i - 1; |
943 | goto fail; | |
39279cc3 | 944 | } |
ccd467d6 | 945 | wait_on_page_writeback(pages[i]); |
39279cc3 | 946 | } |
376cc685 MX |
947 | |
948 | return 0; | |
949 | fail: | |
950 | while (faili >= 0) { | |
951 | unlock_page(pages[faili]); | |
09cbfeaf | 952 | put_page(pages[faili]); |
376cc685 MX |
953 | faili--; |
954 | } | |
955 | return err; | |
956 | ||
957 | } | |
958 | ||
959 | /* | |
960 | * This function locks the extent and properly waits for data=ordered extents | |
961 | * to finish before allowing the pages to be modified if need. | |
962 | * | |
963 | * The return value: | |
964 | * 1 - the extent is locked | |
965 | * 0 - the extent is not locked, and everything is OK | |
966 | * -EAGAIN - need re-prepare the pages | |
967 | * the other < 0 number - Something wrong happens | |
968 | */ | |
969 | static noinline int | |
2cff578c | 970 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 971 | size_t num_pages, loff_t pos, |
2e78c927 | 972 | size_t write_bytes, |
2fcab928 | 973 | u64 *lockstart, u64 *lockend, bool nowait, |
376cc685 MX |
974 | struct extent_state **cached_state) |
975 | { | |
3ffbd68c | 976 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
977 | u64 start_pos; |
978 | u64 last_pos; | |
979 | int i; | |
980 | int ret = 0; | |
981 | ||
0b246afa | 982 | start_pos = round_down(pos, fs_info->sectorsize); |
e21139c6 | 983 | last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1; |
376cc685 | 984 | |
e3b8a485 | 985 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 986 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 987 | |
2fcab928 | 988 | if (nowait) { |
83ae4133 JB |
989 | if (!try_lock_extent(&inode->io_tree, start_pos, last_pos, |
990 | cached_state)) { | |
2fcab928 SR |
991 | for (i = 0; i < num_pages; i++) { |
992 | unlock_page(pages[i]); | |
993 | put_page(pages[i]); | |
994 | pages[i] = NULL; | |
995 | } | |
996 | ||
997 | return -EAGAIN; | |
998 | } | |
999 | } else { | |
1000 | lock_extent(&inode->io_tree, start_pos, last_pos, cached_state); | |
1001 | } | |
1002 | ||
b88935bf MX |
1003 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1004 | last_pos - start_pos + 1); | |
e6dcd2dc | 1005 | if (ordered && |
bffe633e | 1006 | ordered->file_offset + ordered->num_bytes > start_pos && |
376cc685 | 1007 | ordered->file_offset <= last_pos) { |
570eb97b JB |
1008 | unlock_extent(&inode->io_tree, start_pos, last_pos, |
1009 | cached_state); | |
e6dcd2dc CM |
1010 | for (i = 0; i < num_pages; i++) { |
1011 | unlock_page(pages[i]); | |
09cbfeaf | 1012 | put_page(pages[i]); |
e6dcd2dc | 1013 | } |
c0a43603 | 1014 | btrfs_start_ordered_extent(ordered, 1); |
b88935bf MX |
1015 | btrfs_put_ordered_extent(ordered); |
1016 | return -EAGAIN; | |
e6dcd2dc CM |
1017 | } |
1018 | if (ordered) | |
1019 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1020 | |
376cc685 MX |
1021 | *lockstart = start_pos; |
1022 | *lockend = last_pos; | |
1023 | ret = 1; | |
0762704b | 1024 | } |
376cc685 | 1025 | |
7703bdd8 | 1026 | /* |
32443de3 QW |
1027 | * We should be called after prepare_pages() which should have locked |
1028 | * all pages in the range. | |
7703bdd8 | 1029 | */ |
32443de3 | 1030 | for (i = 0; i < num_pages; i++) |
e6dcd2dc | 1031 | WARN_ON(!PageLocked(pages[i])); |
b1bf862e | 1032 | |
376cc685 | 1033 | return ret; |
39279cc3 CM |
1034 | } |
1035 | ||
d7a8ab4e FM |
1036 | /* |
1037 | * Check if we can do nocow write into the range [@pos, @pos + @write_bytes) | |
1038 | * | |
1039 | * @pos: File offset. | |
1040 | * @write_bytes: The length to write, will be updated to the nocow writeable | |
1041 | * range. | |
1042 | * | |
1043 | * This function will flush ordered extents in the range to ensure proper | |
1044 | * nocow checks. | |
1045 | * | |
1046 | * Return: | |
1047 | * > 0 If we can nocow, and updates @write_bytes. | |
1048 | * 0 If we can't do a nocow write. | |
1049 | * -EAGAIN If we can't do a nocow write because snapshoting of the inode's | |
1050 | * root is in progress. | |
1051 | * < 0 If an error happened. | |
1052 | * | |
1053 | * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. | |
1054 | */ | |
1055 | int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, | |
80f9d241 | 1056 | size_t *write_bytes, bool nowait) |
7ee9e440 | 1057 | { |
3ffbd68c | 1058 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1059 | struct btrfs_root *root = inode->root; |
632ddfa2 | 1060 | struct extent_state *cached_state = NULL; |
7ee9e440 JB |
1061 | u64 lockstart, lockend; |
1062 | u64 num_bytes; | |
1063 | int ret; | |
1064 | ||
38d37aa9 QW |
1065 | if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) |
1066 | return 0; | |
1067 | ||
d7a8ab4e | 1068 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) |
5f791ec3 | 1069 | return -EAGAIN; |
8257b2dc | 1070 | |
0b246afa | 1071 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1072 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1073 | fs_info->sectorsize) - 1; |
5dbb75ed | 1074 | num_bytes = lockend - lockstart + 1; |
7ee9e440 | 1075 | |
80f9d241 | 1076 | if (nowait) { |
632ddfa2 JB |
1077 | if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend, |
1078 | &cached_state)) { | |
80f9d241 JB |
1079 | btrfs_drew_write_unlock(&root->snapshot_lock); |
1080 | return -EAGAIN; | |
1081 | } | |
1082 | } else { | |
632ddfa2 JB |
1083 | btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, |
1084 | &cached_state); | |
80f9d241 | 1085 | } |
85b7ab67 | 1086 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
80f9d241 JB |
1087 | NULL, NULL, NULL, nowait, false); |
1088 | if (ret <= 0) | |
d7a8ab4e | 1089 | btrfs_drew_write_unlock(&root->snapshot_lock); |
80f9d241 | 1090 | else |
c933956d MX |
1091 | *write_bytes = min_t(size_t, *write_bytes , |
1092 | num_bytes - pos + lockstart); | |
632ddfa2 | 1093 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
7ee9e440 JB |
1094 | |
1095 | return ret; | |
1096 | } | |
1097 | ||
38d37aa9 QW |
1098 | void btrfs_check_nocow_unlock(struct btrfs_inode *inode) |
1099 | { | |
1100 | btrfs_drew_write_unlock(&inode->root->snapshot_lock); | |
1101 | } | |
1102 | ||
b8d8e1fd GR |
1103 | static void update_time_for_write(struct inode *inode) |
1104 | { | |
1105 | struct timespec64 now; | |
1106 | ||
1107 | if (IS_NOCMTIME(inode)) | |
1108 | return; | |
1109 | ||
1110 | now = current_time(inode); | |
1111 | if (!timespec64_equal(&inode->i_mtime, &now)) | |
1112 | inode->i_mtime = now; | |
1113 | ||
1114 | if (!timespec64_equal(&inode->i_ctime, &now)) | |
1115 | inode->i_ctime = now; | |
1116 | ||
1117 | if (IS_I_VERSION(inode)) | |
1118 | inode_inc_iversion(inode); | |
1119 | } | |
1120 | ||
1121 | static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from, | |
1122 | size_t count) | |
1123 | { | |
1124 | struct file *file = iocb->ki_filp; | |
1125 | struct inode *inode = file_inode(file); | |
1126 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1127 | loff_t pos = iocb->ki_pos; | |
1128 | int ret; | |
1129 | loff_t oldsize; | |
1130 | loff_t start_pos; | |
1131 | ||
d7a8ab4e FM |
1132 | /* |
1133 | * Quickly bail out on NOWAIT writes if we don't have the nodatacow or | |
1134 | * prealloc flags, as without those flags we always have to COW. We will | |
1135 | * later check if we can really COW into the target range (using | |
1136 | * can_nocow_extent() at btrfs_get_blocks_direct_write()). | |
1137 | */ | |
1138 | if ((iocb->ki_flags & IOCB_NOWAIT) && | |
1139 | !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
1140 | return -EAGAIN; | |
b8d8e1fd GR |
1141 | |
1142 | current->backing_dev_info = inode_to_bdi(inode); | |
1143 | ret = file_remove_privs(file); | |
1144 | if (ret) | |
1145 | return ret; | |
1146 | ||
1147 | /* | |
1148 | * We reserve space for updating the inode when we reserve space for the | |
1149 | * extent we are going to write, so we will enospc out there. We don't | |
1150 | * need to start yet another transaction to update the inode as we will | |
1151 | * update the inode when we finish writing whatever data we write. | |
1152 | */ | |
1153 | update_time_for_write(inode); | |
1154 | ||
1155 | start_pos = round_down(pos, fs_info->sectorsize); | |
1156 | oldsize = i_size_read(inode); | |
1157 | if (start_pos > oldsize) { | |
1158 | /* Expand hole size to cover write data, preventing empty gap */ | |
1159 | loff_t end_pos = round_up(pos + count, fs_info->sectorsize); | |
1160 | ||
b06359a3 | 1161 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos); |
b8d8e1fd GR |
1162 | if (ret) { |
1163 | current->backing_dev_info = NULL; | |
1164 | return ret; | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | return 0; | |
1169 | } | |
1170 | ||
e4af400a GR |
1171 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1172 | struct iov_iter *i) | |
4b46fce2 | 1173 | { |
e4af400a | 1174 | struct file *file = iocb->ki_filp; |
c3523706 | 1175 | loff_t pos; |
496ad9aa | 1176 | struct inode *inode = file_inode(file); |
0b246afa | 1177 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1178 | struct page **pages = NULL; |
364ecf36 | 1179 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1180 | u64 release_bytes = 0; |
376cc685 MX |
1181 | u64 lockstart; |
1182 | u64 lockend; | |
d0215f3e JB |
1183 | size_t num_written = 0; |
1184 | int nrptrs; | |
c3523706 | 1185 | ssize_t ret; |
7ee9e440 | 1186 | bool only_release_metadata = false; |
b6316429 | 1187 | bool force_page_uptodate = false; |
5e8b9ef3 | 1188 | loff_t old_isize = i_size_read(inode); |
c3523706 | 1189 | unsigned int ilock_flags = 0; |
304e45ac | 1190 | const bool nowait = (iocb->ki_flags & IOCB_NOWAIT); |
965f47ae | 1191 | unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0); |
c3523706 | 1192 | |
304e45ac | 1193 | if (nowait) |
c3523706 GR |
1194 | ilock_flags |= BTRFS_ILOCK_TRY; |
1195 | ||
29b6352b | 1196 | ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1197 | if (ret < 0) |
1198 | return ret; | |
4b46fce2 | 1199 | |
c3523706 GR |
1200 | ret = generic_write_checks(iocb, i); |
1201 | if (ret <= 0) | |
1202 | goto out; | |
1203 | ||
1204 | ret = btrfs_write_check(iocb, i, ret); | |
1205 | if (ret < 0) | |
1206 | goto out; | |
1207 | ||
1208 | pos = iocb->ki_pos; | |
09cbfeaf KS |
1209 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1210 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1211 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1212 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1213 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
c3523706 GR |
1214 | if (!pages) { |
1215 | ret = -ENOMEM; | |
1216 | goto out; | |
1217 | } | |
ab93dbec | 1218 | |
d0215f3e | 1219 | while (iov_iter_count(i) > 0) { |
c67d970f | 1220 | struct extent_state *cached_state = NULL; |
7073017a | 1221 | size_t offset = offset_in_page(pos); |
2e78c927 | 1222 | size_t sector_offset; |
d0215f3e | 1223 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1224 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1225 | offset); |
eefa45f5 | 1226 | size_t num_pages; |
7ee9e440 | 1227 | size_t reserve_bytes; |
d0215f3e JB |
1228 | size_t dirty_pages; |
1229 | size_t copied; | |
2e78c927 CR |
1230 | size_t dirty_sectors; |
1231 | size_t num_sectors; | |
79f015f2 | 1232 | int extents_locked; |
39279cc3 | 1233 | |
914ee295 XZ |
1234 | /* |
1235 | * Fault pages before locking them in prepare_pages | |
1236 | * to avoid recursive lock | |
1237 | */ | |
a6294593 | 1238 | if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) { |
914ee295 | 1239 | ret = -EFAULT; |
d0215f3e | 1240 | break; |
914ee295 XZ |
1241 | } |
1242 | ||
a0e248bb | 1243 | only_release_metadata = false; |
da17066c | 1244 | sector_offset = pos & (fs_info->sectorsize - 1); |
d9d8b2a5 | 1245 | |
364ecf36 | 1246 | extent_changeset_release(data_reserved); |
36ea6f3e NB |
1247 | ret = btrfs_check_data_free_space(BTRFS_I(inode), |
1248 | &data_reserved, pos, | |
304e45ac | 1249 | write_bytes, nowait); |
c6887cd1 | 1250 | if (ret < 0) { |
80f9d241 JB |
1251 | int can_nocow; |
1252 | ||
304e45ac SR |
1253 | if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) { |
1254 | ret = -EAGAIN; | |
1255 | break; | |
1256 | } | |
1257 | ||
eefa45f5 GR |
1258 | /* |
1259 | * If we don't have to COW at the offset, reserve | |
1260 | * metadata only. write_bytes may get smaller than | |
1261 | * requested here. | |
1262 | */ | |
80f9d241 | 1263 | can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos, |
304e45ac | 1264 | &write_bytes, nowait); |
80f9d241 JB |
1265 | if (can_nocow < 0) |
1266 | ret = can_nocow; | |
1267 | if (can_nocow > 0) | |
1268 | ret = 0; | |
1269 | if (ret) | |
c6887cd1 | 1270 | break; |
80f9d241 | 1271 | only_release_metadata = true; |
c6887cd1 | 1272 | } |
1832a6d5 | 1273 | |
eefa45f5 GR |
1274 | num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE); |
1275 | WARN_ON(num_pages > nrptrs); | |
1276 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1277 | fs_info->sectorsize); | |
8b62f87b | 1278 | WARN_ON(reserve_bytes == 0); |
9f3db423 | 1279 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
28c9b1e7 | 1280 | reserve_bytes, |
304e45ac | 1281 | reserve_bytes, nowait); |
7ee9e440 JB |
1282 | if (ret) { |
1283 | if (!only_release_metadata) | |
25ce28ca | 1284 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
bc42bda2 QW |
1285 | data_reserved, pos, |
1286 | write_bytes); | |
8257b2dc | 1287 | else |
38d37aa9 | 1288 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
a348c8d4 FM |
1289 | |
1290 | if (nowait && ret == -ENOSPC) | |
1291 | ret = -EAGAIN; | |
7ee9e440 JB |
1292 | break; |
1293 | } | |
1294 | ||
1295 | release_bytes = reserve_bytes; | |
376cc685 | 1296 | again: |
965f47ae | 1297 | ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags); |
eb81b682 FM |
1298 | if (ret) { |
1299 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); | |
965f47ae | 1300 | break; |
eb81b682 | 1301 | } |
965f47ae | 1302 | |
4a64001f JB |
1303 | /* |
1304 | * This is going to setup the pages array with the number of | |
1305 | * pages we want, so we don't really need to worry about the | |
1306 | * contents of pages from loop to loop | |
1307 | */ | |
b37392ea | 1308 | ret = prepare_pages(inode, pages, num_pages, |
fc226000 | 1309 | pos, write_bytes, force_page_uptodate, false); |
8b62f87b JB |
1310 | if (ret) { |
1311 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
8702ba93 | 1312 | reserve_bytes); |
d0215f3e | 1313 | break; |
8b62f87b | 1314 | } |
39279cc3 | 1315 | |
79f015f2 GR |
1316 | extents_locked = lock_and_cleanup_extent_if_need( |
1317 | BTRFS_I(inode), pages, | |
2cff578c | 1318 | num_pages, pos, write_bytes, &lockstart, |
304e45ac | 1319 | &lockend, nowait, &cached_state); |
79f015f2 | 1320 | if (extents_locked < 0) { |
304e45ac | 1321 | if (!nowait && extents_locked == -EAGAIN) |
376cc685 | 1322 | goto again; |
304e45ac | 1323 | |
8b62f87b | 1324 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
8702ba93 | 1325 | reserve_bytes); |
79f015f2 | 1326 | ret = extents_locked; |
376cc685 | 1327 | break; |
376cc685 MX |
1328 | } |
1329 | ||
ee22f0c4 | 1330 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1331 | |
0b246afa | 1332 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1333 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1334 | fs_info->sectorsize); |
1335 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1336 | |
b1bf862e CM |
1337 | /* |
1338 | * if we have trouble faulting in the pages, fall | |
1339 | * back to one page at a time | |
1340 | */ | |
1341 | if (copied < write_bytes) | |
1342 | nrptrs = 1; | |
1343 | ||
b6316429 JB |
1344 | if (copied == 0) { |
1345 | force_page_uptodate = true; | |
56244ef1 | 1346 | dirty_sectors = 0; |
b1bf862e | 1347 | dirty_pages = 0; |
b6316429 JB |
1348 | } else { |
1349 | force_page_uptodate = false; | |
ed6078f7 | 1350 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1351 | PAGE_SIZE); |
b6316429 | 1352 | } |
914ee295 | 1353 | |
2e78c927 | 1354 | if (num_sectors > dirty_sectors) { |
8b8b08cb | 1355 | /* release everything except the sectors we dirtied */ |
265fdfa6 | 1356 | release_bytes -= dirty_sectors << fs_info->sectorsize_bits; |
485290a7 | 1357 | if (only_release_metadata) { |
691fa059 | 1358 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1359 | release_bytes, true); |
485290a7 QW |
1360 | } else { |
1361 | u64 __pos; | |
1362 | ||
da17066c | 1363 | __pos = round_down(pos, |
0b246afa | 1364 | fs_info->sectorsize) + |
09cbfeaf | 1365 | (dirty_pages << PAGE_SHIFT); |
86d52921 | 1366 | btrfs_delalloc_release_space(BTRFS_I(inode), |
bc42bda2 | 1367 | data_reserved, __pos, |
43b18595 | 1368 | release_bytes, true); |
485290a7 | 1369 | } |
914ee295 XZ |
1370 | } |
1371 | ||
2e78c927 | 1372 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1373 | fs_info->sectorsize); |
376cc685 | 1374 | |
aa8c1a41 GR |
1375 | ret = btrfs_dirty_pages(BTRFS_I(inode), pages, |
1376 | dirty_pages, pos, copied, | |
1377 | &cached_state, only_release_metadata); | |
c67d970f FM |
1378 | |
1379 | /* | |
1380 | * If we have not locked the extent range, because the range's | |
1381 | * start offset is >= i_size, we might still have a non-NULL | |
1382 | * cached extent state, acquired while marking the extent range | |
1383 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1384 | * possible cached extent state to avoid a memory leak. | |
1385 | */ | |
79f015f2 | 1386 | if (extents_locked) |
570eb97b JB |
1387 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
1388 | lockend, &cached_state); | |
c67d970f FM |
1389 | else |
1390 | free_extent_state(cached_state); | |
1391 | ||
8702ba93 | 1392 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 | 1393 | if (ret) { |
e4f94347 | 1394 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
376cc685 | 1395 | break; |
f1de9683 | 1396 | } |
39279cc3 | 1397 | |
376cc685 | 1398 | release_bytes = 0; |
8257b2dc | 1399 | if (only_release_metadata) |
38d37aa9 | 1400 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
8257b2dc | 1401 | |
e4f94347 | 1402 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
f1de9683 | 1403 | |
d0215f3e JB |
1404 | cond_resched(); |
1405 | ||
914ee295 XZ |
1406 | pos += copied; |
1407 | num_written += copied; | |
d0215f3e | 1408 | } |
39279cc3 | 1409 | |
d0215f3e JB |
1410 | kfree(pages); |
1411 | ||
7ee9e440 | 1412 | if (release_bytes) { |
8257b2dc | 1413 | if (only_release_metadata) { |
38d37aa9 | 1414 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
691fa059 | 1415 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1416 | release_bytes, true); |
8257b2dc | 1417 | } else { |
86d52921 NB |
1418 | btrfs_delalloc_release_space(BTRFS_I(inode), |
1419 | data_reserved, | |
bc42bda2 | 1420 | round_down(pos, fs_info->sectorsize), |
43b18595 | 1421 | release_bytes, true); |
8257b2dc | 1422 | } |
7ee9e440 JB |
1423 | } |
1424 | ||
364ecf36 | 1425 | extent_changeset_free(data_reserved); |
5e8b9ef3 GR |
1426 | if (num_written > 0) { |
1427 | pagecache_isize_extended(inode, old_isize, iocb->ki_pos); | |
1428 | iocb->ki_pos += num_written; | |
1429 | } | |
c3523706 GR |
1430 | out: |
1431 | btrfs_inode_unlock(inode, ilock_flags); | |
d0215f3e JB |
1432 | return num_written ? num_written : ret; |
1433 | } | |
1434 | ||
4e4cabec GR |
1435 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
1436 | const struct iov_iter *iter, loff_t offset) | |
1437 | { | |
1438 | const u32 blocksize_mask = fs_info->sectorsize - 1; | |
1439 | ||
1440 | if (offset & blocksize_mask) | |
1441 | return -EINVAL; | |
1442 | ||
1443 | if (iov_iter_alignment(iter) & blocksize_mask) | |
1444 | return -EINVAL; | |
1445 | ||
1446 | return 0; | |
1447 | } | |
1448 | ||
1449 | static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
d0215f3e JB |
1450 | { |
1451 | struct file *file = iocb->ki_filp; | |
728404da | 1452 | struct inode *inode = file_inode(file); |
4e4cabec | 1453 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c3523706 | 1454 | loff_t pos; |
4e4cabec | 1455 | ssize_t written = 0; |
d0215f3e | 1456 | ssize_t written_buffered; |
51bd9563 | 1457 | size_t prev_left = 0; |
d0215f3e | 1458 | loff_t endbyte; |
c3523706 GR |
1459 | ssize_t err; |
1460 | unsigned int ilock_flags = 0; | |
8184620a | 1461 | struct iomap_dio *dio; |
c3523706 GR |
1462 | |
1463 | if (iocb->ki_flags & IOCB_NOWAIT) | |
1464 | ilock_flags |= BTRFS_ILOCK_TRY; | |
1465 | ||
e9adabb9 GR |
1466 | /* If the write DIO is within EOF, use a shared lock */ |
1467 | if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode)) | |
1468 | ilock_flags |= BTRFS_ILOCK_SHARED; | |
1469 | ||
1470 | relock: | |
29b6352b | 1471 | err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1472 | if (err < 0) |
1473 | return err; | |
1474 | ||
1475 | err = generic_write_checks(iocb, from); | |
1476 | if (err <= 0) { | |
1477 | btrfs_inode_unlock(inode, ilock_flags); | |
1478 | return err; | |
1479 | } | |
d0215f3e | 1480 | |
c3523706 GR |
1481 | err = btrfs_write_check(iocb, from, err); |
1482 | if (err < 0) { | |
1483 | btrfs_inode_unlock(inode, ilock_flags); | |
1484 | goto out; | |
1485 | } | |
1486 | ||
1487 | pos = iocb->ki_pos; | |
e9adabb9 GR |
1488 | /* |
1489 | * Re-check since file size may have changed just before taking the | |
1490 | * lock or pos may have changed because of O_APPEND in generic_write_check() | |
1491 | */ | |
1492 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && | |
1493 | pos + iov_iter_count(from) > i_size_read(inode)) { | |
1494 | btrfs_inode_unlock(inode, ilock_flags); | |
1495 | ilock_flags &= ~BTRFS_ILOCK_SHARED; | |
1496 | goto relock; | |
1497 | } | |
c3523706 GR |
1498 | |
1499 | if (check_direct_IO(fs_info, from, pos)) { | |
1500 | btrfs_inode_unlock(inode, ilock_flags); | |
4e4cabec | 1501 | goto buffered; |
c3523706 | 1502 | } |
4e4cabec | 1503 | |
51bd9563 FM |
1504 | /* |
1505 | * The iov_iter can be mapped to the same file range we are writing to. | |
1506 | * If that's the case, then we will deadlock in the iomap code, because | |
1507 | * it first calls our callback btrfs_dio_iomap_begin(), which will create | |
1508 | * an ordered extent, and after that it will fault in the pages that the | |
1509 | * iov_iter refers to. During the fault in we end up in the readahead | |
1510 | * pages code (starting at btrfs_readahead()), which will lock the range, | |
1511 | * find that ordered extent and then wait for it to complete (at | |
1512 | * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since | |
1513 | * obviously the ordered extent can never complete as we didn't submit | |
1514 | * yet the respective bio(s). This always happens when the buffer is | |
1515 | * memory mapped to the same file range, since the iomap DIO code always | |
1516 | * invalidates pages in the target file range (after starting and waiting | |
1517 | * for any writeback). | |
1518 | * | |
1519 | * So here we disable page faults in the iov_iter and then retry if we | |
1520 | * got -EFAULT, faulting in the pages before the retry. | |
1521 | */ | |
51bd9563 | 1522 | from->nofault = true; |
8184620a | 1523 | dio = btrfs_dio_write(iocb, from, written); |
51bd9563 | 1524 | from->nofault = false; |
d0215f3e | 1525 | |
8184620a FM |
1526 | /* |
1527 | * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync | |
1528 | * iocb, and that needs to lock the inode. So unlock it before calling | |
1529 | * iomap_dio_complete() to avoid a deadlock. | |
1530 | */ | |
1531 | btrfs_inode_unlock(inode, ilock_flags); | |
1532 | ||
1533 | if (IS_ERR_OR_NULL(dio)) | |
1534 | err = PTR_ERR_OR_ZERO(dio); | |
1535 | else | |
1536 | err = iomap_dio_complete(dio); | |
1537 | ||
51bd9563 FM |
1538 | /* No increment (+=) because iomap returns a cumulative value. */ |
1539 | if (err > 0) | |
1540 | written = err; | |
1541 | ||
1542 | if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) { | |
1543 | const size_t left = iov_iter_count(from); | |
1544 | /* | |
1545 | * We have more data left to write. Try to fault in as many as | |
1546 | * possible of the remainder pages and retry. We do this without | |
1547 | * releasing and locking again the inode, to prevent races with | |
1548 | * truncate. | |
1549 | * | |
1550 | * Also, in case the iov refers to pages in the file range of the | |
1551 | * file we want to write to (due to a mmap), we could enter an | |
1552 | * infinite loop if we retry after faulting the pages in, since | |
1553 | * iomap will invalidate any pages in the range early on, before | |
1554 | * it tries to fault in the pages of the iov. So we keep track of | |
1555 | * how much was left of iov in the previous EFAULT and fallback | |
1556 | * to buffered IO in case we haven't made any progress. | |
1557 | */ | |
1558 | if (left == prev_left) { | |
1559 | err = -ENOTBLK; | |
1560 | } else { | |
1561 | fault_in_iov_iter_readable(from, left); | |
1562 | prev_left = left; | |
8184620a | 1563 | goto relock; |
51bd9563 | 1564 | } |
a42fa643 GR |
1565 | } |
1566 | ||
ac5e6669 FM |
1567 | /* |
1568 | * If 'err' is -ENOTBLK or we have not written all data, then it means | |
1569 | * we must fallback to buffered IO. | |
1570 | */ | |
51bd9563 | 1571 | if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from)) |
c3523706 | 1572 | goto out; |
d0215f3e | 1573 | |
4e4cabec | 1574 | buffered: |
ac5e6669 FM |
1575 | /* |
1576 | * If we are in a NOWAIT context, then return -EAGAIN to signal the caller | |
1577 | * it must retry the operation in a context where blocking is acceptable, | |
20af93d9 FM |
1578 | * because even if we end up not blocking during the buffered IO attempt |
1579 | * below, we will block when flushing and waiting for the IO. | |
ac5e6669 FM |
1580 | */ |
1581 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
1582 | err = -EAGAIN; | |
1583 | goto out; | |
1584 | } | |
1585 | ||
e4af400a GR |
1586 | pos = iocb->ki_pos; |
1587 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e JB |
1588 | if (written_buffered < 0) { |
1589 | err = written_buffered; | |
1590 | goto out; | |
39279cc3 | 1591 | } |
075bdbdb FM |
1592 | /* |
1593 | * Ensure all data is persisted. We want the next direct IO read to be | |
1594 | * able to read what was just written. | |
1595 | */ | |
d0215f3e | 1596 | endbyte = pos + written_buffered - 1; |
728404da | 1597 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1598 | if (err) |
1599 | goto out; | |
728404da | 1600 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1601 | if (err) |
1602 | goto out; | |
1603 | written += written_buffered; | |
867c4f93 | 1604 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1605 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1606 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1607 | out: |
51bd9563 | 1608 | return err < 0 ? err : written; |
d0215f3e | 1609 | } |
5b92ee72 | 1610 | |
7c0c7269 OS |
1611 | static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
1612 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
1613 | { | |
1614 | struct file *file = iocb->ki_filp; | |
1615 | struct inode *inode = file_inode(file); | |
1616 | loff_t count; | |
1617 | ssize_t ret; | |
1618 | ||
29b6352b | 1619 | btrfs_inode_lock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1620 | count = encoded->len; |
1621 | ret = generic_write_checks_count(iocb, &count); | |
1622 | if (ret == 0 && count != encoded->len) { | |
1623 | /* | |
1624 | * The write got truncated by generic_write_checks_count(). We | |
1625 | * can't do a partial encoded write. | |
1626 | */ | |
1627 | ret = -EFBIG; | |
1628 | } | |
1629 | if (ret || encoded->len == 0) | |
1630 | goto out; | |
1631 | ||
1632 | ret = btrfs_write_check(iocb, from, encoded->len); | |
1633 | if (ret < 0) | |
1634 | goto out; | |
1635 | ||
1636 | ret = btrfs_do_encoded_write(iocb, from, encoded); | |
1637 | out: | |
1638 | btrfs_inode_unlock(inode, 0); | |
1639 | return ret; | |
1640 | } | |
1641 | ||
1642 | ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from, | |
1643 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
d0215f3e JB |
1644 | { |
1645 | struct file *file = iocb->ki_filp; | |
14971657 | 1646 | struct btrfs_inode *inode = BTRFS_I(file_inode(file)); |
7c0c7269 | 1647 | ssize_t num_written, num_sync; |
91b94c5d | 1648 | const bool sync = iocb_is_dsync(iocb); |
d0215f3e | 1649 | |
c86537a4 GR |
1650 | /* |
1651 | * If the fs flips readonly due to some impossible error, although we | |
1652 | * have opened a file as writable, we have to stop this write operation | |
1653 | * to ensure consistency. | |
1654 | */ | |
84961539 | 1655 | if (BTRFS_FS_ERROR(inode->root->fs_info)) |
c86537a4 GR |
1656 | return -EROFS; |
1657 | ||
926078b2 | 1658 | if (encoded && (iocb->ki_flags & IOCB_NOWAIT)) |
91f9943e CH |
1659 | return -EOPNOTSUPP; |
1660 | ||
b812ce28 | 1661 | if (sync) |
14971657 | 1662 | atomic_inc(&inode->sync_writers); |
b812ce28 | 1663 | |
7c0c7269 OS |
1664 | if (encoded) { |
1665 | num_written = btrfs_encoded_write(iocb, from, encoded); | |
1666 | num_sync = encoded->len; | |
1667 | } else if (iocb->ki_flags & IOCB_DIRECT) { | |
c1867eb3 DS |
1668 | num_written = btrfs_direct_write(iocb, from); |
1669 | num_sync = num_written; | |
7c0c7269 | 1670 | } else { |
c1867eb3 DS |
1671 | num_written = btrfs_buffered_write(iocb, from); |
1672 | num_sync = num_written; | |
7c0c7269 | 1673 | } |
d0215f3e | 1674 | |
bc0939fc FM |
1675 | btrfs_set_inode_last_sub_trans(inode); |
1676 | ||
7c0c7269 OS |
1677 | if (num_sync > 0) { |
1678 | num_sync = generic_write_sync(iocb, num_sync); | |
1679 | if (num_sync < 0) | |
1680 | num_written = num_sync; | |
1681 | } | |
0a3404dc | 1682 | |
b812ce28 | 1683 | if (sync) |
14971657 | 1684 | atomic_dec(&inode->sync_writers); |
b8d8e1fd | 1685 | |
39279cc3 | 1686 | current->backing_dev_info = NULL; |
c3523706 | 1687 | return num_written; |
39279cc3 CM |
1688 | } |
1689 | ||
7c0c7269 OS |
1690 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1691 | { | |
1692 | return btrfs_do_write_iter(iocb, from, NULL); | |
1693 | } | |
1694 | ||
d397712b | 1695 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1696 | { |
23b5ec74 JB |
1697 | struct btrfs_file_private *private = filp->private_data; |
1698 | ||
23b5ec74 JB |
1699 | if (private && private->filldir_buf) |
1700 | kfree(private->filldir_buf); | |
1701 | kfree(private); | |
1702 | filp->private_data = NULL; | |
1703 | ||
f6dc45c7 | 1704 | /* |
1fd4033d NB |
1705 | * Set by setattr when we are about to truncate a file from a non-zero |
1706 | * size to a zero size. This tries to flush down new bytes that may | |
1707 | * have been written if the application were using truncate to replace | |
1708 | * a file in place. | |
f6dc45c7 | 1709 | */ |
1fd4033d | 1710 | if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
f6dc45c7 CM |
1711 | &BTRFS_I(inode)->runtime_flags)) |
1712 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
1713 | return 0; |
1714 | } | |
1715 | ||
669249ee FM |
1716 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
1717 | { | |
1718 | int ret; | |
343e4fc1 | 1719 | struct blk_plug plug; |
669249ee | 1720 | |
343e4fc1 LB |
1721 | /* |
1722 | * This is only called in fsync, which would do synchronous writes, so | |
1723 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
1724 | * multiple disks using raid profile, a large IO can be split to | |
1725 | * several segments of stripe length (currently 64K). | |
1726 | */ | |
1727 | blk_start_plug(&plug); | |
669249ee | 1728 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
728404da | 1729 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee | 1730 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
343e4fc1 | 1731 | blk_finish_plug(&plug); |
669249ee FM |
1732 | |
1733 | return ret; | |
1734 | } | |
1735 | ||
626e9f41 FM |
1736 | static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx) |
1737 | { | |
1738 | struct btrfs_inode *inode = BTRFS_I(ctx->inode); | |
1739 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1740 | ||
1741 | if (btrfs_inode_in_log(inode, fs_info->generation) && | |
1742 | list_empty(&ctx->ordered_extents)) | |
1743 | return true; | |
1744 | ||
1745 | /* | |
1746 | * If we are doing a fast fsync we can not bail out if the inode's | |
1747 | * last_trans is <= then the last committed transaction, because we only | |
1748 | * update the last_trans of the inode during ordered extent completion, | |
1749 | * and for a fast fsync we don't wait for that, we only wait for the | |
1750 | * writeback to complete. | |
1751 | */ | |
1752 | if (inode->last_trans <= fs_info->last_trans_committed && | |
1753 | (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) || | |
1754 | list_empty(&ctx->ordered_extents))) | |
1755 | return true; | |
1756 | ||
1757 | return false; | |
1758 | } | |
1759 | ||
d352ac68 CM |
1760 | /* |
1761 | * fsync call for both files and directories. This logs the inode into | |
1762 | * the tree log instead of forcing full commits whenever possible. | |
1763 | * | |
1764 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1765 | * in the metadata btree are up to date for copying to the log. | |
1766 | * | |
1767 | * It drops the inode mutex before doing the tree log commit. This is an | |
1768 | * important optimization for directories because holding the mutex prevents | |
1769 | * new operations on the dir while we write to disk. | |
1770 | */ | |
02c24a82 | 1771 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1772 | { |
de17e793 | 1773 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 1774 | struct inode *inode = d_inode(dentry); |
0b246afa | 1775 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 1776 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 1777 | struct btrfs_trans_handle *trans; |
8b050d35 | 1778 | struct btrfs_log_ctx ctx; |
333427a5 | 1779 | int ret = 0, err; |
48778179 FM |
1780 | u64 len; |
1781 | bool full_sync; | |
39279cc3 | 1782 | |
1abe9b8a | 1783 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1784 | |
ebb70442 LB |
1785 | btrfs_init_log_ctx(&ctx, inode); |
1786 | ||
95418ed1 | 1787 | /* |
48778179 FM |
1788 | * Always set the range to a full range, otherwise we can get into |
1789 | * several problems, from missing file extent items to represent holes | |
1790 | * when not using the NO_HOLES feature, to log tree corruption due to | |
1791 | * races between hole detection during logging and completion of ordered | |
1792 | * extents outside the range, to missing checksums due to ordered extents | |
1793 | * for which we flushed only a subset of their pages. | |
95418ed1 | 1794 | */ |
48778179 FM |
1795 | start = 0; |
1796 | end = LLONG_MAX; | |
1797 | len = (u64)LLONG_MAX + 1; | |
95418ed1 | 1798 | |
90abccf2 MX |
1799 | /* |
1800 | * We write the dirty pages in the range and wait until they complete | |
1801 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1802 | * multi-task, and make the performance up. See |
1803 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1804 | */ |
669249ee | 1805 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 1806 | if (ret) |
333427a5 | 1807 | goto out; |
90abccf2 | 1808 | |
29b6352b | 1809 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
c495144b | 1810 | |
2ecb7923 | 1811 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 1812 | |
aab15e8e | 1813 | /* |
885f46d8 FM |
1814 | * Before we acquired the inode's lock and the mmap lock, someone may |
1815 | * have dirtied more pages in the target range. We need to make sure | |
1816 | * that writeback for any such pages does not start while we are logging | |
1817 | * the inode, because if it does, any of the following might happen when | |
1818 | * we are not doing a full inode sync: | |
aab15e8e FM |
1819 | * |
1820 | * 1) We log an extent after its writeback finishes but before its | |
1821 | * checksums are added to the csum tree, leading to -EIO errors | |
1822 | * when attempting to read the extent after a log replay. | |
1823 | * | |
1824 | * 2) We can end up logging an extent before its writeback finishes. | |
1825 | * Therefore after the log replay we will have a file extent item | |
1826 | * pointing to an unwritten extent (and no data checksums as well). | |
1827 | * | |
1828 | * So trigger writeback for any eventual new dirty pages and then we | |
1829 | * wait for all ordered extents to complete below. | |
1830 | */ | |
1831 | ret = start_ordered_ops(inode, start, end); | |
1832 | if (ret) { | |
885f46d8 | 1833 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
aab15e8e FM |
1834 | goto out; |
1835 | } | |
1836 | ||
cef7820d FM |
1837 | /* |
1838 | * Always check for the full sync flag while holding the inode's lock, | |
1839 | * to avoid races with other tasks. The flag must be either set all the | |
1840 | * time during logging or always off all the time while logging. | |
1841 | * We check the flag here after starting delalloc above, because when | |
1842 | * running delalloc the full sync flag may be set if we need to drop | |
1843 | * extra extent map ranges due to temporary memory allocation failures. | |
1844 | */ | |
1845 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1846 | &BTRFS_I(inode)->runtime_flags); | |
1847 | ||
669249ee | 1848 | /* |
b5e6c3e1 | 1849 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 1850 | * IO of a lower priority task while holding a transaction open. |
ba0b084a | 1851 | * |
48778179 FM |
1852 | * For a full fsync we wait for the ordered extents to complete while |
1853 | * for a fast fsync we wait just for writeback to complete, and then | |
1854 | * attach the ordered extents to the transaction so that a transaction | |
1855 | * commit waits for their completion, to avoid data loss if we fsync, | |
1856 | * the current transaction commits before the ordered extents complete | |
1857 | * and a power failure happens right after that. | |
d8e3fb10 NA |
1858 | * |
1859 | * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the | |
1860 | * logical address recorded in the ordered extent may change. We need | |
1861 | * to wait for the IO to stabilize the logical address. | |
669249ee | 1862 | */ |
d8e3fb10 | 1863 | if (full_sync || btrfs_is_zoned(fs_info)) { |
48778179 FM |
1864 | ret = btrfs_wait_ordered_range(inode, start, len); |
1865 | } else { | |
1866 | /* | |
1867 | * Get our ordered extents as soon as possible to avoid doing | |
1868 | * checksum lookups in the csum tree, and use instead the | |
1869 | * checksums attached to the ordered extents. | |
1870 | */ | |
1871 | btrfs_get_ordered_extents_for_logging(BTRFS_I(inode), | |
1872 | &ctx.ordered_extents); | |
1873 | ret = filemap_fdatawait_range(inode->i_mapping, start, end); | |
0ef8b726 | 1874 | } |
48778179 FM |
1875 | |
1876 | if (ret) | |
1877 | goto out_release_extents; | |
1878 | ||
2ecb7923 | 1879 | atomic_inc(&root->log_batch); |
257c62e1 | 1880 | |
a4abeea4 | 1881 | smp_mb(); |
626e9f41 | 1882 | if (skip_inode_logging(&ctx)) { |
5dc562c5 | 1883 | /* |
01327610 | 1884 | * We've had everything committed since the last time we were |
5dc562c5 JB |
1885 | * modified so clear this flag in case it was set for whatever |
1886 | * reason, it's no longer relevant. | |
1887 | */ | |
1888 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1889 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
1890 | /* |
1891 | * An ordered extent might have started before and completed | |
1892 | * already with io errors, in which case the inode was not | |
1893 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
1894 | * for any errors that might have happened since we last |
1895 | * checked called fsync. | |
0596a904 | 1896 | */ |
333427a5 | 1897 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
48778179 | 1898 | goto out_release_extents; |
15ee9bc7 | 1899 | } |
15ee9bc7 | 1900 | |
5039eddc JB |
1901 | /* |
1902 | * We use start here because we will need to wait on the IO to complete | |
1903 | * in btrfs_sync_log, which could require joining a transaction (for | |
1904 | * example checking cross references in the nocow path). If we use join | |
1905 | * here we could get into a situation where we're waiting on IO to | |
1906 | * happen that is blocked on a transaction trying to commit. With start | |
1907 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 1908 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
1909 | * from thinking they are super smart and changing this to |
1910 | * btrfs_join_transaction *cough*Josef*cough*. | |
1911 | */ | |
a22285a6 YZ |
1912 | trans = btrfs_start_transaction(root, 0); |
1913 | if (IS_ERR(trans)) { | |
1914 | ret = PTR_ERR(trans); | |
48778179 | 1915 | goto out_release_extents; |
39279cc3 | 1916 | } |
d0c2f4fa | 1917 | trans->in_fsync = true; |
e02119d5 | 1918 | |
48778179 FM |
1919 | ret = btrfs_log_dentry_safe(trans, dentry, &ctx); |
1920 | btrfs_release_log_ctx_extents(&ctx); | |
02c24a82 | 1921 | if (ret < 0) { |
a0634be5 | 1922 | /* Fallthrough and commit/free transaction. */ |
f31f09f6 | 1923 | ret = BTRFS_LOG_FORCE_COMMIT; |
02c24a82 | 1924 | } |
49eb7e46 CM |
1925 | |
1926 | /* we've logged all the items and now have a consistent | |
1927 | * version of the file in the log. It is possible that | |
1928 | * someone will come in and modify the file, but that's | |
1929 | * fine because the log is consistent on disk, and we | |
1930 | * have references to all of the file's extents | |
1931 | * | |
1932 | * It is possible that someone will come in and log the | |
1933 | * file again, but that will end up using the synchronization | |
1934 | * inside btrfs_sync_log to keep things safe. | |
1935 | */ | |
885f46d8 | 1936 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
49eb7e46 | 1937 | |
bf7ba8ee JB |
1938 | if (ret == BTRFS_NO_LOG_SYNC) { |
1939 | ret = btrfs_end_transaction(trans); | |
1940 | goto out; | |
1941 | } | |
1942 | ||
1943 | /* We successfully logged the inode, attempt to sync the log. */ | |
1944 | if (!ret) { | |
1945 | ret = btrfs_sync_log(trans, root, &ctx); | |
0ef8b726 | 1946 | if (!ret) { |
bf7ba8ee JB |
1947 | ret = btrfs_end_transaction(trans); |
1948 | goto out; | |
48778179 | 1949 | } |
bf7ba8ee JB |
1950 | } |
1951 | ||
1952 | /* | |
1953 | * At this point we need to commit the transaction because we had | |
1954 | * btrfs_need_log_full_commit() or some other error. | |
1955 | * | |
1956 | * If we didn't do a full sync we have to stop the trans handle, wait on | |
1957 | * the ordered extents, start it again and commit the transaction. If | |
1958 | * we attempt to wait on the ordered extents here we could deadlock with | |
1959 | * something like fallocate() that is holding the extent lock trying to | |
1960 | * start a transaction while some other thread is trying to commit the | |
1961 | * transaction while we (fsync) are currently holding the transaction | |
1962 | * open. | |
1963 | */ | |
1964 | if (!full_sync) { | |
3a45bb20 | 1965 | ret = btrfs_end_transaction(trans); |
bf7ba8ee JB |
1966 | if (ret) |
1967 | goto out; | |
1968 | ret = btrfs_wait_ordered_range(inode, start, len); | |
1969 | if (ret) | |
1970 | goto out; | |
1971 | ||
1972 | /* | |
1973 | * This is safe to use here because we're only interested in | |
1974 | * making sure the transaction that had the ordered extents is | |
1975 | * committed. We aren't waiting on anything past this point, | |
1976 | * we're purely getting the transaction and committing it. | |
1977 | */ | |
1978 | trans = btrfs_attach_transaction_barrier(root); | |
1979 | if (IS_ERR(trans)) { | |
1980 | ret = PTR_ERR(trans); | |
1981 | ||
1982 | /* | |
1983 | * We committed the transaction and there's no currently | |
1984 | * running transaction, this means everything we care | |
1985 | * about made it to disk and we are done. | |
1986 | */ | |
1987 | if (ret == -ENOENT) | |
1988 | ret = 0; | |
1989 | goto out; | |
1990 | } | |
e02119d5 | 1991 | } |
bf7ba8ee JB |
1992 | |
1993 | ret = btrfs_commit_transaction(trans); | |
39279cc3 | 1994 | out: |
ebb70442 | 1995 | ASSERT(list_empty(&ctx.list)); |
e09d94c9 | 1996 | ASSERT(list_empty(&ctx.conflict_inodes)); |
333427a5 JL |
1997 | err = file_check_and_advance_wb_err(file); |
1998 | if (!ret) | |
1999 | ret = err; | |
014e4ac4 | 2000 | return ret > 0 ? -EIO : ret; |
48778179 FM |
2001 | |
2002 | out_release_extents: | |
2003 | btrfs_release_log_ctx_extents(&ctx); | |
885f46d8 | 2004 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
48778179 | 2005 | goto out; |
39279cc3 CM |
2006 | } |
2007 | ||
f0f37e2f | 2008 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2009 | .fault = filemap_fault, |
f1820361 | 2010 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2011 | .page_mkwrite = btrfs_page_mkwrite, |
2012 | }; | |
2013 | ||
2014 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2015 | { | |
058a457e MX |
2016 | struct address_space *mapping = filp->f_mapping; |
2017 | ||
7e0a1265 | 2018 | if (!mapping->a_ops->read_folio) |
058a457e MX |
2019 | return -ENOEXEC; |
2020 | ||
9ebefb18 | 2021 | file_accessed(filp); |
058a457e | 2022 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2023 | |
9ebefb18 CM |
2024 | return 0; |
2025 | } | |
2026 | ||
35339c24 | 2027 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2028 | int slot, u64 start, u64 end) |
2029 | { | |
2030 | struct btrfs_file_extent_item *fi; | |
2031 | struct btrfs_key key; | |
2032 | ||
2033 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2034 | return 0; | |
2035 | ||
2036 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2037 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2038 | key.type != BTRFS_EXTENT_DATA_KEY) |
2039 | return 0; | |
2040 | ||
2041 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2042 | ||
2043 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2044 | return 0; | |
2045 | ||
2046 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2047 | return 0; | |
2048 | ||
2049 | if (key.offset == end) | |
2050 | return 1; | |
2051 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2052 | return 1; | |
2053 | return 0; | |
2054 | } | |
2055 | ||
a012a74e NB |
2056 | static int fill_holes(struct btrfs_trans_handle *trans, |
2057 | struct btrfs_inode *inode, | |
2058 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2059 | { |
3ffbd68c | 2060 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2061 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2062 | struct extent_buffer *leaf; |
2063 | struct btrfs_file_extent_item *fi; | |
2064 | struct extent_map *hole_em; | |
2aaa6655 JB |
2065 | struct btrfs_key key; |
2066 | int ret; | |
2067 | ||
0b246afa | 2068 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2069 | goto out; |
2070 | ||
a012a74e | 2071 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2072 | key.type = BTRFS_EXTENT_DATA_KEY; |
2073 | key.offset = offset; | |
2074 | ||
2aaa6655 | 2075 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2076 | if (ret <= 0) { |
2077 | /* | |
2078 | * We should have dropped this offset, so if we find it then | |
2079 | * something has gone horribly wrong. | |
2080 | */ | |
2081 | if (ret == 0) | |
2082 | ret = -EINVAL; | |
2aaa6655 | 2083 | return ret; |
f94480bd | 2084 | } |
2aaa6655 JB |
2085 | |
2086 | leaf = path->nodes[0]; | |
a012a74e | 2087 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2088 | u64 num_bytes; |
2089 | ||
2090 | path->slots[0]--; | |
2091 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2092 | struct btrfs_file_extent_item); | |
2093 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2094 | end - offset; | |
2095 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2096 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2097 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2098 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2099 | btrfs_mark_buffer_dirty(leaf); |
2100 | goto out; | |
2101 | } | |
2102 | ||
1707e26d | 2103 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2104 | u64 num_bytes; |
2105 | ||
2aaa6655 | 2106 | key.offset = offset; |
0b246afa | 2107 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2108 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2109 | struct btrfs_file_extent_item); | |
2110 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2111 | offset; | |
2112 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2113 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2114 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2115 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2116 | btrfs_mark_buffer_dirty(leaf); |
2117 | goto out; | |
2118 | } | |
2119 | btrfs_release_path(path); | |
2120 | ||
d1f68ba0 OS |
2121 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, |
2122 | end - offset); | |
2aaa6655 JB |
2123 | if (ret) |
2124 | return ret; | |
2125 | ||
2126 | out: | |
2127 | btrfs_release_path(path); | |
2128 | ||
2129 | hole_em = alloc_extent_map(); | |
2130 | if (!hole_em) { | |
4c0c8cfc | 2131 | btrfs_drop_extent_map_range(inode, offset, end - 1, false); |
23e3337f | 2132 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2133 | } else { |
2134 | hole_em->start = offset; | |
2135 | hole_em->len = end - offset; | |
cc95bef6 | 2136 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2137 | hole_em->orig_start = offset; |
2138 | ||
2139 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2140 | hole_em->block_len = 0; | |
b4939680 | 2141 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2142 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2143 | hole_em->generation = trans->transid; | |
2144 | ||
a1ba4c08 | 2145 | ret = btrfs_replace_extent_map_range(inode, hole_em, true); |
2aaa6655 JB |
2146 | free_extent_map(hole_em); |
2147 | if (ret) | |
23e3337f | 2148 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2149 | } |
2150 | ||
2151 | return 0; | |
2152 | } | |
2153 | ||
d7781546 QW |
2154 | /* |
2155 | * Find a hole extent on given inode and change start/len to the end of hole | |
2156 | * extent.(hole/vacuum extent whose em->start <= start && | |
2157 | * em->start + em->len > start) | |
2158 | * When a hole extent is found, return 1 and modify start/len. | |
2159 | */ | |
dea46d84 | 2160 | static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len) |
d7781546 | 2161 | { |
dea46d84 | 2162 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d7781546 QW |
2163 | struct extent_map *em; |
2164 | int ret = 0; | |
2165 | ||
dea46d84 | 2166 | em = btrfs_get_extent(inode, NULL, 0, |
609805d8 | 2167 | round_down(*start, fs_info->sectorsize), |
39b07b5d | 2168 | round_up(*len, fs_info->sectorsize)); |
9986277e DC |
2169 | if (IS_ERR(em)) |
2170 | return PTR_ERR(em); | |
d7781546 QW |
2171 | |
2172 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2173 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2174 | ret = 1; | |
2175 | *len = em->start + em->len > *start + *len ? | |
2176 | 0 : *start + *len - em->start - em->len; | |
2177 | *start = em->start + em->len; | |
2178 | } | |
2179 | free_extent_map(em); | |
2180 | return ret; | |
2181 | } | |
2182 | ||
55961c8a FM |
2183 | static void btrfs_punch_hole_lock_range(struct inode *inode, |
2184 | const u64 lockstart, | |
2185 | const u64 lockend, | |
2186 | struct extent_state **cached_state) | |
f27451f2 | 2187 | { |
0528476b QW |
2188 | /* |
2189 | * For subpage case, if the range is not at page boundary, we could | |
2190 | * have pages at the leading/tailing part of the range. | |
2191 | * This could lead to dead loop since filemap_range_has_page() | |
2192 | * will always return true. | |
2193 | * So here we need to do extra page alignment for | |
2194 | * filemap_range_has_page(). | |
2195 | */ | |
2196 | const u64 page_lockstart = round_up(lockstart, PAGE_SIZE); | |
2197 | const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1; | |
2198 | ||
f27451f2 | 2199 | while (1) { |
f27451f2 FM |
2200 | truncate_pagecache_range(inode, lockstart, lockend); |
2201 | ||
570eb97b JB |
2202 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2203 | cached_state); | |
f27451f2 | 2204 | /* |
55961c8a FM |
2205 | * We can't have ordered extents in the range, nor dirty/writeback |
2206 | * pages, because we have locked the inode's VFS lock in exclusive | |
2207 | * mode, we have locked the inode's i_mmap_lock in exclusive mode, | |
2208 | * we have flushed all delalloc in the range and we have waited | |
2209 | * for any ordered extents in the range to complete. | |
2210 | * We can race with anyone reading pages from this range, so after | |
2211 | * locking the range check if we have pages in the range, and if | |
2212 | * we do, unlock the range and retry. | |
f27451f2 | 2213 | */ |
55961c8a FM |
2214 | if (!filemap_range_has_page(inode->i_mapping, page_lockstart, |
2215 | page_lockend)) | |
f27451f2 | 2216 | break; |
55961c8a | 2217 | |
570eb97b JB |
2218 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2219 | cached_state); | |
f27451f2 | 2220 | } |
63c34cb4 FM |
2221 | |
2222 | btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend); | |
f27451f2 FM |
2223 | } |
2224 | ||
0cbb5bdf | 2225 | static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans, |
03fcb1ab | 2226 | struct btrfs_inode *inode, |
690a5dbf | 2227 | struct btrfs_path *path, |
bf385648 | 2228 | struct btrfs_replace_extent_info *extent_info, |
2766ff61 FM |
2229 | const u64 replace_len, |
2230 | const u64 bytes_to_drop) | |
690a5dbf | 2231 | { |
03fcb1ab NB |
2232 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2233 | struct btrfs_root *root = inode->root; | |
690a5dbf FM |
2234 | struct btrfs_file_extent_item *extent; |
2235 | struct extent_buffer *leaf; | |
2236 | struct btrfs_key key; | |
2237 | int slot; | |
2238 | struct btrfs_ref ref = { 0 }; | |
690a5dbf FM |
2239 | int ret; |
2240 | ||
bf385648 | 2241 | if (replace_len == 0) |
690a5dbf FM |
2242 | return 0; |
2243 | ||
bf385648 | 2244 | if (extent_info->disk_offset == 0 && |
2766ff61 | 2245 | btrfs_fs_incompat(fs_info, NO_HOLES)) { |
03fcb1ab | 2246 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2247 | return 0; |
2766ff61 | 2248 | } |
690a5dbf | 2249 | |
03fcb1ab | 2250 | key.objectid = btrfs_ino(inode); |
690a5dbf | 2251 | key.type = BTRFS_EXTENT_DATA_KEY; |
bf385648 | 2252 | key.offset = extent_info->file_offset; |
690a5dbf | 2253 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
fb870f6c | 2254 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf FM |
2255 | if (ret) |
2256 | return ret; | |
2257 | leaf = path->nodes[0]; | |
2258 | slot = path->slots[0]; | |
bf385648 | 2259 | write_extent_buffer(leaf, extent_info->extent_buf, |
690a5dbf | 2260 | btrfs_item_ptr_offset(leaf, slot), |
fb870f6c | 2261 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf | 2262 | extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
fb870f6c | 2263 | ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE); |
bf385648 FM |
2264 | btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset); |
2265 | btrfs_set_file_extent_num_bytes(leaf, extent, replace_len); | |
2266 | if (extent_info->is_new_extent) | |
8fccebfa | 2267 | btrfs_set_file_extent_generation(leaf, extent, trans->transid); |
690a5dbf FM |
2268 | btrfs_mark_buffer_dirty(leaf); |
2269 | btrfs_release_path(path); | |
2270 | ||
03fcb1ab NB |
2271 | ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset, |
2272 | replace_len); | |
9ddc959e JB |
2273 | if (ret) |
2274 | return ret; | |
2275 | ||
690a5dbf | 2276 | /* If it's a hole, nothing more needs to be done. */ |
2766ff61 | 2277 | if (extent_info->disk_offset == 0) { |
03fcb1ab | 2278 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2279 | return 0; |
2766ff61 | 2280 | } |
690a5dbf | 2281 | |
03fcb1ab | 2282 | btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop); |
8fccebfa | 2283 | |
bf385648 FM |
2284 | if (extent_info->is_new_extent && extent_info->insertions == 0) { |
2285 | key.objectid = extent_info->disk_offset; | |
8fccebfa | 2286 | key.type = BTRFS_EXTENT_ITEM_KEY; |
bf385648 | 2287 | key.offset = extent_info->disk_len; |
8fccebfa | 2288 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
03fcb1ab | 2289 | btrfs_ino(inode), |
bf385648 FM |
2290 | extent_info->file_offset, |
2291 | extent_info->qgroup_reserved, | |
8fccebfa FM |
2292 | &key); |
2293 | } else { | |
2294 | u64 ref_offset; | |
2295 | ||
2296 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, | |
bf385648 FM |
2297 | extent_info->disk_offset, |
2298 | extent_info->disk_len, 0); | |
2299 | ref_offset = extent_info->file_offset - extent_info->data_offset; | |
8fccebfa | 2300 | btrfs_init_data_ref(&ref, root->root_key.objectid, |
f42c5da6 | 2301 | btrfs_ino(inode), ref_offset, 0, false); |
8fccebfa FM |
2302 | ret = btrfs_inc_extent_ref(trans, &ref); |
2303 | } | |
2304 | ||
bf385648 | 2305 | extent_info->insertions++; |
690a5dbf FM |
2306 | |
2307 | return ret; | |
2308 | } | |
2309 | ||
9cba40a6 FM |
2310 | /* |
2311 | * The respective range must have been previously locked, as well as the inode. | |
2312 | * The end offset is inclusive (last byte of the range). | |
bf385648 FM |
2313 | * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing |
2314 | * the file range with an extent. | |
2315 | * When not punching a hole, we don't want to end up in a state where we dropped | |
2316 | * extents without inserting a new one, so we must abort the transaction to avoid | |
2317 | * a corruption. | |
9cba40a6 | 2318 | */ |
bfc78479 NB |
2319 | int btrfs_replace_file_extents(struct btrfs_inode *inode, |
2320 | struct btrfs_path *path, const u64 start, | |
2321 | const u64 end, | |
2322 | struct btrfs_replace_extent_info *extent_info, | |
2323 | struct btrfs_trans_handle **trans_out) | |
9cba40a6 | 2324 | { |
5893dfb9 | 2325 | struct btrfs_drop_extents_args drop_args = { 0 }; |
bfc78479 NB |
2326 | struct btrfs_root *root = inode->root; |
2327 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2bd36e7b | 2328 | u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1); |
bfc78479 | 2329 | u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize); |
9cba40a6 FM |
2330 | struct btrfs_trans_handle *trans = NULL; |
2331 | struct btrfs_block_rsv *rsv; | |
2332 | unsigned int rsv_count; | |
2333 | u64 cur_offset; | |
9cba40a6 FM |
2334 | u64 len = end - start; |
2335 | int ret = 0; | |
2336 | ||
2337 | if (end <= start) | |
2338 | return -EINVAL; | |
2339 | ||
2340 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); | |
2341 | if (!rsv) { | |
2342 | ret = -ENOMEM; | |
2343 | goto out; | |
2344 | } | |
2bd36e7b | 2345 | rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1); |
710d5921 | 2346 | rsv->failfast = true; |
9cba40a6 FM |
2347 | |
2348 | /* | |
2349 | * 1 - update the inode | |
2350 | * 1 - removing the extents in the range | |
bf385648 FM |
2351 | * 1 - adding the hole extent if no_holes isn't set or if we are |
2352 | * replacing the range with a new extent | |
9cba40a6 | 2353 | */ |
bf385648 | 2354 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info) |
690a5dbf FM |
2355 | rsv_count = 3; |
2356 | else | |
2357 | rsv_count = 2; | |
2358 | ||
9cba40a6 FM |
2359 | trans = btrfs_start_transaction(root, rsv_count); |
2360 | if (IS_ERR(trans)) { | |
2361 | ret = PTR_ERR(trans); | |
2362 | trans = NULL; | |
2363 | goto out_free; | |
2364 | } | |
2365 | ||
2366 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, | |
2367 | min_size, false); | |
650c9cab FM |
2368 | if (WARN_ON(ret)) |
2369 | goto out_trans; | |
9cba40a6 FM |
2370 | trans->block_rsv = rsv; |
2371 | ||
2372 | cur_offset = start; | |
5893dfb9 FM |
2373 | drop_args.path = path; |
2374 | drop_args.end = end + 1; | |
2375 | drop_args.drop_cache = true; | |
9cba40a6 | 2376 | while (cur_offset < end) { |
5893dfb9 | 2377 | drop_args.start = cur_offset; |
bfc78479 | 2378 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
2766ff61 FM |
2379 | /* If we are punching a hole decrement the inode's byte count */ |
2380 | if (!extent_info) | |
bfc78479 | 2381 | btrfs_update_inode_bytes(inode, 0, |
2766ff61 | 2382 | drop_args.bytes_found); |
690a5dbf FM |
2383 | if (ret != -ENOSPC) { |
2384 | /* | |
4afb912f JB |
2385 | * The only time we don't want to abort is if we are |
2386 | * attempting to clone a partial inline extent, in which | |
2387 | * case we'll get EOPNOTSUPP. However if we aren't | |
2388 | * clone we need to abort no matter what, because if we | |
2389 | * got EOPNOTSUPP via prealloc then we messed up and | |
2390 | * need to abort. | |
690a5dbf | 2391 | */ |
4afb912f JB |
2392 | if (ret && |
2393 | (ret != -EOPNOTSUPP || | |
2394 | (extent_info && extent_info->is_new_extent))) | |
690a5dbf | 2395 | btrfs_abort_transaction(trans, ret); |
9cba40a6 | 2396 | break; |
690a5dbf | 2397 | } |
9cba40a6 FM |
2398 | |
2399 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2400 | ||
5893dfb9 | 2401 | if (!extent_info && cur_offset < drop_args.drop_end && |
690a5dbf | 2402 | cur_offset < ino_size) { |
bfc78479 NB |
2403 | ret = fill_holes(trans, inode, path, cur_offset, |
2404 | drop_args.drop_end); | |
9cba40a6 FM |
2405 | if (ret) { |
2406 | /* | |
2407 | * If we failed then we didn't insert our hole | |
2408 | * entries for the area we dropped, so now the | |
2409 | * fs is corrupted, so we must abort the | |
2410 | * transaction. | |
2411 | */ | |
2412 | btrfs_abort_transaction(trans, ret); | |
2413 | break; | |
2414 | } | |
5893dfb9 | 2415 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e JB |
2416 | /* |
2417 | * We are past the i_size here, but since we didn't | |
2418 | * insert holes we need to clear the mapped area so we | |
2419 | * know to not set disk_i_size in this area until a new | |
2420 | * file extent is inserted here. | |
2421 | */ | |
bfc78479 | 2422 | ret = btrfs_inode_clear_file_extent_range(inode, |
5893dfb9 FM |
2423 | cur_offset, |
2424 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2425 | if (ret) { |
2426 | /* | |
2427 | * We couldn't clear our area, so we could | |
2428 | * presumably adjust up and corrupt the fs, so | |
2429 | * we need to abort. | |
2430 | */ | |
2431 | btrfs_abort_transaction(trans, ret); | |
2432 | break; | |
2433 | } | |
9cba40a6 FM |
2434 | } |
2435 | ||
5893dfb9 FM |
2436 | if (extent_info && |
2437 | drop_args.drop_end > extent_info->file_offset) { | |
2438 | u64 replace_len = drop_args.drop_end - | |
2439 | extent_info->file_offset; | |
690a5dbf | 2440 | |
bfc78479 NB |
2441 | ret = btrfs_insert_replace_extent(trans, inode, path, |
2442 | extent_info, replace_len, | |
03fcb1ab | 2443 | drop_args.bytes_found); |
690a5dbf FM |
2444 | if (ret) { |
2445 | btrfs_abort_transaction(trans, ret); | |
2446 | break; | |
2447 | } | |
bf385648 FM |
2448 | extent_info->data_len -= replace_len; |
2449 | extent_info->data_offset += replace_len; | |
2450 | extent_info->file_offset += replace_len; | |
690a5dbf FM |
2451 | } |
2452 | ||
983d8209 FM |
2453 | /* |
2454 | * We are releasing our handle on the transaction, balance the | |
2455 | * dirty pages of the btree inode and flush delayed items, and | |
2456 | * then get a new transaction handle, which may now point to a | |
2457 | * new transaction in case someone else may have committed the | |
2458 | * transaction we used to replace/drop file extent items. So | |
2459 | * bump the inode's iversion and update mtime and ctime except | |
2460 | * if we are called from a dedupe context. This is because a | |
2461 | * power failure/crash may happen after the transaction is | |
2462 | * committed and before we finish replacing/dropping all the | |
2463 | * file extent items we need. | |
2464 | */ | |
2465 | inode_inc_iversion(&inode->vfs_inode); | |
2466 | ||
2467 | if (!extent_info || extent_info->update_times) { | |
2468 | inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode); | |
2469 | inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime; | |
2470 | } | |
2471 | ||
bfc78479 | 2472 | ret = btrfs_update_inode(trans, root, inode); |
9cba40a6 FM |
2473 | if (ret) |
2474 | break; | |
2475 | ||
2476 | btrfs_end_transaction(trans); | |
2477 | btrfs_btree_balance_dirty(fs_info); | |
2478 | ||
2479 | trans = btrfs_start_transaction(root, rsv_count); | |
2480 | if (IS_ERR(trans)) { | |
2481 | ret = PTR_ERR(trans); | |
2482 | trans = NULL; | |
2483 | break; | |
2484 | } | |
2485 | ||
2486 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, | |
2487 | rsv, min_size, false); | |
650c9cab FM |
2488 | if (WARN_ON(ret)) |
2489 | break; | |
9cba40a6 FM |
2490 | trans->block_rsv = rsv; |
2491 | ||
3227788c BC |
2492 | cur_offset = drop_args.drop_end; |
2493 | len = end - cur_offset; | |
2494 | if (!extent_info && len) { | |
bfc78479 | 2495 | ret = find_first_non_hole(inode, &cur_offset, &len); |
690a5dbf FM |
2496 | if (unlikely(ret < 0)) |
2497 | break; | |
2498 | if (ret && !len) { | |
2499 | ret = 0; | |
2500 | break; | |
2501 | } | |
9cba40a6 FM |
2502 | } |
2503 | } | |
2504 | ||
690a5dbf FM |
2505 | /* |
2506 | * If we were cloning, force the next fsync to be a full one since we | |
2507 | * we replaced (or just dropped in the case of cloning holes when | |
e2b84217 FM |
2508 | * NO_HOLES is enabled) file extent items and did not setup new extent |
2509 | * maps for the replacement extents (or holes). | |
690a5dbf | 2510 | */ |
bf385648 | 2511 | if (extent_info && !extent_info->is_new_extent) |
23e3337f | 2512 | btrfs_set_inode_full_sync(inode); |
690a5dbf | 2513 | |
9cba40a6 FM |
2514 | if (ret) |
2515 | goto out_trans; | |
2516 | ||
2517 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2518 | /* | |
2519 | * If we are using the NO_HOLES feature we might have had already an | |
2520 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2521 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2522 | * represent holes, drop_end can be less than lockend and so we must | |
2523 | * make sure we have an extent map representing the existing hole (the | |
2524 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2525 | * map representing the existing hole), otherwise the fast fsync path | |
2526 | * will not record the existence of the hole region | |
2527 | * [existing_hole_start, lockend]. | |
2528 | */ | |
5893dfb9 FM |
2529 | if (drop_args.drop_end <= end) |
2530 | drop_args.drop_end = end + 1; | |
9cba40a6 FM |
2531 | /* |
2532 | * Don't insert file hole extent item if it's for a range beyond eof | |
2533 | * (because it's useless) or if it represents a 0 bytes range (when | |
2534 | * cur_offset == drop_end). | |
2535 | */ | |
5893dfb9 FM |
2536 | if (!extent_info && cur_offset < ino_size && |
2537 | cur_offset < drop_args.drop_end) { | |
bfc78479 NB |
2538 | ret = fill_holes(trans, inode, path, cur_offset, |
2539 | drop_args.drop_end); | |
9cba40a6 FM |
2540 | if (ret) { |
2541 | /* Same comment as above. */ | |
2542 | btrfs_abort_transaction(trans, ret); | |
2543 | goto out_trans; | |
2544 | } | |
5893dfb9 | 2545 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e | 2546 | /* See the comment in the loop above for the reasoning here. */ |
bfc78479 NB |
2547 | ret = btrfs_inode_clear_file_extent_range(inode, cur_offset, |
2548 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2549 | if (ret) { |
2550 | btrfs_abort_transaction(trans, ret); | |
2551 | goto out_trans; | |
2552 | } | |
2553 | ||
9cba40a6 | 2554 | } |
bf385648 | 2555 | if (extent_info) { |
bfc78479 | 2556 | ret = btrfs_insert_replace_extent(trans, inode, path, |
03fcb1ab NB |
2557 | extent_info, extent_info->data_len, |
2558 | drop_args.bytes_found); | |
690a5dbf FM |
2559 | if (ret) { |
2560 | btrfs_abort_transaction(trans, ret); | |
2561 | goto out_trans; | |
2562 | } | |
2563 | } | |
9cba40a6 FM |
2564 | |
2565 | out_trans: | |
2566 | if (!trans) | |
2567 | goto out_free; | |
2568 | ||
2569 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2570 | if (ret) | |
2571 | btrfs_end_transaction(trans); | |
2572 | else | |
2573 | *trans_out = trans; | |
2574 | out_free: | |
2575 | btrfs_free_block_rsv(fs_info, rsv); | |
2576 | out: | |
2577 | return ret; | |
2578 | } | |
2579 | ||
05fd9564 | 2580 | static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len) |
2aaa6655 | 2581 | { |
05fd9564 | 2582 | struct inode *inode = file_inode(file); |
0b246afa | 2583 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2584 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2585 | struct extent_state *cached_state = NULL; | |
2586 | struct btrfs_path *path; | |
9cba40a6 | 2587 | struct btrfs_trans_handle *trans = NULL; |
d7781546 QW |
2588 | u64 lockstart; |
2589 | u64 lockend; | |
2590 | u64 tail_start; | |
2591 | u64 tail_len; | |
2592 | u64 orig_start = offset; | |
2aaa6655 | 2593 | int ret = 0; |
9703fefe | 2594 | bool same_block; |
a1a50f60 | 2595 | u64 ino_size; |
9703fefe | 2596 | bool truncated_block = false; |
e8c1c76e | 2597 | bool updated_inode = false; |
2aaa6655 | 2598 | |
29b6352b | 2599 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
bd6526d0 | 2600 | |
0ef8b726 JB |
2601 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2602 | if (ret) | |
bd6526d0 | 2603 | goto out_only_mutex; |
2aaa6655 | 2604 | |
0b246afa | 2605 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
dea46d84 | 2606 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2607 | if (ret < 0) |
2608 | goto out_only_mutex; | |
2609 | if (ret && !len) { | |
2610 | /* Already in a large hole */ | |
2611 | ret = 0; | |
2612 | goto out_only_mutex; | |
2613 | } | |
2614 | ||
05fd9564 DW |
2615 | ret = file_modified(file); |
2616 | if (ret) | |
2617 | goto out_only_mutex; | |
2618 | ||
ee8ba05c JB |
2619 | lockstart = round_up(offset, fs_info->sectorsize); |
2620 | lockend = round_down(offset + len, fs_info->sectorsize) - 1; | |
0b246afa JM |
2621 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2622 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2623 | /* |
9703fefe | 2624 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2625 | * because we are sure there is no data there. |
2626 | */ | |
2aaa6655 | 2627 | /* |
9703fefe CR |
2628 | * Only do this if we are in the same block and we aren't doing the |
2629 | * entire block. | |
2aaa6655 | 2630 | */ |
0b246afa | 2631 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2632 | if (offset < ino_size) { |
9703fefe | 2633 | truncated_block = true; |
217f42eb NB |
2634 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2635 | 0); | |
e8c1c76e FM |
2636 | } else { |
2637 | ret = 0; | |
2638 | } | |
d7781546 | 2639 | goto out_only_mutex; |
2aaa6655 JB |
2640 | } |
2641 | ||
9703fefe | 2642 | /* zero back part of the first block */ |
12870f1c | 2643 | if (offset < ino_size) { |
9703fefe | 2644 | truncated_block = true; |
217f42eb | 2645 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
7426cc04 | 2646 | if (ret) { |
8d9b4a16 | 2647 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
7426cc04 MX |
2648 | return ret; |
2649 | } | |
2aaa6655 JB |
2650 | } |
2651 | ||
d7781546 QW |
2652 | /* Check the aligned pages after the first unaligned page, |
2653 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2654 | * including several following pages are already in holes, |
d7781546 QW |
2655 | * the extra check can be skipped */ |
2656 | if (offset == orig_start) { | |
2657 | /* after truncate page, check hole again */ | |
2658 | len = offset + len - lockstart; | |
2659 | offset = lockstart; | |
dea46d84 | 2660 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2661 | if (ret < 0) |
2662 | goto out_only_mutex; | |
2663 | if (ret && !len) { | |
2664 | ret = 0; | |
2665 | goto out_only_mutex; | |
2666 | } | |
2667 | lockstart = offset; | |
2668 | } | |
2669 | ||
2670 | /* Check the tail unaligned part is in a hole */ | |
2671 | tail_start = lockend + 1; | |
2672 | tail_len = offset + len - tail_start; | |
2673 | if (tail_len) { | |
dea46d84 | 2674 | ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len); |
d7781546 QW |
2675 | if (unlikely(ret < 0)) |
2676 | goto out_only_mutex; | |
2677 | if (!ret) { | |
2678 | /* zero the front end of the last page */ | |
2679 | if (tail_start + tail_len < ino_size) { | |
9703fefe | 2680 | truncated_block = true; |
217f42eb | 2681 | ret = btrfs_truncate_block(BTRFS_I(inode), |
9703fefe CR |
2682 | tail_start + tail_len, |
2683 | 0, 1); | |
d7781546 QW |
2684 | if (ret) |
2685 | goto out_only_mutex; | |
51f395ad | 2686 | } |
0061280d | 2687 | } |
2aaa6655 JB |
2688 | } |
2689 | ||
2690 | if (lockend < lockstart) { | |
e8c1c76e FM |
2691 | ret = 0; |
2692 | goto out_only_mutex; | |
2aaa6655 JB |
2693 | } |
2694 | ||
55961c8a | 2695 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state); |
2aaa6655 JB |
2696 | |
2697 | path = btrfs_alloc_path(); | |
2698 | if (!path) { | |
2699 | ret = -ENOMEM; | |
2700 | goto out; | |
2701 | } | |
2702 | ||
bfc78479 NB |
2703 | ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart, |
2704 | lockend, NULL, &trans); | |
9cba40a6 FM |
2705 | btrfs_free_path(path); |
2706 | if (ret) | |
2707 | goto out; | |
2aaa6655 | 2708 | |
9cba40a6 | 2709 | ASSERT(trans != NULL); |
e1f5790e | 2710 | inode_inc_iversion(inode); |
c1867eb3 DS |
2711 | inode->i_mtime = current_time(inode); |
2712 | inode->i_ctime = inode->i_mtime; | |
9a56fcd1 | 2713 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e8c1c76e | 2714 | updated_inode = true; |
3a45bb20 | 2715 | btrfs_end_transaction(trans); |
2ff7e61e | 2716 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2717 | out: |
570eb97b JB |
2718 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2719 | &cached_state); | |
d7781546 | 2720 | out_only_mutex: |
9cba40a6 | 2721 | if (!updated_inode && truncated_block && !ret) { |
e8c1c76e FM |
2722 | /* |
2723 | * If we only end up zeroing part of a page, we still need to | |
2724 | * update the inode item, so that all the time fields are | |
2725 | * updated as well as the necessary btrfs inode in memory fields | |
2726 | * for detecting, at fsync time, if the inode isn't yet in the | |
2727 | * log tree or it's there but not up to date. | |
2728 | */ | |
17900668 FM |
2729 | struct timespec64 now = current_time(inode); |
2730 | ||
2731 | inode_inc_iversion(inode); | |
2732 | inode->i_mtime = now; | |
2733 | inode->i_ctime = now; | |
e8c1c76e FM |
2734 | trans = btrfs_start_transaction(root, 1); |
2735 | if (IS_ERR(trans)) { | |
9cba40a6 | 2736 | ret = PTR_ERR(trans); |
e8c1c76e | 2737 | } else { |
9cba40a6 FM |
2738 | int ret2; |
2739 | ||
9a56fcd1 | 2740 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
9cba40a6 FM |
2741 | ret2 = btrfs_end_transaction(trans); |
2742 | if (!ret) | |
2743 | ret = ret2; | |
e8c1c76e FM |
2744 | } |
2745 | } | |
8d9b4a16 | 2746 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
9cba40a6 | 2747 | return ret; |
2aaa6655 JB |
2748 | } |
2749 | ||
14524a84 QW |
2750 | /* Helper structure to record which range is already reserved */ |
2751 | struct falloc_range { | |
2752 | struct list_head list; | |
2753 | u64 start; | |
2754 | u64 len; | |
2755 | }; | |
2756 | ||
2757 | /* | |
2758 | * Helper function to add falloc range | |
2759 | * | |
2760 | * Caller should have locked the larger range of extent containing | |
2761 | * [start, len) | |
2762 | */ | |
2763 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2764 | { | |
14524a84 QW |
2765 | struct falloc_range *range = NULL; |
2766 | ||
77d25534 NB |
2767 | if (!list_empty(head)) { |
2768 | /* | |
2769 | * As fallocate iterates by bytenr order, we only need to check | |
2770 | * the last range. | |
2771 | */ | |
2772 | range = list_last_entry(head, struct falloc_range, list); | |
2773 | if (range->start + range->len == start) { | |
2774 | range->len += len; | |
2775 | return 0; | |
2776 | } | |
14524a84 | 2777 | } |
77d25534 | 2778 | |
32fc932e | 2779 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2780 | if (!range) |
2781 | return -ENOMEM; | |
2782 | range->start = start; | |
2783 | range->len = len; | |
2784 | list_add_tail(&range->list, head); | |
2785 | return 0; | |
2786 | } | |
2787 | ||
f27451f2 FM |
2788 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2789 | const u64 end, | |
2790 | const int mode) | |
2791 | { | |
2792 | struct btrfs_trans_handle *trans; | |
2793 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2794 | int ret; | |
2795 | int ret2; | |
2796 | ||
2797 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2798 | return 0; | |
2799 | ||
2800 | trans = btrfs_start_transaction(root, 1); | |
2801 | if (IS_ERR(trans)) | |
2802 | return PTR_ERR(trans); | |
2803 | ||
2804 | inode->i_ctime = current_time(inode); | |
2805 | i_size_write(inode, end); | |
76aea537 | 2806 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
9a56fcd1 | 2807 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
f27451f2 FM |
2808 | ret2 = btrfs_end_transaction(trans); |
2809 | ||
2810 | return ret ? ret : ret2; | |
2811 | } | |
2812 | ||
81fdf638 | 2813 | enum { |
f262fa8d DS |
2814 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
2815 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
2816 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
2817 | }; |
2818 | ||
948dfeb8 | 2819 | static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode, |
f27451f2 FM |
2820 | u64 offset) |
2821 | { | |
ee8ba05c | 2822 | const u64 sectorsize = inode->root->fs_info->sectorsize; |
f27451f2 | 2823 | struct extent_map *em; |
81fdf638 | 2824 | int ret; |
f27451f2 FM |
2825 | |
2826 | offset = round_down(offset, sectorsize); | |
948dfeb8 | 2827 | em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize); |
f27451f2 FM |
2828 | if (IS_ERR(em)) |
2829 | return PTR_ERR(em); | |
2830 | ||
2831 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
2832 | ret = RANGE_BOUNDARY_HOLE; |
2833 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2834 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
2835 | else | |
2836 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
2837 | |
2838 | free_extent_map(em); | |
2839 | return ret; | |
2840 | } | |
2841 | ||
2842 | static int btrfs_zero_range(struct inode *inode, | |
2843 | loff_t offset, | |
2844 | loff_t len, | |
2845 | const int mode) | |
2846 | { | |
2847 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2848 | struct extent_map *em; | |
2849 | struct extent_changeset *data_reserved = NULL; | |
2850 | int ret; | |
2851 | u64 alloc_hint = 0; | |
ee8ba05c | 2852 | const u64 sectorsize = fs_info->sectorsize; |
f27451f2 FM |
2853 | u64 alloc_start = round_down(offset, sectorsize); |
2854 | u64 alloc_end = round_up(offset + len, sectorsize); | |
2855 | u64 bytes_to_reserve = 0; | |
2856 | bool space_reserved = false; | |
2857 | ||
39b07b5d OS |
2858 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2859 | alloc_end - alloc_start); | |
f27451f2 FM |
2860 | if (IS_ERR(em)) { |
2861 | ret = PTR_ERR(em); | |
2862 | goto out; | |
2863 | } | |
2864 | ||
2865 | /* | |
2866 | * Avoid hole punching and extent allocation for some cases. More cases | |
2867 | * could be considered, but these are unlikely common and we keep things | |
2868 | * as simple as possible for now. Also, intentionally, if the target | |
2869 | * range contains one or more prealloc extents together with regular | |
2870 | * extents and holes, we drop all the existing extents and allocate a | |
2871 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
2872 | */ | |
2873 | if (em->start <= alloc_start && | |
2874 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2875 | const u64 em_end = em->start + em->len; | |
2876 | ||
2877 | if (em_end >= offset + len) { | |
2878 | /* | |
2879 | * The whole range is already a prealloc extent, | |
2880 | * do nothing except updating the inode's i_size if | |
2881 | * needed. | |
2882 | */ | |
2883 | free_extent_map(em); | |
2884 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2885 | mode); | |
2886 | goto out; | |
2887 | } | |
2888 | /* | |
2889 | * Part of the range is already a prealloc extent, so operate | |
2890 | * only on the remaining part of the range. | |
2891 | */ | |
2892 | alloc_start = em_end; | |
2893 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
2894 | len = offset + len - alloc_start; | |
2895 | offset = alloc_start; | |
2896 | alloc_hint = em->block_start + em->len; | |
2897 | } | |
2898 | free_extent_map(em); | |
2899 | ||
2900 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
2901 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
39b07b5d OS |
2902 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2903 | sectorsize); | |
f27451f2 FM |
2904 | if (IS_ERR(em)) { |
2905 | ret = PTR_ERR(em); | |
2906 | goto out; | |
2907 | } | |
2908 | ||
2909 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2910 | free_extent_map(em); | |
2911 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2912 | mode); | |
2913 | goto out; | |
2914 | } | |
2915 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
2916 | free_extent_map(em); | |
217f42eb NB |
2917 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2918 | 0); | |
f27451f2 FM |
2919 | if (!ret) |
2920 | ret = btrfs_fallocate_update_isize(inode, | |
2921 | offset + len, | |
2922 | mode); | |
2923 | return ret; | |
2924 | } | |
2925 | free_extent_map(em); | |
2926 | alloc_start = round_down(offset, sectorsize); | |
2927 | alloc_end = alloc_start + sectorsize; | |
2928 | goto reserve_space; | |
2929 | } | |
2930 | ||
2931 | alloc_start = round_up(offset, sectorsize); | |
2932 | alloc_end = round_down(offset + len, sectorsize); | |
2933 | ||
2934 | /* | |
2935 | * For unaligned ranges, check the pages at the boundaries, they might | |
2936 | * map to an extent, in which case we need to partially zero them, or | |
2937 | * they might map to a hole, in which case we need our allocation range | |
2938 | * to cover them. | |
2939 | */ | |
2940 | if (!IS_ALIGNED(offset, sectorsize)) { | |
948dfeb8 NB |
2941 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
2942 | offset); | |
f27451f2 FM |
2943 | if (ret < 0) |
2944 | goto out; | |
81fdf638 | 2945 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2946 | alloc_start = round_down(offset, sectorsize); |
2947 | ret = 0; | |
81fdf638 | 2948 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb | 2949 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
f27451f2 FM |
2950 | if (ret) |
2951 | goto out; | |
81fdf638 FM |
2952 | } else { |
2953 | ret = 0; | |
f27451f2 FM |
2954 | } |
2955 | } | |
2956 | ||
2957 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
948dfeb8 | 2958 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
f27451f2 FM |
2959 | offset + len); |
2960 | if (ret < 0) | |
2961 | goto out; | |
81fdf638 | 2962 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2963 | alloc_end = round_up(offset + len, sectorsize); |
2964 | ret = 0; | |
81fdf638 | 2965 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb NB |
2966 | ret = btrfs_truncate_block(BTRFS_I(inode), offset + len, |
2967 | 0, 1); | |
f27451f2 FM |
2968 | if (ret) |
2969 | goto out; | |
81fdf638 FM |
2970 | } else { |
2971 | ret = 0; | |
f27451f2 FM |
2972 | } |
2973 | } | |
2974 | ||
2975 | reserve_space: | |
2976 | if (alloc_start < alloc_end) { | |
2977 | struct extent_state *cached_state = NULL; | |
2978 | const u64 lockstart = alloc_start; | |
2979 | const u64 lockend = alloc_end - 1; | |
2980 | ||
2981 | bytes_to_reserve = alloc_end - alloc_start; | |
2982 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
2983 | bytes_to_reserve); | |
2984 | if (ret < 0) | |
2985 | goto out; | |
2986 | space_reserved = true; | |
55961c8a FM |
2987 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
2988 | &cached_state); | |
7661a3e0 | 2989 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved, |
a7f8b1c2 | 2990 | alloc_start, bytes_to_reserve); |
4f6a49de | 2991 | if (ret) { |
570eb97b JB |
2992 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
2993 | lockend, &cached_state); | |
a7f8b1c2 | 2994 | goto out; |
4f6a49de | 2995 | } |
f27451f2 FM |
2996 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, |
2997 | alloc_end - alloc_start, | |
2998 | i_blocksize(inode), | |
2999 | offset + len, &alloc_hint); | |
570eb97b JB |
3000 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3001 | &cached_state); | |
f27451f2 | 3002 | /* btrfs_prealloc_file_range releases reserved space on error */ |
9f13ce74 | 3003 | if (ret) { |
f27451f2 | 3004 | space_reserved = false; |
9f13ce74 FM |
3005 | goto out; |
3006 | } | |
f27451f2 | 3007 | } |
9f13ce74 | 3008 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3009 | out: |
3010 | if (ret && space_reserved) | |
25ce28ca | 3011 | btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved, |
f27451f2 FM |
3012 | alloc_start, bytes_to_reserve); |
3013 | extent_changeset_free(data_reserved); | |
3014 | ||
3015 | return ret; | |
3016 | } | |
3017 | ||
2fe17c10 CH |
3018 | static long btrfs_fallocate(struct file *file, int mode, |
3019 | loff_t offset, loff_t len) | |
3020 | { | |
496ad9aa | 3021 | struct inode *inode = file_inode(file); |
2fe17c10 | 3022 | struct extent_state *cached_state = NULL; |
364ecf36 | 3023 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3024 | struct falloc_range *range; |
3025 | struct falloc_range *tmp; | |
3026 | struct list_head reserve_list; | |
2fe17c10 CH |
3027 | u64 cur_offset; |
3028 | u64 last_byte; | |
3029 | u64 alloc_start; | |
3030 | u64 alloc_end; | |
3031 | u64 alloc_hint = 0; | |
3032 | u64 locked_end; | |
14524a84 | 3033 | u64 actual_end = 0; |
47e1d1c7 FM |
3034 | u64 data_space_needed = 0; |
3035 | u64 data_space_reserved = 0; | |
3036 | u64 qgroup_reserved = 0; | |
2fe17c10 | 3037 | struct extent_map *em; |
ee8ba05c | 3038 | int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize; |
2fe17c10 CH |
3039 | int ret; |
3040 | ||
f1569c4c NA |
3041 | /* Do not allow fallocate in ZONED mode */ |
3042 | if (btrfs_is_zoned(btrfs_sb(inode->i_sb))) | |
3043 | return -EOPNOTSUPP; | |
3044 | ||
797f4277 MX |
3045 | alloc_start = round_down(offset, blocksize); |
3046 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3047 | cur_offset = alloc_start; |
2fe17c10 | 3048 | |
2aaa6655 | 3049 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3050 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3051 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3052 | return -EOPNOTSUPP; |
3053 | ||
2aaa6655 | 3054 | if (mode & FALLOC_FL_PUNCH_HOLE) |
05fd9564 | 3055 | return btrfs_punch_hole(file, offset, len); |
2aaa6655 | 3056 | |
29b6352b | 3057 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
2a162ce9 DI |
3058 | |
3059 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3060 | ret = inode_newsize_ok(inode, offset + len); | |
3061 | if (ret) | |
3062 | goto out; | |
3063 | } | |
2fe17c10 | 3064 | |
05fd9564 DW |
3065 | ret = file_modified(file); |
3066 | if (ret) | |
3067 | goto out; | |
3068 | ||
14524a84 QW |
3069 | /* |
3070 | * TODO: Move these two operations after we have checked | |
3071 | * accurate reserved space, or fallocate can still fail but | |
3072 | * with page truncated or size expanded. | |
3073 | * | |
3074 | * But that's a minor problem and won't do much harm BTW. | |
3075 | */ | |
2fe17c10 | 3076 | if (alloc_start > inode->i_size) { |
b06359a3 | 3077 | ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode), |
a41ad394 | 3078 | alloc_start); |
2fe17c10 CH |
3079 | if (ret) |
3080 | goto out; | |
0f6925fa | 3081 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3082 | /* |
3083 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3084 | * need to zero out the end of the block if i_size lands in the |
3085 | * middle of a block. | |
a71754fc | 3086 | */ |
217f42eb | 3087 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
a71754fc JB |
3088 | if (ret) |
3089 | goto out; | |
2fe17c10 CH |
3090 | } |
3091 | ||
a71754fc | 3092 | /* |
ffa8fc60 FM |
3093 | * We have locked the inode at the VFS level (in exclusive mode) and we |
3094 | * have locked the i_mmap_lock lock (in exclusive mode). Now before | |
3095 | * locking the file range, flush all dealloc in the range and wait for | |
3096 | * all ordered extents in the range to complete. After this we can lock | |
3097 | * the file range and, due to the previous locking we did, we know there | |
3098 | * can't be more delalloc or ordered extents in the range. | |
a71754fc | 3099 | */ |
0ef8b726 JB |
3100 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3101 | alloc_end - alloc_start); | |
3102 | if (ret) | |
3103 | goto out; | |
a71754fc | 3104 | |
f27451f2 FM |
3105 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3106 | ret = btrfs_zero_range(inode, offset, len, mode); | |
8d9b4a16 | 3107 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
f27451f2 FM |
3108 | return ret; |
3109 | } | |
3110 | ||
2fe17c10 | 3111 | locked_end = alloc_end - 1; |
570eb97b JB |
3112 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3113 | &cached_state); | |
2fe17c10 | 3114 | |
63c34cb4 FM |
3115 | btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end); |
3116 | ||
14524a84 QW |
3117 | /* First, check if we exceed the qgroup limit */ |
3118 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 3119 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3120 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
39b07b5d | 3121 | alloc_end - cur_offset); |
9986277e DC |
3122 | if (IS_ERR(em)) { |
3123 | ret = PTR_ERR(em); | |
79787eaa JM |
3124 | break; |
3125 | } | |
2fe17c10 | 3126 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3127 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3128 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3129 | if (em->block_start == EXTENT_MAP_HOLE || |
3130 | (cur_offset >= inode->i_size && | |
3131 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
47e1d1c7 FM |
3132 | const u64 range_len = last_byte - cur_offset; |
3133 | ||
3134 | ret = add_falloc_range(&reserve_list, cur_offset, range_len); | |
14524a84 QW |
3135 | if (ret < 0) { |
3136 | free_extent_map(em); | |
3137 | break; | |
3d850dd4 | 3138 | } |
7661a3e0 | 3139 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), |
47e1d1c7 | 3140 | &data_reserved, cur_offset, range_len); |
be2d253c FM |
3141 | if (ret < 0) { |
3142 | free_extent_map(em); | |
14524a84 | 3143 | break; |
be2d253c | 3144 | } |
47e1d1c7 FM |
3145 | qgroup_reserved += range_len; |
3146 | data_space_needed += range_len; | |
2fe17c10 CH |
3147 | } |
3148 | free_extent_map(em); | |
2fe17c10 | 3149 | cur_offset = last_byte; |
14524a84 QW |
3150 | } |
3151 | ||
47e1d1c7 FM |
3152 | if (!ret && data_space_needed > 0) { |
3153 | /* | |
3154 | * We are safe to reserve space here as we can't have delalloc | |
3155 | * in the range, see above. | |
3156 | */ | |
3157 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3158 | data_space_needed); | |
3159 | if (!ret) | |
3160 | data_space_reserved = data_space_needed; | |
3161 | } | |
3162 | ||
14524a84 QW |
3163 | /* |
3164 | * If ret is still 0, means we're OK to fallocate. | |
3165 | * Or just cleanup the list and exit. | |
3166 | */ | |
3167 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
47e1d1c7 | 3168 | if (!ret) { |
14524a84 QW |
3169 | ret = btrfs_prealloc_file_range(inode, mode, |
3170 | range->start, | |
93407472 | 3171 | range->len, i_blocksize(inode), |
14524a84 | 3172 | offset + len, &alloc_hint); |
47e1d1c7 FM |
3173 | /* |
3174 | * btrfs_prealloc_file_range() releases space even | |
3175 | * if it returns an error. | |
3176 | */ | |
3177 | data_space_reserved -= range->len; | |
3178 | qgroup_reserved -= range->len; | |
3179 | } else if (data_space_reserved > 0) { | |
25ce28ca | 3180 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
47e1d1c7 FM |
3181 | data_reserved, range->start, |
3182 | range->len); | |
3183 | data_space_reserved -= range->len; | |
3184 | qgroup_reserved -= range->len; | |
3185 | } else if (qgroup_reserved > 0) { | |
3186 | btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved, | |
3187 | range->start, range->len); | |
3188 | qgroup_reserved -= range->len; | |
3189 | } | |
14524a84 QW |
3190 | list_del(&range->list); |
3191 | kfree(range); | |
3192 | } | |
3193 | if (ret < 0) | |
3194 | goto out_unlock; | |
3195 | ||
f27451f2 FM |
3196 | /* |
3197 | * We didn't need to allocate any more space, but we still extended the | |
3198 | * size of the file so we need to update i_size and the inode item. | |
3199 | */ | |
3200 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3201 | out_unlock: |
570eb97b JB |
3202 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3203 | &cached_state); | |
2fe17c10 | 3204 | out: |
8d9b4a16 | 3205 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
364ecf36 | 3206 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3207 | return ret; |
3208 | } | |
3209 | ||
b6e83356 | 3210 | /* |
ac3c0d36 FM |
3211 | * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range |
3212 | * that has unflushed and/or flushing delalloc. There might be other adjacent | |
3213 | * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps | |
3214 | * looping while it gets adjacent subranges, and merging them together. | |
b6e83356 FM |
3215 | */ |
3216 | static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end, | |
3217 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) | |
3218 | { | |
3219 | const u64 len = end + 1 - start; | |
3220 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
3221 | struct extent_map *em; | |
3222 | u64 em_end; | |
3223 | u64 delalloc_len; | |
a2853ffc | 3224 | unsigned int outstanding_extents; |
b6e83356 FM |
3225 | |
3226 | /* | |
3227 | * Search the io tree first for EXTENT_DELALLOC. If we find any, it | |
3228 | * means we have delalloc (dirty pages) for which writeback has not | |
3229 | * started yet. | |
3230 | */ | |
a2853ffc FM |
3231 | spin_lock(&inode->lock); |
3232 | outstanding_extents = inode->outstanding_extents; | |
3233 | ||
3234 | if (inode->delalloc_bytes > 0) { | |
3235 | spin_unlock(&inode->lock); | |
3236 | *delalloc_start_ret = start; | |
3237 | delalloc_len = count_range_bits(&inode->io_tree, | |
3238 | delalloc_start_ret, end, | |
3239 | len, EXTENT_DELALLOC, 1); | |
3240 | } else { | |
3241 | spin_unlock(&inode->lock); | |
3242 | delalloc_len = 0; | |
3243 | } | |
3244 | ||
b6e83356 FM |
3245 | /* |
3246 | * If delalloc was found then *delalloc_start_ret has a sector size | |
3247 | * aligned value (rounded down). | |
3248 | */ | |
3249 | if (delalloc_len > 0) | |
3250 | *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1; | |
3251 | ||
a2853ffc FM |
3252 | /* |
3253 | * No outstanding extents means we don't have any delalloc that is | |
3254 | * flushing, so return the unflushed range found in the io tree (if any). | |
3255 | */ | |
3256 | if (outstanding_extents == 0) | |
013f9c70 | 3257 | return (delalloc_len > 0); |
013f9c70 | 3258 | |
b6e83356 FM |
3259 | /* |
3260 | * Now also check if there's any extent map in the range that does not | |
3261 | * map to a hole or prealloc extent. We do this because: | |
3262 | * | |
3263 | * 1) When delalloc is flushed, the file range is locked, we clear the | |
3264 | * EXTENT_DELALLOC bit from the io tree and create an extent map for | |
3265 | * an allocated extent. So we might just have been called after | |
3266 | * delalloc is flushed and before the ordered extent completes and | |
3267 | * inserts the new file extent item in the subvolume's btree; | |
3268 | * | |
3269 | * 2) We may have an extent map created by flushing delalloc for a | |
3270 | * subrange that starts before the subrange we found marked with | |
3271 | * EXTENT_DELALLOC in the io tree. | |
3272 | */ | |
3273 | read_lock(&em_tree->lock); | |
3274 | em = lookup_extent_mapping(em_tree, start, len); | |
d47704bd FM |
3275 | if (!em) { |
3276 | read_unlock(&em_tree->lock); | |
3277 | return (delalloc_len > 0); | |
3278 | } | |
b6e83356 FM |
3279 | |
3280 | /* extent_map_end() returns a non-inclusive end offset. */ | |
d47704bd | 3281 | em_end = extent_map_end(em); |
b6e83356 FM |
3282 | |
3283 | /* | |
3284 | * If we have a hole/prealloc extent map, check the next one if this one | |
3285 | * ends before our range's end. | |
3286 | */ | |
d47704bd FM |
3287 | if ((em->block_start == EXTENT_MAP_HOLE || |
3288 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) && em_end < end) { | |
b6e83356 FM |
3289 | struct extent_map *next_em; |
3290 | ||
d47704bd | 3291 | next_em = btrfs_next_extent_map(em_tree, em); |
b6e83356 | 3292 | free_extent_map(em); |
d47704bd FM |
3293 | |
3294 | /* | |
3295 | * There's no next extent map or the next one starts beyond our | |
3296 | * range, return the range found in the io tree (if any). | |
3297 | */ | |
3298 | if (!next_em || next_em->start > end) { | |
3299 | read_unlock(&em_tree->lock); | |
3300 | free_extent_map(next_em); | |
3301 | return (delalloc_len > 0); | |
3302 | } | |
3303 | ||
3304 | em_end = extent_map_end(next_em); | |
b6e83356 FM |
3305 | em = next_em; |
3306 | } | |
3307 | ||
d47704bd | 3308 | read_unlock(&em_tree->lock); |
b6e83356 FM |
3309 | |
3310 | /* | |
d47704bd FM |
3311 | * We have a hole or prealloc extent that ends at or beyond our range's |
3312 | * end, return the range found in the io tree (if any). | |
b6e83356 | 3313 | */ |
d47704bd FM |
3314 | if (em->block_start == EXTENT_MAP_HOLE || |
3315 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
3316 | free_extent_map(em); | |
b6e83356 | 3317 | return (delalloc_len > 0); |
d47704bd | 3318 | } |
b6e83356 FM |
3319 | |
3320 | /* | |
3321 | * We don't have any range as EXTENT_DELALLOC in the io tree, so the | |
3322 | * extent map is the only subrange representing delalloc. | |
3323 | */ | |
3324 | if (delalloc_len == 0) { | |
3325 | *delalloc_start_ret = em->start; | |
3326 | *delalloc_end_ret = min(end, em_end - 1); | |
3327 | free_extent_map(em); | |
3328 | return true; | |
3329 | } | |
3330 | ||
3331 | /* | |
3332 | * The extent map represents a delalloc range that starts before the | |
3333 | * delalloc range we found in the io tree. | |
3334 | */ | |
3335 | if (em->start < *delalloc_start_ret) { | |
3336 | *delalloc_start_ret = em->start; | |
3337 | /* | |
3338 | * If the ranges are adjacent, return a combined range. | |
3339 | * Otherwise return the extent map's range. | |
3340 | */ | |
3341 | if (em_end < *delalloc_start_ret) | |
3342 | *delalloc_end_ret = min(end, em_end - 1); | |
3343 | ||
3344 | free_extent_map(em); | |
3345 | return true; | |
3346 | } | |
3347 | ||
3348 | /* | |
3349 | * The extent map starts after the delalloc range we found in the io | |
3350 | * tree. If it's adjacent, return a combined range, otherwise return | |
3351 | * the range found in the io tree. | |
3352 | */ | |
3353 | if (*delalloc_end_ret + 1 == em->start) | |
3354 | *delalloc_end_ret = min(end, em_end - 1); | |
3355 | ||
3356 | free_extent_map(em); | |
3357 | return true; | |
3358 | } | |
3359 | ||
3360 | /* | |
3361 | * Check if there's delalloc in a given range. | |
3362 | * | |
3363 | * @inode: The inode. | |
3364 | * @start: The start offset of the range. It does not need to be | |
3365 | * sector size aligned. | |
3366 | * @end: The end offset (inclusive value) of the search range. | |
3367 | * It does not need to be sector size aligned. | |
3368 | * @delalloc_start_ret: Output argument, set to the start offset of the | |
3369 | * subrange found with delalloc (may not be sector size | |
3370 | * aligned). | |
3371 | * @delalloc_end_ret: Output argument, set to he end offset (inclusive value) | |
3372 | * of the subrange found with delalloc. | |
3373 | * | |
3374 | * Returns true if a subrange with delalloc is found within the given range, and | |
3375 | * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and | |
3376 | * end offsets of the subrange. | |
3377 | */ | |
ac3c0d36 FM |
3378 | bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end, |
3379 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) | |
b6e83356 FM |
3380 | { |
3381 | u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize); | |
3382 | u64 prev_delalloc_end = 0; | |
3383 | bool ret = false; | |
3384 | ||
3385 | while (cur_offset < end) { | |
3386 | u64 delalloc_start; | |
3387 | u64 delalloc_end; | |
3388 | bool delalloc; | |
3389 | ||
3390 | delalloc = find_delalloc_subrange(inode, cur_offset, end, | |
3391 | &delalloc_start, | |
3392 | &delalloc_end); | |
3393 | if (!delalloc) | |
3394 | break; | |
3395 | ||
3396 | if (prev_delalloc_end == 0) { | |
3397 | /* First subrange found. */ | |
3398 | *delalloc_start_ret = max(delalloc_start, start); | |
3399 | *delalloc_end_ret = delalloc_end; | |
3400 | ret = true; | |
3401 | } else if (delalloc_start == prev_delalloc_end + 1) { | |
3402 | /* Subrange adjacent to the previous one, merge them. */ | |
3403 | *delalloc_end_ret = delalloc_end; | |
3404 | } else { | |
3405 | /* Subrange not adjacent to the previous one, exit. */ | |
3406 | break; | |
3407 | } | |
3408 | ||
3409 | prev_delalloc_end = delalloc_end; | |
3410 | cur_offset = delalloc_end + 1; | |
3411 | cond_resched(); | |
3412 | } | |
3413 | ||
3414 | return ret; | |
3415 | } | |
3416 | ||
3417 | /* | |
3418 | * Check if there's a hole or delalloc range in a range representing a hole (or | |
3419 | * prealloc extent) found in the inode's subvolume btree. | |
3420 | * | |
3421 | * @inode: The inode. | |
3422 | * @whence: Seek mode (SEEK_DATA or SEEK_HOLE). | |
3423 | * @start: Start offset of the hole region. It does not need to be sector | |
3424 | * size aligned. | |
3425 | * @end: End offset (inclusive value) of the hole region. It does not | |
3426 | * need to be sector size aligned. | |
3427 | * @start_ret: Return parameter, used to set the start of the subrange in the | |
3428 | * hole that matches the search criteria (seek mode), if such | |
3429 | * subrange is found (return value of the function is true). | |
3430 | * The value returned here may not be sector size aligned. | |
3431 | * | |
3432 | * Returns true if a subrange matching the given seek mode is found, and if one | |
3433 | * is found, it updates @start_ret with the start of the subrange. | |
3434 | */ | |
3435 | static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence, | |
3436 | u64 start, u64 end, u64 *start_ret) | |
3437 | { | |
3438 | u64 delalloc_start; | |
3439 | u64 delalloc_end; | |
3440 | bool delalloc; | |
3441 | ||
ac3c0d36 FM |
3442 | delalloc = btrfs_find_delalloc_in_range(inode, start, end, |
3443 | &delalloc_start, &delalloc_end); | |
b6e83356 FM |
3444 | if (delalloc && whence == SEEK_DATA) { |
3445 | *start_ret = delalloc_start; | |
3446 | return true; | |
3447 | } | |
3448 | ||
3449 | if (delalloc && whence == SEEK_HOLE) { | |
3450 | /* | |
3451 | * We found delalloc but it starts after out start offset. So we | |
3452 | * have a hole between our start offset and the delalloc start. | |
3453 | */ | |
3454 | if (start < delalloc_start) { | |
3455 | *start_ret = start; | |
3456 | return true; | |
3457 | } | |
3458 | /* | |
3459 | * Delalloc range starts at our start offset. | |
3460 | * If the delalloc range's length is smaller than our range, | |
3461 | * then it means we have a hole that starts where the delalloc | |
3462 | * subrange ends. | |
3463 | */ | |
3464 | if (delalloc_end < end) { | |
3465 | *start_ret = delalloc_end + 1; | |
3466 | return true; | |
3467 | } | |
3468 | ||
3469 | /* There's delalloc for the whole range. */ | |
3470 | return false; | |
3471 | } | |
3472 | ||
3473 | if (!delalloc && whence == SEEK_HOLE) { | |
3474 | *start_ret = start; | |
3475 | return true; | |
3476 | } | |
3477 | ||
3478 | /* | |
3479 | * No delalloc in the range and we are seeking for data. The caller has | |
3480 | * to iterate to the next extent item in the subvolume btree. | |
3481 | */ | |
3482 | return false; | |
3483 | } | |
3484 | ||
cca5de97 | 3485 | static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset, |
bc80230e | 3486 | int whence) |
b2675157 | 3487 | { |
cca5de97 | 3488 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
b2675157 | 3489 | struct extent_state *cached_state = NULL; |
b6e83356 FM |
3490 | const loff_t i_size = i_size_read(&inode->vfs_inode); |
3491 | const u64 ino = btrfs_ino(inode); | |
3492 | struct btrfs_root *root = inode->root; | |
3493 | struct btrfs_path *path; | |
3494 | struct btrfs_key key; | |
3495 | u64 last_extent_end; | |
4d1a40c6 LB |
3496 | u64 lockstart; |
3497 | u64 lockend; | |
3498 | u64 start; | |
b6e83356 FM |
3499 | int ret; |
3500 | bool found = false; | |
b2675157 | 3501 | |
bc80230e | 3502 | if (i_size == 0 || offset >= i_size) |
4d1a40c6 LB |
3503 | return -ENXIO; |
3504 | ||
b6e83356 FM |
3505 | /* |
3506 | * Quick path. If the inode has no prealloc extents and its number of | |
3507 | * bytes used matches its i_size, then it can not have holes. | |
3508 | */ | |
3509 | if (whence == SEEK_HOLE && | |
3510 | !(inode->flags & BTRFS_INODE_PREALLOC) && | |
3511 | inode_get_bytes(&inode->vfs_inode) == i_size) | |
3512 | return i_size; | |
3513 | ||
4d1a40c6 | 3514 | /* |
bc80230e | 3515 | * offset can be negative, in this case we start finding DATA/HOLE from |
4d1a40c6 LB |
3516 | * the very start of the file. |
3517 | */ | |
bc80230e | 3518 | start = max_t(loff_t, 0, offset); |
4d1a40c6 | 3519 | |
0b246afa | 3520 | lockstart = round_down(start, fs_info->sectorsize); |
d79b7c26 | 3521 | lockend = round_up(i_size, fs_info->sectorsize); |
b2675157 | 3522 | if (lockend <= lockstart) |
0b246afa | 3523 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3524 | lockend--; |
b6e83356 FM |
3525 | |
3526 | path = btrfs_alloc_path(); | |
3527 | if (!path) | |
3528 | return -ENOMEM; | |
3529 | path->reada = READA_FORWARD; | |
3530 | ||
3531 | key.objectid = ino; | |
3532 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3533 | key.offset = start; | |
3534 | ||
3535 | last_extent_end = lockstart; | |
b2675157 | 3536 | |
570eb97b | 3537 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b2675157 | 3538 | |
b6e83356 FM |
3539 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3540 | if (ret < 0) { | |
3541 | goto out; | |
3542 | } else if (ret > 0 && path->slots[0] > 0) { | |
3543 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3544 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3545 | path->slots[0]--; | |
3546 | } | |
3547 | ||
d79b7c26 | 3548 | while (start < i_size) { |
b6e83356 FM |
3549 | struct extent_buffer *leaf = path->nodes[0]; |
3550 | struct btrfs_file_extent_item *extent; | |
3551 | u64 extent_end; | |
3552 | ||
3553 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3554 | ret = btrfs_next_leaf(root, path); | |
3555 | if (ret < 0) | |
3556 | goto out; | |
3557 | else if (ret > 0) | |
3558 | break; | |
3559 | ||
3560 | leaf = path->nodes[0]; | |
b2675157 JB |
3561 | } |
3562 | ||
b6e83356 FM |
3563 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3564 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
7f4ca37c | 3565 | break; |
b2675157 | 3566 | |
b6e83356 FM |
3567 | extent_end = btrfs_file_extent_end(path); |
3568 | ||
3569 | /* | |
3570 | * In the first iteration we may have a slot that points to an | |
3571 | * extent that ends before our start offset, so skip it. | |
3572 | */ | |
3573 | if (extent_end <= start) { | |
3574 | path->slots[0]++; | |
3575 | continue; | |
3576 | } | |
3577 | ||
3578 | /* We have an implicit hole, NO_HOLES feature is likely set. */ | |
3579 | if (last_extent_end < key.offset) { | |
3580 | u64 search_start = last_extent_end; | |
3581 | u64 found_start; | |
3582 | ||
3583 | /* | |
3584 | * First iteration, @start matches @offset and it's | |
3585 | * within the hole. | |
3586 | */ | |
3587 | if (start == offset) | |
3588 | search_start = offset; | |
3589 | ||
3590 | found = find_desired_extent_in_hole(inode, whence, | |
3591 | search_start, | |
3592 | key.offset - 1, | |
3593 | &found_start); | |
3594 | if (found) { | |
3595 | start = found_start; | |
3596 | break; | |
3597 | } | |
3598 | /* | |
3599 | * Didn't find data or a hole (due to delalloc) in the | |
3600 | * implicit hole range, so need to analyze the extent. | |
3601 | */ | |
3602 | } | |
3603 | ||
3604 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
3605 | struct btrfs_file_extent_item); | |
3606 | ||
3607 | if (btrfs_file_extent_disk_bytenr(leaf, extent) == 0 || | |
3608 | btrfs_file_extent_type(leaf, extent) == | |
3609 | BTRFS_FILE_EXTENT_PREALLOC) { | |
3610 | /* | |
3611 | * Explicit hole or prealloc extent, search for delalloc. | |
3612 | * A prealloc extent is treated like a hole. | |
3613 | */ | |
3614 | u64 search_start = key.offset; | |
3615 | u64 found_start; | |
3616 | ||
3617 | /* | |
3618 | * First iteration, @start matches @offset and it's | |
3619 | * within the hole. | |
3620 | */ | |
3621 | if (start == offset) | |
3622 | search_start = offset; | |
3623 | ||
3624 | found = find_desired_extent_in_hole(inode, whence, | |
3625 | search_start, | |
3626 | extent_end - 1, | |
3627 | &found_start); | |
3628 | if (found) { | |
3629 | start = found_start; | |
3630 | break; | |
3631 | } | |
3632 | /* | |
3633 | * Didn't find data or a hole (due to delalloc) in the | |
3634 | * implicit hole range, so need to analyze the next | |
3635 | * extent item. | |
3636 | */ | |
3637 | } else { | |
3638 | /* | |
3639 | * Found a regular or inline extent. | |
3640 | * If we are seeking for data, adjust the start offset | |
3641 | * and stop, we're done. | |
3642 | */ | |
3643 | if (whence == SEEK_DATA) { | |
3644 | start = max_t(u64, key.offset, offset); | |
3645 | found = true; | |
3646 | break; | |
3647 | } | |
3648 | /* | |
3649 | * Else, we are seeking for a hole, check the next file | |
3650 | * extent item. | |
3651 | */ | |
3652 | } | |
3653 | ||
3654 | start = extent_end; | |
3655 | last_extent_end = extent_end; | |
3656 | path->slots[0]++; | |
aed0ca18 FM |
3657 | if (fatal_signal_pending(current)) { |
3658 | ret = -EINTR; | |
b6e83356 | 3659 | goto out; |
aed0ca18 | 3660 | } |
b2675157 JB |
3661 | cond_resched(); |
3662 | } | |
b6e83356 FM |
3663 | |
3664 | /* We have an implicit hole from the last extent found up to i_size. */ | |
3665 | if (!found && start < i_size) { | |
3666 | found = find_desired_extent_in_hole(inode, whence, start, | |
3667 | i_size - 1, &start); | |
3668 | if (!found) | |
3669 | start = i_size; | |
3670 | } | |
3671 | ||
3672 | out: | |
570eb97b | 3673 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b6e83356 FM |
3674 | btrfs_free_path(path); |
3675 | ||
3676 | if (ret < 0) | |
3677 | return ret; | |
3678 | ||
3679 | if (whence == SEEK_DATA && start >= i_size) | |
3680 | return -ENXIO; | |
bc80230e | 3681 | |
b6e83356 | 3682 | return min_t(loff_t, start, i_size); |
b2675157 JB |
3683 | } |
3684 | ||
965c8e59 | 3685 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3686 | { |
3687 | struct inode *inode = file->f_mapping->host; | |
b2675157 | 3688 | |
965c8e59 | 3689 | switch (whence) { |
2034f3b4 NB |
3690 | default: |
3691 | return generic_file_llseek(file, offset, whence); | |
b2675157 JB |
3692 | case SEEK_DATA: |
3693 | case SEEK_HOLE: | |
29b6352b | 3694 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
cca5de97 | 3695 | offset = find_desired_extent(BTRFS_I(inode), offset, whence); |
a14b78ad | 3696 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
bc80230e | 3697 | break; |
b2675157 JB |
3698 | } |
3699 | ||
bc80230e NB |
3700 | if (offset < 0) |
3701 | return offset; | |
3702 | ||
2034f3b4 | 3703 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
3704 | } |
3705 | ||
edf064e7 GR |
3706 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3707 | { | |
14605409 BB |
3708 | int ret; |
3709 | ||
926078b2 | 3710 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC; |
14605409 BB |
3711 | |
3712 | ret = fsverity_file_open(inode, filp); | |
3713 | if (ret) | |
3714 | return ret; | |
edf064e7 GR |
3715 | return generic_file_open(inode, filp); |
3716 | } | |
3717 | ||
4e4cabec GR |
3718 | static int check_direct_read(struct btrfs_fs_info *fs_info, |
3719 | const struct iov_iter *iter, loff_t offset) | |
3720 | { | |
3721 | int ret; | |
3722 | int i, seg; | |
3723 | ||
3724 | ret = check_direct_IO(fs_info, iter, offset); | |
3725 | if (ret < 0) | |
3726 | return ret; | |
3727 | ||
3728 | if (!iter_is_iovec(iter)) | |
3729 | return 0; | |
3730 | ||
3731 | for (seg = 0; seg < iter->nr_segs; seg++) | |
3732 | for (i = seg + 1; i < iter->nr_segs; i++) | |
3733 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
3734 | return -EINVAL; | |
3735 | return 0; | |
3736 | } | |
3737 | ||
3738 | static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to) | |
3739 | { | |
3740 | struct inode *inode = file_inode(iocb->ki_filp); | |
51bd9563 FM |
3741 | size_t prev_left = 0; |
3742 | ssize_t read = 0; | |
4e4cabec GR |
3743 | ssize_t ret; |
3744 | ||
14605409 BB |
3745 | if (fsverity_active(inode)) |
3746 | return 0; | |
3747 | ||
4e4cabec GR |
3748 | if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos)) |
3749 | return 0; | |
3750 | ||
29b6352b | 3751 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 FM |
3752 | again: |
3753 | /* | |
3754 | * This is similar to what we do for direct IO writes, see the comment | |
3755 | * at btrfs_direct_write(), but we also disable page faults in addition | |
3756 | * to disabling them only at the iov_iter level. This is because when | |
3757 | * reading from a hole or prealloc extent, iomap calls iov_iter_zero(), | |
3758 | * which can still trigger page fault ins despite having set ->nofault | |
3759 | * to true of our 'to' iov_iter. | |
3760 | * | |
3761 | * The difference to direct IO writes is that we deadlock when trying | |
3762 | * to lock the extent range in the inode's tree during he page reads | |
3763 | * triggered by the fault in (while for writes it is due to waiting for | |
3764 | * our own ordered extent). This is because for direct IO reads, | |
3765 | * btrfs_dio_iomap_begin() returns with the extent range locked, which | |
3766 | * is only unlocked in the endio callback (end_bio_extent_readpage()). | |
3767 | */ | |
3768 | pagefault_disable(); | |
3769 | to->nofault = true; | |
8184620a | 3770 | ret = btrfs_dio_read(iocb, to, read); |
51bd9563 FM |
3771 | to->nofault = false; |
3772 | pagefault_enable(); | |
3773 | ||
3774 | /* No increment (+=) because iomap returns a cumulative value. */ | |
3775 | if (ret > 0) | |
3776 | read = ret; | |
3777 | ||
3778 | if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) { | |
3779 | const size_t left = iov_iter_count(to); | |
3780 | ||
3781 | if (left == prev_left) { | |
3782 | /* | |
3783 | * We didn't make any progress since the last attempt, | |
3784 | * fallback to a buffered read for the remainder of the | |
3785 | * range. This is just to avoid any possibility of looping | |
3786 | * for too long. | |
3787 | */ | |
3788 | ret = read; | |
3789 | } else { | |
3790 | /* | |
3791 | * We made some progress since the last retry or this is | |
3792 | * the first time we are retrying. Fault in as many pages | |
3793 | * as possible and retry. | |
3794 | */ | |
3795 | fault_in_iov_iter_writeable(to, left); | |
3796 | prev_left = left; | |
3797 | goto again; | |
3798 | } | |
3799 | } | |
a14b78ad | 3800 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
51bd9563 | 3801 | return ret < 0 ? ret : read; |
4e4cabec GR |
3802 | } |
3803 | ||
f85781fb GR |
3804 | static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
3805 | { | |
3806 | ssize_t ret = 0; | |
3807 | ||
3808 | if (iocb->ki_flags & IOCB_DIRECT) { | |
4e4cabec | 3809 | ret = btrfs_direct_read(iocb, to); |
0425e7ba JT |
3810 | if (ret < 0 || !iov_iter_count(to) || |
3811 | iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp))) | |
f85781fb GR |
3812 | return ret; |
3813 | } | |
3814 | ||
87fa0f3e | 3815 | return filemap_read(iocb, to, ret); |
f85781fb GR |
3816 | } |
3817 | ||
828c0950 | 3818 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3819 | .llseek = btrfs_file_llseek, |
f85781fb | 3820 | .read_iter = btrfs_file_read_iter, |
e9906a98 | 3821 | .splice_read = generic_file_splice_read, |
b30ac0fc | 3822 | .write_iter = btrfs_file_write_iter, |
d7776591 | 3823 | .splice_write = iter_file_splice_write, |
9ebefb18 | 3824 | .mmap = btrfs_file_mmap, |
edf064e7 | 3825 | .open = btrfs_file_open, |
e1b81e67 | 3826 | .release = btrfs_release_file, |
b0c58223 | 3827 | .get_unmapped_area = thp_get_unmapped_area, |
39279cc3 | 3828 | .fsync = btrfs_sync_file, |
2fe17c10 | 3829 | .fallocate = btrfs_fallocate, |
34287aa3 | 3830 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3831 | #ifdef CONFIG_COMPAT |
4c63c245 | 3832 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3833 | #endif |
2e5dfc99 | 3834 | .remap_file_range = btrfs_remap_file_range, |
39279cc3 | 3835 | }; |
9247f317 | 3836 | |
728404da FM |
3837 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) |
3838 | { | |
3839 | int ret; | |
3840 | ||
3841 | /* | |
3842 | * So with compression we will find and lock a dirty page and clear the | |
3843 | * first one as dirty, setup an async extent, and immediately return | |
3844 | * with the entire range locked but with nobody actually marked with | |
3845 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3846 | * expect it to work since it will just kick off a thread to do the | |
3847 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3848 | * since it will wait on the page lock, which won't be unlocked until | |
3849 | * after the pages have been marked as writeback and so we're good to go | |
3850 | * from there. We have to do this otherwise we'll miss the ordered | |
3851 | * extents and that results in badness. Please Josef, do not think you | |
3852 | * know better and pull this out at some point in the future, it is | |
3853 | * right and you are wrong. | |
3854 | */ | |
3855 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3856 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3857 | &BTRFS_I(inode)->runtime_flags)) | |
3858 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3859 | ||
3860 | return ret; | |
3861 | } |