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); |
162d053e FM |
383 | if (ret) { |
384 | btrfs_abort_transaction(trans, ret); | |
385 | break; | |
386 | } | |
771ed689 | 387 | } |
5893dfb9 | 388 | key.offset = args->start; |
6643558d | 389 | } |
62fe51c1 JB |
390 | /* |
391 | * From here on out we will have actually dropped something, so | |
392 | * last_end can be updated. | |
393 | */ | |
394 | last_end = extent_end; | |
395 | ||
920bbbfb YZ |
396 | /* |
397 | * | ---- range to drop ----- | | |
398 | * | -------- extent -------- | | |
399 | */ | |
5893dfb9 | 400 | if (args->start <= key.offset && args->end < extent_end) { |
00fdf13a | 401 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 402 | ret = -EOPNOTSUPP; |
00fdf13a LB |
403 | break; |
404 | } | |
6643558d | 405 | |
920bbbfb | 406 | memcpy(&new_key, &key, sizeof(new_key)); |
5893dfb9 | 407 | new_key.offset = args->end; |
0b246afa | 408 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 409 | |
5893dfb9 | 410 | extent_offset += args->end - key.offset; |
920bbbfb YZ |
411 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
412 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 413 | extent_end - args->end); |
920bbbfb | 414 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 415 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 416 | args->bytes_found += args->end - key.offset; |
920bbbfb | 417 | break; |
39279cc3 | 418 | } |
771ed689 | 419 | |
920bbbfb YZ |
420 | search_start = extent_end; |
421 | /* | |
422 | * | ---- range to drop ----- | | |
423 | * | -------- extent -------- | | |
424 | */ | |
5893dfb9 | 425 | if (args->start > key.offset && args->end >= extent_end) { |
920bbbfb | 426 | BUG_ON(del_nr > 0); |
00fdf13a | 427 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 428 | ret = -EOPNOTSUPP; |
00fdf13a LB |
429 | break; |
430 | } | |
8c2383c3 | 431 | |
920bbbfb | 432 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5893dfb9 | 433 | args->start - key.offset); |
920bbbfb | 434 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 435 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 436 | args->bytes_found += extent_end - args->start; |
5893dfb9 | 437 | if (args->end == extent_end) |
920bbbfb | 438 | break; |
c8b97818 | 439 | |
920bbbfb YZ |
440 | path->slots[0]++; |
441 | goto next_slot; | |
31840ae1 ZY |
442 | } |
443 | ||
920bbbfb YZ |
444 | /* |
445 | * | ---- range to drop ----- | | |
446 | * | ------ extent ------ | | |
447 | */ | |
5893dfb9 | 448 | if (args->start <= key.offset && args->end >= extent_end) { |
fc19c5e7 | 449 | delete_extent_item: |
920bbbfb YZ |
450 | if (del_nr == 0) { |
451 | del_slot = path->slots[0]; | |
452 | del_nr = 1; | |
453 | } else { | |
454 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
455 | del_nr++; | |
456 | } | |
31840ae1 | 457 | |
5dc562c5 JB |
458 | if (update_refs && |
459 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
2766ff61 | 460 | args->bytes_found += extent_end - key.offset; |
920bbbfb | 461 | extent_end = ALIGN(extent_end, |
0b246afa | 462 | fs_info->sectorsize); |
5dc562c5 | 463 | } else if (update_refs && disk_bytenr > 0) { |
ffd4bb2a QW |
464 | btrfs_init_generic_ref(&ref, |
465 | BTRFS_DROP_DELAYED_REF, | |
466 | disk_bytenr, num_bytes, 0); | |
467 | btrfs_init_data_ref(&ref, | |
920bbbfb | 468 | root->root_key.objectid, |
ffd4bb2a | 469 | key.objectid, |
f42c5da6 NB |
470 | key.offset - extent_offset, 0, |
471 | false); | |
ffd4bb2a | 472 | ret = btrfs_free_extent(trans, &ref); |
162d053e FM |
473 | if (ret) { |
474 | btrfs_abort_transaction(trans, ret); | |
475 | break; | |
476 | } | |
2766ff61 | 477 | args->bytes_found += extent_end - key.offset; |
31840ae1 | 478 | } |
31840ae1 | 479 | |
5893dfb9 | 480 | if (args->end == extent_end) |
920bbbfb YZ |
481 | break; |
482 | ||
483 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
484 | path->slots[0]++; | |
485 | goto next_slot; | |
486 | } | |
487 | ||
488 | ret = btrfs_del_items(trans, root, path, del_slot, | |
489 | del_nr); | |
79787eaa | 490 | if (ret) { |
66642832 | 491 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 492 | break; |
79787eaa | 493 | } |
920bbbfb YZ |
494 | |
495 | del_nr = 0; | |
496 | del_slot = 0; | |
497 | ||
b3b4aa74 | 498 | btrfs_release_path(path); |
920bbbfb | 499 | continue; |
39279cc3 | 500 | } |
920bbbfb | 501 | |
290342f6 | 502 | BUG(); |
39279cc3 | 503 | } |
920bbbfb | 504 | |
79787eaa | 505 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
506 | /* |
507 | * Set path->slots[0] to first slot, so that after the delete | |
508 | * if items are move off from our leaf to its immediate left or | |
509 | * right neighbor leafs, we end up with a correct and adjusted | |
5893dfb9 | 510 | * path->slots[0] for our insertion (if args->replace_extent). |
1acae57b FDBM |
511 | */ |
512 | path->slots[0] = del_slot; | |
920bbbfb | 513 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 514 | if (ret) |
66642832 | 515 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 516 | } |
1acae57b | 517 | |
d5f37527 FDBM |
518 | leaf = path->nodes[0]; |
519 | /* | |
520 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
521 | * which case it unlocked our path, so check path->locks[0] matches a | |
522 | * write lock. | |
523 | */ | |
7ecb4c31 | 524 | if (!ret && args->replace_extent && |
ac5887c8 | 525 | path->locks[0] == BTRFS_WRITE_LOCK && |
e902baac | 526 | btrfs_leaf_free_space(leaf) >= |
5893dfb9 | 527 | sizeof(struct btrfs_item) + args->extent_item_size) { |
d5f37527 FDBM |
528 | |
529 | key.objectid = ino; | |
530 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5893dfb9 | 531 | key.offset = args->start; |
d5f37527 FDBM |
532 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
533 | struct btrfs_key slot_key; | |
534 | ||
535 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
536 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
537 | path->slots[0]++; | |
1acae57b | 538 | } |
f0641656 | 539 | btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size); |
5893dfb9 | 540 | args->extent_inserted = true; |
6643558d | 541 | } |
920bbbfb | 542 | |
5893dfb9 FM |
543 | if (!args->path) |
544 | btrfs_free_path(path); | |
545 | else if (!args->extent_inserted) | |
1acae57b | 546 | btrfs_release_path(path); |
5893dfb9 FM |
547 | out: |
548 | args->drop_end = found ? min(args->end, last_end) : args->end; | |
5dc562c5 | 549 | |
39279cc3 CM |
550 | return ret; |
551 | } | |
552 | ||
d899e052 | 553 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
554 | u64 objectid, u64 bytenr, u64 orig_offset, |
555 | u64 *start, u64 *end) | |
d899e052 YZ |
556 | { |
557 | struct btrfs_file_extent_item *fi; | |
558 | struct btrfs_key key; | |
559 | u64 extent_end; | |
560 | ||
561 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
562 | return 0; | |
563 | ||
564 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
565 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
566 | return 0; | |
567 | ||
568 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
569 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
570 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 571 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
572 | btrfs_file_extent_compression(leaf, fi) || |
573 | btrfs_file_extent_encryption(leaf, fi) || | |
574 | btrfs_file_extent_other_encoding(leaf, fi)) | |
575 | return 0; | |
576 | ||
577 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
578 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
579 | return 0; | |
580 | ||
581 | *start = key.offset; | |
582 | *end = extent_end; | |
583 | return 1; | |
584 | } | |
585 | ||
586 | /* | |
587 | * Mark extent in the range start - end as written. | |
588 | * | |
589 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
590 | * part of extent is marked as written, the extent will be split into | |
591 | * two or three. | |
592 | */ | |
593 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 594 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 595 | { |
3ffbd68c | 596 | struct btrfs_fs_info *fs_info = trans->fs_info; |
7a6d7067 | 597 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
598 | struct extent_buffer *leaf; |
599 | struct btrfs_path *path; | |
600 | struct btrfs_file_extent_item *fi; | |
82fa113f | 601 | struct btrfs_ref ref = { 0 }; |
d899e052 | 602 | struct btrfs_key key; |
920bbbfb | 603 | struct btrfs_key new_key; |
d899e052 YZ |
604 | u64 bytenr; |
605 | u64 num_bytes; | |
606 | u64 extent_end; | |
5d4f98a2 | 607 | u64 orig_offset; |
d899e052 YZ |
608 | u64 other_start; |
609 | u64 other_end; | |
920bbbfb YZ |
610 | u64 split; |
611 | int del_nr = 0; | |
612 | int del_slot = 0; | |
6c7d54ac | 613 | int recow; |
e7b2ec3d | 614 | int ret = 0; |
7a6d7067 | 615 | u64 ino = btrfs_ino(inode); |
d899e052 | 616 | |
d899e052 | 617 | path = btrfs_alloc_path(); |
d8926bb3 MF |
618 | if (!path) |
619 | return -ENOMEM; | |
d899e052 | 620 | again: |
6c7d54ac | 621 | recow = 0; |
920bbbfb | 622 | split = start; |
33345d01 | 623 | key.objectid = ino; |
d899e052 | 624 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 625 | key.offset = split; |
d899e052 YZ |
626 | |
627 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
628 | if (ret < 0) |
629 | goto out; | |
d899e052 YZ |
630 | if (ret > 0 && path->slots[0] > 0) |
631 | path->slots[0]--; | |
632 | ||
633 | leaf = path->nodes[0]; | |
634 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
635 | if (key.objectid != ino || |
636 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
637 | ret = -EINVAL; | |
638 | btrfs_abort_transaction(trans, ret); | |
639 | goto out; | |
640 | } | |
d899e052 YZ |
641 | fi = btrfs_item_ptr(leaf, path->slots[0], |
642 | struct btrfs_file_extent_item); | |
9c8e63db JB |
643 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
644 | ret = -EINVAL; | |
645 | btrfs_abort_transaction(trans, ret); | |
646 | goto out; | |
647 | } | |
d899e052 | 648 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
649 | if (key.offset > start || extent_end < end) { |
650 | ret = -EINVAL; | |
651 | btrfs_abort_transaction(trans, ret); | |
652 | goto out; | |
653 | } | |
d899e052 YZ |
654 | |
655 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
656 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 657 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
658 | memcpy(&new_key, &key, sizeof(new_key)); |
659 | ||
660 | if (start == key.offset && end < extent_end) { | |
661 | other_start = 0; | |
662 | other_end = start; | |
663 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 664 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
665 | &other_start, &other_end)) { |
666 | new_key.offset = end; | |
0b246afa | 667 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
668 | fi = btrfs_item_ptr(leaf, path->slots[0], |
669 | struct btrfs_file_extent_item); | |
224ecce5 JB |
670 | btrfs_set_file_extent_generation(leaf, fi, |
671 | trans->transid); | |
6c7d54ac YZ |
672 | btrfs_set_file_extent_num_bytes(leaf, fi, |
673 | extent_end - end); | |
674 | btrfs_set_file_extent_offset(leaf, fi, | |
675 | end - orig_offset); | |
676 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
677 | struct btrfs_file_extent_item); | |
224ecce5 JB |
678 | btrfs_set_file_extent_generation(leaf, fi, |
679 | trans->transid); | |
6c7d54ac YZ |
680 | btrfs_set_file_extent_num_bytes(leaf, fi, |
681 | end - other_start); | |
682 | btrfs_mark_buffer_dirty(leaf); | |
683 | goto out; | |
684 | } | |
685 | } | |
686 | ||
687 | if (start > key.offset && end == extent_end) { | |
688 | other_start = end; | |
689 | other_end = 0; | |
690 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 691 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
692 | &other_start, &other_end)) { |
693 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
694 | struct btrfs_file_extent_item); | |
695 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
696 | start - key.offset); | |
224ecce5 JB |
697 | btrfs_set_file_extent_generation(leaf, fi, |
698 | trans->transid); | |
6c7d54ac YZ |
699 | path->slots[0]++; |
700 | new_key.offset = start; | |
0b246afa | 701 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
702 | |
703 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
704 | struct btrfs_file_extent_item); | |
224ecce5 JB |
705 | btrfs_set_file_extent_generation(leaf, fi, |
706 | trans->transid); | |
6c7d54ac YZ |
707 | btrfs_set_file_extent_num_bytes(leaf, fi, |
708 | other_end - start); | |
709 | btrfs_set_file_extent_offset(leaf, fi, | |
710 | start - orig_offset); | |
711 | btrfs_mark_buffer_dirty(leaf); | |
712 | goto out; | |
713 | } | |
714 | } | |
d899e052 | 715 | |
920bbbfb YZ |
716 | while (start > key.offset || end < extent_end) { |
717 | if (key.offset == start) | |
718 | split = end; | |
719 | ||
920bbbfb YZ |
720 | new_key.offset = split; |
721 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
722 | if (ret == -EAGAIN) { | |
b3b4aa74 | 723 | btrfs_release_path(path); |
920bbbfb | 724 | goto again; |
d899e052 | 725 | } |
79787eaa | 726 | if (ret < 0) { |
66642832 | 727 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
728 | goto out; |
729 | } | |
d899e052 | 730 | |
920bbbfb YZ |
731 | leaf = path->nodes[0]; |
732 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 733 | struct btrfs_file_extent_item); |
224ecce5 | 734 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 735 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
736 | split - key.offset); |
737 | ||
738 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
739 | struct btrfs_file_extent_item); | |
740 | ||
224ecce5 | 741 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
742 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
743 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
744 | extent_end - split); | |
d899e052 YZ |
745 | btrfs_mark_buffer_dirty(leaf); |
746 | ||
82fa113f QW |
747 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr, |
748 | num_bytes, 0); | |
749 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, | |
f42c5da6 | 750 | orig_offset, 0, false); |
82fa113f | 751 | ret = btrfs_inc_extent_ref(trans, &ref); |
9c8e63db JB |
752 | if (ret) { |
753 | btrfs_abort_transaction(trans, ret); | |
754 | goto out; | |
755 | } | |
d899e052 | 756 | |
920bbbfb YZ |
757 | if (split == start) { |
758 | key.offset = start; | |
759 | } else { | |
9c8e63db JB |
760 | if (start != key.offset) { |
761 | ret = -EINVAL; | |
762 | btrfs_abort_transaction(trans, ret); | |
763 | goto out; | |
764 | } | |
d899e052 | 765 | path->slots[0]--; |
920bbbfb | 766 | extent_end = end; |
d899e052 | 767 | } |
6c7d54ac | 768 | recow = 1; |
d899e052 YZ |
769 | } |
770 | ||
920bbbfb YZ |
771 | other_start = end; |
772 | other_end = 0; | |
ffd4bb2a QW |
773 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
774 | num_bytes, 0); | |
f42c5da6 NB |
775 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset, |
776 | 0, false); | |
6c7d54ac | 777 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 778 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
779 | &other_start, &other_end)) { |
780 | if (recow) { | |
b3b4aa74 | 781 | btrfs_release_path(path); |
6c7d54ac YZ |
782 | goto again; |
783 | } | |
920bbbfb YZ |
784 | extent_end = other_end; |
785 | del_slot = path->slots[0] + 1; | |
786 | del_nr++; | |
ffd4bb2a | 787 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
788 | if (ret) { |
789 | btrfs_abort_transaction(trans, ret); | |
790 | goto out; | |
791 | } | |
d899e052 | 792 | } |
920bbbfb YZ |
793 | other_start = 0; |
794 | other_end = start; | |
6c7d54ac | 795 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 796 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
797 | &other_start, &other_end)) { |
798 | if (recow) { | |
b3b4aa74 | 799 | btrfs_release_path(path); |
6c7d54ac YZ |
800 | goto again; |
801 | } | |
920bbbfb YZ |
802 | key.offset = other_start; |
803 | del_slot = path->slots[0]; | |
804 | del_nr++; | |
ffd4bb2a | 805 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
806 | if (ret) { |
807 | btrfs_abort_transaction(trans, ret); | |
808 | goto out; | |
809 | } | |
920bbbfb YZ |
810 | } |
811 | if (del_nr == 0) { | |
3f6fae95 SL |
812 | fi = btrfs_item_ptr(leaf, path->slots[0], |
813 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
814 | btrfs_set_file_extent_type(leaf, fi, |
815 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 816 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 817 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 818 | } else { |
3f6fae95 SL |
819 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
820 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
821 | btrfs_set_file_extent_type(leaf, fi, |
822 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 823 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
824 | btrfs_set_file_extent_num_bytes(leaf, fi, |
825 | extent_end - key.offset); | |
826 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 827 | |
6c7d54ac | 828 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 829 | if (ret < 0) { |
66642832 | 830 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
831 | goto out; |
832 | } | |
6c7d54ac | 833 | } |
920bbbfb | 834 | out: |
d899e052 | 835 | btrfs_free_path(path); |
e7b2ec3d | 836 | return ret; |
d899e052 YZ |
837 | } |
838 | ||
b1bf862e CM |
839 | /* |
840 | * on error we return an unlocked page and the error value | |
841 | * on success we return a locked page and 0 | |
842 | */ | |
bb1591b4 CM |
843 | static int prepare_uptodate_page(struct inode *inode, |
844 | struct page *page, u64 pos, | |
b6316429 | 845 | bool force_uptodate) |
b1bf862e | 846 | { |
fb12489b | 847 | struct folio *folio = page_folio(page); |
b1bf862e CM |
848 | int ret = 0; |
849 | ||
09cbfeaf | 850 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 851 | !PageUptodate(page)) { |
fb12489b | 852 | ret = btrfs_read_folio(NULL, folio); |
b1bf862e CM |
853 | if (ret) |
854 | return ret; | |
855 | lock_page(page); | |
856 | if (!PageUptodate(page)) { | |
857 | unlock_page(page); | |
858 | return -EIO; | |
859 | } | |
e0467866 QW |
860 | |
861 | /* | |
fb12489b | 862 | * Since btrfs_read_folio() will unlock the folio before it |
f913cff3 | 863 | * returns, there is a window where btrfs_release_folio() can be |
7c11d0ae QW |
864 | * called to release the page. Here we check both inode |
865 | * mapping and PagePrivate() to make sure the page was not | |
866 | * released. | |
e0467866 QW |
867 | * |
868 | * The private flag check is essential for subpage as we need | |
869 | * to store extra bitmap using page->private. | |
870 | */ | |
871 | if (page->mapping != inode->i_mapping || !PagePrivate(page)) { | |
bb1591b4 CM |
872 | unlock_page(page); |
873 | return -EAGAIN; | |
874 | } | |
b1bf862e CM |
875 | } |
876 | return 0; | |
877 | } | |
878 | ||
fc226000 SR |
879 | static unsigned int get_prepare_fgp_flags(bool nowait) |
880 | { | |
881 | unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT; | |
882 | ||
883 | if (nowait) | |
884 | fgp_flags |= FGP_NOWAIT; | |
885 | ||
886 | return fgp_flags; | |
887 | } | |
888 | ||
889 | static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait) | |
890 | { | |
891 | gfp_t gfp; | |
892 | ||
893 | gfp = btrfs_alloc_write_mask(inode->i_mapping); | |
894 | if (nowait) { | |
895 | gfp &= ~__GFP_DIRECT_RECLAIM; | |
896 | gfp |= GFP_NOWAIT; | |
897 | } | |
898 | ||
899 | return gfp; | |
900 | } | |
901 | ||
39279cc3 | 902 | /* |
376cc685 | 903 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 904 | */ |
b37392ea MX |
905 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
906 | size_t num_pages, loff_t pos, | |
fc226000 SR |
907 | size_t write_bytes, bool force_uptodate, |
908 | bool nowait) | |
39279cc3 CM |
909 | { |
910 | int i; | |
09cbfeaf | 911 | unsigned long index = pos >> PAGE_SHIFT; |
fc226000 SR |
912 | gfp_t mask = get_prepare_gfp_flags(inode, nowait); |
913 | unsigned int fgp_flags = get_prepare_fgp_flags(nowait); | |
fc28b62d | 914 | int err = 0; |
376cc685 | 915 | int faili; |
8c2383c3 | 916 | |
39279cc3 | 917 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 918 | again: |
fc226000 SR |
919 | pages[i] = pagecache_get_page(inode->i_mapping, index + i, |
920 | fgp_flags, mask | __GFP_WRITE); | |
39279cc3 | 921 | if (!pages[i]) { |
b1bf862e | 922 | faili = i - 1; |
fc226000 SR |
923 | if (nowait) |
924 | err = -EAGAIN; | |
925 | else | |
926 | err = -ENOMEM; | |
b1bf862e CM |
927 | goto fail; |
928 | } | |
929 | ||
32443de3 QW |
930 | err = set_page_extent_mapped(pages[i]); |
931 | if (err < 0) { | |
932 | faili = i; | |
933 | goto fail; | |
934 | } | |
935 | ||
b1bf862e | 936 | if (i == 0) |
bb1591b4 | 937 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 938 | force_uptodate); |
bb1591b4 CM |
939 | if (!err && i == num_pages - 1) |
940 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 941 | pos + write_bytes, false); |
b1bf862e | 942 | if (err) { |
09cbfeaf | 943 | put_page(pages[i]); |
fc226000 | 944 | if (!nowait && err == -EAGAIN) { |
bb1591b4 CM |
945 | err = 0; |
946 | goto again; | |
947 | } | |
b1bf862e CM |
948 | faili = i - 1; |
949 | goto fail; | |
39279cc3 | 950 | } |
ccd467d6 | 951 | wait_on_page_writeback(pages[i]); |
39279cc3 | 952 | } |
376cc685 MX |
953 | |
954 | return 0; | |
955 | fail: | |
956 | while (faili >= 0) { | |
957 | unlock_page(pages[faili]); | |
09cbfeaf | 958 | put_page(pages[faili]); |
376cc685 MX |
959 | faili--; |
960 | } | |
961 | return err; | |
962 | ||
963 | } | |
964 | ||
965 | /* | |
966 | * This function locks the extent and properly waits for data=ordered extents | |
967 | * to finish before allowing the pages to be modified if need. | |
968 | * | |
969 | * The return value: | |
970 | * 1 - the extent is locked | |
971 | * 0 - the extent is not locked, and everything is OK | |
972 | * -EAGAIN - need re-prepare the pages | |
973 | * the other < 0 number - Something wrong happens | |
974 | */ | |
975 | static noinline int | |
2cff578c | 976 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 977 | size_t num_pages, loff_t pos, |
2e78c927 | 978 | size_t write_bytes, |
2fcab928 | 979 | u64 *lockstart, u64 *lockend, bool nowait, |
376cc685 MX |
980 | struct extent_state **cached_state) |
981 | { | |
3ffbd68c | 982 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
983 | u64 start_pos; |
984 | u64 last_pos; | |
985 | int i; | |
986 | int ret = 0; | |
987 | ||
0b246afa | 988 | start_pos = round_down(pos, fs_info->sectorsize); |
e21139c6 | 989 | last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1; |
376cc685 | 990 | |
e3b8a485 | 991 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 992 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 993 | |
2fcab928 | 994 | if (nowait) { |
83ae4133 JB |
995 | if (!try_lock_extent(&inode->io_tree, start_pos, last_pos, |
996 | cached_state)) { | |
2fcab928 SR |
997 | for (i = 0; i < num_pages; i++) { |
998 | unlock_page(pages[i]); | |
999 | put_page(pages[i]); | |
1000 | pages[i] = NULL; | |
1001 | } | |
1002 | ||
1003 | return -EAGAIN; | |
1004 | } | |
1005 | } else { | |
1006 | lock_extent(&inode->io_tree, start_pos, last_pos, cached_state); | |
1007 | } | |
1008 | ||
b88935bf MX |
1009 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1010 | last_pos - start_pos + 1); | |
e6dcd2dc | 1011 | if (ordered && |
bffe633e | 1012 | ordered->file_offset + ordered->num_bytes > start_pos && |
376cc685 | 1013 | ordered->file_offset <= last_pos) { |
570eb97b JB |
1014 | unlock_extent(&inode->io_tree, start_pos, last_pos, |
1015 | cached_state); | |
e6dcd2dc CM |
1016 | for (i = 0; i < num_pages; i++) { |
1017 | unlock_page(pages[i]); | |
09cbfeaf | 1018 | put_page(pages[i]); |
e6dcd2dc | 1019 | } |
36d45567 | 1020 | btrfs_start_ordered_extent(ordered); |
b88935bf MX |
1021 | btrfs_put_ordered_extent(ordered); |
1022 | return -EAGAIN; | |
e6dcd2dc CM |
1023 | } |
1024 | if (ordered) | |
1025 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1026 | |
376cc685 MX |
1027 | *lockstart = start_pos; |
1028 | *lockend = last_pos; | |
1029 | ret = 1; | |
0762704b | 1030 | } |
376cc685 | 1031 | |
7703bdd8 | 1032 | /* |
32443de3 QW |
1033 | * We should be called after prepare_pages() which should have locked |
1034 | * all pages in the range. | |
7703bdd8 | 1035 | */ |
32443de3 | 1036 | for (i = 0; i < num_pages; i++) |
e6dcd2dc | 1037 | WARN_ON(!PageLocked(pages[i])); |
b1bf862e | 1038 | |
376cc685 | 1039 | return ret; |
39279cc3 CM |
1040 | } |
1041 | ||
d7a8ab4e FM |
1042 | /* |
1043 | * Check if we can do nocow write into the range [@pos, @pos + @write_bytes) | |
1044 | * | |
1045 | * @pos: File offset. | |
1046 | * @write_bytes: The length to write, will be updated to the nocow writeable | |
1047 | * range. | |
1048 | * | |
1049 | * This function will flush ordered extents in the range to ensure proper | |
1050 | * nocow checks. | |
1051 | * | |
1052 | * Return: | |
1053 | * > 0 If we can nocow, and updates @write_bytes. | |
1054 | * 0 If we can't do a nocow write. | |
1055 | * -EAGAIN If we can't do a nocow write because snapshoting of the inode's | |
1056 | * root is in progress. | |
1057 | * < 0 If an error happened. | |
1058 | * | |
1059 | * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. | |
1060 | */ | |
1061 | int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, | |
80f9d241 | 1062 | size_t *write_bytes, bool nowait) |
7ee9e440 | 1063 | { |
3ffbd68c | 1064 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1065 | struct btrfs_root *root = inode->root; |
632ddfa2 | 1066 | struct extent_state *cached_state = NULL; |
7ee9e440 JB |
1067 | u64 lockstart, lockend; |
1068 | u64 num_bytes; | |
1069 | int ret; | |
1070 | ||
38d37aa9 QW |
1071 | if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) |
1072 | return 0; | |
1073 | ||
d7a8ab4e | 1074 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) |
5f791ec3 | 1075 | return -EAGAIN; |
8257b2dc | 1076 | |
0b246afa | 1077 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1078 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1079 | fs_info->sectorsize) - 1; |
5dbb75ed | 1080 | num_bytes = lockend - lockstart + 1; |
7ee9e440 | 1081 | |
80f9d241 | 1082 | if (nowait) { |
632ddfa2 JB |
1083 | if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend, |
1084 | &cached_state)) { | |
80f9d241 JB |
1085 | btrfs_drew_write_unlock(&root->snapshot_lock); |
1086 | return -EAGAIN; | |
1087 | } | |
1088 | } else { | |
632ddfa2 JB |
1089 | btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, |
1090 | &cached_state); | |
80f9d241 | 1091 | } |
85b7ab67 | 1092 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
80f9d241 JB |
1093 | NULL, NULL, NULL, nowait, false); |
1094 | if (ret <= 0) | |
d7a8ab4e | 1095 | btrfs_drew_write_unlock(&root->snapshot_lock); |
80f9d241 | 1096 | else |
c933956d MX |
1097 | *write_bytes = min_t(size_t, *write_bytes , |
1098 | num_bytes - pos + lockstart); | |
632ddfa2 | 1099 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
7ee9e440 JB |
1100 | |
1101 | return ret; | |
1102 | } | |
1103 | ||
38d37aa9 QW |
1104 | void btrfs_check_nocow_unlock(struct btrfs_inode *inode) |
1105 | { | |
1106 | btrfs_drew_write_unlock(&inode->root->snapshot_lock); | |
1107 | } | |
1108 | ||
b8d8e1fd GR |
1109 | static void update_time_for_write(struct inode *inode) |
1110 | { | |
1111 | struct timespec64 now; | |
1112 | ||
1113 | if (IS_NOCMTIME(inode)) | |
1114 | return; | |
1115 | ||
1116 | now = current_time(inode); | |
1117 | if (!timespec64_equal(&inode->i_mtime, &now)) | |
1118 | inode->i_mtime = now; | |
1119 | ||
1120 | if (!timespec64_equal(&inode->i_ctime, &now)) | |
1121 | inode->i_ctime = now; | |
1122 | ||
1123 | if (IS_I_VERSION(inode)) | |
1124 | inode_inc_iversion(inode); | |
1125 | } | |
1126 | ||
1127 | static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from, | |
1128 | size_t count) | |
1129 | { | |
1130 | struct file *file = iocb->ki_filp; | |
1131 | struct inode *inode = file_inode(file); | |
1132 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1133 | loff_t pos = iocb->ki_pos; | |
1134 | int ret; | |
1135 | loff_t oldsize; | |
1136 | loff_t start_pos; | |
1137 | ||
d7a8ab4e FM |
1138 | /* |
1139 | * Quickly bail out on NOWAIT writes if we don't have the nodatacow or | |
1140 | * prealloc flags, as without those flags we always have to COW. We will | |
1141 | * later check if we can really COW into the target range (using | |
1142 | * can_nocow_extent() at btrfs_get_blocks_direct_write()). | |
1143 | */ | |
1144 | if ((iocb->ki_flags & IOCB_NOWAIT) && | |
1145 | !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
1146 | return -EAGAIN; | |
b8d8e1fd GR |
1147 | |
1148 | current->backing_dev_info = inode_to_bdi(inode); | |
1149 | ret = file_remove_privs(file); | |
1150 | if (ret) | |
1151 | return ret; | |
1152 | ||
1153 | /* | |
1154 | * We reserve space for updating the inode when we reserve space for the | |
1155 | * extent we are going to write, so we will enospc out there. We don't | |
1156 | * need to start yet another transaction to update the inode as we will | |
1157 | * update the inode when we finish writing whatever data we write. | |
1158 | */ | |
1159 | update_time_for_write(inode); | |
1160 | ||
1161 | start_pos = round_down(pos, fs_info->sectorsize); | |
1162 | oldsize = i_size_read(inode); | |
1163 | if (start_pos > oldsize) { | |
1164 | /* Expand hole size to cover write data, preventing empty gap */ | |
1165 | loff_t end_pos = round_up(pos + count, fs_info->sectorsize); | |
1166 | ||
b06359a3 | 1167 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos); |
b8d8e1fd GR |
1168 | if (ret) { |
1169 | current->backing_dev_info = NULL; | |
1170 | return ret; | |
1171 | } | |
1172 | } | |
1173 | ||
1174 | return 0; | |
1175 | } | |
1176 | ||
e4af400a GR |
1177 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1178 | struct iov_iter *i) | |
4b46fce2 | 1179 | { |
e4af400a | 1180 | struct file *file = iocb->ki_filp; |
c3523706 | 1181 | loff_t pos; |
496ad9aa | 1182 | struct inode *inode = file_inode(file); |
0b246afa | 1183 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1184 | struct page **pages = NULL; |
364ecf36 | 1185 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1186 | u64 release_bytes = 0; |
376cc685 MX |
1187 | u64 lockstart; |
1188 | u64 lockend; | |
d0215f3e JB |
1189 | size_t num_written = 0; |
1190 | int nrptrs; | |
c3523706 | 1191 | ssize_t ret; |
7ee9e440 | 1192 | bool only_release_metadata = false; |
b6316429 | 1193 | bool force_page_uptodate = false; |
5e8b9ef3 | 1194 | loff_t old_isize = i_size_read(inode); |
c3523706 | 1195 | unsigned int ilock_flags = 0; |
304e45ac | 1196 | const bool nowait = (iocb->ki_flags & IOCB_NOWAIT); |
965f47ae | 1197 | unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0); |
c3523706 | 1198 | |
304e45ac | 1199 | if (nowait) |
c3523706 GR |
1200 | ilock_flags |= BTRFS_ILOCK_TRY; |
1201 | ||
29b6352b | 1202 | ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1203 | if (ret < 0) |
1204 | return ret; | |
4b46fce2 | 1205 | |
c3523706 GR |
1206 | ret = generic_write_checks(iocb, i); |
1207 | if (ret <= 0) | |
1208 | goto out; | |
1209 | ||
1210 | ret = btrfs_write_check(iocb, i, ret); | |
1211 | if (ret < 0) | |
1212 | goto out; | |
1213 | ||
1214 | pos = iocb->ki_pos; | |
09cbfeaf KS |
1215 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1216 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1217 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1218 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1219 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
c3523706 GR |
1220 | if (!pages) { |
1221 | ret = -ENOMEM; | |
1222 | goto out; | |
1223 | } | |
ab93dbec | 1224 | |
d0215f3e | 1225 | while (iov_iter_count(i) > 0) { |
c67d970f | 1226 | struct extent_state *cached_state = NULL; |
7073017a | 1227 | size_t offset = offset_in_page(pos); |
2e78c927 | 1228 | size_t sector_offset; |
d0215f3e | 1229 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1230 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1231 | offset); |
eefa45f5 | 1232 | size_t num_pages; |
7ee9e440 | 1233 | size_t reserve_bytes; |
d0215f3e JB |
1234 | size_t dirty_pages; |
1235 | size_t copied; | |
2e78c927 CR |
1236 | size_t dirty_sectors; |
1237 | size_t num_sectors; | |
79f015f2 | 1238 | int extents_locked; |
39279cc3 | 1239 | |
914ee295 XZ |
1240 | /* |
1241 | * Fault pages before locking them in prepare_pages | |
1242 | * to avoid recursive lock | |
1243 | */ | |
a6294593 | 1244 | if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) { |
914ee295 | 1245 | ret = -EFAULT; |
d0215f3e | 1246 | break; |
914ee295 XZ |
1247 | } |
1248 | ||
a0e248bb | 1249 | only_release_metadata = false; |
da17066c | 1250 | sector_offset = pos & (fs_info->sectorsize - 1); |
d9d8b2a5 | 1251 | |
364ecf36 | 1252 | extent_changeset_release(data_reserved); |
36ea6f3e NB |
1253 | ret = btrfs_check_data_free_space(BTRFS_I(inode), |
1254 | &data_reserved, pos, | |
304e45ac | 1255 | write_bytes, nowait); |
c6887cd1 | 1256 | if (ret < 0) { |
80f9d241 JB |
1257 | int can_nocow; |
1258 | ||
304e45ac SR |
1259 | if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) { |
1260 | ret = -EAGAIN; | |
1261 | break; | |
1262 | } | |
1263 | ||
eefa45f5 GR |
1264 | /* |
1265 | * If we don't have to COW at the offset, reserve | |
1266 | * metadata only. write_bytes may get smaller than | |
1267 | * requested here. | |
1268 | */ | |
80f9d241 | 1269 | can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos, |
304e45ac | 1270 | &write_bytes, nowait); |
80f9d241 JB |
1271 | if (can_nocow < 0) |
1272 | ret = can_nocow; | |
1273 | if (can_nocow > 0) | |
1274 | ret = 0; | |
1275 | if (ret) | |
c6887cd1 | 1276 | break; |
80f9d241 | 1277 | only_release_metadata = true; |
c6887cd1 | 1278 | } |
1832a6d5 | 1279 | |
eefa45f5 GR |
1280 | num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE); |
1281 | WARN_ON(num_pages > nrptrs); | |
1282 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1283 | fs_info->sectorsize); | |
8b62f87b | 1284 | WARN_ON(reserve_bytes == 0); |
9f3db423 | 1285 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
28c9b1e7 | 1286 | reserve_bytes, |
304e45ac | 1287 | reserve_bytes, nowait); |
7ee9e440 JB |
1288 | if (ret) { |
1289 | if (!only_release_metadata) | |
25ce28ca | 1290 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
bc42bda2 QW |
1291 | data_reserved, pos, |
1292 | write_bytes); | |
8257b2dc | 1293 | else |
38d37aa9 | 1294 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
a348c8d4 FM |
1295 | |
1296 | if (nowait && ret == -ENOSPC) | |
1297 | ret = -EAGAIN; | |
7ee9e440 JB |
1298 | break; |
1299 | } | |
1300 | ||
1301 | release_bytes = reserve_bytes; | |
376cc685 | 1302 | again: |
965f47ae | 1303 | ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags); |
eb81b682 FM |
1304 | if (ret) { |
1305 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); | |
965f47ae | 1306 | break; |
eb81b682 | 1307 | } |
965f47ae | 1308 | |
4a64001f JB |
1309 | /* |
1310 | * This is going to setup the pages array with the number of | |
1311 | * pages we want, so we don't really need to worry about the | |
1312 | * contents of pages from loop to loop | |
1313 | */ | |
b37392ea | 1314 | ret = prepare_pages(inode, pages, num_pages, |
fc226000 | 1315 | pos, write_bytes, force_page_uptodate, false); |
8b62f87b JB |
1316 | if (ret) { |
1317 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
8702ba93 | 1318 | reserve_bytes); |
d0215f3e | 1319 | break; |
8b62f87b | 1320 | } |
39279cc3 | 1321 | |
79f015f2 GR |
1322 | extents_locked = lock_and_cleanup_extent_if_need( |
1323 | BTRFS_I(inode), pages, | |
2cff578c | 1324 | num_pages, pos, write_bytes, &lockstart, |
304e45ac | 1325 | &lockend, nowait, &cached_state); |
79f015f2 | 1326 | if (extents_locked < 0) { |
304e45ac | 1327 | if (!nowait && extents_locked == -EAGAIN) |
376cc685 | 1328 | goto again; |
304e45ac | 1329 | |
8b62f87b | 1330 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
8702ba93 | 1331 | reserve_bytes); |
79f015f2 | 1332 | ret = extents_locked; |
376cc685 | 1333 | break; |
376cc685 MX |
1334 | } |
1335 | ||
ee22f0c4 | 1336 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1337 | |
0b246afa | 1338 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1339 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1340 | fs_info->sectorsize); |
1341 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1342 | |
b1bf862e CM |
1343 | /* |
1344 | * if we have trouble faulting in the pages, fall | |
1345 | * back to one page at a time | |
1346 | */ | |
1347 | if (copied < write_bytes) | |
1348 | nrptrs = 1; | |
1349 | ||
b6316429 JB |
1350 | if (copied == 0) { |
1351 | force_page_uptodate = true; | |
56244ef1 | 1352 | dirty_sectors = 0; |
b1bf862e | 1353 | dirty_pages = 0; |
b6316429 JB |
1354 | } else { |
1355 | force_page_uptodate = false; | |
ed6078f7 | 1356 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1357 | PAGE_SIZE); |
b6316429 | 1358 | } |
914ee295 | 1359 | |
2e78c927 | 1360 | if (num_sectors > dirty_sectors) { |
8b8b08cb | 1361 | /* release everything except the sectors we dirtied */ |
265fdfa6 | 1362 | release_bytes -= dirty_sectors << fs_info->sectorsize_bits; |
485290a7 | 1363 | if (only_release_metadata) { |
691fa059 | 1364 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1365 | release_bytes, true); |
485290a7 QW |
1366 | } else { |
1367 | u64 __pos; | |
1368 | ||
da17066c | 1369 | __pos = round_down(pos, |
0b246afa | 1370 | fs_info->sectorsize) + |
09cbfeaf | 1371 | (dirty_pages << PAGE_SHIFT); |
86d52921 | 1372 | btrfs_delalloc_release_space(BTRFS_I(inode), |
bc42bda2 | 1373 | data_reserved, __pos, |
43b18595 | 1374 | release_bytes, true); |
485290a7 | 1375 | } |
914ee295 XZ |
1376 | } |
1377 | ||
2e78c927 | 1378 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1379 | fs_info->sectorsize); |
376cc685 | 1380 | |
aa8c1a41 GR |
1381 | ret = btrfs_dirty_pages(BTRFS_I(inode), pages, |
1382 | dirty_pages, pos, copied, | |
1383 | &cached_state, only_release_metadata); | |
c67d970f FM |
1384 | |
1385 | /* | |
1386 | * If we have not locked the extent range, because the range's | |
1387 | * start offset is >= i_size, we might still have a non-NULL | |
1388 | * cached extent state, acquired while marking the extent range | |
1389 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1390 | * possible cached extent state to avoid a memory leak. | |
1391 | */ | |
79f015f2 | 1392 | if (extents_locked) |
570eb97b JB |
1393 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
1394 | lockend, &cached_state); | |
c67d970f FM |
1395 | else |
1396 | free_extent_state(cached_state); | |
1397 | ||
8702ba93 | 1398 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 | 1399 | if (ret) { |
e4f94347 | 1400 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
376cc685 | 1401 | break; |
f1de9683 | 1402 | } |
39279cc3 | 1403 | |
376cc685 | 1404 | release_bytes = 0; |
8257b2dc | 1405 | if (only_release_metadata) |
38d37aa9 | 1406 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
8257b2dc | 1407 | |
e4f94347 | 1408 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
f1de9683 | 1409 | |
d0215f3e JB |
1410 | cond_resched(); |
1411 | ||
914ee295 XZ |
1412 | pos += copied; |
1413 | num_written += copied; | |
d0215f3e | 1414 | } |
39279cc3 | 1415 | |
d0215f3e JB |
1416 | kfree(pages); |
1417 | ||
7ee9e440 | 1418 | if (release_bytes) { |
8257b2dc | 1419 | if (only_release_metadata) { |
38d37aa9 | 1420 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
691fa059 | 1421 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1422 | release_bytes, true); |
8257b2dc | 1423 | } else { |
86d52921 NB |
1424 | btrfs_delalloc_release_space(BTRFS_I(inode), |
1425 | data_reserved, | |
bc42bda2 | 1426 | round_down(pos, fs_info->sectorsize), |
43b18595 | 1427 | release_bytes, true); |
8257b2dc | 1428 | } |
7ee9e440 JB |
1429 | } |
1430 | ||
364ecf36 | 1431 | extent_changeset_free(data_reserved); |
5e8b9ef3 GR |
1432 | if (num_written > 0) { |
1433 | pagecache_isize_extended(inode, old_isize, iocb->ki_pos); | |
1434 | iocb->ki_pos += num_written; | |
1435 | } | |
c3523706 | 1436 | out: |
e5d4d75b | 1437 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
d0215f3e JB |
1438 | return num_written ? num_written : ret; |
1439 | } | |
1440 | ||
4e4cabec GR |
1441 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
1442 | const struct iov_iter *iter, loff_t offset) | |
1443 | { | |
1444 | const u32 blocksize_mask = fs_info->sectorsize - 1; | |
1445 | ||
1446 | if (offset & blocksize_mask) | |
1447 | return -EINVAL; | |
1448 | ||
1449 | if (iov_iter_alignment(iter) & blocksize_mask) | |
1450 | return -EINVAL; | |
1451 | ||
1452 | return 0; | |
1453 | } | |
1454 | ||
1455 | static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
d0215f3e JB |
1456 | { |
1457 | struct file *file = iocb->ki_filp; | |
728404da | 1458 | struct inode *inode = file_inode(file); |
4e4cabec | 1459 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c3523706 | 1460 | loff_t pos; |
4e4cabec | 1461 | ssize_t written = 0; |
d0215f3e | 1462 | ssize_t written_buffered; |
51bd9563 | 1463 | size_t prev_left = 0; |
d0215f3e | 1464 | loff_t endbyte; |
c3523706 GR |
1465 | ssize_t err; |
1466 | unsigned int ilock_flags = 0; | |
8184620a | 1467 | struct iomap_dio *dio; |
c3523706 GR |
1468 | |
1469 | if (iocb->ki_flags & IOCB_NOWAIT) | |
1470 | ilock_flags |= BTRFS_ILOCK_TRY; | |
1471 | ||
e9adabb9 GR |
1472 | /* If the write DIO is within EOF, use a shared lock */ |
1473 | if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode)) | |
1474 | ilock_flags |= BTRFS_ILOCK_SHARED; | |
1475 | ||
1476 | relock: | |
29b6352b | 1477 | err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1478 | if (err < 0) |
1479 | return err; | |
1480 | ||
1481 | err = generic_write_checks(iocb, from); | |
1482 | if (err <= 0) { | |
e5d4d75b | 1483 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1484 | return err; |
1485 | } | |
d0215f3e | 1486 | |
c3523706 GR |
1487 | err = btrfs_write_check(iocb, from, err); |
1488 | if (err < 0) { | |
e5d4d75b | 1489 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1490 | goto out; |
1491 | } | |
1492 | ||
1493 | pos = iocb->ki_pos; | |
e9adabb9 GR |
1494 | /* |
1495 | * Re-check since file size may have changed just before taking the | |
1496 | * lock or pos may have changed because of O_APPEND in generic_write_check() | |
1497 | */ | |
1498 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && | |
1499 | pos + iov_iter_count(from) > i_size_read(inode)) { | |
e5d4d75b | 1500 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
e9adabb9 GR |
1501 | ilock_flags &= ~BTRFS_ILOCK_SHARED; |
1502 | goto relock; | |
1503 | } | |
c3523706 GR |
1504 | |
1505 | if (check_direct_IO(fs_info, from, pos)) { | |
e5d4d75b | 1506 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
4e4cabec | 1507 | goto buffered; |
c3523706 | 1508 | } |
4e4cabec | 1509 | |
51bd9563 FM |
1510 | /* |
1511 | * The iov_iter can be mapped to the same file range we are writing to. | |
1512 | * If that's the case, then we will deadlock in the iomap code, because | |
1513 | * it first calls our callback btrfs_dio_iomap_begin(), which will create | |
1514 | * an ordered extent, and after that it will fault in the pages that the | |
1515 | * iov_iter refers to. During the fault in we end up in the readahead | |
1516 | * pages code (starting at btrfs_readahead()), which will lock the range, | |
1517 | * find that ordered extent and then wait for it to complete (at | |
1518 | * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since | |
1519 | * obviously the ordered extent can never complete as we didn't submit | |
1520 | * yet the respective bio(s). This always happens when the buffer is | |
1521 | * memory mapped to the same file range, since the iomap DIO code always | |
1522 | * invalidates pages in the target file range (after starting and waiting | |
1523 | * for any writeback). | |
1524 | * | |
1525 | * So here we disable page faults in the iov_iter and then retry if we | |
1526 | * got -EFAULT, faulting in the pages before the retry. | |
1527 | */ | |
51bd9563 | 1528 | from->nofault = true; |
8184620a | 1529 | dio = btrfs_dio_write(iocb, from, written); |
51bd9563 | 1530 | from->nofault = false; |
d0215f3e | 1531 | |
8184620a FM |
1532 | /* |
1533 | * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync | |
1534 | * iocb, and that needs to lock the inode. So unlock it before calling | |
1535 | * iomap_dio_complete() to avoid a deadlock. | |
1536 | */ | |
e5d4d75b | 1537 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
8184620a FM |
1538 | |
1539 | if (IS_ERR_OR_NULL(dio)) | |
1540 | err = PTR_ERR_OR_ZERO(dio); | |
1541 | else | |
1542 | err = iomap_dio_complete(dio); | |
1543 | ||
51bd9563 FM |
1544 | /* No increment (+=) because iomap returns a cumulative value. */ |
1545 | if (err > 0) | |
1546 | written = err; | |
1547 | ||
1548 | if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) { | |
1549 | const size_t left = iov_iter_count(from); | |
1550 | /* | |
1551 | * We have more data left to write. Try to fault in as many as | |
1552 | * possible of the remainder pages and retry. We do this without | |
1553 | * releasing and locking again the inode, to prevent races with | |
1554 | * truncate. | |
1555 | * | |
1556 | * Also, in case the iov refers to pages in the file range of the | |
1557 | * file we want to write to (due to a mmap), we could enter an | |
1558 | * infinite loop if we retry after faulting the pages in, since | |
1559 | * iomap will invalidate any pages in the range early on, before | |
1560 | * it tries to fault in the pages of the iov. So we keep track of | |
1561 | * how much was left of iov in the previous EFAULT and fallback | |
1562 | * to buffered IO in case we haven't made any progress. | |
1563 | */ | |
1564 | if (left == prev_left) { | |
1565 | err = -ENOTBLK; | |
1566 | } else { | |
1567 | fault_in_iov_iter_readable(from, left); | |
1568 | prev_left = left; | |
8184620a | 1569 | goto relock; |
51bd9563 | 1570 | } |
a42fa643 GR |
1571 | } |
1572 | ||
ac5e6669 FM |
1573 | /* |
1574 | * If 'err' is -ENOTBLK or we have not written all data, then it means | |
1575 | * we must fallback to buffered IO. | |
1576 | */ | |
51bd9563 | 1577 | if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from)) |
c3523706 | 1578 | goto out; |
d0215f3e | 1579 | |
4e4cabec | 1580 | buffered: |
ac5e6669 FM |
1581 | /* |
1582 | * If we are in a NOWAIT context, then return -EAGAIN to signal the caller | |
1583 | * it must retry the operation in a context where blocking is acceptable, | |
20af93d9 FM |
1584 | * because even if we end up not blocking during the buffered IO attempt |
1585 | * below, we will block when flushing and waiting for the IO. | |
ac5e6669 FM |
1586 | */ |
1587 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
1588 | err = -EAGAIN; | |
1589 | goto out; | |
1590 | } | |
1591 | ||
e4af400a GR |
1592 | pos = iocb->ki_pos; |
1593 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e JB |
1594 | if (written_buffered < 0) { |
1595 | err = written_buffered; | |
1596 | goto out; | |
39279cc3 | 1597 | } |
075bdbdb FM |
1598 | /* |
1599 | * Ensure all data is persisted. We want the next direct IO read to be | |
1600 | * able to read what was just written. | |
1601 | */ | |
d0215f3e | 1602 | endbyte = pos + written_buffered - 1; |
728404da | 1603 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1604 | if (err) |
1605 | goto out; | |
728404da | 1606 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1607 | if (err) |
1608 | goto out; | |
1609 | written += written_buffered; | |
867c4f93 | 1610 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1611 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1612 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1613 | out: |
51bd9563 | 1614 | return err < 0 ? err : written; |
d0215f3e | 1615 | } |
5b92ee72 | 1616 | |
7c0c7269 OS |
1617 | static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
1618 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
1619 | { | |
1620 | struct file *file = iocb->ki_filp; | |
1621 | struct inode *inode = file_inode(file); | |
1622 | loff_t count; | |
1623 | ssize_t ret; | |
1624 | ||
29b6352b | 1625 | btrfs_inode_lock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1626 | count = encoded->len; |
1627 | ret = generic_write_checks_count(iocb, &count); | |
1628 | if (ret == 0 && count != encoded->len) { | |
1629 | /* | |
1630 | * The write got truncated by generic_write_checks_count(). We | |
1631 | * can't do a partial encoded write. | |
1632 | */ | |
1633 | ret = -EFBIG; | |
1634 | } | |
1635 | if (ret || encoded->len == 0) | |
1636 | goto out; | |
1637 | ||
1638 | ret = btrfs_write_check(iocb, from, encoded->len); | |
1639 | if (ret < 0) | |
1640 | goto out; | |
1641 | ||
1642 | ret = btrfs_do_encoded_write(iocb, from, encoded); | |
1643 | out: | |
e5d4d75b | 1644 | btrfs_inode_unlock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1645 | return ret; |
1646 | } | |
1647 | ||
1648 | ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from, | |
1649 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
d0215f3e JB |
1650 | { |
1651 | struct file *file = iocb->ki_filp; | |
14971657 | 1652 | struct btrfs_inode *inode = BTRFS_I(file_inode(file)); |
7c0c7269 | 1653 | ssize_t num_written, num_sync; |
d0215f3e | 1654 | |
c86537a4 GR |
1655 | /* |
1656 | * If the fs flips readonly due to some impossible error, although we | |
1657 | * have opened a file as writable, we have to stop this write operation | |
1658 | * to ensure consistency. | |
1659 | */ | |
84961539 | 1660 | if (BTRFS_FS_ERROR(inode->root->fs_info)) |
c86537a4 GR |
1661 | return -EROFS; |
1662 | ||
926078b2 | 1663 | if (encoded && (iocb->ki_flags & IOCB_NOWAIT)) |
91f9943e CH |
1664 | return -EOPNOTSUPP; |
1665 | ||
7c0c7269 OS |
1666 | if (encoded) { |
1667 | num_written = btrfs_encoded_write(iocb, from, encoded); | |
1668 | num_sync = encoded->len; | |
1669 | } else if (iocb->ki_flags & IOCB_DIRECT) { | |
c1867eb3 DS |
1670 | num_written = btrfs_direct_write(iocb, from); |
1671 | num_sync = num_written; | |
7c0c7269 | 1672 | } else { |
c1867eb3 DS |
1673 | num_written = btrfs_buffered_write(iocb, from); |
1674 | num_sync = num_written; | |
7c0c7269 | 1675 | } |
d0215f3e | 1676 | |
bc0939fc FM |
1677 | btrfs_set_inode_last_sub_trans(inode); |
1678 | ||
7c0c7269 OS |
1679 | if (num_sync > 0) { |
1680 | num_sync = generic_write_sync(iocb, num_sync); | |
1681 | if (num_sync < 0) | |
1682 | num_written = num_sync; | |
1683 | } | |
0a3404dc | 1684 | |
39279cc3 | 1685 | current->backing_dev_info = NULL; |
c3523706 | 1686 | return num_written; |
39279cc3 CM |
1687 | } |
1688 | ||
7c0c7269 OS |
1689 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1690 | { | |
1691 | return btrfs_do_write_iter(iocb, from, NULL); | |
1692 | } | |
1693 | ||
d397712b | 1694 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1695 | { |
23b5ec74 JB |
1696 | struct btrfs_file_private *private = filp->private_data; |
1697 | ||
3c32c721 | 1698 | if (private) { |
23b5ec74 | 1699 | kfree(private->filldir_buf); |
3c32c721 FM |
1700 | free_extent_state(private->llseek_cached_state); |
1701 | kfree(private); | |
1702 | filp->private_data = NULL; | |
1703 | } | |
23b5ec74 | 1704 | |
f6dc45c7 | 1705 | /* |
1fd4033d NB |
1706 | * Set by setattr when we are about to truncate a file from a non-zero |
1707 | * size to a zero size. This tries to flush down new bytes that may | |
1708 | * have been written if the application were using truncate to replace | |
1709 | * a file in place. | |
f6dc45c7 | 1710 | */ |
1fd4033d | 1711 | if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
f6dc45c7 CM |
1712 | &BTRFS_I(inode)->runtime_flags)) |
1713 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
1714 | return 0; |
1715 | } | |
1716 | ||
669249ee FM |
1717 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
1718 | { | |
1719 | int ret; | |
343e4fc1 | 1720 | struct blk_plug plug; |
669249ee | 1721 | |
343e4fc1 LB |
1722 | /* |
1723 | * This is only called in fsync, which would do synchronous writes, so | |
1724 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
1725 | * multiple disks using raid profile, a large IO can be split to | |
1726 | * several segments of stripe length (currently 64K). | |
1727 | */ | |
1728 | blk_start_plug(&plug); | |
728404da | 1729 | ret = btrfs_fdatawrite_range(inode, start, end); |
343e4fc1 | 1730 | blk_finish_plug(&plug); |
669249ee FM |
1731 | |
1732 | return ret; | |
1733 | } | |
1734 | ||
626e9f41 FM |
1735 | static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx) |
1736 | { | |
1737 | struct btrfs_inode *inode = BTRFS_I(ctx->inode); | |
1738 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1739 | ||
1740 | if (btrfs_inode_in_log(inode, fs_info->generation) && | |
1741 | list_empty(&ctx->ordered_extents)) | |
1742 | return true; | |
1743 | ||
1744 | /* | |
1745 | * If we are doing a fast fsync we can not bail out if the inode's | |
1746 | * last_trans is <= then the last committed transaction, because we only | |
1747 | * update the last_trans of the inode during ordered extent completion, | |
1748 | * and for a fast fsync we don't wait for that, we only wait for the | |
1749 | * writeback to complete. | |
1750 | */ | |
1751 | if (inode->last_trans <= fs_info->last_trans_committed && | |
1752 | (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) || | |
1753 | list_empty(&ctx->ordered_extents))) | |
1754 | return true; | |
1755 | ||
1756 | return false; | |
1757 | } | |
1758 | ||
d352ac68 CM |
1759 | /* |
1760 | * fsync call for both files and directories. This logs the inode into | |
1761 | * the tree log instead of forcing full commits whenever possible. | |
1762 | * | |
1763 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1764 | * in the metadata btree are up to date for copying to the log. | |
1765 | * | |
1766 | * It drops the inode mutex before doing the tree log commit. This is an | |
1767 | * important optimization for directories because holding the mutex prevents | |
1768 | * new operations on the dir while we write to disk. | |
1769 | */ | |
02c24a82 | 1770 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1771 | { |
de17e793 | 1772 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 1773 | struct inode *inode = d_inode(dentry); |
0b246afa | 1774 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 1775 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 1776 | struct btrfs_trans_handle *trans; |
8b050d35 | 1777 | struct btrfs_log_ctx ctx; |
333427a5 | 1778 | int ret = 0, err; |
48778179 FM |
1779 | u64 len; |
1780 | bool full_sync; | |
39279cc3 | 1781 | |
1abe9b8a | 1782 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1783 | |
ebb70442 LB |
1784 | btrfs_init_log_ctx(&ctx, inode); |
1785 | ||
95418ed1 | 1786 | /* |
48778179 FM |
1787 | * Always set the range to a full range, otherwise we can get into |
1788 | * several problems, from missing file extent items to represent holes | |
1789 | * when not using the NO_HOLES feature, to log tree corruption due to | |
1790 | * races between hole detection during logging and completion of ordered | |
1791 | * extents outside the range, to missing checksums due to ordered extents | |
1792 | * for which we flushed only a subset of their pages. | |
95418ed1 | 1793 | */ |
48778179 FM |
1794 | start = 0; |
1795 | end = LLONG_MAX; | |
1796 | len = (u64)LLONG_MAX + 1; | |
95418ed1 | 1797 | |
90abccf2 MX |
1798 | /* |
1799 | * We write the dirty pages in the range and wait until they complete | |
1800 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1801 | * multi-task, and make the performance up. See |
1802 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1803 | */ |
669249ee | 1804 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 1805 | if (ret) |
333427a5 | 1806 | goto out; |
90abccf2 | 1807 | |
29b6352b | 1808 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
c495144b | 1809 | |
2ecb7923 | 1810 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 1811 | |
aab15e8e | 1812 | /* |
885f46d8 FM |
1813 | * Before we acquired the inode's lock and the mmap lock, someone may |
1814 | * have dirtied more pages in the target range. We need to make sure | |
1815 | * that writeback for any such pages does not start while we are logging | |
1816 | * the inode, because if it does, any of the following might happen when | |
1817 | * we are not doing a full inode sync: | |
aab15e8e FM |
1818 | * |
1819 | * 1) We log an extent after its writeback finishes but before its | |
1820 | * checksums are added to the csum tree, leading to -EIO errors | |
1821 | * when attempting to read the extent after a log replay. | |
1822 | * | |
1823 | * 2) We can end up logging an extent before its writeback finishes. | |
1824 | * Therefore after the log replay we will have a file extent item | |
1825 | * pointing to an unwritten extent (and no data checksums as well). | |
1826 | * | |
1827 | * So trigger writeback for any eventual new dirty pages and then we | |
1828 | * wait for all ordered extents to complete below. | |
1829 | */ | |
1830 | ret = start_ordered_ops(inode, start, end); | |
1831 | if (ret) { | |
e5d4d75b | 1832 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
aab15e8e FM |
1833 | goto out; |
1834 | } | |
1835 | ||
cef7820d FM |
1836 | /* |
1837 | * Always check for the full sync flag while holding the inode's lock, | |
1838 | * to avoid races with other tasks. The flag must be either set all the | |
1839 | * time during logging or always off all the time while logging. | |
1840 | * We check the flag here after starting delalloc above, because when | |
1841 | * running delalloc the full sync flag may be set if we need to drop | |
1842 | * extra extent map ranges due to temporary memory allocation failures. | |
1843 | */ | |
1844 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1845 | &BTRFS_I(inode)->runtime_flags); | |
1846 | ||
669249ee | 1847 | /* |
b5e6c3e1 | 1848 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 1849 | * IO of a lower priority task while holding a transaction open. |
ba0b084a | 1850 | * |
48778179 FM |
1851 | * For a full fsync we wait for the ordered extents to complete while |
1852 | * for a fast fsync we wait just for writeback to complete, and then | |
1853 | * attach the ordered extents to the transaction so that a transaction | |
1854 | * commit waits for their completion, to avoid data loss if we fsync, | |
1855 | * the current transaction commits before the ordered extents complete | |
1856 | * and a power failure happens right after that. | |
d8e3fb10 NA |
1857 | * |
1858 | * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the | |
1859 | * logical address recorded in the ordered extent may change. We need | |
1860 | * to wait for the IO to stabilize the logical address. | |
669249ee | 1861 | */ |
d8e3fb10 | 1862 | if (full_sync || btrfs_is_zoned(fs_info)) { |
48778179 FM |
1863 | ret = btrfs_wait_ordered_range(inode, start, len); |
1864 | } else { | |
1865 | /* | |
1866 | * Get our ordered extents as soon as possible to avoid doing | |
1867 | * checksum lookups in the csum tree, and use instead the | |
1868 | * checksums attached to the ordered extents. | |
1869 | */ | |
1870 | btrfs_get_ordered_extents_for_logging(BTRFS_I(inode), | |
1871 | &ctx.ordered_extents); | |
1872 | ret = filemap_fdatawait_range(inode->i_mapping, start, end); | |
0ef8b726 | 1873 | } |
48778179 FM |
1874 | |
1875 | if (ret) | |
1876 | goto out_release_extents; | |
1877 | ||
2ecb7923 | 1878 | atomic_inc(&root->log_batch); |
257c62e1 | 1879 | |
a4abeea4 | 1880 | smp_mb(); |
626e9f41 | 1881 | if (skip_inode_logging(&ctx)) { |
5dc562c5 | 1882 | /* |
01327610 | 1883 | * We've had everything committed since the last time we were |
5dc562c5 JB |
1884 | * modified so clear this flag in case it was set for whatever |
1885 | * reason, it's no longer relevant. | |
1886 | */ | |
1887 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1888 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
1889 | /* |
1890 | * An ordered extent might have started before and completed | |
1891 | * already with io errors, in which case the inode was not | |
1892 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
1893 | * for any errors that might have happened since we last |
1894 | * checked called fsync. | |
0596a904 | 1895 | */ |
333427a5 | 1896 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
48778179 | 1897 | goto out_release_extents; |
15ee9bc7 | 1898 | } |
15ee9bc7 | 1899 | |
5039eddc JB |
1900 | /* |
1901 | * We use start here because we will need to wait on the IO to complete | |
1902 | * in btrfs_sync_log, which could require joining a transaction (for | |
1903 | * example checking cross references in the nocow path). If we use join | |
1904 | * here we could get into a situation where we're waiting on IO to | |
1905 | * happen that is blocked on a transaction trying to commit. With start | |
1906 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 1907 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
1908 | * from thinking they are super smart and changing this to |
1909 | * btrfs_join_transaction *cough*Josef*cough*. | |
1910 | */ | |
a22285a6 YZ |
1911 | trans = btrfs_start_transaction(root, 0); |
1912 | if (IS_ERR(trans)) { | |
1913 | ret = PTR_ERR(trans); | |
48778179 | 1914 | goto out_release_extents; |
39279cc3 | 1915 | } |
d0c2f4fa | 1916 | trans->in_fsync = true; |
e02119d5 | 1917 | |
48778179 FM |
1918 | ret = btrfs_log_dentry_safe(trans, dentry, &ctx); |
1919 | btrfs_release_log_ctx_extents(&ctx); | |
02c24a82 | 1920 | if (ret < 0) { |
a0634be5 | 1921 | /* Fallthrough and commit/free transaction. */ |
f31f09f6 | 1922 | ret = BTRFS_LOG_FORCE_COMMIT; |
02c24a82 | 1923 | } |
49eb7e46 CM |
1924 | |
1925 | /* we've logged all the items and now have a consistent | |
1926 | * version of the file in the log. It is possible that | |
1927 | * someone will come in and modify the file, but that's | |
1928 | * fine because the log is consistent on disk, and we | |
1929 | * have references to all of the file's extents | |
1930 | * | |
1931 | * It is possible that someone will come in and log the | |
1932 | * file again, but that will end up using the synchronization | |
1933 | * inside btrfs_sync_log to keep things safe. | |
1934 | */ | |
e5d4d75b | 1935 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
49eb7e46 | 1936 | |
bf7ba8ee JB |
1937 | if (ret == BTRFS_NO_LOG_SYNC) { |
1938 | ret = btrfs_end_transaction(trans); | |
1939 | goto out; | |
1940 | } | |
1941 | ||
1942 | /* We successfully logged the inode, attempt to sync the log. */ | |
1943 | if (!ret) { | |
1944 | ret = btrfs_sync_log(trans, root, &ctx); | |
0ef8b726 | 1945 | if (!ret) { |
bf7ba8ee JB |
1946 | ret = btrfs_end_transaction(trans); |
1947 | goto out; | |
48778179 | 1948 | } |
bf7ba8ee JB |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * At this point we need to commit the transaction because we had | |
1953 | * btrfs_need_log_full_commit() or some other error. | |
1954 | * | |
1955 | * If we didn't do a full sync we have to stop the trans handle, wait on | |
1956 | * the ordered extents, start it again and commit the transaction. If | |
1957 | * we attempt to wait on the ordered extents here we could deadlock with | |
1958 | * something like fallocate() that is holding the extent lock trying to | |
1959 | * start a transaction while some other thread is trying to commit the | |
1960 | * transaction while we (fsync) are currently holding the transaction | |
1961 | * open. | |
1962 | */ | |
1963 | if (!full_sync) { | |
3a45bb20 | 1964 | ret = btrfs_end_transaction(trans); |
bf7ba8ee JB |
1965 | if (ret) |
1966 | goto out; | |
1967 | ret = btrfs_wait_ordered_range(inode, start, len); | |
1968 | if (ret) | |
1969 | goto out; | |
1970 | ||
1971 | /* | |
1972 | * This is safe to use here because we're only interested in | |
1973 | * making sure the transaction that had the ordered extents is | |
1974 | * committed. We aren't waiting on anything past this point, | |
1975 | * we're purely getting the transaction and committing it. | |
1976 | */ | |
1977 | trans = btrfs_attach_transaction_barrier(root); | |
1978 | if (IS_ERR(trans)) { | |
1979 | ret = PTR_ERR(trans); | |
1980 | ||
1981 | /* | |
1982 | * We committed the transaction and there's no currently | |
1983 | * running transaction, this means everything we care | |
1984 | * about made it to disk and we are done. | |
1985 | */ | |
1986 | if (ret == -ENOENT) | |
1987 | ret = 0; | |
1988 | goto out; | |
1989 | } | |
e02119d5 | 1990 | } |
bf7ba8ee JB |
1991 | |
1992 | ret = btrfs_commit_transaction(trans); | |
39279cc3 | 1993 | out: |
ebb70442 | 1994 | ASSERT(list_empty(&ctx.list)); |
e09d94c9 | 1995 | ASSERT(list_empty(&ctx.conflict_inodes)); |
333427a5 JL |
1996 | err = file_check_and_advance_wb_err(file); |
1997 | if (!ret) | |
1998 | ret = err; | |
014e4ac4 | 1999 | return ret > 0 ? -EIO : ret; |
48778179 FM |
2000 | |
2001 | out_release_extents: | |
2002 | btrfs_release_log_ctx_extents(&ctx); | |
e5d4d75b | 2003 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
48778179 | 2004 | goto out; |
39279cc3 CM |
2005 | } |
2006 | ||
f0f37e2f | 2007 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2008 | .fault = filemap_fault, |
f1820361 | 2009 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2010 | .page_mkwrite = btrfs_page_mkwrite, |
2011 | }; | |
2012 | ||
2013 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2014 | { | |
058a457e MX |
2015 | struct address_space *mapping = filp->f_mapping; |
2016 | ||
7e0a1265 | 2017 | if (!mapping->a_ops->read_folio) |
058a457e MX |
2018 | return -ENOEXEC; |
2019 | ||
9ebefb18 | 2020 | file_accessed(filp); |
058a457e | 2021 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2022 | |
9ebefb18 CM |
2023 | return 0; |
2024 | } | |
2025 | ||
35339c24 | 2026 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2027 | int slot, u64 start, u64 end) |
2028 | { | |
2029 | struct btrfs_file_extent_item *fi; | |
2030 | struct btrfs_key key; | |
2031 | ||
2032 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2033 | return 0; | |
2034 | ||
2035 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2036 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2037 | key.type != BTRFS_EXTENT_DATA_KEY) |
2038 | return 0; | |
2039 | ||
2040 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2041 | ||
2042 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2043 | return 0; | |
2044 | ||
2045 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2046 | return 0; | |
2047 | ||
2048 | if (key.offset == end) | |
2049 | return 1; | |
2050 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2051 | return 1; | |
2052 | return 0; | |
2053 | } | |
2054 | ||
a012a74e NB |
2055 | static int fill_holes(struct btrfs_trans_handle *trans, |
2056 | struct btrfs_inode *inode, | |
2057 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2058 | { |
3ffbd68c | 2059 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2060 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2061 | struct extent_buffer *leaf; |
2062 | struct btrfs_file_extent_item *fi; | |
2063 | struct extent_map *hole_em; | |
2aaa6655 JB |
2064 | struct btrfs_key key; |
2065 | int ret; | |
2066 | ||
0b246afa | 2067 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2068 | goto out; |
2069 | ||
a012a74e | 2070 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2071 | key.type = BTRFS_EXTENT_DATA_KEY; |
2072 | key.offset = offset; | |
2073 | ||
2aaa6655 | 2074 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2075 | if (ret <= 0) { |
2076 | /* | |
2077 | * We should have dropped this offset, so if we find it then | |
2078 | * something has gone horribly wrong. | |
2079 | */ | |
2080 | if (ret == 0) | |
2081 | ret = -EINVAL; | |
2aaa6655 | 2082 | return ret; |
f94480bd | 2083 | } |
2aaa6655 JB |
2084 | |
2085 | leaf = path->nodes[0]; | |
a012a74e | 2086 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2087 | u64 num_bytes; |
2088 | ||
2089 | path->slots[0]--; | |
2090 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2091 | struct btrfs_file_extent_item); | |
2092 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2093 | end - offset; | |
2094 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2095 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2096 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2097 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2098 | btrfs_mark_buffer_dirty(leaf); |
2099 | goto out; | |
2100 | } | |
2101 | ||
1707e26d | 2102 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2103 | u64 num_bytes; |
2104 | ||
2aaa6655 | 2105 | key.offset = offset; |
0b246afa | 2106 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2107 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2108 | struct btrfs_file_extent_item); | |
2109 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2110 | offset; | |
2111 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2112 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2113 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2114 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2115 | btrfs_mark_buffer_dirty(leaf); |
2116 | goto out; | |
2117 | } | |
2118 | btrfs_release_path(path); | |
2119 | ||
d1f68ba0 OS |
2120 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, |
2121 | end - offset); | |
2aaa6655 JB |
2122 | if (ret) |
2123 | return ret; | |
2124 | ||
2125 | out: | |
2126 | btrfs_release_path(path); | |
2127 | ||
2128 | hole_em = alloc_extent_map(); | |
2129 | if (!hole_em) { | |
4c0c8cfc | 2130 | btrfs_drop_extent_map_range(inode, offset, end - 1, false); |
23e3337f | 2131 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2132 | } else { |
2133 | hole_em->start = offset; | |
2134 | hole_em->len = end - offset; | |
cc95bef6 | 2135 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2136 | hole_em->orig_start = offset; |
2137 | ||
2138 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2139 | hole_em->block_len = 0; | |
b4939680 | 2140 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2141 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2142 | hole_em->generation = trans->transid; | |
2143 | ||
a1ba4c08 | 2144 | ret = btrfs_replace_extent_map_range(inode, hole_em, true); |
2aaa6655 JB |
2145 | free_extent_map(hole_em); |
2146 | if (ret) | |
23e3337f | 2147 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2148 | } |
2149 | ||
2150 | return 0; | |
2151 | } | |
2152 | ||
d7781546 QW |
2153 | /* |
2154 | * Find a hole extent on given inode and change start/len to the end of hole | |
2155 | * extent.(hole/vacuum extent whose em->start <= start && | |
2156 | * em->start + em->len > start) | |
2157 | * When a hole extent is found, return 1 and modify start/len. | |
2158 | */ | |
dea46d84 | 2159 | static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len) |
d7781546 | 2160 | { |
dea46d84 | 2161 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d7781546 QW |
2162 | struct extent_map *em; |
2163 | int ret = 0; | |
2164 | ||
dea46d84 | 2165 | em = btrfs_get_extent(inode, NULL, 0, |
609805d8 | 2166 | round_down(*start, fs_info->sectorsize), |
39b07b5d | 2167 | round_up(*len, fs_info->sectorsize)); |
9986277e DC |
2168 | if (IS_ERR(em)) |
2169 | return PTR_ERR(em); | |
d7781546 QW |
2170 | |
2171 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2172 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2173 | ret = 1; | |
2174 | *len = em->start + em->len > *start + *len ? | |
2175 | 0 : *start + *len - em->start - em->len; | |
2176 | *start = em->start + em->len; | |
2177 | } | |
2178 | free_extent_map(em); | |
2179 | return ret; | |
2180 | } | |
2181 | ||
55961c8a FM |
2182 | static void btrfs_punch_hole_lock_range(struct inode *inode, |
2183 | const u64 lockstart, | |
2184 | const u64 lockend, | |
2185 | struct extent_state **cached_state) | |
f27451f2 | 2186 | { |
0528476b QW |
2187 | /* |
2188 | * For subpage case, if the range is not at page boundary, we could | |
2189 | * have pages at the leading/tailing part of the range. | |
2190 | * This could lead to dead loop since filemap_range_has_page() | |
2191 | * will always return true. | |
2192 | * So here we need to do extra page alignment for | |
2193 | * filemap_range_has_page(). | |
2194 | */ | |
2195 | const u64 page_lockstart = round_up(lockstart, PAGE_SIZE); | |
2196 | const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1; | |
2197 | ||
f27451f2 | 2198 | while (1) { |
f27451f2 FM |
2199 | truncate_pagecache_range(inode, lockstart, lockend); |
2200 | ||
570eb97b JB |
2201 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2202 | cached_state); | |
f27451f2 | 2203 | /* |
55961c8a FM |
2204 | * We can't have ordered extents in the range, nor dirty/writeback |
2205 | * pages, because we have locked the inode's VFS lock in exclusive | |
2206 | * mode, we have locked the inode's i_mmap_lock in exclusive mode, | |
2207 | * we have flushed all delalloc in the range and we have waited | |
2208 | * for any ordered extents in the range to complete. | |
2209 | * We can race with anyone reading pages from this range, so after | |
2210 | * locking the range check if we have pages in the range, and if | |
2211 | * we do, unlock the range and retry. | |
f27451f2 | 2212 | */ |
55961c8a FM |
2213 | if (!filemap_range_has_page(inode->i_mapping, page_lockstart, |
2214 | page_lockend)) | |
f27451f2 | 2215 | break; |
55961c8a | 2216 | |
570eb97b JB |
2217 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2218 | cached_state); | |
f27451f2 | 2219 | } |
63c34cb4 FM |
2220 | |
2221 | btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend); | |
f27451f2 FM |
2222 | } |
2223 | ||
0cbb5bdf | 2224 | static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans, |
03fcb1ab | 2225 | struct btrfs_inode *inode, |
690a5dbf | 2226 | struct btrfs_path *path, |
bf385648 | 2227 | struct btrfs_replace_extent_info *extent_info, |
2766ff61 FM |
2228 | const u64 replace_len, |
2229 | const u64 bytes_to_drop) | |
690a5dbf | 2230 | { |
03fcb1ab NB |
2231 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2232 | struct btrfs_root *root = inode->root; | |
690a5dbf FM |
2233 | struct btrfs_file_extent_item *extent; |
2234 | struct extent_buffer *leaf; | |
2235 | struct btrfs_key key; | |
2236 | int slot; | |
2237 | struct btrfs_ref ref = { 0 }; | |
690a5dbf FM |
2238 | int ret; |
2239 | ||
bf385648 | 2240 | if (replace_len == 0) |
690a5dbf FM |
2241 | return 0; |
2242 | ||
bf385648 | 2243 | if (extent_info->disk_offset == 0 && |
2766ff61 | 2244 | btrfs_fs_incompat(fs_info, NO_HOLES)) { |
03fcb1ab | 2245 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2246 | return 0; |
2766ff61 | 2247 | } |
690a5dbf | 2248 | |
03fcb1ab | 2249 | key.objectid = btrfs_ino(inode); |
690a5dbf | 2250 | key.type = BTRFS_EXTENT_DATA_KEY; |
bf385648 | 2251 | key.offset = extent_info->file_offset; |
690a5dbf | 2252 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
fb870f6c | 2253 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf FM |
2254 | if (ret) |
2255 | return ret; | |
2256 | leaf = path->nodes[0]; | |
2257 | slot = path->slots[0]; | |
bf385648 | 2258 | write_extent_buffer(leaf, extent_info->extent_buf, |
690a5dbf | 2259 | btrfs_item_ptr_offset(leaf, slot), |
fb870f6c | 2260 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf | 2261 | extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
fb870f6c | 2262 | ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE); |
bf385648 FM |
2263 | btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset); |
2264 | btrfs_set_file_extent_num_bytes(leaf, extent, replace_len); | |
2265 | if (extent_info->is_new_extent) | |
8fccebfa | 2266 | btrfs_set_file_extent_generation(leaf, extent, trans->transid); |
690a5dbf FM |
2267 | btrfs_mark_buffer_dirty(leaf); |
2268 | btrfs_release_path(path); | |
2269 | ||
03fcb1ab NB |
2270 | ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset, |
2271 | replace_len); | |
9ddc959e JB |
2272 | if (ret) |
2273 | return ret; | |
2274 | ||
690a5dbf | 2275 | /* If it's a hole, nothing more needs to be done. */ |
2766ff61 | 2276 | if (extent_info->disk_offset == 0) { |
03fcb1ab | 2277 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2278 | return 0; |
2766ff61 | 2279 | } |
690a5dbf | 2280 | |
03fcb1ab | 2281 | btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop); |
8fccebfa | 2282 | |
bf385648 FM |
2283 | if (extent_info->is_new_extent && extent_info->insertions == 0) { |
2284 | key.objectid = extent_info->disk_offset; | |
8fccebfa | 2285 | key.type = BTRFS_EXTENT_ITEM_KEY; |
bf385648 | 2286 | key.offset = extent_info->disk_len; |
8fccebfa | 2287 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
03fcb1ab | 2288 | btrfs_ino(inode), |
bf385648 FM |
2289 | extent_info->file_offset, |
2290 | extent_info->qgroup_reserved, | |
8fccebfa FM |
2291 | &key); |
2292 | } else { | |
2293 | u64 ref_offset; | |
2294 | ||
2295 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, | |
bf385648 FM |
2296 | extent_info->disk_offset, |
2297 | extent_info->disk_len, 0); | |
2298 | ref_offset = extent_info->file_offset - extent_info->data_offset; | |
8fccebfa | 2299 | btrfs_init_data_ref(&ref, root->root_key.objectid, |
f42c5da6 | 2300 | btrfs_ino(inode), ref_offset, 0, false); |
8fccebfa FM |
2301 | ret = btrfs_inc_extent_ref(trans, &ref); |
2302 | } | |
2303 | ||
bf385648 | 2304 | extent_info->insertions++; |
690a5dbf FM |
2305 | |
2306 | return ret; | |
2307 | } | |
2308 | ||
9cba40a6 FM |
2309 | /* |
2310 | * The respective range must have been previously locked, as well as the inode. | |
2311 | * The end offset is inclusive (last byte of the range). | |
bf385648 FM |
2312 | * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing |
2313 | * the file range with an extent. | |
2314 | * When not punching a hole, we don't want to end up in a state where we dropped | |
2315 | * extents without inserting a new one, so we must abort the transaction to avoid | |
2316 | * a corruption. | |
9cba40a6 | 2317 | */ |
bfc78479 NB |
2318 | int btrfs_replace_file_extents(struct btrfs_inode *inode, |
2319 | struct btrfs_path *path, const u64 start, | |
2320 | const u64 end, | |
2321 | struct btrfs_replace_extent_info *extent_info, | |
2322 | struct btrfs_trans_handle **trans_out) | |
9cba40a6 | 2323 | { |
5893dfb9 | 2324 | struct btrfs_drop_extents_args drop_args = { 0 }; |
bfc78479 NB |
2325 | struct btrfs_root *root = inode->root; |
2326 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2bd36e7b | 2327 | u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1); |
bfc78479 | 2328 | u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize); |
9cba40a6 FM |
2329 | struct btrfs_trans_handle *trans = NULL; |
2330 | struct btrfs_block_rsv *rsv; | |
2331 | unsigned int rsv_count; | |
2332 | u64 cur_offset; | |
9cba40a6 FM |
2333 | u64 len = end - start; |
2334 | int ret = 0; | |
2335 | ||
2336 | if (end <= start) | |
2337 | return -EINVAL; | |
2338 | ||
2339 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); | |
2340 | if (!rsv) { | |
2341 | ret = -ENOMEM; | |
2342 | goto out; | |
2343 | } | |
2bd36e7b | 2344 | rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1); |
710d5921 | 2345 | rsv->failfast = true; |
9cba40a6 FM |
2346 | |
2347 | /* | |
2348 | * 1 - update the inode | |
2349 | * 1 - removing the extents in the range | |
bf385648 FM |
2350 | * 1 - adding the hole extent if no_holes isn't set or if we are |
2351 | * replacing the range with a new extent | |
9cba40a6 | 2352 | */ |
bf385648 | 2353 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info) |
690a5dbf FM |
2354 | rsv_count = 3; |
2355 | else | |
2356 | rsv_count = 2; | |
2357 | ||
9cba40a6 FM |
2358 | trans = btrfs_start_transaction(root, rsv_count); |
2359 | if (IS_ERR(trans)) { | |
2360 | ret = PTR_ERR(trans); | |
2361 | trans = NULL; | |
2362 | goto out_free; | |
2363 | } | |
2364 | ||
2365 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, | |
2366 | min_size, false); | |
650c9cab FM |
2367 | if (WARN_ON(ret)) |
2368 | goto out_trans; | |
9cba40a6 FM |
2369 | trans->block_rsv = rsv; |
2370 | ||
2371 | cur_offset = start; | |
5893dfb9 FM |
2372 | drop_args.path = path; |
2373 | drop_args.end = end + 1; | |
2374 | drop_args.drop_cache = true; | |
9cba40a6 | 2375 | while (cur_offset < end) { |
5893dfb9 | 2376 | drop_args.start = cur_offset; |
bfc78479 | 2377 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
2766ff61 FM |
2378 | /* If we are punching a hole decrement the inode's byte count */ |
2379 | if (!extent_info) | |
bfc78479 | 2380 | btrfs_update_inode_bytes(inode, 0, |
2766ff61 | 2381 | drop_args.bytes_found); |
690a5dbf FM |
2382 | if (ret != -ENOSPC) { |
2383 | /* | |
4afb912f JB |
2384 | * The only time we don't want to abort is if we are |
2385 | * attempting to clone a partial inline extent, in which | |
2386 | * case we'll get EOPNOTSUPP. However if we aren't | |
2387 | * clone we need to abort no matter what, because if we | |
2388 | * got EOPNOTSUPP via prealloc then we messed up and | |
2389 | * need to abort. | |
690a5dbf | 2390 | */ |
4afb912f JB |
2391 | if (ret && |
2392 | (ret != -EOPNOTSUPP || | |
2393 | (extent_info && extent_info->is_new_extent))) | |
690a5dbf | 2394 | btrfs_abort_transaction(trans, ret); |
9cba40a6 | 2395 | break; |
690a5dbf | 2396 | } |
9cba40a6 FM |
2397 | |
2398 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2399 | ||
5893dfb9 | 2400 | if (!extent_info && cur_offset < drop_args.drop_end && |
690a5dbf | 2401 | cur_offset < ino_size) { |
bfc78479 NB |
2402 | ret = fill_holes(trans, inode, path, cur_offset, |
2403 | drop_args.drop_end); | |
9cba40a6 FM |
2404 | if (ret) { |
2405 | /* | |
2406 | * If we failed then we didn't insert our hole | |
2407 | * entries for the area we dropped, so now the | |
2408 | * fs is corrupted, so we must abort the | |
2409 | * transaction. | |
2410 | */ | |
2411 | btrfs_abort_transaction(trans, ret); | |
2412 | break; | |
2413 | } | |
5893dfb9 | 2414 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e JB |
2415 | /* |
2416 | * We are past the i_size here, but since we didn't | |
2417 | * insert holes we need to clear the mapped area so we | |
2418 | * know to not set disk_i_size in this area until a new | |
2419 | * file extent is inserted here. | |
2420 | */ | |
bfc78479 | 2421 | ret = btrfs_inode_clear_file_extent_range(inode, |
5893dfb9 FM |
2422 | cur_offset, |
2423 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2424 | if (ret) { |
2425 | /* | |
2426 | * We couldn't clear our area, so we could | |
2427 | * presumably adjust up and corrupt the fs, so | |
2428 | * we need to abort. | |
2429 | */ | |
2430 | btrfs_abort_transaction(trans, ret); | |
2431 | break; | |
2432 | } | |
9cba40a6 FM |
2433 | } |
2434 | ||
5893dfb9 FM |
2435 | if (extent_info && |
2436 | drop_args.drop_end > extent_info->file_offset) { | |
2437 | u64 replace_len = drop_args.drop_end - | |
2438 | extent_info->file_offset; | |
690a5dbf | 2439 | |
bfc78479 NB |
2440 | ret = btrfs_insert_replace_extent(trans, inode, path, |
2441 | extent_info, replace_len, | |
03fcb1ab | 2442 | drop_args.bytes_found); |
690a5dbf FM |
2443 | if (ret) { |
2444 | btrfs_abort_transaction(trans, ret); | |
2445 | break; | |
2446 | } | |
bf385648 FM |
2447 | extent_info->data_len -= replace_len; |
2448 | extent_info->data_offset += replace_len; | |
2449 | extent_info->file_offset += replace_len; | |
690a5dbf FM |
2450 | } |
2451 | ||
983d8209 FM |
2452 | /* |
2453 | * We are releasing our handle on the transaction, balance the | |
2454 | * dirty pages of the btree inode and flush delayed items, and | |
2455 | * then get a new transaction handle, which may now point to a | |
2456 | * new transaction in case someone else may have committed the | |
2457 | * transaction we used to replace/drop file extent items. So | |
2458 | * bump the inode's iversion and update mtime and ctime except | |
2459 | * if we are called from a dedupe context. This is because a | |
2460 | * power failure/crash may happen after the transaction is | |
2461 | * committed and before we finish replacing/dropping all the | |
2462 | * file extent items we need. | |
2463 | */ | |
2464 | inode_inc_iversion(&inode->vfs_inode); | |
2465 | ||
2466 | if (!extent_info || extent_info->update_times) { | |
2467 | inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode); | |
2468 | inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime; | |
2469 | } | |
2470 | ||
bfc78479 | 2471 | ret = btrfs_update_inode(trans, root, inode); |
9cba40a6 FM |
2472 | if (ret) |
2473 | break; | |
2474 | ||
2475 | btrfs_end_transaction(trans); | |
2476 | btrfs_btree_balance_dirty(fs_info); | |
2477 | ||
2478 | trans = btrfs_start_transaction(root, rsv_count); | |
2479 | if (IS_ERR(trans)) { | |
2480 | ret = PTR_ERR(trans); | |
2481 | trans = NULL; | |
2482 | break; | |
2483 | } | |
2484 | ||
2485 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, | |
2486 | rsv, min_size, false); | |
650c9cab FM |
2487 | if (WARN_ON(ret)) |
2488 | break; | |
9cba40a6 FM |
2489 | trans->block_rsv = rsv; |
2490 | ||
3227788c BC |
2491 | cur_offset = drop_args.drop_end; |
2492 | len = end - cur_offset; | |
2493 | if (!extent_info && len) { | |
bfc78479 | 2494 | ret = find_first_non_hole(inode, &cur_offset, &len); |
690a5dbf FM |
2495 | if (unlikely(ret < 0)) |
2496 | break; | |
2497 | if (ret && !len) { | |
2498 | ret = 0; | |
2499 | break; | |
2500 | } | |
9cba40a6 FM |
2501 | } |
2502 | } | |
2503 | ||
690a5dbf FM |
2504 | /* |
2505 | * If we were cloning, force the next fsync to be a full one since we | |
2506 | * we replaced (or just dropped in the case of cloning holes when | |
e2b84217 FM |
2507 | * NO_HOLES is enabled) file extent items and did not setup new extent |
2508 | * maps for the replacement extents (or holes). | |
690a5dbf | 2509 | */ |
bf385648 | 2510 | if (extent_info && !extent_info->is_new_extent) |
23e3337f | 2511 | btrfs_set_inode_full_sync(inode); |
690a5dbf | 2512 | |
9cba40a6 FM |
2513 | if (ret) |
2514 | goto out_trans; | |
2515 | ||
2516 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2517 | /* | |
2518 | * If we are using the NO_HOLES feature we might have had already an | |
2519 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2520 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2521 | * represent holes, drop_end can be less than lockend and so we must | |
2522 | * make sure we have an extent map representing the existing hole (the | |
2523 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2524 | * map representing the existing hole), otherwise the fast fsync path | |
2525 | * will not record the existence of the hole region | |
2526 | * [existing_hole_start, lockend]. | |
2527 | */ | |
5893dfb9 FM |
2528 | if (drop_args.drop_end <= end) |
2529 | drop_args.drop_end = end + 1; | |
9cba40a6 FM |
2530 | /* |
2531 | * Don't insert file hole extent item if it's for a range beyond eof | |
2532 | * (because it's useless) or if it represents a 0 bytes range (when | |
2533 | * cur_offset == drop_end). | |
2534 | */ | |
5893dfb9 FM |
2535 | if (!extent_info && cur_offset < ino_size && |
2536 | cur_offset < drop_args.drop_end) { | |
bfc78479 NB |
2537 | ret = fill_holes(trans, inode, path, cur_offset, |
2538 | drop_args.drop_end); | |
9cba40a6 FM |
2539 | if (ret) { |
2540 | /* Same comment as above. */ | |
2541 | btrfs_abort_transaction(trans, ret); | |
2542 | goto out_trans; | |
2543 | } | |
5893dfb9 | 2544 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e | 2545 | /* See the comment in the loop above for the reasoning here. */ |
bfc78479 NB |
2546 | ret = btrfs_inode_clear_file_extent_range(inode, cur_offset, |
2547 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2548 | if (ret) { |
2549 | btrfs_abort_transaction(trans, ret); | |
2550 | goto out_trans; | |
2551 | } | |
2552 | ||
9cba40a6 | 2553 | } |
bf385648 | 2554 | if (extent_info) { |
bfc78479 | 2555 | ret = btrfs_insert_replace_extent(trans, inode, path, |
03fcb1ab NB |
2556 | extent_info, extent_info->data_len, |
2557 | drop_args.bytes_found); | |
690a5dbf FM |
2558 | if (ret) { |
2559 | btrfs_abort_transaction(trans, ret); | |
2560 | goto out_trans; | |
2561 | } | |
2562 | } | |
9cba40a6 FM |
2563 | |
2564 | out_trans: | |
2565 | if (!trans) | |
2566 | goto out_free; | |
2567 | ||
2568 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2569 | if (ret) | |
2570 | btrfs_end_transaction(trans); | |
2571 | else | |
2572 | *trans_out = trans; | |
2573 | out_free: | |
2574 | btrfs_free_block_rsv(fs_info, rsv); | |
2575 | out: | |
2576 | return ret; | |
2577 | } | |
2578 | ||
05fd9564 | 2579 | static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len) |
2aaa6655 | 2580 | { |
05fd9564 | 2581 | struct inode *inode = file_inode(file); |
0b246afa | 2582 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2583 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2584 | struct extent_state *cached_state = NULL; | |
2585 | struct btrfs_path *path; | |
9cba40a6 | 2586 | struct btrfs_trans_handle *trans = NULL; |
d7781546 QW |
2587 | u64 lockstart; |
2588 | u64 lockend; | |
2589 | u64 tail_start; | |
2590 | u64 tail_len; | |
2591 | u64 orig_start = offset; | |
2aaa6655 | 2592 | int ret = 0; |
9703fefe | 2593 | bool same_block; |
a1a50f60 | 2594 | u64 ino_size; |
9703fefe | 2595 | bool truncated_block = false; |
e8c1c76e | 2596 | bool updated_inode = false; |
2aaa6655 | 2597 | |
29b6352b | 2598 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
bd6526d0 | 2599 | |
0ef8b726 JB |
2600 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2601 | if (ret) | |
bd6526d0 | 2602 | goto out_only_mutex; |
2aaa6655 | 2603 | |
0b246afa | 2604 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
dea46d84 | 2605 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2606 | if (ret < 0) |
2607 | goto out_only_mutex; | |
2608 | if (ret && !len) { | |
2609 | /* Already in a large hole */ | |
2610 | ret = 0; | |
2611 | goto out_only_mutex; | |
2612 | } | |
2613 | ||
05fd9564 DW |
2614 | ret = file_modified(file); |
2615 | if (ret) | |
2616 | goto out_only_mutex; | |
2617 | ||
ee8ba05c JB |
2618 | lockstart = round_up(offset, fs_info->sectorsize); |
2619 | lockend = round_down(offset + len, fs_info->sectorsize) - 1; | |
0b246afa JM |
2620 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2621 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2622 | /* |
9703fefe | 2623 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2624 | * because we are sure there is no data there. |
2625 | */ | |
2aaa6655 | 2626 | /* |
9703fefe CR |
2627 | * Only do this if we are in the same block and we aren't doing the |
2628 | * entire block. | |
2aaa6655 | 2629 | */ |
0b246afa | 2630 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2631 | if (offset < ino_size) { |
9703fefe | 2632 | truncated_block = true; |
217f42eb NB |
2633 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2634 | 0); | |
e8c1c76e FM |
2635 | } else { |
2636 | ret = 0; | |
2637 | } | |
d7781546 | 2638 | goto out_only_mutex; |
2aaa6655 JB |
2639 | } |
2640 | ||
9703fefe | 2641 | /* zero back part of the first block */ |
12870f1c | 2642 | if (offset < ino_size) { |
9703fefe | 2643 | truncated_block = true; |
217f42eb | 2644 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
7426cc04 | 2645 | if (ret) { |
e5d4d75b | 2646 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
7426cc04 MX |
2647 | return ret; |
2648 | } | |
2aaa6655 JB |
2649 | } |
2650 | ||
d7781546 QW |
2651 | /* Check the aligned pages after the first unaligned page, |
2652 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2653 | * including several following pages are already in holes, |
d7781546 QW |
2654 | * the extra check can be skipped */ |
2655 | if (offset == orig_start) { | |
2656 | /* after truncate page, check hole again */ | |
2657 | len = offset + len - lockstart; | |
2658 | offset = lockstart; | |
dea46d84 | 2659 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2660 | if (ret < 0) |
2661 | goto out_only_mutex; | |
2662 | if (ret && !len) { | |
2663 | ret = 0; | |
2664 | goto out_only_mutex; | |
2665 | } | |
2666 | lockstart = offset; | |
2667 | } | |
2668 | ||
2669 | /* Check the tail unaligned part is in a hole */ | |
2670 | tail_start = lockend + 1; | |
2671 | tail_len = offset + len - tail_start; | |
2672 | if (tail_len) { | |
dea46d84 | 2673 | ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len); |
d7781546 QW |
2674 | if (unlikely(ret < 0)) |
2675 | goto out_only_mutex; | |
2676 | if (!ret) { | |
2677 | /* zero the front end of the last page */ | |
2678 | if (tail_start + tail_len < ino_size) { | |
9703fefe | 2679 | truncated_block = true; |
217f42eb | 2680 | ret = btrfs_truncate_block(BTRFS_I(inode), |
9703fefe CR |
2681 | tail_start + tail_len, |
2682 | 0, 1); | |
d7781546 QW |
2683 | if (ret) |
2684 | goto out_only_mutex; | |
51f395ad | 2685 | } |
0061280d | 2686 | } |
2aaa6655 JB |
2687 | } |
2688 | ||
2689 | if (lockend < lockstart) { | |
e8c1c76e FM |
2690 | ret = 0; |
2691 | goto out_only_mutex; | |
2aaa6655 JB |
2692 | } |
2693 | ||
55961c8a | 2694 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state); |
2aaa6655 JB |
2695 | |
2696 | path = btrfs_alloc_path(); | |
2697 | if (!path) { | |
2698 | ret = -ENOMEM; | |
2699 | goto out; | |
2700 | } | |
2701 | ||
bfc78479 NB |
2702 | ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart, |
2703 | lockend, NULL, &trans); | |
9cba40a6 FM |
2704 | btrfs_free_path(path); |
2705 | if (ret) | |
2706 | goto out; | |
2aaa6655 | 2707 | |
9cba40a6 | 2708 | ASSERT(trans != NULL); |
e1f5790e | 2709 | inode_inc_iversion(inode); |
c1867eb3 DS |
2710 | inode->i_mtime = current_time(inode); |
2711 | inode->i_ctime = inode->i_mtime; | |
9a56fcd1 | 2712 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e8c1c76e | 2713 | updated_inode = true; |
3a45bb20 | 2714 | btrfs_end_transaction(trans); |
2ff7e61e | 2715 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2716 | out: |
570eb97b JB |
2717 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2718 | &cached_state); | |
d7781546 | 2719 | out_only_mutex: |
9cba40a6 | 2720 | if (!updated_inode && truncated_block && !ret) { |
e8c1c76e FM |
2721 | /* |
2722 | * If we only end up zeroing part of a page, we still need to | |
2723 | * update the inode item, so that all the time fields are | |
2724 | * updated as well as the necessary btrfs inode in memory fields | |
2725 | * for detecting, at fsync time, if the inode isn't yet in the | |
2726 | * log tree or it's there but not up to date. | |
2727 | */ | |
17900668 FM |
2728 | struct timespec64 now = current_time(inode); |
2729 | ||
2730 | inode_inc_iversion(inode); | |
2731 | inode->i_mtime = now; | |
2732 | inode->i_ctime = now; | |
e8c1c76e FM |
2733 | trans = btrfs_start_transaction(root, 1); |
2734 | if (IS_ERR(trans)) { | |
9cba40a6 | 2735 | ret = PTR_ERR(trans); |
e8c1c76e | 2736 | } else { |
9cba40a6 FM |
2737 | int ret2; |
2738 | ||
9a56fcd1 | 2739 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
9cba40a6 FM |
2740 | ret2 = btrfs_end_transaction(trans); |
2741 | if (!ret) | |
2742 | ret = ret2; | |
e8c1c76e FM |
2743 | } |
2744 | } | |
e5d4d75b | 2745 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
9cba40a6 | 2746 | return ret; |
2aaa6655 JB |
2747 | } |
2748 | ||
14524a84 QW |
2749 | /* Helper structure to record which range is already reserved */ |
2750 | struct falloc_range { | |
2751 | struct list_head list; | |
2752 | u64 start; | |
2753 | u64 len; | |
2754 | }; | |
2755 | ||
2756 | /* | |
2757 | * Helper function to add falloc range | |
2758 | * | |
2759 | * Caller should have locked the larger range of extent containing | |
2760 | * [start, len) | |
2761 | */ | |
2762 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2763 | { | |
14524a84 QW |
2764 | struct falloc_range *range = NULL; |
2765 | ||
77d25534 NB |
2766 | if (!list_empty(head)) { |
2767 | /* | |
2768 | * As fallocate iterates by bytenr order, we only need to check | |
2769 | * the last range. | |
2770 | */ | |
2771 | range = list_last_entry(head, struct falloc_range, list); | |
2772 | if (range->start + range->len == start) { | |
2773 | range->len += len; | |
2774 | return 0; | |
2775 | } | |
14524a84 | 2776 | } |
77d25534 | 2777 | |
32fc932e | 2778 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2779 | if (!range) |
2780 | return -ENOMEM; | |
2781 | range->start = start; | |
2782 | range->len = len; | |
2783 | list_add_tail(&range->list, head); | |
2784 | return 0; | |
2785 | } | |
2786 | ||
f27451f2 FM |
2787 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2788 | const u64 end, | |
2789 | const int mode) | |
2790 | { | |
2791 | struct btrfs_trans_handle *trans; | |
2792 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2793 | int ret; | |
2794 | int ret2; | |
2795 | ||
2796 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2797 | return 0; | |
2798 | ||
2799 | trans = btrfs_start_transaction(root, 1); | |
2800 | if (IS_ERR(trans)) | |
2801 | return PTR_ERR(trans); | |
2802 | ||
2803 | inode->i_ctime = current_time(inode); | |
2804 | i_size_write(inode, end); | |
76aea537 | 2805 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
9a56fcd1 | 2806 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
f27451f2 FM |
2807 | ret2 = btrfs_end_transaction(trans); |
2808 | ||
2809 | return ret ? ret : ret2; | |
2810 | } | |
2811 | ||
81fdf638 | 2812 | enum { |
f262fa8d DS |
2813 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
2814 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
2815 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
2816 | }; |
2817 | ||
948dfeb8 | 2818 | static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode, |
f27451f2 FM |
2819 | u64 offset) |
2820 | { | |
ee8ba05c | 2821 | const u64 sectorsize = inode->root->fs_info->sectorsize; |
f27451f2 | 2822 | struct extent_map *em; |
81fdf638 | 2823 | int ret; |
f27451f2 FM |
2824 | |
2825 | offset = round_down(offset, sectorsize); | |
948dfeb8 | 2826 | em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize); |
f27451f2 FM |
2827 | if (IS_ERR(em)) |
2828 | return PTR_ERR(em); | |
2829 | ||
2830 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
2831 | ret = RANGE_BOUNDARY_HOLE; |
2832 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2833 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
2834 | else | |
2835 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
2836 | |
2837 | free_extent_map(em); | |
2838 | return ret; | |
2839 | } | |
2840 | ||
2841 | static int btrfs_zero_range(struct inode *inode, | |
2842 | loff_t offset, | |
2843 | loff_t len, | |
2844 | const int mode) | |
2845 | { | |
2846 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2847 | struct extent_map *em; | |
2848 | struct extent_changeset *data_reserved = NULL; | |
2849 | int ret; | |
2850 | u64 alloc_hint = 0; | |
ee8ba05c | 2851 | const u64 sectorsize = fs_info->sectorsize; |
f27451f2 FM |
2852 | u64 alloc_start = round_down(offset, sectorsize); |
2853 | u64 alloc_end = round_up(offset + len, sectorsize); | |
2854 | u64 bytes_to_reserve = 0; | |
2855 | bool space_reserved = false; | |
2856 | ||
39b07b5d OS |
2857 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2858 | alloc_end - alloc_start); | |
f27451f2 FM |
2859 | if (IS_ERR(em)) { |
2860 | ret = PTR_ERR(em); | |
2861 | goto out; | |
2862 | } | |
2863 | ||
2864 | /* | |
2865 | * Avoid hole punching and extent allocation for some cases. More cases | |
2866 | * could be considered, but these are unlikely common and we keep things | |
2867 | * as simple as possible for now. Also, intentionally, if the target | |
2868 | * range contains one or more prealloc extents together with regular | |
2869 | * extents and holes, we drop all the existing extents and allocate a | |
2870 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
2871 | */ | |
2872 | if (em->start <= alloc_start && | |
2873 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2874 | const u64 em_end = em->start + em->len; | |
2875 | ||
2876 | if (em_end >= offset + len) { | |
2877 | /* | |
2878 | * The whole range is already a prealloc extent, | |
2879 | * do nothing except updating the inode's i_size if | |
2880 | * needed. | |
2881 | */ | |
2882 | free_extent_map(em); | |
2883 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2884 | mode); | |
2885 | goto out; | |
2886 | } | |
2887 | /* | |
2888 | * Part of the range is already a prealloc extent, so operate | |
2889 | * only on the remaining part of the range. | |
2890 | */ | |
2891 | alloc_start = em_end; | |
2892 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
2893 | len = offset + len - alloc_start; | |
2894 | offset = alloc_start; | |
2895 | alloc_hint = em->block_start + em->len; | |
2896 | } | |
2897 | free_extent_map(em); | |
2898 | ||
2899 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
2900 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
39b07b5d OS |
2901 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2902 | sectorsize); | |
f27451f2 FM |
2903 | if (IS_ERR(em)) { |
2904 | ret = PTR_ERR(em); | |
2905 | goto out; | |
2906 | } | |
2907 | ||
2908 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2909 | free_extent_map(em); | |
2910 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2911 | mode); | |
2912 | goto out; | |
2913 | } | |
2914 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
2915 | free_extent_map(em); | |
217f42eb NB |
2916 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2917 | 0); | |
f27451f2 FM |
2918 | if (!ret) |
2919 | ret = btrfs_fallocate_update_isize(inode, | |
2920 | offset + len, | |
2921 | mode); | |
2922 | return ret; | |
2923 | } | |
2924 | free_extent_map(em); | |
2925 | alloc_start = round_down(offset, sectorsize); | |
2926 | alloc_end = alloc_start + sectorsize; | |
2927 | goto reserve_space; | |
2928 | } | |
2929 | ||
2930 | alloc_start = round_up(offset, sectorsize); | |
2931 | alloc_end = round_down(offset + len, sectorsize); | |
2932 | ||
2933 | /* | |
2934 | * For unaligned ranges, check the pages at the boundaries, they might | |
2935 | * map to an extent, in which case we need to partially zero them, or | |
2936 | * they might map to a hole, in which case we need our allocation range | |
2937 | * to cover them. | |
2938 | */ | |
2939 | if (!IS_ALIGNED(offset, sectorsize)) { | |
948dfeb8 NB |
2940 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
2941 | offset); | |
f27451f2 FM |
2942 | if (ret < 0) |
2943 | goto out; | |
81fdf638 | 2944 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2945 | alloc_start = round_down(offset, sectorsize); |
2946 | ret = 0; | |
81fdf638 | 2947 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb | 2948 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
f27451f2 FM |
2949 | if (ret) |
2950 | goto out; | |
81fdf638 FM |
2951 | } else { |
2952 | ret = 0; | |
f27451f2 FM |
2953 | } |
2954 | } | |
2955 | ||
2956 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
948dfeb8 | 2957 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
f27451f2 FM |
2958 | offset + len); |
2959 | if (ret < 0) | |
2960 | goto out; | |
81fdf638 | 2961 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2962 | alloc_end = round_up(offset + len, sectorsize); |
2963 | ret = 0; | |
81fdf638 | 2964 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb NB |
2965 | ret = btrfs_truncate_block(BTRFS_I(inode), offset + len, |
2966 | 0, 1); | |
f27451f2 FM |
2967 | if (ret) |
2968 | goto out; | |
81fdf638 FM |
2969 | } else { |
2970 | ret = 0; | |
f27451f2 FM |
2971 | } |
2972 | } | |
2973 | ||
2974 | reserve_space: | |
2975 | if (alloc_start < alloc_end) { | |
2976 | struct extent_state *cached_state = NULL; | |
2977 | const u64 lockstart = alloc_start; | |
2978 | const u64 lockend = alloc_end - 1; | |
2979 | ||
2980 | bytes_to_reserve = alloc_end - alloc_start; | |
2981 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
2982 | bytes_to_reserve); | |
2983 | if (ret < 0) | |
2984 | goto out; | |
2985 | space_reserved = true; | |
55961c8a FM |
2986 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
2987 | &cached_state); | |
7661a3e0 | 2988 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved, |
a7f8b1c2 | 2989 | alloc_start, bytes_to_reserve); |
4f6a49de | 2990 | if (ret) { |
570eb97b JB |
2991 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
2992 | lockend, &cached_state); | |
a7f8b1c2 | 2993 | goto out; |
4f6a49de | 2994 | } |
f27451f2 FM |
2995 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, |
2996 | alloc_end - alloc_start, | |
2997 | i_blocksize(inode), | |
2998 | offset + len, &alloc_hint); | |
570eb97b JB |
2999 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3000 | &cached_state); | |
f27451f2 | 3001 | /* btrfs_prealloc_file_range releases reserved space on error */ |
9f13ce74 | 3002 | if (ret) { |
f27451f2 | 3003 | space_reserved = false; |
9f13ce74 FM |
3004 | goto out; |
3005 | } | |
f27451f2 | 3006 | } |
9f13ce74 | 3007 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3008 | out: |
3009 | if (ret && space_reserved) | |
25ce28ca | 3010 | btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved, |
f27451f2 FM |
3011 | alloc_start, bytes_to_reserve); |
3012 | extent_changeset_free(data_reserved); | |
3013 | ||
3014 | return ret; | |
3015 | } | |
3016 | ||
2fe17c10 CH |
3017 | static long btrfs_fallocate(struct file *file, int mode, |
3018 | loff_t offset, loff_t len) | |
3019 | { | |
496ad9aa | 3020 | struct inode *inode = file_inode(file); |
2fe17c10 | 3021 | struct extent_state *cached_state = NULL; |
364ecf36 | 3022 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3023 | struct falloc_range *range; |
3024 | struct falloc_range *tmp; | |
3025 | struct list_head reserve_list; | |
2fe17c10 CH |
3026 | u64 cur_offset; |
3027 | u64 last_byte; | |
3028 | u64 alloc_start; | |
3029 | u64 alloc_end; | |
3030 | u64 alloc_hint = 0; | |
3031 | u64 locked_end; | |
14524a84 | 3032 | u64 actual_end = 0; |
47e1d1c7 FM |
3033 | u64 data_space_needed = 0; |
3034 | u64 data_space_reserved = 0; | |
3035 | u64 qgroup_reserved = 0; | |
2fe17c10 | 3036 | struct extent_map *em; |
ee8ba05c | 3037 | int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize; |
2fe17c10 CH |
3038 | int ret; |
3039 | ||
f1569c4c NA |
3040 | /* Do not allow fallocate in ZONED mode */ |
3041 | if (btrfs_is_zoned(btrfs_sb(inode->i_sb))) | |
3042 | return -EOPNOTSUPP; | |
3043 | ||
797f4277 MX |
3044 | alloc_start = round_down(offset, blocksize); |
3045 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3046 | cur_offset = alloc_start; |
2fe17c10 | 3047 | |
2aaa6655 | 3048 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3049 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3050 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3051 | return -EOPNOTSUPP; |
3052 | ||
2aaa6655 | 3053 | if (mode & FALLOC_FL_PUNCH_HOLE) |
05fd9564 | 3054 | return btrfs_punch_hole(file, offset, len); |
2aaa6655 | 3055 | |
29b6352b | 3056 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
2a162ce9 DI |
3057 | |
3058 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3059 | ret = inode_newsize_ok(inode, offset + len); | |
3060 | if (ret) | |
3061 | goto out; | |
3062 | } | |
2fe17c10 | 3063 | |
05fd9564 DW |
3064 | ret = file_modified(file); |
3065 | if (ret) | |
3066 | goto out; | |
3067 | ||
14524a84 QW |
3068 | /* |
3069 | * TODO: Move these two operations after we have checked | |
3070 | * accurate reserved space, or fallocate can still fail but | |
3071 | * with page truncated or size expanded. | |
3072 | * | |
3073 | * But that's a minor problem and won't do much harm BTW. | |
3074 | */ | |
2fe17c10 | 3075 | if (alloc_start > inode->i_size) { |
b06359a3 | 3076 | ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode), |
a41ad394 | 3077 | alloc_start); |
2fe17c10 CH |
3078 | if (ret) |
3079 | goto out; | |
0f6925fa | 3080 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3081 | /* |
3082 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3083 | * need to zero out the end of the block if i_size lands in the |
3084 | * middle of a block. | |
a71754fc | 3085 | */ |
217f42eb | 3086 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
a71754fc JB |
3087 | if (ret) |
3088 | goto out; | |
2fe17c10 CH |
3089 | } |
3090 | ||
a71754fc | 3091 | /* |
ffa8fc60 FM |
3092 | * We have locked the inode at the VFS level (in exclusive mode) and we |
3093 | * have locked the i_mmap_lock lock (in exclusive mode). Now before | |
3094 | * locking the file range, flush all dealloc in the range and wait for | |
3095 | * all ordered extents in the range to complete. After this we can lock | |
3096 | * the file range and, due to the previous locking we did, we know there | |
3097 | * can't be more delalloc or ordered extents in the range. | |
a71754fc | 3098 | */ |
0ef8b726 JB |
3099 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3100 | alloc_end - alloc_start); | |
3101 | if (ret) | |
3102 | goto out; | |
a71754fc | 3103 | |
f27451f2 FM |
3104 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3105 | ret = btrfs_zero_range(inode, offset, len, mode); | |
e5d4d75b | 3106 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
f27451f2 FM |
3107 | return ret; |
3108 | } | |
3109 | ||
2fe17c10 | 3110 | locked_end = alloc_end - 1; |
570eb97b JB |
3111 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3112 | &cached_state); | |
2fe17c10 | 3113 | |
63c34cb4 FM |
3114 | btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end); |
3115 | ||
14524a84 QW |
3116 | /* First, check if we exceed the qgroup limit */ |
3117 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 3118 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3119 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
39b07b5d | 3120 | alloc_end - cur_offset); |
9986277e DC |
3121 | if (IS_ERR(em)) { |
3122 | ret = PTR_ERR(em); | |
79787eaa JM |
3123 | break; |
3124 | } | |
2fe17c10 | 3125 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3126 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3127 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3128 | if (em->block_start == EXTENT_MAP_HOLE || |
3129 | (cur_offset >= inode->i_size && | |
3130 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
47e1d1c7 FM |
3131 | const u64 range_len = last_byte - cur_offset; |
3132 | ||
3133 | ret = add_falloc_range(&reserve_list, cur_offset, range_len); | |
14524a84 QW |
3134 | if (ret < 0) { |
3135 | free_extent_map(em); | |
3136 | break; | |
3d850dd4 | 3137 | } |
7661a3e0 | 3138 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), |
47e1d1c7 | 3139 | &data_reserved, cur_offset, range_len); |
be2d253c FM |
3140 | if (ret < 0) { |
3141 | free_extent_map(em); | |
14524a84 | 3142 | break; |
be2d253c | 3143 | } |
47e1d1c7 FM |
3144 | qgroup_reserved += range_len; |
3145 | data_space_needed += range_len; | |
2fe17c10 CH |
3146 | } |
3147 | free_extent_map(em); | |
2fe17c10 | 3148 | cur_offset = last_byte; |
14524a84 QW |
3149 | } |
3150 | ||
47e1d1c7 FM |
3151 | if (!ret && data_space_needed > 0) { |
3152 | /* | |
3153 | * We are safe to reserve space here as we can't have delalloc | |
3154 | * in the range, see above. | |
3155 | */ | |
3156 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3157 | data_space_needed); | |
3158 | if (!ret) | |
3159 | data_space_reserved = data_space_needed; | |
3160 | } | |
3161 | ||
14524a84 QW |
3162 | /* |
3163 | * If ret is still 0, means we're OK to fallocate. | |
3164 | * Or just cleanup the list and exit. | |
3165 | */ | |
3166 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
47e1d1c7 | 3167 | if (!ret) { |
14524a84 QW |
3168 | ret = btrfs_prealloc_file_range(inode, mode, |
3169 | range->start, | |
93407472 | 3170 | range->len, i_blocksize(inode), |
14524a84 | 3171 | offset + len, &alloc_hint); |
47e1d1c7 FM |
3172 | /* |
3173 | * btrfs_prealloc_file_range() releases space even | |
3174 | * if it returns an error. | |
3175 | */ | |
3176 | data_space_reserved -= range->len; | |
3177 | qgroup_reserved -= range->len; | |
3178 | } else if (data_space_reserved > 0) { | |
25ce28ca | 3179 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
47e1d1c7 FM |
3180 | data_reserved, range->start, |
3181 | range->len); | |
3182 | data_space_reserved -= range->len; | |
3183 | qgroup_reserved -= range->len; | |
3184 | } else if (qgroup_reserved > 0) { | |
3185 | btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved, | |
3186 | range->start, range->len); | |
3187 | qgroup_reserved -= range->len; | |
3188 | } | |
14524a84 QW |
3189 | list_del(&range->list); |
3190 | kfree(range); | |
3191 | } | |
3192 | if (ret < 0) | |
3193 | goto out_unlock; | |
3194 | ||
f27451f2 FM |
3195 | /* |
3196 | * We didn't need to allocate any more space, but we still extended the | |
3197 | * size of the file so we need to update i_size and the inode item. | |
3198 | */ | |
3199 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3200 | out_unlock: |
570eb97b JB |
3201 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3202 | &cached_state); | |
2fe17c10 | 3203 | out: |
e5d4d75b | 3204 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
364ecf36 | 3205 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3206 | return ret; |
3207 | } | |
3208 | ||
b6e83356 | 3209 | /* |
ac3c0d36 FM |
3210 | * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range |
3211 | * that has unflushed and/or flushing delalloc. There might be other adjacent | |
3212 | * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps | |
3213 | * looping while it gets adjacent subranges, and merging them together. | |
b6e83356 FM |
3214 | */ |
3215 | static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end, | |
b3e744fe | 3216 | struct extent_state **cached_state, |
af979fd6 | 3217 | bool *search_io_tree, |
b6e83356 FM |
3218 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
3219 | { | |
40daf3e0 | 3220 | u64 len = end + 1 - start; |
8ddc8274 FM |
3221 | u64 delalloc_len = 0; |
3222 | struct btrfs_ordered_extent *oe; | |
3223 | u64 oe_start; | |
3224 | u64 oe_end; | |
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 | */ | |
8ddc8274 FM |
3231 | if (*search_io_tree) { |
3232 | spin_lock(&inode->lock); | |
3233 | if (inode->delalloc_bytes > 0) { | |
3234 | spin_unlock(&inode->lock); | |
3235 | *delalloc_start_ret = start; | |
3236 | delalloc_len = count_range_bits(&inode->io_tree, | |
3237 | delalloc_start_ret, end, | |
8c6e53a7 | 3238 | len, EXTENT_DELALLOC, 1, |
b3e744fe | 3239 | cached_state); |
8ddc8274 FM |
3240 | } else { |
3241 | spin_unlock(&inode->lock); | |
3242 | } | |
a2853ffc FM |
3243 | } |
3244 | ||
40daf3e0 FM |
3245 | if (delalloc_len > 0) { |
3246 | /* | |
3247 | * If delalloc was found then *delalloc_start_ret has a sector size | |
3248 | * aligned value (rounded down). | |
3249 | */ | |
b6e83356 FM |
3250 | *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1; |
3251 | ||
40daf3e0 FM |
3252 | if (*delalloc_start_ret == start) { |
3253 | /* Delalloc for the whole range, nothing more to do. */ | |
3254 | if (*delalloc_end_ret == end) | |
3255 | return true; | |
8ddc8274 | 3256 | /* Else trim our search range for ordered extents. */ |
40daf3e0 FM |
3257 | start = *delalloc_end_ret + 1; |
3258 | len = end + 1 - start; | |
3259 | } | |
af979fd6 FM |
3260 | } else { |
3261 | /* No delalloc, future calls don't need to search again. */ | |
3262 | *search_io_tree = false; | |
40daf3e0 FM |
3263 | } |
3264 | ||
a2853ffc | 3265 | /* |
8ddc8274 FM |
3266 | * Now also check if there's any ordered extent in the range. |
3267 | * We do this because: | |
b6e83356 FM |
3268 | * |
3269 | * 1) When delalloc is flushed, the file range is locked, we clear the | |
8ddc8274 FM |
3270 | * EXTENT_DELALLOC bit from the io tree and create an extent map and |
3271 | * an ordered extent for the write. So we might just have been called | |
3272 | * after delalloc is flushed and before the ordered extent completes | |
3273 | * and inserts the new file extent item in the subvolume's btree; | |
b6e83356 | 3274 | * |
8ddc8274 | 3275 | * 2) We may have an ordered extent created by flushing delalloc for a |
b6e83356 FM |
3276 | * subrange that starts before the subrange we found marked with |
3277 | * EXTENT_DELALLOC in the io tree. | |
8ddc8274 FM |
3278 | * |
3279 | * We could also use the extent map tree to find such delalloc that is | |
3280 | * being flushed, but using the ordered extents tree is more efficient | |
3281 | * because it's usually much smaller as ordered extents are removed from | |
3282 | * the tree once they complete. With the extent maps, we mau have them | |
3283 | * in the extent map tree for a very long time, and they were either | |
3284 | * created by previous writes or loaded by read operations. | |
b6e83356 | 3285 | */ |
8ddc8274 FM |
3286 | oe = btrfs_lookup_first_ordered_range(inode, start, len); |
3287 | if (!oe) | |
d47704bd | 3288 | return (delalloc_len > 0); |
d47704bd | 3289 | |
8ddc8274 FM |
3290 | /* The ordered extent may span beyond our search range. */ |
3291 | oe_start = max(oe->file_offset, start); | |
3292 | oe_end = min(oe->file_offset + oe->num_bytes - 1, end); | |
b6e83356 | 3293 | |
8ddc8274 | 3294 | btrfs_put_ordered_extent(oe); |
b6e83356 | 3295 | |
8ddc8274 | 3296 | /* Don't have unflushed delalloc, return the ordered extent range. */ |
b6e83356 | 3297 | if (delalloc_len == 0) { |
8ddc8274 FM |
3298 | *delalloc_start_ret = oe_start; |
3299 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3300 | return true; |
3301 | } | |
3302 | ||
3303 | /* | |
8ddc8274 FM |
3304 | * We have both unflushed delalloc (io_tree) and an ordered extent. |
3305 | * If the ranges are adjacent returned a combined range, otherwise | |
3306 | * return the leftmost range. | |
b6e83356 | 3307 | */ |
8ddc8274 FM |
3308 | if (oe_start < *delalloc_start_ret) { |
3309 | if (oe_end < *delalloc_start_ret) | |
3310 | *delalloc_end_ret = oe_end; | |
3311 | *delalloc_start_ret = oe_start; | |
3312 | } else if (*delalloc_end_ret + 1 == oe_start) { | |
3313 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3314 | } |
3315 | ||
b6e83356 FM |
3316 | return true; |
3317 | } | |
3318 | ||
3319 | /* | |
3320 | * Check if there's delalloc in a given range. | |
3321 | * | |
3322 | * @inode: The inode. | |
3323 | * @start: The start offset of the range. It does not need to be | |
3324 | * sector size aligned. | |
3325 | * @end: The end offset (inclusive value) of the search range. | |
3326 | * It does not need to be sector size aligned. | |
b3e744fe FM |
3327 | * @cached_state: Extent state record used for speeding up delalloc |
3328 | * searches in the inode's io_tree. Can be NULL. | |
b6e83356 FM |
3329 | * @delalloc_start_ret: Output argument, set to the start offset of the |
3330 | * subrange found with delalloc (may not be sector size | |
3331 | * aligned). | |
3332 | * @delalloc_end_ret: Output argument, set to he end offset (inclusive value) | |
3333 | * of the subrange found with delalloc. | |
3334 | * | |
3335 | * Returns true if a subrange with delalloc is found within the given range, and | |
3336 | * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and | |
3337 | * end offsets of the subrange. | |
3338 | */ | |
ac3c0d36 | 3339 | bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end, |
b3e744fe | 3340 | struct extent_state **cached_state, |
ac3c0d36 | 3341 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
b6e83356 FM |
3342 | { |
3343 | u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize); | |
3344 | u64 prev_delalloc_end = 0; | |
af979fd6 | 3345 | bool search_io_tree = true; |
b6e83356 FM |
3346 | bool ret = false; |
3347 | ||
2f2e84ca | 3348 | while (cur_offset <= end) { |
b6e83356 FM |
3349 | u64 delalloc_start; |
3350 | u64 delalloc_end; | |
3351 | bool delalloc; | |
3352 | ||
3353 | delalloc = find_delalloc_subrange(inode, cur_offset, end, | |
b3e744fe | 3354 | cached_state, &search_io_tree, |
b6e83356 FM |
3355 | &delalloc_start, |
3356 | &delalloc_end); | |
3357 | if (!delalloc) | |
3358 | break; | |
3359 | ||
3360 | if (prev_delalloc_end == 0) { | |
3361 | /* First subrange found. */ | |
3362 | *delalloc_start_ret = max(delalloc_start, start); | |
3363 | *delalloc_end_ret = delalloc_end; | |
3364 | ret = true; | |
3365 | } else if (delalloc_start == prev_delalloc_end + 1) { | |
3366 | /* Subrange adjacent to the previous one, merge them. */ | |
3367 | *delalloc_end_ret = delalloc_end; | |
3368 | } else { | |
3369 | /* Subrange not adjacent to the previous one, exit. */ | |
3370 | break; | |
3371 | } | |
3372 | ||
3373 | prev_delalloc_end = delalloc_end; | |
3374 | cur_offset = delalloc_end + 1; | |
3375 | cond_resched(); | |
3376 | } | |
3377 | ||
3378 | return ret; | |
3379 | } | |
3380 | ||
3381 | /* | |
3382 | * Check if there's a hole or delalloc range in a range representing a hole (or | |
3383 | * prealloc extent) found in the inode's subvolume btree. | |
3384 | * | |
3385 | * @inode: The inode. | |
3386 | * @whence: Seek mode (SEEK_DATA or SEEK_HOLE). | |
3387 | * @start: Start offset of the hole region. It does not need to be sector | |
3388 | * size aligned. | |
3389 | * @end: End offset (inclusive value) of the hole region. It does not | |
3390 | * need to be sector size aligned. | |
3391 | * @start_ret: Return parameter, used to set the start of the subrange in the | |
3392 | * hole that matches the search criteria (seek mode), if such | |
3393 | * subrange is found (return value of the function is true). | |
3394 | * The value returned here may not be sector size aligned. | |
3395 | * | |
3396 | * Returns true if a subrange matching the given seek mode is found, and if one | |
3397 | * is found, it updates @start_ret with the start of the subrange. | |
3398 | */ | |
3399 | static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence, | |
3c32c721 | 3400 | struct extent_state **cached_state, |
b6e83356 FM |
3401 | u64 start, u64 end, u64 *start_ret) |
3402 | { | |
3403 | u64 delalloc_start; | |
3404 | u64 delalloc_end; | |
3405 | bool delalloc; | |
3406 | ||
3c32c721 | 3407 | delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state, |
ac3c0d36 | 3408 | &delalloc_start, &delalloc_end); |
b6e83356 FM |
3409 | if (delalloc && whence == SEEK_DATA) { |
3410 | *start_ret = delalloc_start; | |
3411 | return true; | |
3412 | } | |
3413 | ||
3414 | if (delalloc && whence == SEEK_HOLE) { | |
3415 | /* | |
3416 | * We found delalloc but it starts after out start offset. So we | |
3417 | * have a hole between our start offset and the delalloc start. | |
3418 | */ | |
3419 | if (start < delalloc_start) { | |
3420 | *start_ret = start; | |
3421 | return true; | |
3422 | } | |
3423 | /* | |
3424 | * Delalloc range starts at our start offset. | |
3425 | * If the delalloc range's length is smaller than our range, | |
3426 | * then it means we have a hole that starts where the delalloc | |
3427 | * subrange ends. | |
3428 | */ | |
3429 | if (delalloc_end < end) { | |
3430 | *start_ret = delalloc_end + 1; | |
3431 | return true; | |
3432 | } | |
3433 | ||
3434 | /* There's delalloc for the whole range. */ | |
3435 | return false; | |
3436 | } | |
3437 | ||
3438 | if (!delalloc && whence == SEEK_HOLE) { | |
3439 | *start_ret = start; | |
3440 | return true; | |
3441 | } | |
3442 | ||
3443 | /* | |
3444 | * No delalloc in the range and we are seeking for data. The caller has | |
3445 | * to iterate to the next extent item in the subvolume btree. | |
3446 | */ | |
3447 | return false; | |
3448 | } | |
3449 | ||
3c32c721 | 3450 | static loff_t find_desired_extent(struct file *file, loff_t offset, int whence) |
b2675157 | 3451 | { |
3c32c721 FM |
3452 | struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host); |
3453 | struct btrfs_file_private *private = file->private_data; | |
cca5de97 | 3454 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
b2675157 | 3455 | struct extent_state *cached_state = NULL; |
3c32c721 | 3456 | struct extent_state **delalloc_cached_state; |
b6e83356 FM |
3457 | const loff_t i_size = i_size_read(&inode->vfs_inode); |
3458 | const u64 ino = btrfs_ino(inode); | |
3459 | struct btrfs_root *root = inode->root; | |
3460 | struct btrfs_path *path; | |
3461 | struct btrfs_key key; | |
3462 | u64 last_extent_end; | |
4d1a40c6 LB |
3463 | u64 lockstart; |
3464 | u64 lockend; | |
3465 | u64 start; | |
b6e83356 FM |
3466 | int ret; |
3467 | bool found = false; | |
b2675157 | 3468 | |
bc80230e | 3469 | if (i_size == 0 || offset >= i_size) |
4d1a40c6 LB |
3470 | return -ENXIO; |
3471 | ||
b6e83356 FM |
3472 | /* |
3473 | * Quick path. If the inode has no prealloc extents and its number of | |
3474 | * bytes used matches its i_size, then it can not have holes. | |
3475 | */ | |
3476 | if (whence == SEEK_HOLE && | |
3477 | !(inode->flags & BTRFS_INODE_PREALLOC) && | |
3478 | inode_get_bytes(&inode->vfs_inode) == i_size) | |
3479 | return i_size; | |
3480 | ||
3c32c721 FM |
3481 | if (!private) { |
3482 | private = kzalloc(sizeof(*private), GFP_KERNEL); | |
3483 | /* | |
3484 | * No worries if memory allocation failed. | |
3485 | * The private structure is used only for speeding up multiple | |
3486 | * lseek SEEK_HOLE/DATA calls to a file when there's delalloc, | |
3487 | * so everything will still be correct. | |
3488 | */ | |
3489 | file->private_data = private; | |
3490 | } | |
3491 | ||
3492 | if (private) | |
3493 | delalloc_cached_state = &private->llseek_cached_state; | |
3494 | else | |
3495 | delalloc_cached_state = NULL; | |
3496 | ||
4d1a40c6 | 3497 | /* |
bc80230e | 3498 | * offset can be negative, in this case we start finding DATA/HOLE from |
4d1a40c6 LB |
3499 | * the very start of the file. |
3500 | */ | |
bc80230e | 3501 | start = max_t(loff_t, 0, offset); |
4d1a40c6 | 3502 | |
0b246afa | 3503 | lockstart = round_down(start, fs_info->sectorsize); |
d79b7c26 | 3504 | lockend = round_up(i_size, fs_info->sectorsize); |
b2675157 | 3505 | if (lockend <= lockstart) |
0b246afa | 3506 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3507 | lockend--; |
b6e83356 FM |
3508 | |
3509 | path = btrfs_alloc_path(); | |
3510 | if (!path) | |
3511 | return -ENOMEM; | |
3512 | path->reada = READA_FORWARD; | |
3513 | ||
3514 | key.objectid = ino; | |
3515 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3516 | key.offset = start; | |
3517 | ||
3518 | last_extent_end = lockstart; | |
b2675157 | 3519 | |
570eb97b | 3520 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b2675157 | 3521 | |
b6e83356 FM |
3522 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3523 | if (ret < 0) { | |
3524 | goto out; | |
3525 | } else if (ret > 0 && path->slots[0] > 0) { | |
3526 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3527 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3528 | path->slots[0]--; | |
3529 | } | |
3530 | ||
d79b7c26 | 3531 | while (start < i_size) { |
b6e83356 FM |
3532 | struct extent_buffer *leaf = path->nodes[0]; |
3533 | struct btrfs_file_extent_item *extent; | |
3534 | u64 extent_end; | |
1f55ee6d | 3535 | u8 type; |
b6e83356 FM |
3536 | |
3537 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3538 | ret = btrfs_next_leaf(root, path); | |
3539 | if (ret < 0) | |
3540 | goto out; | |
3541 | else if (ret > 0) | |
3542 | break; | |
3543 | ||
3544 | leaf = path->nodes[0]; | |
b2675157 JB |
3545 | } |
3546 | ||
b6e83356 FM |
3547 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3548 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
7f4ca37c | 3549 | break; |
b2675157 | 3550 | |
b6e83356 FM |
3551 | extent_end = btrfs_file_extent_end(path); |
3552 | ||
3553 | /* | |
3554 | * In the first iteration we may have a slot that points to an | |
3555 | * extent that ends before our start offset, so skip it. | |
3556 | */ | |
3557 | if (extent_end <= start) { | |
3558 | path->slots[0]++; | |
3559 | continue; | |
3560 | } | |
3561 | ||
3562 | /* We have an implicit hole, NO_HOLES feature is likely set. */ | |
3563 | if (last_extent_end < key.offset) { | |
3564 | u64 search_start = last_extent_end; | |
3565 | u64 found_start; | |
3566 | ||
3567 | /* | |
3568 | * First iteration, @start matches @offset and it's | |
3569 | * within the hole. | |
3570 | */ | |
3571 | if (start == offset) | |
3572 | search_start = offset; | |
3573 | ||
3574 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3575 | delalloc_cached_state, |
b6e83356 FM |
3576 | search_start, |
3577 | key.offset - 1, | |
3578 | &found_start); | |
3579 | if (found) { | |
3580 | start = found_start; | |
3581 | break; | |
3582 | } | |
3583 | /* | |
3584 | * Didn't find data or a hole (due to delalloc) in the | |
3585 | * implicit hole range, so need to analyze the extent. | |
3586 | */ | |
3587 | } | |
3588 | ||
3589 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
3590 | struct btrfs_file_extent_item); | |
1f55ee6d | 3591 | type = btrfs_file_extent_type(leaf, extent); |
b6e83356 | 3592 | |
1f55ee6d FM |
3593 | /* |
3594 | * Can't access the extent's disk_bytenr field if this is an | |
3595 | * inline extent, since at that offset, it's where the extent | |
3596 | * data starts. | |
3597 | */ | |
3598 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
3599 | (type == BTRFS_FILE_EXTENT_REG && | |
3600 | btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) { | |
b6e83356 FM |
3601 | /* |
3602 | * Explicit hole or prealloc extent, search for delalloc. | |
3603 | * A prealloc extent is treated like a hole. | |
3604 | */ | |
3605 | u64 search_start = key.offset; | |
3606 | u64 found_start; | |
3607 | ||
3608 | /* | |
3609 | * First iteration, @start matches @offset and it's | |
3610 | * within the hole. | |
3611 | */ | |
3612 | if (start == offset) | |
3613 | search_start = offset; | |
3614 | ||
3615 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3616 | delalloc_cached_state, |
b6e83356 FM |
3617 | search_start, |
3618 | extent_end - 1, | |
3619 | &found_start); | |
3620 | if (found) { | |
3621 | start = found_start; | |
3622 | break; | |
3623 | } | |
3624 | /* | |
3625 | * Didn't find data or a hole (due to delalloc) in the | |
3626 | * implicit hole range, so need to analyze the next | |
3627 | * extent item. | |
3628 | */ | |
3629 | } else { | |
3630 | /* | |
3631 | * Found a regular or inline extent. | |
3632 | * If we are seeking for data, adjust the start offset | |
3633 | * and stop, we're done. | |
3634 | */ | |
3635 | if (whence == SEEK_DATA) { | |
3636 | start = max_t(u64, key.offset, offset); | |
3637 | found = true; | |
3638 | break; | |
3639 | } | |
3640 | /* | |
3641 | * Else, we are seeking for a hole, check the next file | |
3642 | * extent item. | |
3643 | */ | |
3644 | } | |
3645 | ||
3646 | start = extent_end; | |
3647 | last_extent_end = extent_end; | |
3648 | path->slots[0]++; | |
aed0ca18 FM |
3649 | if (fatal_signal_pending(current)) { |
3650 | ret = -EINTR; | |
b6e83356 | 3651 | goto out; |
aed0ca18 | 3652 | } |
b2675157 JB |
3653 | cond_resched(); |
3654 | } | |
b6e83356 FM |
3655 | |
3656 | /* We have an implicit hole from the last extent found up to i_size. */ | |
3657 | if (!found && start < i_size) { | |
3c32c721 FM |
3658 | found = find_desired_extent_in_hole(inode, whence, |
3659 | delalloc_cached_state, start, | |
b6e83356 FM |
3660 | i_size - 1, &start); |
3661 | if (!found) | |
3662 | start = i_size; | |
3663 | } | |
3664 | ||
3665 | out: | |
570eb97b | 3666 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b6e83356 FM |
3667 | btrfs_free_path(path); |
3668 | ||
3669 | if (ret < 0) | |
3670 | return ret; | |
3671 | ||
3672 | if (whence == SEEK_DATA && start >= i_size) | |
3673 | return -ENXIO; | |
bc80230e | 3674 | |
b6e83356 | 3675 | return min_t(loff_t, start, i_size); |
b2675157 JB |
3676 | } |
3677 | ||
965c8e59 | 3678 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3679 | { |
3680 | struct inode *inode = file->f_mapping->host; | |
b2675157 | 3681 | |
965c8e59 | 3682 | switch (whence) { |
2034f3b4 NB |
3683 | default: |
3684 | return generic_file_llseek(file, offset, whence); | |
b2675157 JB |
3685 | case SEEK_DATA: |
3686 | case SEEK_HOLE: | |
29b6352b | 3687 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
3c32c721 | 3688 | offset = find_desired_extent(file, offset, whence); |
e5d4d75b | 3689 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
bc80230e | 3690 | break; |
b2675157 JB |
3691 | } |
3692 | ||
bc80230e NB |
3693 | if (offset < 0) |
3694 | return offset; | |
3695 | ||
2034f3b4 | 3696 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
3697 | } |
3698 | ||
edf064e7 GR |
3699 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3700 | { | |
14605409 BB |
3701 | int ret; |
3702 | ||
f02c75e6 CH |
3703 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC | |
3704 | FMODE_CAN_ODIRECT; | |
14605409 BB |
3705 | |
3706 | ret = fsverity_file_open(inode, filp); | |
3707 | if (ret) | |
3708 | return ret; | |
edf064e7 GR |
3709 | return generic_file_open(inode, filp); |
3710 | } | |
3711 | ||
4e4cabec GR |
3712 | static int check_direct_read(struct btrfs_fs_info *fs_info, |
3713 | const struct iov_iter *iter, loff_t offset) | |
3714 | { | |
3715 | int ret; | |
3716 | int i, seg; | |
3717 | ||
3718 | ret = check_direct_IO(fs_info, iter, offset); | |
3719 | if (ret < 0) | |
3720 | return ret; | |
3721 | ||
3722 | if (!iter_is_iovec(iter)) | |
3723 | return 0; | |
3724 | ||
de4f5fed JA |
3725 | for (seg = 0; seg < iter->nr_segs; seg++) { |
3726 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
3727 | const struct iovec *iov1 = iter_iov(iter) + seg; | |
3728 | const struct iovec *iov2 = iter_iov(iter) + i; | |
3729 | ||
3730 | if (iov1->iov_base == iov2->iov_base) | |
4e4cabec | 3731 | return -EINVAL; |
de4f5fed JA |
3732 | } |
3733 | } | |
4e4cabec GR |
3734 | return 0; |
3735 | } | |
3736 | ||
3737 | static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to) | |
3738 | { | |
3739 | struct inode *inode = file_inode(iocb->ki_filp); | |
51bd9563 FM |
3740 | size_t prev_left = 0; |
3741 | ssize_t read = 0; | |
4e4cabec GR |
3742 | ssize_t ret; |
3743 | ||
14605409 BB |
3744 | if (fsverity_active(inode)) |
3745 | return 0; | |
3746 | ||
4e4cabec GR |
3747 | if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos)) |
3748 | return 0; | |
3749 | ||
29b6352b | 3750 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 FM |
3751 | again: |
3752 | /* | |
3753 | * This is similar to what we do for direct IO writes, see the comment | |
3754 | * at btrfs_direct_write(), but we also disable page faults in addition | |
3755 | * to disabling them only at the iov_iter level. This is because when | |
3756 | * reading from a hole or prealloc extent, iomap calls iov_iter_zero(), | |
3757 | * which can still trigger page fault ins despite having set ->nofault | |
3758 | * to true of our 'to' iov_iter. | |
3759 | * | |
3760 | * The difference to direct IO writes is that we deadlock when trying | |
3761 | * to lock the extent range in the inode's tree during he page reads | |
3762 | * triggered by the fault in (while for writes it is due to waiting for | |
3763 | * our own ordered extent). This is because for direct IO reads, | |
3764 | * btrfs_dio_iomap_begin() returns with the extent range locked, which | |
3765 | * is only unlocked in the endio callback (end_bio_extent_readpage()). | |
3766 | */ | |
3767 | pagefault_disable(); | |
3768 | to->nofault = true; | |
8184620a | 3769 | ret = btrfs_dio_read(iocb, to, read); |
51bd9563 FM |
3770 | to->nofault = false; |
3771 | pagefault_enable(); | |
3772 | ||
3773 | /* No increment (+=) because iomap returns a cumulative value. */ | |
3774 | if (ret > 0) | |
3775 | read = ret; | |
3776 | ||
3777 | if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) { | |
3778 | const size_t left = iov_iter_count(to); | |
3779 | ||
3780 | if (left == prev_left) { | |
3781 | /* | |
3782 | * We didn't make any progress since the last attempt, | |
3783 | * fallback to a buffered read for the remainder of the | |
3784 | * range. This is just to avoid any possibility of looping | |
3785 | * for too long. | |
3786 | */ | |
3787 | ret = read; | |
3788 | } else { | |
3789 | /* | |
3790 | * We made some progress since the last retry or this is | |
3791 | * the first time we are retrying. Fault in as many pages | |
3792 | * as possible and retry. | |
3793 | */ | |
3794 | fault_in_iov_iter_writeable(to, left); | |
3795 | prev_left = left; | |
3796 | goto again; | |
3797 | } | |
3798 | } | |
e5d4d75b | 3799 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 | 3800 | return ret < 0 ? ret : read; |
4e4cabec GR |
3801 | } |
3802 | ||
f85781fb GR |
3803 | static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
3804 | { | |
3805 | ssize_t ret = 0; | |
3806 | ||
3807 | if (iocb->ki_flags & IOCB_DIRECT) { | |
4e4cabec | 3808 | ret = btrfs_direct_read(iocb, to); |
0425e7ba JT |
3809 | if (ret < 0 || !iov_iter_count(to) || |
3810 | iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp))) | |
f85781fb GR |
3811 | return ret; |
3812 | } | |
3813 | ||
87fa0f3e | 3814 | return filemap_read(iocb, to, ret); |
f85781fb GR |
3815 | } |
3816 | ||
828c0950 | 3817 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3818 | .llseek = btrfs_file_llseek, |
f85781fb | 3819 | .read_iter = btrfs_file_read_iter, |
2cb1e089 | 3820 | .splice_read = filemap_splice_read, |
b30ac0fc | 3821 | .write_iter = btrfs_file_write_iter, |
d7776591 | 3822 | .splice_write = iter_file_splice_write, |
9ebefb18 | 3823 | .mmap = btrfs_file_mmap, |
edf064e7 | 3824 | .open = btrfs_file_open, |
e1b81e67 | 3825 | .release = btrfs_release_file, |
b0c58223 | 3826 | .get_unmapped_area = thp_get_unmapped_area, |
39279cc3 | 3827 | .fsync = btrfs_sync_file, |
2fe17c10 | 3828 | .fallocate = btrfs_fallocate, |
34287aa3 | 3829 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3830 | #ifdef CONFIG_COMPAT |
4c63c245 | 3831 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3832 | #endif |
2e5dfc99 | 3833 | .remap_file_range = btrfs_remap_file_range, |
39279cc3 | 3834 | }; |
9247f317 | 3835 | |
728404da FM |
3836 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) |
3837 | { | |
3838 | int ret; | |
3839 | ||
3840 | /* | |
3841 | * So with compression we will find and lock a dirty page and clear the | |
3842 | * first one as dirty, setup an async extent, and immediately return | |
3843 | * with the entire range locked but with nobody actually marked with | |
3844 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3845 | * expect it to work since it will just kick off a thread to do the | |
3846 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3847 | * since it will wait on the page lock, which won't be unlocked until | |
3848 | * after the pages have been marked as writeback and so we're good to go | |
3849 | * from there. We have to do this otherwise we'll miss the ordered | |
3850 | * extents and that results in badness. Please Josef, do not think you | |
3851 | * know better and pull this out at some point in the future, it is | |
3852 | * right and you are wrong. | |
3853 | */ | |
3854 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3855 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3856 | &BTRFS_I(inode)->runtime_flags)) | |
3857 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3858 | ||
3859 | return ret; | |
3860 | } |