Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
39279cc3 CM |
6 | #include <linux/fs.h> |
7 | #include <linux/pagemap.h> | |
39279cc3 CM |
8 | #include <linux/time.h> |
9 | #include <linux/init.h> | |
10 | #include <linux/string.h> | |
39279cc3 | 11 | #include <linux/backing-dev.h> |
2fe17c10 | 12 | #include <linux/falloc.h> |
39279cc3 | 13 | #include <linux/writeback.h> |
39279cc3 | 14 | #include <linux/compat.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
55e301fd | 16 | #include <linux/btrfs.h> |
e2e40f2c | 17 | #include <linux/uio.h> |
ae5e165d | 18 | #include <linux/iversion.h> |
14605409 | 19 | #include <linux/fsverity.h> |
3ecb43cb | 20 | #include <linux/iomap.h> |
39279cc3 CM |
21 | #include "ctree.h" |
22 | #include "disk-io.h" | |
23 | #include "transaction.h" | |
24 | #include "btrfs_inode.h" | |
39279cc3 | 25 | #include "print-tree.h" |
e02119d5 CM |
26 | #include "tree-log.h" |
27 | #include "locking.h" | |
2aaa6655 | 28 | #include "volumes.h" |
fcebe456 | 29 | #include "qgroup.h" |
ebb8765b | 30 | #include "compression.h" |
86736342 | 31 | #include "delalloc-space.h" |
6a177381 | 32 | #include "reflink.h" |
f02a85d2 | 33 | #include "subpage.h" |
c7f13d42 | 34 | #include "fs.h" |
07e81dc9 | 35 | #include "accessors.h" |
a0231804 | 36 | #include "extent-tree.h" |
7c8ede16 | 37 | #include "file-item.h" |
7572dec8 | 38 | #include "ioctl.h" |
af142b6f | 39 | #include "file.h" |
7f0add25 | 40 | #include "super.h" |
39279cc3 | 41 | |
d352ac68 CM |
42 | /* simple helper to fault in pages and copy. This should go away |
43 | * and be replaced with calls into generic code. | |
44 | */ | |
ee22f0c4 | 45 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 46 | struct page **prepared_pages, |
11c65dcc | 47 | struct iov_iter *i) |
39279cc3 | 48 | { |
914ee295 | 49 | size_t copied = 0; |
d0215f3e | 50 | size_t total_copied = 0; |
11c65dcc | 51 | int pg = 0; |
7073017a | 52 | int offset = offset_in_page(pos); |
39279cc3 | 53 | |
11c65dcc | 54 | while (write_bytes > 0) { |
39279cc3 | 55 | size_t count = min_t(size_t, |
09cbfeaf | 56 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 57 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
58 | /* |
59 | * Copy data from userspace to the current page | |
914ee295 | 60 | */ |
f0b65f39 | 61 | copied = copy_page_from_iter_atomic(page, offset, count, i); |
11c65dcc | 62 | |
39279cc3 CM |
63 | /* Flush processor's dcache for this page */ |
64 | flush_dcache_page(page); | |
31339acd CM |
65 | |
66 | /* | |
67 | * if we get a partial write, we can end up with | |
68 | * partially up to date pages. These add | |
69 | * a lot of complexity, so make sure they don't | |
70 | * happen by forcing this copy to be retried. | |
71 | * | |
72 | * The rest of the btrfs_file_write code will fall | |
73 | * back to page at a time copies after we return 0. | |
74 | */ | |
f0b65f39 AV |
75 | if (unlikely(copied < count)) { |
76 | if (!PageUptodate(page)) { | |
77 | iov_iter_revert(i, copied); | |
78 | copied = 0; | |
79 | } | |
80 | if (!copied) | |
81 | break; | |
82 | } | |
31339acd | 83 | |
11c65dcc | 84 | write_bytes -= copied; |
914ee295 | 85 | total_copied += copied; |
f0b65f39 AV |
86 | offset += copied; |
87 | if (offset == PAGE_SIZE) { | |
11c65dcc JB |
88 | pg++; |
89 | offset = 0; | |
90 | } | |
39279cc3 | 91 | } |
914ee295 | 92 | return total_copied; |
39279cc3 CM |
93 | } |
94 | ||
d352ac68 CM |
95 | /* |
96 | * unlocks pages after btrfs_file_write is done with them | |
97 | */ | |
e4f94347 QW |
98 | static void btrfs_drop_pages(struct btrfs_fs_info *fs_info, |
99 | struct page **pages, size_t num_pages, | |
100 | u64 pos, u64 copied) | |
39279cc3 CM |
101 | { |
102 | size_t i; | |
e4f94347 QW |
103 | u64 block_start = round_down(pos, fs_info->sectorsize); |
104 | u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start; | |
105 | ||
106 | ASSERT(block_len <= U32_MAX); | |
39279cc3 | 107 | for (i = 0; i < num_pages; i++) { |
d352ac68 CM |
108 | /* page checked is some magic around finding pages that |
109 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
110 | * clear it here. There should be no need to mark the pages |
111 | * accessed as prepare_pages should have marked them accessed | |
112 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 113 | */ |
e4f94347 QW |
114 | btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start, |
115 | block_len); | |
39279cc3 | 116 | unlock_page(pages[i]); |
09cbfeaf | 117 | put_page(pages[i]); |
39279cc3 CM |
118 | } |
119 | } | |
120 | ||
d352ac68 | 121 | /* |
c0fab480 QW |
122 | * After btrfs_copy_from_user(), update the following things for delalloc: |
123 | * - Mark newly dirtied pages as DELALLOC in the io tree. | |
124 | * Used to advise which range is to be written back. | |
125 | * - Mark modified pages as Uptodate/Dirty and not needing COW fixup | |
126 | * - Update inode size for past EOF write | |
d352ac68 | 127 | */ |
088545f6 | 128 | int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages, |
2ff7e61e | 129 | size_t num_pages, loff_t pos, size_t write_bytes, |
aa8c1a41 | 130 | struct extent_state **cached, bool noreserve) |
39279cc3 | 131 | { |
088545f6 | 132 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
39279cc3 | 133 | int err = 0; |
a52d9a80 | 134 | int i; |
db94535d | 135 | u64 num_bytes; |
a52d9a80 CM |
136 | u64 start_pos; |
137 | u64 end_of_last_block; | |
138 | u64 end_pos = pos + write_bytes; | |
088545f6 | 139 | loff_t isize = i_size_read(&inode->vfs_inode); |
e3b8a485 | 140 | unsigned int extra_bits = 0; |
39279cc3 | 141 | |
aa8c1a41 GR |
142 | if (write_bytes == 0) |
143 | return 0; | |
144 | ||
145 | if (noreserve) | |
146 | extra_bits |= EXTENT_NORESERVE; | |
147 | ||
13f0dd8f | 148 | start_pos = round_down(pos, fs_info->sectorsize); |
da17066c | 149 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 150 | fs_info->sectorsize); |
f02a85d2 | 151 | ASSERT(num_bytes <= U32_MAX); |
39279cc3 | 152 | |
db94535d | 153 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 | 154 | |
7703bdd8 CM |
155 | /* |
156 | * The pages may have already been dirty, clear out old accounting so | |
157 | * we can set things up properly | |
158 | */ | |
088545f6 | 159 | clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block, |
e182163d | 160 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 161 | cached); |
7703bdd8 | 162 | |
088545f6 | 163 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
330a5827 | 164 | extra_bits, cached); |
d0215f3e JB |
165 | if (err) |
166 | return err; | |
9ed74f2d | 167 | |
c8b97818 CM |
168 | for (i = 0; i < num_pages; i++) { |
169 | struct page *p = pages[i]; | |
f02a85d2 QW |
170 | |
171 | btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes); | |
e4f94347 | 172 | btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes); |
f02a85d2 | 173 | btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes); |
a52d9a80 | 174 | } |
9f570b8d JB |
175 | |
176 | /* | |
177 | * we've only changed i_size in ram, and we haven't updated | |
178 | * the disk i_size. There is no need to log the inode | |
179 | * at this time. | |
180 | */ | |
181 | if (end_pos > isize) | |
088545f6 | 182 | i_size_write(&inode->vfs_inode, end_pos); |
a22285a6 | 183 | return 0; |
39279cc3 CM |
184 | } |
185 | ||
186 | /* | |
187 | * this is very complex, but the basic idea is to drop all extents | |
188 | * in the range start - end. hint_block is filled in with a block number | |
189 | * that would be a good hint to the block allocator for this file. | |
190 | * | |
191 | * If an extent intersects the range but is not entirely inside the range | |
192 | * it is either truncated or split. Anything entirely inside the range | |
193 | * is deleted from the tree. | |
2766ff61 FM |
194 | * |
195 | * Note: the VFS' inode number of bytes is not updated, it's up to the caller | |
196 | * to deal with that. We set the field 'bytes_found' of the arguments structure | |
197 | * with the number of allocated bytes found in the target range, so that the | |
198 | * caller can update the inode's number of bytes in an atomic way when | |
199 | * replacing extents in a range to avoid races with stat(2). | |
39279cc3 | 200 | */ |
5893dfb9 FM |
201 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, |
202 | struct btrfs_root *root, struct btrfs_inode *inode, | |
203 | struct btrfs_drop_extents_args *args) | |
39279cc3 | 204 | { |
0b246afa | 205 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 206 | struct extent_buffer *leaf; |
920bbbfb | 207 | struct btrfs_file_extent_item *fi; |
82fa113f | 208 | struct btrfs_ref ref = { 0 }; |
00f5c795 | 209 | struct btrfs_key key; |
920bbbfb | 210 | struct btrfs_key new_key; |
906c448c | 211 | u64 ino = btrfs_ino(inode); |
5893dfb9 | 212 | u64 search_start = args->start; |
920bbbfb YZ |
213 | u64 disk_bytenr = 0; |
214 | u64 num_bytes = 0; | |
215 | u64 extent_offset = 0; | |
216 | u64 extent_end = 0; | |
5893dfb9 | 217 | u64 last_end = args->start; |
920bbbfb YZ |
218 | int del_nr = 0; |
219 | int del_slot = 0; | |
220 | int extent_type; | |
ccd467d6 | 221 | int recow; |
00f5c795 | 222 | int ret; |
dc7fdde3 | 223 | int modify_tree = -1; |
27cdeb70 | 224 | int update_refs; |
c3308f84 | 225 | int found = 0; |
5893dfb9 FM |
226 | struct btrfs_path *path = args->path; |
227 | ||
2766ff61 | 228 | args->bytes_found = 0; |
5893dfb9 FM |
229 | args->extent_inserted = false; |
230 | ||
231 | /* Must always have a path if ->replace_extent is true */ | |
232 | ASSERT(!(args->replace_extent && !args->path)); | |
233 | ||
234 | if (!path) { | |
235 | path = btrfs_alloc_path(); | |
236 | if (!path) { | |
237 | ret = -ENOMEM; | |
238 | goto out; | |
239 | } | |
240 | } | |
39279cc3 | 241 | |
5893dfb9 | 242 | if (args->drop_cache) |
4c0c8cfc | 243 | btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false); |
a52d9a80 | 244 | |
5893dfb9 | 245 | if (args->start >= inode->disk_i_size && !args->replace_extent) |
dc7fdde3 CM |
246 | modify_tree = 0; |
247 | ||
d175209b | 248 | update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID); |
d397712b | 249 | while (1) { |
ccd467d6 | 250 | recow = 0; |
33345d01 | 251 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 252 | search_start, modify_tree); |
39279cc3 | 253 | if (ret < 0) |
920bbbfb | 254 | break; |
5893dfb9 | 255 | if (ret > 0 && path->slots[0] > 0 && search_start == args->start) { |
920bbbfb YZ |
256 | leaf = path->nodes[0]; |
257 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 258 | if (key.objectid == ino && |
920bbbfb YZ |
259 | key.type == BTRFS_EXTENT_DATA_KEY) |
260 | path->slots[0]--; | |
39279cc3 | 261 | } |
920bbbfb | 262 | ret = 0; |
8c2383c3 | 263 | next_slot: |
5f39d397 | 264 | leaf = path->nodes[0]; |
920bbbfb YZ |
265 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
266 | BUG_ON(del_nr > 0); | |
267 | ret = btrfs_next_leaf(root, path); | |
268 | if (ret < 0) | |
269 | break; | |
270 | if (ret > 0) { | |
271 | ret = 0; | |
272 | break; | |
8c2383c3 | 273 | } |
920bbbfb YZ |
274 | leaf = path->nodes[0]; |
275 | recow = 1; | |
276 | } | |
277 | ||
278 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
279 | |
280 | if (key.objectid > ino) | |
281 | break; | |
282 | if (WARN_ON_ONCE(key.objectid < ino) || | |
283 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
284 | ASSERT(del_nr == 0); | |
285 | path->slots[0]++; | |
286 | goto next_slot; | |
287 | } | |
5893dfb9 | 288 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end) |
920bbbfb YZ |
289 | break; |
290 | ||
291 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
292 | struct btrfs_file_extent_item); | |
293 | extent_type = btrfs_file_extent_type(leaf, fi); | |
294 | ||
295 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
296 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
297 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
298 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
299 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
300 | extent_end = key.offset + | |
301 | btrfs_file_extent_num_bytes(leaf, fi); | |
302 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
303 | extent_end = key.offset + | |
e41ca589 | 304 | btrfs_file_extent_ram_bytes(leaf, fi); |
8c2383c3 | 305 | } else { |
aeafbf84 FM |
306 | /* can't happen */ |
307 | BUG(); | |
39279cc3 CM |
308 | } |
309 | ||
fc19c5e7 FM |
310 | /* |
311 | * Don't skip extent items representing 0 byte lengths. They | |
312 | * used to be created (bug) if while punching holes we hit | |
313 | * -ENOSPC condition. So if we find one here, just ensure we | |
314 | * delete it, otherwise we would insert a new file extent item | |
315 | * with the same key (offset) as that 0 bytes length file | |
316 | * extent item in the call to setup_items_for_insert() later | |
317 | * in this function. | |
318 | */ | |
62fe51c1 JB |
319 | if (extent_end == key.offset && extent_end >= search_start) { |
320 | last_end = extent_end; | |
fc19c5e7 | 321 | goto delete_extent_item; |
62fe51c1 | 322 | } |
fc19c5e7 | 323 | |
920bbbfb YZ |
324 | if (extent_end <= search_start) { |
325 | path->slots[0]++; | |
8c2383c3 | 326 | goto next_slot; |
39279cc3 CM |
327 | } |
328 | ||
c3308f84 | 329 | found = 1; |
5893dfb9 | 330 | search_start = max(key.offset, args->start); |
dc7fdde3 CM |
331 | if (recow || !modify_tree) { |
332 | modify_tree = -1; | |
b3b4aa74 | 333 | btrfs_release_path(path); |
920bbbfb | 334 | continue; |
39279cc3 | 335 | } |
6643558d | 336 | |
920bbbfb YZ |
337 | /* |
338 | * | - range to drop - | | |
339 | * | -------- extent -------- | | |
340 | */ | |
5893dfb9 | 341 | if (args->start > key.offset && args->end < extent_end) { |
920bbbfb | 342 | BUG_ON(del_nr > 0); |
00fdf13a | 343 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 344 | ret = -EOPNOTSUPP; |
00fdf13a LB |
345 | break; |
346 | } | |
920bbbfb YZ |
347 | |
348 | memcpy(&new_key, &key, sizeof(new_key)); | |
5893dfb9 | 349 | new_key.offset = args->start; |
920bbbfb YZ |
350 | ret = btrfs_duplicate_item(trans, root, path, |
351 | &new_key); | |
352 | if (ret == -EAGAIN) { | |
b3b4aa74 | 353 | btrfs_release_path(path); |
920bbbfb | 354 | continue; |
6643558d | 355 | } |
920bbbfb YZ |
356 | if (ret < 0) |
357 | break; | |
358 | ||
359 | leaf = path->nodes[0]; | |
360 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
361 | struct btrfs_file_extent_item); | |
362 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 363 | args->start - key.offset); |
920bbbfb YZ |
364 | |
365 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
366 | struct btrfs_file_extent_item); | |
367 | ||
5893dfb9 | 368 | extent_offset += args->start - key.offset; |
920bbbfb YZ |
369 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
370 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 371 | extent_end - args->start); |
50564b65 | 372 | btrfs_mark_buffer_dirty(trans, leaf); |
920bbbfb | 373 | |
5dc562c5 | 374 | if (update_refs && disk_bytenr > 0) { |
82fa113f QW |
375 | btrfs_init_generic_ref(&ref, |
376 | BTRFS_ADD_DELAYED_REF, | |
457cb1dd BB |
377 | disk_bytenr, num_bytes, 0, |
378 | root->root_key.objectid); | |
82fa113f | 379 | btrfs_init_data_ref(&ref, |
920bbbfb YZ |
380 | root->root_key.objectid, |
381 | new_key.objectid, | |
f42c5da6 NB |
382 | args->start - extent_offset, |
383 | 0, false); | |
82fa113f | 384 | ret = btrfs_inc_extent_ref(trans, &ref); |
162d053e FM |
385 | if (ret) { |
386 | btrfs_abort_transaction(trans, ret); | |
387 | break; | |
388 | } | |
771ed689 | 389 | } |
5893dfb9 | 390 | key.offset = args->start; |
6643558d | 391 | } |
62fe51c1 JB |
392 | /* |
393 | * From here on out we will have actually dropped something, so | |
394 | * last_end can be updated. | |
395 | */ | |
396 | last_end = extent_end; | |
397 | ||
920bbbfb YZ |
398 | /* |
399 | * | ---- range to drop ----- | | |
400 | * | -------- extent -------- | | |
401 | */ | |
5893dfb9 | 402 | if (args->start <= key.offset && args->end < extent_end) { |
00fdf13a | 403 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 404 | ret = -EOPNOTSUPP; |
00fdf13a LB |
405 | break; |
406 | } | |
6643558d | 407 | |
920bbbfb | 408 | memcpy(&new_key, &key, sizeof(new_key)); |
5893dfb9 | 409 | new_key.offset = args->end; |
50564b65 | 410 | btrfs_set_item_key_safe(trans, path, &new_key); |
6643558d | 411 | |
5893dfb9 | 412 | extent_offset += args->end - key.offset; |
920bbbfb YZ |
413 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
414 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 415 | extent_end - args->end); |
50564b65 | 416 | btrfs_mark_buffer_dirty(trans, leaf); |
2671485d | 417 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 418 | args->bytes_found += args->end - key.offset; |
920bbbfb | 419 | break; |
39279cc3 | 420 | } |
771ed689 | 421 | |
920bbbfb YZ |
422 | search_start = extent_end; |
423 | /* | |
424 | * | ---- range to drop ----- | | |
425 | * | -------- extent -------- | | |
426 | */ | |
5893dfb9 | 427 | if (args->start > key.offset && args->end >= extent_end) { |
920bbbfb | 428 | BUG_ON(del_nr > 0); |
00fdf13a | 429 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 430 | ret = -EOPNOTSUPP; |
00fdf13a LB |
431 | break; |
432 | } | |
8c2383c3 | 433 | |
920bbbfb | 434 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5893dfb9 | 435 | args->start - key.offset); |
50564b65 | 436 | btrfs_mark_buffer_dirty(trans, leaf); |
2671485d | 437 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 438 | args->bytes_found += extent_end - args->start; |
5893dfb9 | 439 | if (args->end == extent_end) |
920bbbfb | 440 | break; |
c8b97818 | 441 | |
920bbbfb YZ |
442 | path->slots[0]++; |
443 | goto next_slot; | |
31840ae1 ZY |
444 | } |
445 | ||
920bbbfb YZ |
446 | /* |
447 | * | ---- range to drop ----- | | |
448 | * | ------ extent ------ | | |
449 | */ | |
5893dfb9 | 450 | if (args->start <= key.offset && args->end >= extent_end) { |
fc19c5e7 | 451 | delete_extent_item: |
920bbbfb YZ |
452 | if (del_nr == 0) { |
453 | del_slot = path->slots[0]; | |
454 | del_nr = 1; | |
455 | } else { | |
456 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
457 | del_nr++; | |
458 | } | |
31840ae1 | 459 | |
5dc562c5 JB |
460 | if (update_refs && |
461 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
2766ff61 | 462 | args->bytes_found += extent_end - key.offset; |
920bbbfb | 463 | extent_end = ALIGN(extent_end, |
0b246afa | 464 | fs_info->sectorsize); |
5dc562c5 | 465 | } else if (update_refs && disk_bytenr > 0) { |
ffd4bb2a QW |
466 | btrfs_init_generic_ref(&ref, |
467 | BTRFS_DROP_DELAYED_REF, | |
457cb1dd BB |
468 | disk_bytenr, num_bytes, 0, |
469 | root->root_key.objectid); | |
ffd4bb2a | 470 | btrfs_init_data_ref(&ref, |
920bbbfb | 471 | root->root_key.objectid, |
ffd4bb2a | 472 | key.objectid, |
f42c5da6 NB |
473 | key.offset - extent_offset, 0, |
474 | false); | |
ffd4bb2a | 475 | ret = btrfs_free_extent(trans, &ref); |
162d053e FM |
476 | if (ret) { |
477 | btrfs_abort_transaction(trans, ret); | |
478 | break; | |
479 | } | |
2766ff61 | 480 | args->bytes_found += extent_end - key.offset; |
31840ae1 | 481 | } |
31840ae1 | 482 | |
5893dfb9 | 483 | if (args->end == extent_end) |
920bbbfb YZ |
484 | break; |
485 | ||
486 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
487 | path->slots[0]++; | |
488 | goto next_slot; | |
489 | } | |
490 | ||
491 | ret = btrfs_del_items(trans, root, path, del_slot, | |
492 | del_nr); | |
79787eaa | 493 | if (ret) { |
66642832 | 494 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 495 | break; |
79787eaa | 496 | } |
920bbbfb YZ |
497 | |
498 | del_nr = 0; | |
499 | del_slot = 0; | |
500 | ||
b3b4aa74 | 501 | btrfs_release_path(path); |
920bbbfb | 502 | continue; |
39279cc3 | 503 | } |
920bbbfb | 504 | |
290342f6 | 505 | BUG(); |
39279cc3 | 506 | } |
920bbbfb | 507 | |
79787eaa | 508 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
509 | /* |
510 | * Set path->slots[0] to first slot, so that after the delete | |
511 | * if items are move off from our leaf to its immediate left or | |
512 | * right neighbor leafs, we end up with a correct and adjusted | |
5893dfb9 | 513 | * path->slots[0] for our insertion (if args->replace_extent). |
1acae57b FDBM |
514 | */ |
515 | path->slots[0] = del_slot; | |
920bbbfb | 516 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 517 | if (ret) |
66642832 | 518 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 519 | } |
1acae57b | 520 | |
d5f37527 FDBM |
521 | leaf = path->nodes[0]; |
522 | /* | |
523 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
524 | * which case it unlocked our path, so check path->locks[0] matches a | |
525 | * write lock. | |
526 | */ | |
7ecb4c31 | 527 | if (!ret && args->replace_extent && |
ac5887c8 | 528 | path->locks[0] == BTRFS_WRITE_LOCK && |
e902baac | 529 | btrfs_leaf_free_space(leaf) >= |
5893dfb9 | 530 | sizeof(struct btrfs_item) + args->extent_item_size) { |
d5f37527 FDBM |
531 | |
532 | key.objectid = ino; | |
533 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5893dfb9 | 534 | key.offset = args->start; |
d5f37527 FDBM |
535 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
536 | struct btrfs_key slot_key; | |
537 | ||
538 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
539 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
540 | path->slots[0]++; | |
1acae57b | 541 | } |
50564b65 FM |
542 | btrfs_setup_item_for_insert(trans, root, path, &key, |
543 | args->extent_item_size); | |
5893dfb9 | 544 | args->extent_inserted = true; |
6643558d | 545 | } |
920bbbfb | 546 | |
5893dfb9 FM |
547 | if (!args->path) |
548 | btrfs_free_path(path); | |
549 | else if (!args->extent_inserted) | |
1acae57b | 550 | btrfs_release_path(path); |
5893dfb9 FM |
551 | out: |
552 | args->drop_end = found ? min(args->end, last_end) : args->end; | |
5dc562c5 | 553 | |
39279cc3 CM |
554 | return ret; |
555 | } | |
556 | ||
d899e052 | 557 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
558 | u64 objectid, u64 bytenr, u64 orig_offset, |
559 | u64 *start, u64 *end) | |
d899e052 YZ |
560 | { |
561 | struct btrfs_file_extent_item *fi; | |
562 | struct btrfs_key key; | |
563 | u64 extent_end; | |
564 | ||
565 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
566 | return 0; | |
567 | ||
568 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
569 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
570 | return 0; | |
571 | ||
572 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
573 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
574 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 575 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
576 | btrfs_file_extent_compression(leaf, fi) || |
577 | btrfs_file_extent_encryption(leaf, fi) || | |
578 | btrfs_file_extent_other_encoding(leaf, fi)) | |
579 | return 0; | |
580 | ||
581 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
582 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
583 | return 0; | |
584 | ||
585 | *start = key.offset; | |
586 | *end = extent_end; | |
587 | return 1; | |
588 | } | |
589 | ||
590 | /* | |
591 | * Mark extent in the range start - end as written. | |
592 | * | |
593 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
594 | * part of extent is marked as written, the extent will be split into | |
595 | * two or three. | |
596 | */ | |
597 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 598 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 599 | { |
7a6d7067 | 600 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
601 | struct extent_buffer *leaf; |
602 | struct btrfs_path *path; | |
603 | struct btrfs_file_extent_item *fi; | |
82fa113f | 604 | struct btrfs_ref ref = { 0 }; |
d899e052 | 605 | struct btrfs_key key; |
920bbbfb | 606 | struct btrfs_key new_key; |
d899e052 YZ |
607 | u64 bytenr; |
608 | u64 num_bytes; | |
609 | u64 extent_end; | |
5d4f98a2 | 610 | u64 orig_offset; |
d899e052 YZ |
611 | u64 other_start; |
612 | u64 other_end; | |
920bbbfb YZ |
613 | u64 split; |
614 | int del_nr = 0; | |
615 | int del_slot = 0; | |
6c7d54ac | 616 | int recow; |
e7b2ec3d | 617 | int ret = 0; |
7a6d7067 | 618 | u64 ino = btrfs_ino(inode); |
d899e052 | 619 | |
d899e052 | 620 | path = btrfs_alloc_path(); |
d8926bb3 MF |
621 | if (!path) |
622 | return -ENOMEM; | |
d899e052 | 623 | again: |
6c7d54ac | 624 | recow = 0; |
920bbbfb | 625 | split = start; |
33345d01 | 626 | key.objectid = ino; |
d899e052 | 627 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 628 | key.offset = split; |
d899e052 YZ |
629 | |
630 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
631 | if (ret < 0) |
632 | goto out; | |
d899e052 YZ |
633 | if (ret > 0 && path->slots[0] > 0) |
634 | path->slots[0]--; | |
635 | ||
636 | leaf = path->nodes[0]; | |
637 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
638 | if (key.objectid != ino || |
639 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
640 | ret = -EINVAL; | |
641 | btrfs_abort_transaction(trans, ret); | |
642 | goto out; | |
643 | } | |
d899e052 YZ |
644 | fi = btrfs_item_ptr(leaf, path->slots[0], |
645 | struct btrfs_file_extent_item); | |
9c8e63db JB |
646 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
647 | ret = -EINVAL; | |
648 | btrfs_abort_transaction(trans, ret); | |
649 | goto out; | |
650 | } | |
d899e052 | 651 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
652 | if (key.offset > start || extent_end < end) { |
653 | ret = -EINVAL; | |
654 | btrfs_abort_transaction(trans, ret); | |
655 | goto out; | |
656 | } | |
d899e052 YZ |
657 | |
658 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
659 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 660 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
661 | memcpy(&new_key, &key, sizeof(new_key)); |
662 | ||
663 | if (start == key.offset && end < extent_end) { | |
664 | other_start = 0; | |
665 | other_end = start; | |
666 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 667 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
668 | &other_start, &other_end)) { |
669 | new_key.offset = end; | |
50564b65 | 670 | btrfs_set_item_key_safe(trans, path, &new_key); |
6c7d54ac YZ |
671 | fi = btrfs_item_ptr(leaf, path->slots[0], |
672 | struct btrfs_file_extent_item); | |
224ecce5 JB |
673 | btrfs_set_file_extent_generation(leaf, fi, |
674 | trans->transid); | |
6c7d54ac YZ |
675 | btrfs_set_file_extent_num_bytes(leaf, fi, |
676 | extent_end - end); | |
677 | btrfs_set_file_extent_offset(leaf, fi, | |
678 | end - orig_offset); | |
679 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
680 | struct btrfs_file_extent_item); | |
224ecce5 JB |
681 | btrfs_set_file_extent_generation(leaf, fi, |
682 | trans->transid); | |
6c7d54ac YZ |
683 | btrfs_set_file_extent_num_bytes(leaf, fi, |
684 | end - other_start); | |
50564b65 | 685 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac YZ |
686 | goto out; |
687 | } | |
688 | } | |
689 | ||
690 | if (start > key.offset && end == extent_end) { | |
691 | other_start = end; | |
692 | other_end = 0; | |
693 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 694 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
695 | &other_start, &other_end)) { |
696 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
697 | struct btrfs_file_extent_item); | |
698 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
699 | start - key.offset); | |
224ecce5 JB |
700 | btrfs_set_file_extent_generation(leaf, fi, |
701 | trans->transid); | |
6c7d54ac YZ |
702 | path->slots[0]++; |
703 | new_key.offset = start; | |
50564b65 | 704 | btrfs_set_item_key_safe(trans, path, &new_key); |
6c7d54ac YZ |
705 | |
706 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
707 | struct btrfs_file_extent_item); | |
224ecce5 JB |
708 | btrfs_set_file_extent_generation(leaf, fi, |
709 | trans->transid); | |
6c7d54ac YZ |
710 | btrfs_set_file_extent_num_bytes(leaf, fi, |
711 | other_end - start); | |
712 | btrfs_set_file_extent_offset(leaf, fi, | |
713 | start - orig_offset); | |
50564b65 | 714 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac YZ |
715 | goto out; |
716 | } | |
717 | } | |
d899e052 | 718 | |
920bbbfb YZ |
719 | while (start > key.offset || end < extent_end) { |
720 | if (key.offset == start) | |
721 | split = end; | |
722 | ||
920bbbfb YZ |
723 | new_key.offset = split; |
724 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
725 | if (ret == -EAGAIN) { | |
b3b4aa74 | 726 | btrfs_release_path(path); |
920bbbfb | 727 | goto again; |
d899e052 | 728 | } |
79787eaa | 729 | if (ret < 0) { |
66642832 | 730 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
731 | goto out; |
732 | } | |
d899e052 | 733 | |
920bbbfb YZ |
734 | leaf = path->nodes[0]; |
735 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 736 | struct btrfs_file_extent_item); |
224ecce5 | 737 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 738 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
739 | split - key.offset); |
740 | ||
741 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
742 | struct btrfs_file_extent_item); | |
743 | ||
224ecce5 | 744 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
745 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
746 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
747 | extent_end - split); | |
50564b65 | 748 | btrfs_mark_buffer_dirty(trans, leaf); |
d899e052 | 749 | |
82fa113f | 750 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr, |
457cb1dd | 751 | num_bytes, 0, root->root_key.objectid); |
82fa113f | 752 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, |
f42c5da6 | 753 | orig_offset, 0, false); |
82fa113f | 754 | ret = btrfs_inc_extent_ref(trans, &ref); |
9c8e63db JB |
755 | if (ret) { |
756 | btrfs_abort_transaction(trans, ret); | |
757 | goto out; | |
758 | } | |
d899e052 | 759 | |
920bbbfb YZ |
760 | if (split == start) { |
761 | key.offset = start; | |
762 | } else { | |
9c8e63db JB |
763 | if (start != key.offset) { |
764 | ret = -EINVAL; | |
765 | btrfs_abort_transaction(trans, ret); | |
766 | goto out; | |
767 | } | |
d899e052 | 768 | path->slots[0]--; |
920bbbfb | 769 | extent_end = end; |
d899e052 | 770 | } |
6c7d54ac | 771 | recow = 1; |
d899e052 YZ |
772 | } |
773 | ||
920bbbfb YZ |
774 | other_start = end; |
775 | other_end = 0; | |
ffd4bb2a | 776 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
457cb1dd | 777 | num_bytes, 0, root->root_key.objectid); |
f42c5da6 NB |
778 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset, |
779 | 0, false); | |
6c7d54ac | 780 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 781 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
782 | &other_start, &other_end)) { |
783 | if (recow) { | |
b3b4aa74 | 784 | btrfs_release_path(path); |
6c7d54ac YZ |
785 | goto again; |
786 | } | |
920bbbfb YZ |
787 | extent_end = other_end; |
788 | del_slot = path->slots[0] + 1; | |
789 | del_nr++; | |
ffd4bb2a | 790 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
791 | if (ret) { |
792 | btrfs_abort_transaction(trans, ret); | |
793 | goto out; | |
794 | } | |
d899e052 | 795 | } |
920bbbfb YZ |
796 | other_start = 0; |
797 | other_end = start; | |
6c7d54ac | 798 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 799 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
800 | &other_start, &other_end)) { |
801 | if (recow) { | |
b3b4aa74 | 802 | btrfs_release_path(path); |
6c7d54ac YZ |
803 | goto again; |
804 | } | |
920bbbfb YZ |
805 | key.offset = other_start; |
806 | del_slot = path->slots[0]; | |
807 | del_nr++; | |
ffd4bb2a | 808 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
809 | if (ret) { |
810 | btrfs_abort_transaction(trans, ret); | |
811 | goto out; | |
812 | } | |
920bbbfb YZ |
813 | } |
814 | if (del_nr == 0) { | |
3f6fae95 SL |
815 | fi = btrfs_item_ptr(leaf, path->slots[0], |
816 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
817 | btrfs_set_file_extent_type(leaf, fi, |
818 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 819 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 820 | btrfs_mark_buffer_dirty(trans, leaf); |
6c7d54ac | 821 | } else { |
3f6fae95 SL |
822 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
823 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
824 | btrfs_set_file_extent_type(leaf, fi, |
825 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 826 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
827 | btrfs_set_file_extent_num_bytes(leaf, fi, |
828 | extent_end - key.offset); | |
50564b65 | 829 | btrfs_mark_buffer_dirty(trans, leaf); |
920bbbfb | 830 | |
6c7d54ac | 831 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 832 | if (ret < 0) { |
66642832 | 833 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
834 | goto out; |
835 | } | |
6c7d54ac | 836 | } |
920bbbfb | 837 | out: |
d899e052 | 838 | btrfs_free_path(path); |
e7b2ec3d | 839 | return ret; |
d899e052 YZ |
840 | } |
841 | ||
b1bf862e CM |
842 | /* |
843 | * on error we return an unlocked page and the error value | |
844 | * on success we return a locked page and 0 | |
845 | */ | |
bb1591b4 CM |
846 | static int prepare_uptodate_page(struct inode *inode, |
847 | struct page *page, u64 pos, | |
b6316429 | 848 | bool force_uptodate) |
b1bf862e | 849 | { |
fb12489b | 850 | struct folio *folio = page_folio(page); |
b1bf862e CM |
851 | int ret = 0; |
852 | ||
09cbfeaf | 853 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 854 | !PageUptodate(page)) { |
fb12489b | 855 | ret = btrfs_read_folio(NULL, folio); |
b1bf862e CM |
856 | if (ret) |
857 | return ret; | |
858 | lock_page(page); | |
859 | if (!PageUptodate(page)) { | |
860 | unlock_page(page); | |
861 | return -EIO; | |
862 | } | |
e0467866 QW |
863 | |
864 | /* | |
fb12489b | 865 | * Since btrfs_read_folio() will unlock the folio before it |
f913cff3 | 866 | * returns, there is a window where btrfs_release_folio() can be |
7c11d0ae QW |
867 | * called to release the page. Here we check both inode |
868 | * mapping and PagePrivate() to make sure the page was not | |
869 | * released. | |
e0467866 QW |
870 | * |
871 | * The private flag check is essential for subpage as we need | |
872 | * to store extra bitmap using page->private. | |
873 | */ | |
874 | if (page->mapping != inode->i_mapping || !PagePrivate(page)) { | |
bb1591b4 CM |
875 | unlock_page(page); |
876 | return -EAGAIN; | |
877 | } | |
b1bf862e CM |
878 | } |
879 | return 0; | |
880 | } | |
881 | ||
ffc143db | 882 | static fgf_t get_prepare_fgp_flags(bool nowait) |
fc226000 | 883 | { |
ffc143db | 884 | fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT; |
fc226000 SR |
885 | |
886 | if (nowait) | |
887 | fgp_flags |= FGP_NOWAIT; | |
888 | ||
889 | return fgp_flags; | |
890 | } | |
891 | ||
892 | static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait) | |
893 | { | |
894 | gfp_t gfp; | |
895 | ||
896 | gfp = btrfs_alloc_write_mask(inode->i_mapping); | |
897 | if (nowait) { | |
898 | gfp &= ~__GFP_DIRECT_RECLAIM; | |
899 | gfp |= GFP_NOWAIT; | |
900 | } | |
901 | ||
902 | return gfp; | |
903 | } | |
904 | ||
39279cc3 | 905 | /* |
376cc685 | 906 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 907 | */ |
b37392ea MX |
908 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
909 | size_t num_pages, loff_t pos, | |
fc226000 SR |
910 | size_t write_bytes, bool force_uptodate, |
911 | bool nowait) | |
39279cc3 CM |
912 | { |
913 | int i; | |
09cbfeaf | 914 | unsigned long index = pos >> PAGE_SHIFT; |
fc226000 | 915 | gfp_t mask = get_prepare_gfp_flags(inode, nowait); |
ffc143db | 916 | fgf_t fgp_flags = get_prepare_fgp_flags(nowait); |
fc28b62d | 917 | int err = 0; |
376cc685 | 918 | int faili; |
8c2383c3 | 919 | |
39279cc3 | 920 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 921 | again: |
fc226000 SR |
922 | pages[i] = pagecache_get_page(inode->i_mapping, index + i, |
923 | fgp_flags, mask | __GFP_WRITE); | |
39279cc3 | 924 | if (!pages[i]) { |
b1bf862e | 925 | faili = i - 1; |
fc226000 SR |
926 | if (nowait) |
927 | err = -EAGAIN; | |
928 | else | |
929 | err = -ENOMEM; | |
b1bf862e CM |
930 | goto fail; |
931 | } | |
932 | ||
32443de3 QW |
933 | err = set_page_extent_mapped(pages[i]); |
934 | if (err < 0) { | |
935 | faili = i; | |
936 | goto fail; | |
937 | } | |
938 | ||
b1bf862e | 939 | if (i == 0) |
bb1591b4 | 940 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 941 | force_uptodate); |
bb1591b4 CM |
942 | if (!err && i == num_pages - 1) |
943 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 944 | pos + write_bytes, false); |
b1bf862e | 945 | if (err) { |
09cbfeaf | 946 | put_page(pages[i]); |
fc226000 | 947 | if (!nowait && err == -EAGAIN) { |
bb1591b4 CM |
948 | err = 0; |
949 | goto again; | |
950 | } | |
b1bf862e CM |
951 | faili = i - 1; |
952 | goto fail; | |
39279cc3 | 953 | } |
ccd467d6 | 954 | wait_on_page_writeback(pages[i]); |
39279cc3 | 955 | } |
376cc685 MX |
956 | |
957 | return 0; | |
958 | fail: | |
959 | while (faili >= 0) { | |
960 | unlock_page(pages[faili]); | |
09cbfeaf | 961 | put_page(pages[faili]); |
376cc685 MX |
962 | faili--; |
963 | } | |
964 | return err; | |
965 | ||
966 | } | |
967 | ||
968 | /* | |
969 | * This function locks the extent and properly waits for data=ordered extents | |
970 | * to finish before allowing the pages to be modified if need. | |
971 | * | |
972 | * The return value: | |
973 | * 1 - the extent is locked | |
974 | * 0 - the extent is not locked, and everything is OK | |
975 | * -EAGAIN - need re-prepare the pages | |
976 | * the other < 0 number - Something wrong happens | |
977 | */ | |
978 | static noinline int | |
2cff578c | 979 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 980 | size_t num_pages, loff_t pos, |
2e78c927 | 981 | size_t write_bytes, |
2fcab928 | 982 | u64 *lockstart, u64 *lockend, bool nowait, |
376cc685 MX |
983 | struct extent_state **cached_state) |
984 | { | |
3ffbd68c | 985 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
986 | u64 start_pos; |
987 | u64 last_pos; | |
988 | int i; | |
989 | int ret = 0; | |
990 | ||
0b246afa | 991 | start_pos = round_down(pos, fs_info->sectorsize); |
e21139c6 | 992 | last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1; |
376cc685 | 993 | |
e3b8a485 | 994 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 995 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 996 | |
2fcab928 | 997 | if (nowait) { |
83ae4133 JB |
998 | if (!try_lock_extent(&inode->io_tree, start_pos, last_pos, |
999 | cached_state)) { | |
2fcab928 SR |
1000 | for (i = 0; i < num_pages; i++) { |
1001 | unlock_page(pages[i]); | |
1002 | put_page(pages[i]); | |
1003 | pages[i] = NULL; | |
1004 | } | |
1005 | ||
1006 | return -EAGAIN; | |
1007 | } | |
1008 | } else { | |
1009 | lock_extent(&inode->io_tree, start_pos, last_pos, cached_state); | |
1010 | } | |
1011 | ||
b88935bf MX |
1012 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1013 | last_pos - start_pos + 1); | |
e6dcd2dc | 1014 | if (ordered && |
bffe633e | 1015 | ordered->file_offset + ordered->num_bytes > start_pos && |
376cc685 | 1016 | ordered->file_offset <= last_pos) { |
570eb97b JB |
1017 | unlock_extent(&inode->io_tree, start_pos, last_pos, |
1018 | cached_state); | |
e6dcd2dc CM |
1019 | for (i = 0; i < num_pages; i++) { |
1020 | unlock_page(pages[i]); | |
09cbfeaf | 1021 | put_page(pages[i]); |
e6dcd2dc | 1022 | } |
36d45567 | 1023 | btrfs_start_ordered_extent(ordered); |
b88935bf MX |
1024 | btrfs_put_ordered_extent(ordered); |
1025 | return -EAGAIN; | |
e6dcd2dc CM |
1026 | } |
1027 | if (ordered) | |
1028 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1029 | |
376cc685 MX |
1030 | *lockstart = start_pos; |
1031 | *lockend = last_pos; | |
1032 | ret = 1; | |
0762704b | 1033 | } |
376cc685 | 1034 | |
7703bdd8 | 1035 | /* |
32443de3 QW |
1036 | * We should be called after prepare_pages() which should have locked |
1037 | * all pages in the range. | |
7703bdd8 | 1038 | */ |
32443de3 | 1039 | for (i = 0; i < num_pages; i++) |
e6dcd2dc | 1040 | WARN_ON(!PageLocked(pages[i])); |
b1bf862e | 1041 | |
376cc685 | 1042 | return ret; |
39279cc3 CM |
1043 | } |
1044 | ||
d7a8ab4e FM |
1045 | /* |
1046 | * Check if we can do nocow write into the range [@pos, @pos + @write_bytes) | |
1047 | * | |
1048 | * @pos: File offset. | |
1049 | * @write_bytes: The length to write, will be updated to the nocow writeable | |
1050 | * range. | |
1051 | * | |
1052 | * This function will flush ordered extents in the range to ensure proper | |
1053 | * nocow checks. | |
1054 | * | |
1055 | * Return: | |
1056 | * > 0 If we can nocow, and updates @write_bytes. | |
1057 | * 0 If we can't do a nocow write. | |
1058 | * -EAGAIN If we can't do a nocow write because snapshoting of the inode's | |
1059 | * root is in progress. | |
1060 | * < 0 If an error happened. | |
1061 | * | |
1062 | * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. | |
1063 | */ | |
1064 | int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, | |
80f9d241 | 1065 | size_t *write_bytes, bool nowait) |
7ee9e440 | 1066 | { |
3ffbd68c | 1067 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1068 | struct btrfs_root *root = inode->root; |
632ddfa2 | 1069 | struct extent_state *cached_state = NULL; |
7ee9e440 JB |
1070 | u64 lockstart, lockend; |
1071 | u64 num_bytes; | |
1072 | int ret; | |
1073 | ||
38d37aa9 QW |
1074 | if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) |
1075 | return 0; | |
1076 | ||
d7a8ab4e | 1077 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) |
5f791ec3 | 1078 | return -EAGAIN; |
8257b2dc | 1079 | |
0b246afa | 1080 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1081 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1082 | fs_info->sectorsize) - 1; |
5dbb75ed | 1083 | num_bytes = lockend - lockstart + 1; |
7ee9e440 | 1084 | |
80f9d241 | 1085 | if (nowait) { |
632ddfa2 JB |
1086 | if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend, |
1087 | &cached_state)) { | |
80f9d241 JB |
1088 | btrfs_drew_write_unlock(&root->snapshot_lock); |
1089 | return -EAGAIN; | |
1090 | } | |
1091 | } else { | |
632ddfa2 JB |
1092 | btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, |
1093 | &cached_state); | |
80f9d241 | 1094 | } |
85b7ab67 | 1095 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
80f9d241 JB |
1096 | NULL, NULL, NULL, nowait, false); |
1097 | if (ret <= 0) | |
d7a8ab4e | 1098 | btrfs_drew_write_unlock(&root->snapshot_lock); |
80f9d241 | 1099 | else |
c933956d MX |
1100 | *write_bytes = min_t(size_t, *write_bytes , |
1101 | num_bytes - pos + lockstart); | |
632ddfa2 | 1102 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
7ee9e440 JB |
1103 | |
1104 | return ret; | |
1105 | } | |
1106 | ||
38d37aa9 QW |
1107 | void btrfs_check_nocow_unlock(struct btrfs_inode *inode) |
1108 | { | |
1109 | btrfs_drew_write_unlock(&inode->root->snapshot_lock); | |
1110 | } | |
1111 | ||
efd34f03 CB |
1112 | static void update_time_for_write(struct inode *inode) |
1113 | { | |
1114 | struct timespec64 now, ctime; | |
1115 | ||
1116 | if (IS_NOCMTIME(inode)) | |
1117 | return; | |
1118 | ||
1119 | now = current_time(inode); | |
1120 | if (!timespec64_equal(&inode->i_mtime, &now)) | |
1121 | inode->i_mtime = now; | |
1122 | ||
1123 | ctime = inode_get_ctime(inode); | |
1124 | if (!timespec64_equal(&ctime, &now)) | |
1125 | inode_set_ctime_to_ts(inode, now); | |
1126 | ||
1127 | if (IS_I_VERSION(inode)) | |
1128 | inode_inc_iversion(inode); | |
1129 | } | |
1130 | ||
b8d8e1fd GR |
1131 | static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from, |
1132 | size_t count) | |
1133 | { | |
1134 | struct file *file = iocb->ki_filp; | |
1135 | struct inode *inode = file_inode(file); | |
1136 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1137 | loff_t pos = iocb->ki_pos; | |
1138 | int ret; | |
1139 | loff_t oldsize; | |
1140 | loff_t start_pos; | |
1141 | ||
d7a8ab4e FM |
1142 | /* |
1143 | * Quickly bail out on NOWAIT writes if we don't have the nodatacow or | |
1144 | * prealloc flags, as without those flags we always have to COW. We will | |
1145 | * later check if we can really COW into the target range (using | |
1146 | * can_nocow_extent() at btrfs_get_blocks_direct_write()). | |
1147 | */ | |
1148 | if ((iocb->ki_flags & IOCB_NOWAIT) && | |
1149 | !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
1150 | return -EAGAIN; | |
b8d8e1fd | 1151 | |
b8d8e1fd GR |
1152 | ret = file_remove_privs(file); |
1153 | if (ret) | |
1154 | return ret; | |
1155 | ||
1156 | /* | |
1157 | * We reserve space for updating the inode when we reserve space for the | |
1158 | * extent we are going to write, so we will enospc out there. We don't | |
1159 | * need to start yet another transaction to update the inode as we will | |
1160 | * update the inode when we finish writing whatever data we write. | |
1161 | */ | |
efd34f03 | 1162 | update_time_for_write(inode); |
b8d8e1fd GR |
1163 | |
1164 | start_pos = round_down(pos, fs_info->sectorsize); | |
1165 | oldsize = i_size_read(inode); | |
1166 | if (start_pos > oldsize) { | |
1167 | /* Expand hole size to cover write data, preventing empty gap */ | |
1168 | loff_t end_pos = round_up(pos + count, fs_info->sectorsize); | |
1169 | ||
b06359a3 | 1170 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos); |
0d625446 | 1171 | if (ret) |
b8d8e1fd | 1172 | return ret; |
b8d8e1fd GR |
1173 | } |
1174 | ||
1175 | return 0; | |
1176 | } | |
1177 | ||
e4af400a GR |
1178 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1179 | struct iov_iter *i) | |
4b46fce2 | 1180 | { |
e4af400a | 1181 | struct file *file = iocb->ki_filp; |
c3523706 | 1182 | loff_t pos; |
496ad9aa | 1183 | struct inode *inode = file_inode(file); |
0b246afa | 1184 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1185 | struct page **pages = NULL; |
364ecf36 | 1186 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1187 | u64 release_bytes = 0; |
376cc685 MX |
1188 | u64 lockstart; |
1189 | u64 lockend; | |
d0215f3e JB |
1190 | size_t num_written = 0; |
1191 | int nrptrs; | |
c3523706 | 1192 | ssize_t ret; |
7ee9e440 | 1193 | bool only_release_metadata = false; |
b6316429 | 1194 | bool force_page_uptodate = false; |
5e8b9ef3 | 1195 | loff_t old_isize = i_size_read(inode); |
c3523706 | 1196 | unsigned int ilock_flags = 0; |
304e45ac | 1197 | const bool nowait = (iocb->ki_flags & IOCB_NOWAIT); |
965f47ae | 1198 | unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0); |
c3523706 | 1199 | |
304e45ac | 1200 | if (nowait) |
c3523706 GR |
1201 | ilock_flags |= BTRFS_ILOCK_TRY; |
1202 | ||
29b6352b | 1203 | ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1204 | if (ret < 0) |
1205 | return ret; | |
4b46fce2 | 1206 | |
c3523706 GR |
1207 | ret = generic_write_checks(iocb, i); |
1208 | if (ret <= 0) | |
1209 | goto out; | |
1210 | ||
1211 | ret = btrfs_write_check(iocb, i, ret); | |
1212 | if (ret < 0) | |
1213 | goto out; | |
1214 | ||
1215 | pos = iocb->ki_pos; | |
09cbfeaf KS |
1216 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1217 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1218 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1219 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1220 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
c3523706 GR |
1221 | if (!pages) { |
1222 | ret = -ENOMEM; | |
1223 | goto out; | |
1224 | } | |
ab93dbec | 1225 | |
d0215f3e | 1226 | while (iov_iter_count(i) > 0) { |
c67d970f | 1227 | struct extent_state *cached_state = NULL; |
7073017a | 1228 | size_t offset = offset_in_page(pos); |
2e78c927 | 1229 | size_t sector_offset; |
d0215f3e | 1230 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1231 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1232 | offset); |
eefa45f5 | 1233 | size_t num_pages; |
7ee9e440 | 1234 | size_t reserve_bytes; |
d0215f3e JB |
1235 | size_t dirty_pages; |
1236 | size_t copied; | |
2e78c927 CR |
1237 | size_t dirty_sectors; |
1238 | size_t num_sectors; | |
79f015f2 | 1239 | int extents_locked; |
39279cc3 | 1240 | |
914ee295 XZ |
1241 | /* |
1242 | * Fault pages before locking them in prepare_pages | |
1243 | * to avoid recursive lock | |
1244 | */ | |
a6294593 | 1245 | if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) { |
914ee295 | 1246 | ret = -EFAULT; |
d0215f3e | 1247 | break; |
914ee295 XZ |
1248 | } |
1249 | ||
a0e248bb | 1250 | only_release_metadata = false; |
da17066c | 1251 | sector_offset = pos & (fs_info->sectorsize - 1); |
d9d8b2a5 | 1252 | |
364ecf36 | 1253 | extent_changeset_release(data_reserved); |
36ea6f3e NB |
1254 | ret = btrfs_check_data_free_space(BTRFS_I(inode), |
1255 | &data_reserved, pos, | |
304e45ac | 1256 | write_bytes, nowait); |
c6887cd1 | 1257 | if (ret < 0) { |
80f9d241 JB |
1258 | int can_nocow; |
1259 | ||
304e45ac SR |
1260 | if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) { |
1261 | ret = -EAGAIN; | |
1262 | break; | |
1263 | } | |
1264 | ||
eefa45f5 GR |
1265 | /* |
1266 | * If we don't have to COW at the offset, reserve | |
1267 | * metadata only. write_bytes may get smaller than | |
1268 | * requested here. | |
1269 | */ | |
80f9d241 | 1270 | can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos, |
304e45ac | 1271 | &write_bytes, nowait); |
80f9d241 JB |
1272 | if (can_nocow < 0) |
1273 | ret = can_nocow; | |
1274 | if (can_nocow > 0) | |
1275 | ret = 0; | |
1276 | if (ret) | |
c6887cd1 | 1277 | break; |
80f9d241 | 1278 | only_release_metadata = true; |
c6887cd1 | 1279 | } |
1832a6d5 | 1280 | |
eefa45f5 GR |
1281 | num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE); |
1282 | WARN_ON(num_pages > nrptrs); | |
1283 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1284 | fs_info->sectorsize); | |
8b62f87b | 1285 | WARN_ON(reserve_bytes == 0); |
9f3db423 | 1286 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
28c9b1e7 | 1287 | reserve_bytes, |
304e45ac | 1288 | reserve_bytes, nowait); |
7ee9e440 JB |
1289 | if (ret) { |
1290 | if (!only_release_metadata) | |
25ce28ca | 1291 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
bc42bda2 QW |
1292 | data_reserved, pos, |
1293 | write_bytes); | |
8257b2dc | 1294 | else |
38d37aa9 | 1295 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
a348c8d4 FM |
1296 | |
1297 | if (nowait && ret == -ENOSPC) | |
1298 | ret = -EAGAIN; | |
7ee9e440 JB |
1299 | break; |
1300 | } | |
1301 | ||
1302 | release_bytes = reserve_bytes; | |
376cc685 | 1303 | again: |
965f47ae | 1304 | ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags); |
eb81b682 FM |
1305 | if (ret) { |
1306 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); | |
965f47ae | 1307 | break; |
eb81b682 | 1308 | } |
965f47ae | 1309 | |
4a64001f JB |
1310 | /* |
1311 | * This is going to setup the pages array with the number of | |
1312 | * pages we want, so we don't really need to worry about the | |
1313 | * contents of pages from loop to loop | |
1314 | */ | |
b37392ea | 1315 | ret = prepare_pages(inode, pages, num_pages, |
fc226000 | 1316 | pos, write_bytes, force_page_uptodate, false); |
8b62f87b JB |
1317 | if (ret) { |
1318 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
8702ba93 | 1319 | reserve_bytes); |
d0215f3e | 1320 | break; |
8b62f87b | 1321 | } |
39279cc3 | 1322 | |
79f015f2 GR |
1323 | extents_locked = lock_and_cleanup_extent_if_need( |
1324 | BTRFS_I(inode), pages, | |
2cff578c | 1325 | num_pages, pos, write_bytes, &lockstart, |
304e45ac | 1326 | &lockend, nowait, &cached_state); |
79f015f2 | 1327 | if (extents_locked < 0) { |
304e45ac | 1328 | if (!nowait && extents_locked == -EAGAIN) |
376cc685 | 1329 | goto again; |
304e45ac | 1330 | |
8b62f87b | 1331 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
8702ba93 | 1332 | reserve_bytes); |
79f015f2 | 1333 | ret = extents_locked; |
376cc685 | 1334 | break; |
376cc685 MX |
1335 | } |
1336 | ||
ee22f0c4 | 1337 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1338 | |
0b246afa | 1339 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1340 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1341 | fs_info->sectorsize); |
1342 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1343 | |
b1bf862e CM |
1344 | /* |
1345 | * if we have trouble faulting in the pages, fall | |
1346 | * back to one page at a time | |
1347 | */ | |
1348 | if (copied < write_bytes) | |
1349 | nrptrs = 1; | |
1350 | ||
b6316429 JB |
1351 | if (copied == 0) { |
1352 | force_page_uptodate = true; | |
56244ef1 | 1353 | dirty_sectors = 0; |
b1bf862e | 1354 | dirty_pages = 0; |
b6316429 JB |
1355 | } else { |
1356 | force_page_uptodate = false; | |
ed6078f7 | 1357 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1358 | PAGE_SIZE); |
b6316429 | 1359 | } |
914ee295 | 1360 | |
2e78c927 | 1361 | if (num_sectors > dirty_sectors) { |
8b8b08cb | 1362 | /* release everything except the sectors we dirtied */ |
265fdfa6 | 1363 | release_bytes -= dirty_sectors << fs_info->sectorsize_bits; |
485290a7 | 1364 | if (only_release_metadata) { |
691fa059 | 1365 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1366 | release_bytes, true); |
485290a7 QW |
1367 | } else { |
1368 | u64 __pos; | |
1369 | ||
da17066c | 1370 | __pos = round_down(pos, |
0b246afa | 1371 | fs_info->sectorsize) + |
09cbfeaf | 1372 | (dirty_pages << PAGE_SHIFT); |
86d52921 | 1373 | btrfs_delalloc_release_space(BTRFS_I(inode), |
bc42bda2 | 1374 | data_reserved, __pos, |
43b18595 | 1375 | release_bytes, true); |
485290a7 | 1376 | } |
914ee295 XZ |
1377 | } |
1378 | ||
2e78c927 | 1379 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1380 | fs_info->sectorsize); |
376cc685 | 1381 | |
aa8c1a41 GR |
1382 | ret = btrfs_dirty_pages(BTRFS_I(inode), pages, |
1383 | dirty_pages, pos, copied, | |
1384 | &cached_state, only_release_metadata); | |
c67d970f FM |
1385 | |
1386 | /* | |
1387 | * If we have not locked the extent range, because the range's | |
1388 | * start offset is >= i_size, we might still have a non-NULL | |
1389 | * cached extent state, acquired while marking the extent range | |
1390 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1391 | * possible cached extent state to avoid a memory leak. | |
1392 | */ | |
79f015f2 | 1393 | if (extents_locked) |
570eb97b JB |
1394 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
1395 | lockend, &cached_state); | |
c67d970f FM |
1396 | else |
1397 | free_extent_state(cached_state); | |
1398 | ||
8702ba93 | 1399 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 | 1400 | if (ret) { |
e4f94347 | 1401 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
376cc685 | 1402 | break; |
f1de9683 | 1403 | } |
39279cc3 | 1404 | |
376cc685 | 1405 | release_bytes = 0; |
8257b2dc | 1406 | if (only_release_metadata) |
38d37aa9 | 1407 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
8257b2dc | 1408 | |
e4f94347 | 1409 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
f1de9683 | 1410 | |
d0215f3e JB |
1411 | cond_resched(); |
1412 | ||
914ee295 XZ |
1413 | pos += copied; |
1414 | num_written += copied; | |
d0215f3e | 1415 | } |
39279cc3 | 1416 | |
d0215f3e JB |
1417 | kfree(pages); |
1418 | ||
7ee9e440 | 1419 | if (release_bytes) { |
8257b2dc | 1420 | if (only_release_metadata) { |
38d37aa9 | 1421 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
691fa059 | 1422 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1423 | release_bytes, true); |
8257b2dc | 1424 | } else { |
86d52921 NB |
1425 | btrfs_delalloc_release_space(BTRFS_I(inode), |
1426 | data_reserved, | |
bc42bda2 | 1427 | round_down(pos, fs_info->sectorsize), |
43b18595 | 1428 | release_bytes, true); |
8257b2dc | 1429 | } |
7ee9e440 JB |
1430 | } |
1431 | ||
364ecf36 | 1432 | extent_changeset_free(data_reserved); |
5e8b9ef3 GR |
1433 | if (num_written > 0) { |
1434 | pagecache_isize_extended(inode, old_isize, iocb->ki_pos); | |
1435 | iocb->ki_pos += num_written; | |
1436 | } | |
c3523706 | 1437 | out: |
e5d4d75b | 1438 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
d0215f3e JB |
1439 | return num_written ? num_written : ret; |
1440 | } | |
1441 | ||
4e4cabec GR |
1442 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
1443 | const struct iov_iter *iter, loff_t offset) | |
1444 | { | |
1445 | const u32 blocksize_mask = fs_info->sectorsize - 1; | |
1446 | ||
1447 | if (offset & blocksize_mask) | |
1448 | return -EINVAL; | |
1449 | ||
1450 | if (iov_iter_alignment(iter) & blocksize_mask) | |
1451 | return -EINVAL; | |
1452 | ||
1453 | return 0; | |
1454 | } | |
1455 | ||
1456 | static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
d0215f3e JB |
1457 | { |
1458 | struct file *file = iocb->ki_filp; | |
728404da | 1459 | struct inode *inode = file_inode(file); |
4e4cabec | 1460 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c3523706 | 1461 | loff_t pos; |
4e4cabec | 1462 | ssize_t written = 0; |
d0215f3e | 1463 | ssize_t written_buffered; |
51bd9563 | 1464 | size_t prev_left = 0; |
d0215f3e | 1465 | loff_t endbyte; |
c3523706 GR |
1466 | ssize_t err; |
1467 | unsigned int ilock_flags = 0; | |
8184620a | 1468 | struct iomap_dio *dio; |
c3523706 GR |
1469 | |
1470 | if (iocb->ki_flags & IOCB_NOWAIT) | |
1471 | ilock_flags |= BTRFS_ILOCK_TRY; | |
1472 | ||
9af86694 BS |
1473 | /* |
1474 | * If the write DIO is within EOF, use a shared lock and also only if | |
1475 | * security bits will likely not be dropped by file_remove_privs() called | |
1476 | * from btrfs_write_check(). Either will need to be rechecked after the | |
1477 | * lock was acquired. | |
1478 | */ | |
1479 | if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode)) | |
e9adabb9 GR |
1480 | ilock_flags |= BTRFS_ILOCK_SHARED; |
1481 | ||
1482 | relock: | |
29b6352b | 1483 | err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1484 | if (err < 0) |
1485 | return err; | |
1486 | ||
9af86694 BS |
1487 | /* Shared lock cannot be used with security bits set. */ |
1488 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) { | |
1489 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); | |
1490 | ilock_flags &= ~BTRFS_ILOCK_SHARED; | |
1491 | goto relock; | |
1492 | } | |
1493 | ||
c3523706 GR |
1494 | err = generic_write_checks(iocb, from); |
1495 | if (err <= 0) { | |
e5d4d75b | 1496 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1497 | return err; |
1498 | } | |
d0215f3e | 1499 | |
c3523706 GR |
1500 | err = btrfs_write_check(iocb, from, err); |
1501 | if (err < 0) { | |
e5d4d75b | 1502 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
c3523706 GR |
1503 | goto out; |
1504 | } | |
1505 | ||
1506 | pos = iocb->ki_pos; | |
e9adabb9 GR |
1507 | /* |
1508 | * Re-check since file size may have changed just before taking the | |
1509 | * lock or pos may have changed because of O_APPEND in generic_write_check() | |
1510 | */ | |
1511 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && | |
1512 | pos + iov_iter_count(from) > i_size_read(inode)) { | |
e5d4d75b | 1513 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
e9adabb9 GR |
1514 | ilock_flags &= ~BTRFS_ILOCK_SHARED; |
1515 | goto relock; | |
1516 | } | |
c3523706 GR |
1517 | |
1518 | if (check_direct_IO(fs_info, from, pos)) { | |
e5d4d75b | 1519 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
4e4cabec | 1520 | goto buffered; |
c3523706 | 1521 | } |
4e4cabec | 1522 | |
51bd9563 FM |
1523 | /* |
1524 | * The iov_iter can be mapped to the same file range we are writing to. | |
1525 | * If that's the case, then we will deadlock in the iomap code, because | |
1526 | * it first calls our callback btrfs_dio_iomap_begin(), which will create | |
1527 | * an ordered extent, and after that it will fault in the pages that the | |
1528 | * iov_iter refers to. During the fault in we end up in the readahead | |
1529 | * pages code (starting at btrfs_readahead()), which will lock the range, | |
1530 | * find that ordered extent and then wait for it to complete (at | |
1531 | * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since | |
1532 | * obviously the ordered extent can never complete as we didn't submit | |
1533 | * yet the respective bio(s). This always happens when the buffer is | |
1534 | * memory mapped to the same file range, since the iomap DIO code always | |
1535 | * invalidates pages in the target file range (after starting and waiting | |
1536 | * for any writeback). | |
1537 | * | |
1538 | * So here we disable page faults in the iov_iter and then retry if we | |
1539 | * got -EFAULT, faulting in the pages before the retry. | |
1540 | */ | |
51bd9563 | 1541 | from->nofault = true; |
8184620a | 1542 | dio = btrfs_dio_write(iocb, from, written); |
51bd9563 | 1543 | from->nofault = false; |
d0215f3e | 1544 | |
8184620a FM |
1545 | /* |
1546 | * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync | |
1547 | * iocb, and that needs to lock the inode. So unlock it before calling | |
1548 | * iomap_dio_complete() to avoid a deadlock. | |
1549 | */ | |
e5d4d75b | 1550 | btrfs_inode_unlock(BTRFS_I(inode), ilock_flags); |
8184620a FM |
1551 | |
1552 | if (IS_ERR_OR_NULL(dio)) | |
1553 | err = PTR_ERR_OR_ZERO(dio); | |
1554 | else | |
1555 | err = iomap_dio_complete(dio); | |
1556 | ||
51bd9563 FM |
1557 | /* No increment (+=) because iomap returns a cumulative value. */ |
1558 | if (err > 0) | |
1559 | written = err; | |
1560 | ||
1561 | if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) { | |
1562 | const size_t left = iov_iter_count(from); | |
1563 | /* | |
1564 | * We have more data left to write. Try to fault in as many as | |
1565 | * possible of the remainder pages and retry. We do this without | |
1566 | * releasing and locking again the inode, to prevent races with | |
1567 | * truncate. | |
1568 | * | |
1569 | * Also, in case the iov refers to pages in the file range of the | |
1570 | * file we want to write to (due to a mmap), we could enter an | |
1571 | * infinite loop if we retry after faulting the pages in, since | |
1572 | * iomap will invalidate any pages in the range early on, before | |
1573 | * it tries to fault in the pages of the iov. So we keep track of | |
1574 | * how much was left of iov in the previous EFAULT and fallback | |
1575 | * to buffered IO in case we haven't made any progress. | |
1576 | */ | |
1577 | if (left == prev_left) { | |
1578 | err = -ENOTBLK; | |
1579 | } else { | |
1580 | fault_in_iov_iter_readable(from, left); | |
1581 | prev_left = left; | |
8184620a | 1582 | goto relock; |
51bd9563 | 1583 | } |
a42fa643 GR |
1584 | } |
1585 | ||
ac5e6669 FM |
1586 | /* |
1587 | * If 'err' is -ENOTBLK or we have not written all data, then it means | |
1588 | * we must fallback to buffered IO. | |
1589 | */ | |
51bd9563 | 1590 | if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from)) |
c3523706 | 1591 | goto out; |
d0215f3e | 1592 | |
4e4cabec | 1593 | buffered: |
ac5e6669 FM |
1594 | /* |
1595 | * If we are in a NOWAIT context, then return -EAGAIN to signal the caller | |
1596 | * it must retry the operation in a context where blocking is acceptable, | |
20af93d9 FM |
1597 | * because even if we end up not blocking during the buffered IO attempt |
1598 | * below, we will block when flushing and waiting for the IO. | |
ac5e6669 FM |
1599 | */ |
1600 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
1601 | err = -EAGAIN; | |
1602 | goto out; | |
1603 | } | |
1604 | ||
e4af400a GR |
1605 | pos = iocb->ki_pos; |
1606 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e JB |
1607 | if (written_buffered < 0) { |
1608 | err = written_buffered; | |
1609 | goto out; | |
39279cc3 | 1610 | } |
075bdbdb FM |
1611 | /* |
1612 | * Ensure all data is persisted. We want the next direct IO read to be | |
1613 | * able to read what was just written. | |
1614 | */ | |
d0215f3e | 1615 | endbyte = pos + written_buffered - 1; |
728404da | 1616 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1617 | if (err) |
1618 | goto out; | |
728404da | 1619 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1620 | if (err) |
1621 | goto out; | |
1622 | written += written_buffered; | |
867c4f93 | 1623 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1624 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1625 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1626 | out: |
51bd9563 | 1627 | return err < 0 ? err : written; |
d0215f3e | 1628 | } |
5b92ee72 | 1629 | |
7c0c7269 OS |
1630 | static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
1631 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
1632 | { | |
1633 | struct file *file = iocb->ki_filp; | |
1634 | struct inode *inode = file_inode(file); | |
1635 | loff_t count; | |
1636 | ssize_t ret; | |
1637 | ||
29b6352b | 1638 | btrfs_inode_lock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1639 | count = encoded->len; |
1640 | ret = generic_write_checks_count(iocb, &count); | |
1641 | if (ret == 0 && count != encoded->len) { | |
1642 | /* | |
1643 | * The write got truncated by generic_write_checks_count(). We | |
1644 | * can't do a partial encoded write. | |
1645 | */ | |
1646 | ret = -EFBIG; | |
1647 | } | |
1648 | if (ret || encoded->len == 0) | |
1649 | goto out; | |
1650 | ||
1651 | ret = btrfs_write_check(iocb, from, encoded->len); | |
1652 | if (ret < 0) | |
1653 | goto out; | |
1654 | ||
1655 | ret = btrfs_do_encoded_write(iocb, from, encoded); | |
1656 | out: | |
e5d4d75b | 1657 | btrfs_inode_unlock(BTRFS_I(inode), 0); |
7c0c7269 OS |
1658 | return ret; |
1659 | } | |
1660 | ||
1661 | ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from, | |
1662 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
d0215f3e JB |
1663 | { |
1664 | struct file *file = iocb->ki_filp; | |
14971657 | 1665 | struct btrfs_inode *inode = BTRFS_I(file_inode(file)); |
7c0c7269 | 1666 | ssize_t num_written, num_sync; |
d0215f3e | 1667 | |
c86537a4 GR |
1668 | /* |
1669 | * If the fs flips readonly due to some impossible error, although we | |
1670 | * have opened a file as writable, we have to stop this write operation | |
1671 | * to ensure consistency. | |
1672 | */ | |
84961539 | 1673 | if (BTRFS_FS_ERROR(inode->root->fs_info)) |
c86537a4 GR |
1674 | return -EROFS; |
1675 | ||
926078b2 | 1676 | if (encoded && (iocb->ki_flags & IOCB_NOWAIT)) |
91f9943e CH |
1677 | return -EOPNOTSUPP; |
1678 | ||
7c0c7269 OS |
1679 | if (encoded) { |
1680 | num_written = btrfs_encoded_write(iocb, from, encoded); | |
1681 | num_sync = encoded->len; | |
1682 | } else if (iocb->ki_flags & IOCB_DIRECT) { | |
c1867eb3 DS |
1683 | num_written = btrfs_direct_write(iocb, from); |
1684 | num_sync = num_written; | |
7c0c7269 | 1685 | } else { |
c1867eb3 DS |
1686 | num_written = btrfs_buffered_write(iocb, from); |
1687 | num_sync = num_written; | |
7c0c7269 | 1688 | } |
d0215f3e | 1689 | |
bc0939fc FM |
1690 | btrfs_set_inode_last_sub_trans(inode); |
1691 | ||
7c0c7269 OS |
1692 | if (num_sync > 0) { |
1693 | num_sync = generic_write_sync(iocb, num_sync); | |
1694 | if (num_sync < 0) | |
1695 | num_written = num_sync; | |
1696 | } | |
0a3404dc | 1697 | |
c3523706 | 1698 | return num_written; |
39279cc3 CM |
1699 | } |
1700 | ||
7c0c7269 OS |
1701 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1702 | { | |
1703 | return btrfs_do_write_iter(iocb, from, NULL); | |
1704 | } | |
1705 | ||
d397712b | 1706 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1707 | { |
23b5ec74 JB |
1708 | struct btrfs_file_private *private = filp->private_data; |
1709 | ||
3c32c721 | 1710 | if (private) { |
23b5ec74 | 1711 | kfree(private->filldir_buf); |
3c32c721 FM |
1712 | free_extent_state(private->llseek_cached_state); |
1713 | kfree(private); | |
1714 | filp->private_data = NULL; | |
1715 | } | |
23b5ec74 | 1716 | |
f6dc45c7 | 1717 | /* |
1fd4033d NB |
1718 | * Set by setattr when we are about to truncate a file from a non-zero |
1719 | * size to a zero size. This tries to flush down new bytes that may | |
1720 | * have been written if the application were using truncate to replace | |
1721 | * a file in place. | |
f6dc45c7 | 1722 | */ |
1fd4033d | 1723 | if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
f6dc45c7 CM |
1724 | &BTRFS_I(inode)->runtime_flags)) |
1725 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
1726 | return 0; |
1727 | } | |
1728 | ||
669249ee FM |
1729 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
1730 | { | |
1731 | int ret; | |
343e4fc1 | 1732 | struct blk_plug plug; |
669249ee | 1733 | |
343e4fc1 LB |
1734 | /* |
1735 | * This is only called in fsync, which would do synchronous writes, so | |
1736 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
1737 | * multiple disks using raid profile, a large IO can be split to | |
1738 | * several segments of stripe length (currently 64K). | |
1739 | */ | |
1740 | blk_start_plug(&plug); | |
728404da | 1741 | ret = btrfs_fdatawrite_range(inode, start, end); |
343e4fc1 | 1742 | blk_finish_plug(&plug); |
669249ee FM |
1743 | |
1744 | return ret; | |
1745 | } | |
1746 | ||
626e9f41 FM |
1747 | static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx) |
1748 | { | |
1749 | struct btrfs_inode *inode = BTRFS_I(ctx->inode); | |
1750 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1751 | ||
4a4f8fe2 | 1752 | if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) && |
626e9f41 FM |
1753 | list_empty(&ctx->ordered_extents)) |
1754 | return true; | |
1755 | ||
1756 | /* | |
1757 | * If we are doing a fast fsync we can not bail out if the inode's | |
1758 | * last_trans is <= then the last committed transaction, because we only | |
1759 | * update the last_trans of the inode during ordered extent completion, | |
1760 | * and for a fast fsync we don't wait for that, we only wait for the | |
1761 | * writeback to complete. | |
1762 | */ | |
0124855f | 1763 | if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) && |
626e9f41 FM |
1764 | (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) || |
1765 | list_empty(&ctx->ordered_extents))) | |
1766 | return true; | |
1767 | ||
1768 | return false; | |
1769 | } | |
1770 | ||
d352ac68 CM |
1771 | /* |
1772 | * fsync call for both files and directories. This logs the inode into | |
1773 | * the tree log instead of forcing full commits whenever possible. | |
1774 | * | |
1775 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1776 | * in the metadata btree are up to date for copying to the log. | |
1777 | * | |
1778 | * It drops the inode mutex before doing the tree log commit. This is an | |
1779 | * important optimization for directories because holding the mutex prevents | |
1780 | * new operations on the dir while we write to disk. | |
1781 | */ | |
02c24a82 | 1782 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1783 | { |
de17e793 | 1784 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 1785 | struct inode *inode = d_inode(dentry); |
0b246afa | 1786 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 1787 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 1788 | struct btrfs_trans_handle *trans; |
8b050d35 | 1789 | struct btrfs_log_ctx ctx; |
333427a5 | 1790 | int ret = 0, err; |
48778179 FM |
1791 | u64 len; |
1792 | bool full_sync; | |
39279cc3 | 1793 | |
1abe9b8a | 1794 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1795 | |
ebb70442 LB |
1796 | btrfs_init_log_ctx(&ctx, inode); |
1797 | ||
95418ed1 | 1798 | /* |
48778179 FM |
1799 | * Always set the range to a full range, otherwise we can get into |
1800 | * several problems, from missing file extent items to represent holes | |
1801 | * when not using the NO_HOLES feature, to log tree corruption due to | |
1802 | * races between hole detection during logging and completion of ordered | |
1803 | * extents outside the range, to missing checksums due to ordered extents | |
1804 | * for which we flushed only a subset of their pages. | |
95418ed1 | 1805 | */ |
48778179 FM |
1806 | start = 0; |
1807 | end = LLONG_MAX; | |
1808 | len = (u64)LLONG_MAX + 1; | |
95418ed1 | 1809 | |
90abccf2 MX |
1810 | /* |
1811 | * We write the dirty pages in the range and wait until they complete | |
1812 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1813 | * multi-task, and make the performance up. See |
1814 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1815 | */ |
669249ee | 1816 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 1817 | if (ret) |
333427a5 | 1818 | goto out; |
90abccf2 | 1819 | |
29b6352b | 1820 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
c495144b | 1821 | |
2ecb7923 | 1822 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 1823 | |
aab15e8e | 1824 | /* |
885f46d8 FM |
1825 | * Before we acquired the inode's lock and the mmap lock, someone may |
1826 | * have dirtied more pages in the target range. We need to make sure | |
1827 | * that writeback for any such pages does not start while we are logging | |
1828 | * the inode, because if it does, any of the following might happen when | |
1829 | * we are not doing a full inode sync: | |
aab15e8e FM |
1830 | * |
1831 | * 1) We log an extent after its writeback finishes but before its | |
1832 | * checksums are added to the csum tree, leading to -EIO errors | |
1833 | * when attempting to read the extent after a log replay. | |
1834 | * | |
1835 | * 2) We can end up logging an extent before its writeback finishes. | |
1836 | * Therefore after the log replay we will have a file extent item | |
1837 | * pointing to an unwritten extent (and no data checksums as well). | |
1838 | * | |
1839 | * So trigger writeback for any eventual new dirty pages and then we | |
1840 | * wait for all ordered extents to complete below. | |
1841 | */ | |
1842 | ret = start_ordered_ops(inode, start, end); | |
1843 | if (ret) { | |
e5d4d75b | 1844 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
aab15e8e FM |
1845 | goto out; |
1846 | } | |
1847 | ||
cef7820d FM |
1848 | /* |
1849 | * Always check for the full sync flag while holding the inode's lock, | |
1850 | * to avoid races with other tasks. The flag must be either set all the | |
1851 | * time during logging or always off all the time while logging. | |
1852 | * We check the flag here after starting delalloc above, because when | |
1853 | * running delalloc the full sync flag may be set if we need to drop | |
1854 | * extra extent map ranges due to temporary memory allocation failures. | |
1855 | */ | |
1856 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1857 | &BTRFS_I(inode)->runtime_flags); | |
1858 | ||
669249ee | 1859 | /* |
b5e6c3e1 | 1860 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 1861 | * IO of a lower priority task while holding a transaction open. |
ba0b084a | 1862 | * |
48778179 FM |
1863 | * For a full fsync we wait for the ordered extents to complete while |
1864 | * for a fast fsync we wait just for writeback to complete, and then | |
1865 | * attach the ordered extents to the transaction so that a transaction | |
1866 | * commit waits for their completion, to avoid data loss if we fsync, | |
1867 | * the current transaction commits before the ordered extents complete | |
1868 | * and a power failure happens right after that. | |
d8e3fb10 NA |
1869 | * |
1870 | * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the | |
1871 | * logical address recorded in the ordered extent may change. We need | |
1872 | * to wait for the IO to stabilize the logical address. | |
669249ee | 1873 | */ |
d8e3fb10 | 1874 | if (full_sync || btrfs_is_zoned(fs_info)) { |
48778179 FM |
1875 | ret = btrfs_wait_ordered_range(inode, start, len); |
1876 | } else { | |
1877 | /* | |
1878 | * Get our ordered extents as soon as possible to avoid doing | |
1879 | * checksum lookups in the csum tree, and use instead the | |
1880 | * checksums attached to the ordered extents. | |
1881 | */ | |
1882 | btrfs_get_ordered_extents_for_logging(BTRFS_I(inode), | |
1883 | &ctx.ordered_extents); | |
1884 | ret = filemap_fdatawait_range(inode->i_mapping, start, end); | |
0ef8b726 | 1885 | } |
48778179 FM |
1886 | |
1887 | if (ret) | |
1888 | goto out_release_extents; | |
1889 | ||
2ecb7923 | 1890 | atomic_inc(&root->log_batch); |
257c62e1 | 1891 | |
626e9f41 | 1892 | if (skip_inode_logging(&ctx)) { |
5dc562c5 | 1893 | /* |
01327610 | 1894 | * We've had everything committed since the last time we were |
5dc562c5 JB |
1895 | * modified so clear this flag in case it was set for whatever |
1896 | * reason, it's no longer relevant. | |
1897 | */ | |
1898 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1899 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
1900 | /* |
1901 | * An ordered extent might have started before and completed | |
1902 | * already with io errors, in which case the inode was not | |
1903 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
1904 | * for any errors that might have happened since we last |
1905 | * checked called fsync. | |
0596a904 | 1906 | */ |
333427a5 | 1907 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
48778179 | 1908 | goto out_release_extents; |
15ee9bc7 | 1909 | } |
15ee9bc7 | 1910 | |
5039eddc JB |
1911 | /* |
1912 | * We use start here because we will need to wait on the IO to complete | |
1913 | * in btrfs_sync_log, which could require joining a transaction (for | |
1914 | * example checking cross references in the nocow path). If we use join | |
1915 | * here we could get into a situation where we're waiting on IO to | |
1916 | * happen that is blocked on a transaction trying to commit. With start | |
1917 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 1918 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
1919 | * from thinking they are super smart and changing this to |
1920 | * btrfs_join_transaction *cough*Josef*cough*. | |
1921 | */ | |
a22285a6 YZ |
1922 | trans = btrfs_start_transaction(root, 0); |
1923 | if (IS_ERR(trans)) { | |
1924 | ret = PTR_ERR(trans); | |
48778179 | 1925 | goto out_release_extents; |
39279cc3 | 1926 | } |
d0c2f4fa | 1927 | trans->in_fsync = true; |
e02119d5 | 1928 | |
48778179 FM |
1929 | ret = btrfs_log_dentry_safe(trans, dentry, &ctx); |
1930 | btrfs_release_log_ctx_extents(&ctx); | |
02c24a82 | 1931 | if (ret < 0) { |
a0634be5 | 1932 | /* Fallthrough and commit/free transaction. */ |
f31f09f6 | 1933 | ret = BTRFS_LOG_FORCE_COMMIT; |
02c24a82 | 1934 | } |
49eb7e46 CM |
1935 | |
1936 | /* we've logged all the items and now have a consistent | |
1937 | * version of the file in the log. It is possible that | |
1938 | * someone will come in and modify the file, but that's | |
1939 | * fine because the log is consistent on disk, and we | |
1940 | * have references to all of the file's extents | |
1941 | * | |
1942 | * It is possible that someone will come in and log the | |
1943 | * file again, but that will end up using the synchronization | |
1944 | * inside btrfs_sync_log to keep things safe. | |
1945 | */ | |
e5d4d75b | 1946 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
49eb7e46 | 1947 | |
bf7ba8ee JB |
1948 | if (ret == BTRFS_NO_LOG_SYNC) { |
1949 | ret = btrfs_end_transaction(trans); | |
1950 | goto out; | |
1951 | } | |
1952 | ||
1953 | /* We successfully logged the inode, attempt to sync the log. */ | |
1954 | if (!ret) { | |
1955 | ret = btrfs_sync_log(trans, root, &ctx); | |
0ef8b726 | 1956 | if (!ret) { |
bf7ba8ee JB |
1957 | ret = btrfs_end_transaction(trans); |
1958 | goto out; | |
48778179 | 1959 | } |
bf7ba8ee JB |
1960 | } |
1961 | ||
1962 | /* | |
1963 | * At this point we need to commit the transaction because we had | |
1964 | * btrfs_need_log_full_commit() or some other error. | |
1965 | * | |
1966 | * If we didn't do a full sync we have to stop the trans handle, wait on | |
1967 | * the ordered extents, start it again and commit the transaction. If | |
1968 | * we attempt to wait on the ordered extents here we could deadlock with | |
1969 | * something like fallocate() that is holding the extent lock trying to | |
1970 | * start a transaction while some other thread is trying to commit the | |
1971 | * transaction while we (fsync) are currently holding the transaction | |
1972 | * open. | |
1973 | */ | |
1974 | if (!full_sync) { | |
3a45bb20 | 1975 | ret = btrfs_end_transaction(trans); |
bf7ba8ee JB |
1976 | if (ret) |
1977 | goto out; | |
1978 | ret = btrfs_wait_ordered_range(inode, start, len); | |
1979 | if (ret) | |
1980 | goto out; | |
1981 | ||
1982 | /* | |
1983 | * This is safe to use here because we're only interested in | |
1984 | * making sure the transaction that had the ordered extents is | |
1985 | * committed. We aren't waiting on anything past this point, | |
1986 | * we're purely getting the transaction and committing it. | |
1987 | */ | |
1988 | trans = btrfs_attach_transaction_barrier(root); | |
1989 | if (IS_ERR(trans)) { | |
1990 | ret = PTR_ERR(trans); | |
1991 | ||
1992 | /* | |
1993 | * We committed the transaction and there's no currently | |
1994 | * running transaction, this means everything we care | |
1995 | * about made it to disk and we are done. | |
1996 | */ | |
1997 | if (ret == -ENOENT) | |
1998 | ret = 0; | |
1999 | goto out; | |
2000 | } | |
e02119d5 | 2001 | } |
bf7ba8ee JB |
2002 | |
2003 | ret = btrfs_commit_transaction(trans); | |
39279cc3 | 2004 | out: |
ebb70442 | 2005 | ASSERT(list_empty(&ctx.list)); |
e09d94c9 | 2006 | ASSERT(list_empty(&ctx.conflict_inodes)); |
333427a5 JL |
2007 | err = file_check_and_advance_wb_err(file); |
2008 | if (!ret) | |
2009 | ret = err; | |
014e4ac4 | 2010 | return ret > 0 ? -EIO : ret; |
48778179 FM |
2011 | |
2012 | out_release_extents: | |
2013 | btrfs_release_log_ctx_extents(&ctx); | |
e5d4d75b | 2014 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
48778179 | 2015 | goto out; |
39279cc3 CM |
2016 | } |
2017 | ||
f0f37e2f | 2018 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2019 | .fault = filemap_fault, |
f1820361 | 2020 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2021 | .page_mkwrite = btrfs_page_mkwrite, |
2022 | }; | |
2023 | ||
2024 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2025 | { | |
058a457e MX |
2026 | struct address_space *mapping = filp->f_mapping; |
2027 | ||
7e0a1265 | 2028 | if (!mapping->a_ops->read_folio) |
058a457e MX |
2029 | return -ENOEXEC; |
2030 | ||
9ebefb18 | 2031 | file_accessed(filp); |
058a457e | 2032 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2033 | |
9ebefb18 CM |
2034 | return 0; |
2035 | } | |
2036 | ||
35339c24 | 2037 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2038 | int slot, u64 start, u64 end) |
2039 | { | |
2040 | struct btrfs_file_extent_item *fi; | |
2041 | struct btrfs_key key; | |
2042 | ||
2043 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2044 | return 0; | |
2045 | ||
2046 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2047 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2048 | key.type != BTRFS_EXTENT_DATA_KEY) |
2049 | return 0; | |
2050 | ||
2051 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2052 | ||
2053 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2054 | return 0; | |
2055 | ||
2056 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2057 | return 0; | |
2058 | ||
2059 | if (key.offset == end) | |
2060 | return 1; | |
2061 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2062 | return 1; | |
2063 | return 0; | |
2064 | } | |
2065 | ||
a012a74e NB |
2066 | static int fill_holes(struct btrfs_trans_handle *trans, |
2067 | struct btrfs_inode *inode, | |
2068 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2069 | { |
3ffbd68c | 2070 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2071 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2072 | struct extent_buffer *leaf; |
2073 | struct btrfs_file_extent_item *fi; | |
2074 | struct extent_map *hole_em; | |
2aaa6655 JB |
2075 | struct btrfs_key key; |
2076 | int ret; | |
2077 | ||
0b246afa | 2078 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2079 | goto out; |
2080 | ||
a012a74e | 2081 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2082 | key.type = BTRFS_EXTENT_DATA_KEY; |
2083 | key.offset = offset; | |
2084 | ||
2aaa6655 | 2085 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2086 | if (ret <= 0) { |
2087 | /* | |
2088 | * We should have dropped this offset, so if we find it then | |
2089 | * something has gone horribly wrong. | |
2090 | */ | |
2091 | if (ret == 0) | |
2092 | ret = -EINVAL; | |
2aaa6655 | 2093 | return ret; |
f94480bd | 2094 | } |
2aaa6655 JB |
2095 | |
2096 | leaf = path->nodes[0]; | |
a012a74e | 2097 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2098 | u64 num_bytes; |
2099 | ||
2100 | path->slots[0]--; | |
2101 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2102 | struct btrfs_file_extent_item); | |
2103 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2104 | end - offset; | |
2105 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2106 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2107 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2108 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 2109 | btrfs_mark_buffer_dirty(trans, leaf); |
2aaa6655 JB |
2110 | goto out; |
2111 | } | |
2112 | ||
1707e26d | 2113 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2114 | u64 num_bytes; |
2115 | ||
2aaa6655 | 2116 | key.offset = offset; |
50564b65 | 2117 | btrfs_set_item_key_safe(trans, path, &key); |
2aaa6655 JB |
2118 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2119 | struct btrfs_file_extent_item); | |
2120 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2121 | offset; | |
2122 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2123 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2124 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2125 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
50564b65 | 2126 | btrfs_mark_buffer_dirty(trans, leaf); |
2aaa6655 JB |
2127 | goto out; |
2128 | } | |
2129 | btrfs_release_path(path); | |
2130 | ||
d1f68ba0 OS |
2131 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, |
2132 | end - offset); | |
2aaa6655 JB |
2133 | if (ret) |
2134 | return ret; | |
2135 | ||
2136 | out: | |
2137 | btrfs_release_path(path); | |
2138 | ||
2139 | hole_em = alloc_extent_map(); | |
2140 | if (!hole_em) { | |
4c0c8cfc | 2141 | btrfs_drop_extent_map_range(inode, offset, end - 1, false); |
23e3337f | 2142 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2143 | } else { |
2144 | hole_em->start = offset; | |
2145 | hole_em->len = end - offset; | |
cc95bef6 | 2146 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2147 | hole_em->orig_start = offset; |
2148 | ||
2149 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2150 | hole_em->block_len = 0; | |
b4939680 | 2151 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2152 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2153 | hole_em->generation = trans->transid; | |
2154 | ||
a1ba4c08 | 2155 | ret = btrfs_replace_extent_map_range(inode, hole_em, true); |
2aaa6655 JB |
2156 | free_extent_map(hole_em); |
2157 | if (ret) | |
23e3337f | 2158 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2159 | } |
2160 | ||
2161 | return 0; | |
2162 | } | |
2163 | ||
d7781546 QW |
2164 | /* |
2165 | * Find a hole extent on given inode and change start/len to the end of hole | |
2166 | * extent.(hole/vacuum extent whose em->start <= start && | |
2167 | * em->start + em->len > start) | |
2168 | * When a hole extent is found, return 1 and modify start/len. | |
2169 | */ | |
dea46d84 | 2170 | static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len) |
d7781546 | 2171 | { |
dea46d84 | 2172 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d7781546 QW |
2173 | struct extent_map *em; |
2174 | int ret = 0; | |
2175 | ||
dea46d84 | 2176 | em = btrfs_get_extent(inode, NULL, 0, |
609805d8 | 2177 | round_down(*start, fs_info->sectorsize), |
39b07b5d | 2178 | round_up(*len, fs_info->sectorsize)); |
9986277e DC |
2179 | if (IS_ERR(em)) |
2180 | return PTR_ERR(em); | |
d7781546 QW |
2181 | |
2182 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2183 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2184 | ret = 1; | |
2185 | *len = em->start + em->len > *start + *len ? | |
2186 | 0 : *start + *len - em->start - em->len; | |
2187 | *start = em->start + em->len; | |
2188 | } | |
2189 | free_extent_map(em); | |
2190 | return ret; | |
2191 | } | |
2192 | ||
55961c8a FM |
2193 | static void btrfs_punch_hole_lock_range(struct inode *inode, |
2194 | const u64 lockstart, | |
2195 | const u64 lockend, | |
2196 | struct extent_state **cached_state) | |
f27451f2 | 2197 | { |
0528476b QW |
2198 | /* |
2199 | * For subpage case, if the range is not at page boundary, we could | |
2200 | * have pages at the leading/tailing part of the range. | |
2201 | * This could lead to dead loop since filemap_range_has_page() | |
2202 | * will always return true. | |
2203 | * So here we need to do extra page alignment for | |
2204 | * filemap_range_has_page(). | |
2205 | */ | |
2206 | const u64 page_lockstart = round_up(lockstart, PAGE_SIZE); | |
2207 | const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1; | |
2208 | ||
f27451f2 | 2209 | while (1) { |
f27451f2 FM |
2210 | truncate_pagecache_range(inode, lockstart, lockend); |
2211 | ||
570eb97b JB |
2212 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2213 | cached_state); | |
f27451f2 | 2214 | /* |
55961c8a FM |
2215 | * We can't have ordered extents in the range, nor dirty/writeback |
2216 | * pages, because we have locked the inode's VFS lock in exclusive | |
2217 | * mode, we have locked the inode's i_mmap_lock in exclusive mode, | |
2218 | * we have flushed all delalloc in the range and we have waited | |
2219 | * for any ordered extents in the range to complete. | |
2220 | * We can race with anyone reading pages from this range, so after | |
2221 | * locking the range check if we have pages in the range, and if | |
2222 | * we do, unlock the range and retry. | |
f27451f2 | 2223 | */ |
55961c8a FM |
2224 | if (!filemap_range_has_page(inode->i_mapping, page_lockstart, |
2225 | page_lockend)) | |
f27451f2 | 2226 | break; |
55961c8a | 2227 | |
570eb97b JB |
2228 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2229 | cached_state); | |
f27451f2 | 2230 | } |
63c34cb4 FM |
2231 | |
2232 | btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend); | |
f27451f2 FM |
2233 | } |
2234 | ||
0cbb5bdf | 2235 | static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans, |
03fcb1ab | 2236 | struct btrfs_inode *inode, |
690a5dbf | 2237 | struct btrfs_path *path, |
bf385648 | 2238 | struct btrfs_replace_extent_info *extent_info, |
2766ff61 FM |
2239 | const u64 replace_len, |
2240 | const u64 bytes_to_drop) | |
690a5dbf | 2241 | { |
03fcb1ab NB |
2242 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2243 | struct btrfs_root *root = inode->root; | |
690a5dbf FM |
2244 | struct btrfs_file_extent_item *extent; |
2245 | struct extent_buffer *leaf; | |
2246 | struct btrfs_key key; | |
2247 | int slot; | |
2248 | struct btrfs_ref ref = { 0 }; | |
690a5dbf FM |
2249 | int ret; |
2250 | ||
bf385648 | 2251 | if (replace_len == 0) |
690a5dbf FM |
2252 | return 0; |
2253 | ||
bf385648 | 2254 | if (extent_info->disk_offset == 0 && |
2766ff61 | 2255 | btrfs_fs_incompat(fs_info, NO_HOLES)) { |
03fcb1ab | 2256 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2257 | return 0; |
2766ff61 | 2258 | } |
690a5dbf | 2259 | |
03fcb1ab | 2260 | key.objectid = btrfs_ino(inode); |
690a5dbf | 2261 | key.type = BTRFS_EXTENT_DATA_KEY; |
bf385648 | 2262 | key.offset = extent_info->file_offset; |
690a5dbf | 2263 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
fb870f6c | 2264 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf FM |
2265 | if (ret) |
2266 | return ret; | |
2267 | leaf = path->nodes[0]; | |
2268 | slot = path->slots[0]; | |
bf385648 | 2269 | write_extent_buffer(leaf, extent_info->extent_buf, |
690a5dbf | 2270 | btrfs_item_ptr_offset(leaf, slot), |
fb870f6c | 2271 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf | 2272 | extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
fb870f6c | 2273 | ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE); |
bf385648 FM |
2274 | btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset); |
2275 | btrfs_set_file_extent_num_bytes(leaf, extent, replace_len); | |
2276 | if (extent_info->is_new_extent) | |
8fccebfa | 2277 | btrfs_set_file_extent_generation(leaf, extent, trans->transid); |
50564b65 | 2278 | btrfs_mark_buffer_dirty(trans, leaf); |
690a5dbf FM |
2279 | btrfs_release_path(path); |
2280 | ||
03fcb1ab NB |
2281 | ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset, |
2282 | replace_len); | |
9ddc959e JB |
2283 | if (ret) |
2284 | return ret; | |
2285 | ||
690a5dbf | 2286 | /* If it's a hole, nothing more needs to be done. */ |
2766ff61 | 2287 | if (extent_info->disk_offset == 0) { |
03fcb1ab | 2288 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2289 | return 0; |
2766ff61 | 2290 | } |
690a5dbf | 2291 | |
03fcb1ab | 2292 | btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop); |
8fccebfa | 2293 | |
bf385648 FM |
2294 | if (extent_info->is_new_extent && extent_info->insertions == 0) { |
2295 | key.objectid = extent_info->disk_offset; | |
8fccebfa | 2296 | key.type = BTRFS_EXTENT_ITEM_KEY; |
bf385648 | 2297 | key.offset = extent_info->disk_len; |
8fccebfa | 2298 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
03fcb1ab | 2299 | btrfs_ino(inode), |
bf385648 FM |
2300 | extent_info->file_offset, |
2301 | extent_info->qgroup_reserved, | |
8fccebfa FM |
2302 | &key); |
2303 | } else { | |
2304 | u64 ref_offset; | |
2305 | ||
2306 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, | |
bf385648 | 2307 | extent_info->disk_offset, |
457cb1dd BB |
2308 | extent_info->disk_len, 0, |
2309 | root->root_key.objectid); | |
bf385648 | 2310 | ref_offset = extent_info->file_offset - extent_info->data_offset; |
8fccebfa | 2311 | btrfs_init_data_ref(&ref, root->root_key.objectid, |
f42c5da6 | 2312 | btrfs_ino(inode), ref_offset, 0, false); |
8fccebfa FM |
2313 | ret = btrfs_inc_extent_ref(trans, &ref); |
2314 | } | |
2315 | ||
bf385648 | 2316 | extent_info->insertions++; |
690a5dbf FM |
2317 | |
2318 | return ret; | |
2319 | } | |
2320 | ||
9cba40a6 FM |
2321 | /* |
2322 | * The respective range must have been previously locked, as well as the inode. | |
2323 | * The end offset is inclusive (last byte of the range). | |
bf385648 FM |
2324 | * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing |
2325 | * the file range with an extent. | |
2326 | * When not punching a hole, we don't want to end up in a state where we dropped | |
2327 | * extents without inserting a new one, so we must abort the transaction to avoid | |
2328 | * a corruption. | |
9cba40a6 | 2329 | */ |
bfc78479 NB |
2330 | int btrfs_replace_file_extents(struct btrfs_inode *inode, |
2331 | struct btrfs_path *path, const u64 start, | |
2332 | const u64 end, | |
2333 | struct btrfs_replace_extent_info *extent_info, | |
2334 | struct btrfs_trans_handle **trans_out) | |
9cba40a6 | 2335 | { |
5893dfb9 | 2336 | struct btrfs_drop_extents_args drop_args = { 0 }; |
bfc78479 NB |
2337 | struct btrfs_root *root = inode->root; |
2338 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2bd36e7b | 2339 | u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1); |
bfc78479 | 2340 | u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize); |
9cba40a6 FM |
2341 | struct btrfs_trans_handle *trans = NULL; |
2342 | struct btrfs_block_rsv *rsv; | |
2343 | unsigned int rsv_count; | |
2344 | u64 cur_offset; | |
9cba40a6 FM |
2345 | u64 len = end - start; |
2346 | int ret = 0; | |
2347 | ||
2348 | if (end <= start) | |
2349 | return -EINVAL; | |
2350 | ||
2351 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); | |
2352 | if (!rsv) { | |
2353 | ret = -ENOMEM; | |
2354 | goto out; | |
2355 | } | |
2bd36e7b | 2356 | rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1); |
710d5921 | 2357 | rsv->failfast = true; |
9cba40a6 FM |
2358 | |
2359 | /* | |
2360 | * 1 - update the inode | |
2361 | * 1 - removing the extents in the range | |
bf385648 FM |
2362 | * 1 - adding the hole extent if no_holes isn't set or if we are |
2363 | * replacing the range with a new extent | |
9cba40a6 | 2364 | */ |
bf385648 | 2365 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info) |
690a5dbf FM |
2366 | rsv_count = 3; |
2367 | else | |
2368 | rsv_count = 2; | |
2369 | ||
9cba40a6 FM |
2370 | trans = btrfs_start_transaction(root, rsv_count); |
2371 | if (IS_ERR(trans)) { | |
2372 | ret = PTR_ERR(trans); | |
2373 | trans = NULL; | |
2374 | goto out_free; | |
2375 | } | |
2376 | ||
2377 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, | |
2378 | min_size, false); | |
650c9cab FM |
2379 | if (WARN_ON(ret)) |
2380 | goto out_trans; | |
9cba40a6 FM |
2381 | trans->block_rsv = rsv; |
2382 | ||
2383 | cur_offset = start; | |
5893dfb9 FM |
2384 | drop_args.path = path; |
2385 | drop_args.end = end + 1; | |
2386 | drop_args.drop_cache = true; | |
9cba40a6 | 2387 | while (cur_offset < end) { |
5893dfb9 | 2388 | drop_args.start = cur_offset; |
bfc78479 | 2389 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
2766ff61 FM |
2390 | /* If we are punching a hole decrement the inode's byte count */ |
2391 | if (!extent_info) | |
bfc78479 | 2392 | btrfs_update_inode_bytes(inode, 0, |
2766ff61 | 2393 | drop_args.bytes_found); |
690a5dbf FM |
2394 | if (ret != -ENOSPC) { |
2395 | /* | |
4afb912f JB |
2396 | * The only time we don't want to abort is if we are |
2397 | * attempting to clone a partial inline extent, in which | |
2398 | * case we'll get EOPNOTSUPP. However if we aren't | |
2399 | * clone we need to abort no matter what, because if we | |
2400 | * got EOPNOTSUPP via prealloc then we messed up and | |
2401 | * need to abort. | |
690a5dbf | 2402 | */ |
4afb912f JB |
2403 | if (ret && |
2404 | (ret != -EOPNOTSUPP || | |
2405 | (extent_info && extent_info->is_new_extent))) | |
690a5dbf | 2406 | btrfs_abort_transaction(trans, ret); |
9cba40a6 | 2407 | break; |
690a5dbf | 2408 | } |
9cba40a6 FM |
2409 | |
2410 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2411 | ||
5893dfb9 | 2412 | if (!extent_info && cur_offset < drop_args.drop_end && |
690a5dbf | 2413 | cur_offset < ino_size) { |
bfc78479 NB |
2414 | ret = fill_holes(trans, inode, path, cur_offset, |
2415 | drop_args.drop_end); | |
9cba40a6 FM |
2416 | if (ret) { |
2417 | /* | |
2418 | * If we failed then we didn't insert our hole | |
2419 | * entries for the area we dropped, so now the | |
2420 | * fs is corrupted, so we must abort the | |
2421 | * transaction. | |
2422 | */ | |
2423 | btrfs_abort_transaction(trans, ret); | |
2424 | break; | |
2425 | } | |
5893dfb9 | 2426 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e JB |
2427 | /* |
2428 | * We are past the i_size here, but since we didn't | |
2429 | * insert holes we need to clear the mapped area so we | |
2430 | * know to not set disk_i_size in this area until a new | |
2431 | * file extent is inserted here. | |
2432 | */ | |
bfc78479 | 2433 | ret = btrfs_inode_clear_file_extent_range(inode, |
5893dfb9 FM |
2434 | cur_offset, |
2435 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2436 | if (ret) { |
2437 | /* | |
2438 | * We couldn't clear our area, so we could | |
2439 | * presumably adjust up and corrupt the fs, so | |
2440 | * we need to abort. | |
2441 | */ | |
2442 | btrfs_abort_transaction(trans, ret); | |
2443 | break; | |
2444 | } | |
9cba40a6 FM |
2445 | } |
2446 | ||
5893dfb9 FM |
2447 | if (extent_info && |
2448 | drop_args.drop_end > extent_info->file_offset) { | |
2449 | u64 replace_len = drop_args.drop_end - | |
2450 | extent_info->file_offset; | |
690a5dbf | 2451 | |
bfc78479 NB |
2452 | ret = btrfs_insert_replace_extent(trans, inode, path, |
2453 | extent_info, replace_len, | |
03fcb1ab | 2454 | drop_args.bytes_found); |
690a5dbf FM |
2455 | if (ret) { |
2456 | btrfs_abort_transaction(trans, ret); | |
2457 | break; | |
2458 | } | |
bf385648 FM |
2459 | extent_info->data_len -= replace_len; |
2460 | extent_info->data_offset += replace_len; | |
2461 | extent_info->file_offset += replace_len; | |
690a5dbf FM |
2462 | } |
2463 | ||
983d8209 FM |
2464 | /* |
2465 | * We are releasing our handle on the transaction, balance the | |
2466 | * dirty pages of the btree inode and flush delayed items, and | |
2467 | * then get a new transaction handle, which may now point to a | |
2468 | * new transaction in case someone else may have committed the | |
2469 | * transaction we used to replace/drop file extent items. So | |
2470 | * bump the inode's iversion and update mtime and ctime except | |
2471 | * if we are called from a dedupe context. This is because a | |
2472 | * power failure/crash may happen after the transaction is | |
2473 | * committed and before we finish replacing/dropping all the | |
2474 | * file extent items we need. | |
2475 | */ | |
2476 | inode_inc_iversion(&inode->vfs_inode); | |
2477 | ||
2a9462de JL |
2478 | if (!extent_info || extent_info->update_times) |
2479 | inode->vfs_inode.i_mtime = inode_set_ctime_current(&inode->vfs_inode); | |
983d8209 | 2480 | |
8b9d0322 | 2481 | ret = btrfs_update_inode(trans, inode); |
9cba40a6 FM |
2482 | if (ret) |
2483 | break; | |
2484 | ||
2485 | btrfs_end_transaction(trans); | |
2486 | btrfs_btree_balance_dirty(fs_info); | |
2487 | ||
2488 | trans = btrfs_start_transaction(root, rsv_count); | |
2489 | if (IS_ERR(trans)) { | |
2490 | ret = PTR_ERR(trans); | |
2491 | trans = NULL; | |
2492 | break; | |
2493 | } | |
2494 | ||
2495 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, | |
2496 | rsv, min_size, false); | |
650c9cab FM |
2497 | if (WARN_ON(ret)) |
2498 | break; | |
9cba40a6 FM |
2499 | trans->block_rsv = rsv; |
2500 | ||
3227788c BC |
2501 | cur_offset = drop_args.drop_end; |
2502 | len = end - cur_offset; | |
2503 | if (!extent_info && len) { | |
bfc78479 | 2504 | ret = find_first_non_hole(inode, &cur_offset, &len); |
690a5dbf FM |
2505 | if (unlikely(ret < 0)) |
2506 | break; | |
2507 | if (ret && !len) { | |
2508 | ret = 0; | |
2509 | break; | |
2510 | } | |
9cba40a6 FM |
2511 | } |
2512 | } | |
2513 | ||
690a5dbf FM |
2514 | /* |
2515 | * If we were cloning, force the next fsync to be a full one since we | |
2516 | * we replaced (or just dropped in the case of cloning holes when | |
e2b84217 FM |
2517 | * NO_HOLES is enabled) file extent items and did not setup new extent |
2518 | * maps for the replacement extents (or holes). | |
690a5dbf | 2519 | */ |
bf385648 | 2520 | if (extent_info && !extent_info->is_new_extent) |
23e3337f | 2521 | btrfs_set_inode_full_sync(inode); |
690a5dbf | 2522 | |
9cba40a6 FM |
2523 | if (ret) |
2524 | goto out_trans; | |
2525 | ||
2526 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2527 | /* | |
2528 | * If we are using the NO_HOLES feature we might have had already an | |
2529 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2530 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2531 | * represent holes, drop_end can be less than lockend and so we must | |
2532 | * make sure we have an extent map representing the existing hole (the | |
2533 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2534 | * map representing the existing hole), otherwise the fast fsync path | |
2535 | * will not record the existence of the hole region | |
2536 | * [existing_hole_start, lockend]. | |
2537 | */ | |
5893dfb9 FM |
2538 | if (drop_args.drop_end <= end) |
2539 | drop_args.drop_end = end + 1; | |
9cba40a6 FM |
2540 | /* |
2541 | * Don't insert file hole extent item if it's for a range beyond eof | |
2542 | * (because it's useless) or if it represents a 0 bytes range (when | |
2543 | * cur_offset == drop_end). | |
2544 | */ | |
5893dfb9 FM |
2545 | if (!extent_info && cur_offset < ino_size && |
2546 | cur_offset < drop_args.drop_end) { | |
bfc78479 NB |
2547 | ret = fill_holes(trans, inode, path, cur_offset, |
2548 | drop_args.drop_end); | |
9cba40a6 FM |
2549 | if (ret) { |
2550 | /* Same comment as above. */ | |
2551 | btrfs_abort_transaction(trans, ret); | |
2552 | goto out_trans; | |
2553 | } | |
5893dfb9 | 2554 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e | 2555 | /* See the comment in the loop above for the reasoning here. */ |
bfc78479 NB |
2556 | ret = btrfs_inode_clear_file_extent_range(inode, cur_offset, |
2557 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2558 | if (ret) { |
2559 | btrfs_abort_transaction(trans, ret); | |
2560 | goto out_trans; | |
2561 | } | |
2562 | ||
9cba40a6 | 2563 | } |
bf385648 | 2564 | if (extent_info) { |
bfc78479 | 2565 | ret = btrfs_insert_replace_extent(trans, inode, path, |
03fcb1ab NB |
2566 | extent_info, extent_info->data_len, |
2567 | drop_args.bytes_found); | |
690a5dbf FM |
2568 | if (ret) { |
2569 | btrfs_abort_transaction(trans, ret); | |
2570 | goto out_trans; | |
2571 | } | |
2572 | } | |
9cba40a6 FM |
2573 | |
2574 | out_trans: | |
2575 | if (!trans) | |
2576 | goto out_free; | |
2577 | ||
2578 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2579 | if (ret) | |
2580 | btrfs_end_transaction(trans); | |
2581 | else | |
2582 | *trans_out = trans; | |
2583 | out_free: | |
2584 | btrfs_free_block_rsv(fs_info, rsv); | |
2585 | out: | |
2586 | return ret; | |
2587 | } | |
2588 | ||
05fd9564 | 2589 | static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len) |
2aaa6655 | 2590 | { |
05fd9564 | 2591 | struct inode *inode = file_inode(file); |
0b246afa | 2592 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2593 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2594 | struct extent_state *cached_state = NULL; | |
2595 | struct btrfs_path *path; | |
9cba40a6 | 2596 | struct btrfs_trans_handle *trans = NULL; |
d7781546 QW |
2597 | u64 lockstart; |
2598 | u64 lockend; | |
2599 | u64 tail_start; | |
2600 | u64 tail_len; | |
2601 | u64 orig_start = offset; | |
2aaa6655 | 2602 | int ret = 0; |
9703fefe | 2603 | bool same_block; |
a1a50f60 | 2604 | u64 ino_size; |
9703fefe | 2605 | bool truncated_block = false; |
e8c1c76e | 2606 | bool updated_inode = false; |
2aaa6655 | 2607 | |
29b6352b | 2608 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
bd6526d0 | 2609 | |
0ef8b726 JB |
2610 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2611 | if (ret) | |
bd6526d0 | 2612 | goto out_only_mutex; |
2aaa6655 | 2613 | |
0b246afa | 2614 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
dea46d84 | 2615 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2616 | if (ret < 0) |
2617 | goto out_only_mutex; | |
2618 | if (ret && !len) { | |
2619 | /* Already in a large hole */ | |
2620 | ret = 0; | |
2621 | goto out_only_mutex; | |
2622 | } | |
2623 | ||
05fd9564 DW |
2624 | ret = file_modified(file); |
2625 | if (ret) | |
2626 | goto out_only_mutex; | |
2627 | ||
ee8ba05c JB |
2628 | lockstart = round_up(offset, fs_info->sectorsize); |
2629 | lockend = round_down(offset + len, fs_info->sectorsize) - 1; | |
0b246afa JM |
2630 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2631 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2632 | /* |
9703fefe | 2633 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2634 | * because we are sure there is no data there. |
2635 | */ | |
2aaa6655 | 2636 | /* |
9703fefe CR |
2637 | * Only do this if we are in the same block and we aren't doing the |
2638 | * entire block. | |
2aaa6655 | 2639 | */ |
0b246afa | 2640 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2641 | if (offset < ino_size) { |
9703fefe | 2642 | truncated_block = true; |
217f42eb NB |
2643 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2644 | 0); | |
e8c1c76e FM |
2645 | } else { |
2646 | ret = 0; | |
2647 | } | |
d7781546 | 2648 | goto out_only_mutex; |
2aaa6655 JB |
2649 | } |
2650 | ||
9703fefe | 2651 | /* zero back part of the first block */ |
12870f1c | 2652 | if (offset < ino_size) { |
9703fefe | 2653 | truncated_block = true; |
217f42eb | 2654 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
7426cc04 | 2655 | if (ret) { |
e5d4d75b | 2656 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
7426cc04 MX |
2657 | return ret; |
2658 | } | |
2aaa6655 JB |
2659 | } |
2660 | ||
d7781546 QW |
2661 | /* Check the aligned pages after the first unaligned page, |
2662 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2663 | * including several following pages are already in holes, |
d7781546 QW |
2664 | * the extra check can be skipped */ |
2665 | if (offset == orig_start) { | |
2666 | /* after truncate page, check hole again */ | |
2667 | len = offset + len - lockstart; | |
2668 | offset = lockstart; | |
dea46d84 | 2669 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
2670 | if (ret < 0) |
2671 | goto out_only_mutex; | |
2672 | if (ret && !len) { | |
2673 | ret = 0; | |
2674 | goto out_only_mutex; | |
2675 | } | |
2676 | lockstart = offset; | |
2677 | } | |
2678 | ||
2679 | /* Check the tail unaligned part is in a hole */ | |
2680 | tail_start = lockend + 1; | |
2681 | tail_len = offset + len - tail_start; | |
2682 | if (tail_len) { | |
dea46d84 | 2683 | ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len); |
d7781546 QW |
2684 | if (unlikely(ret < 0)) |
2685 | goto out_only_mutex; | |
2686 | if (!ret) { | |
2687 | /* zero the front end of the last page */ | |
2688 | if (tail_start + tail_len < ino_size) { | |
9703fefe | 2689 | truncated_block = true; |
217f42eb | 2690 | ret = btrfs_truncate_block(BTRFS_I(inode), |
9703fefe CR |
2691 | tail_start + tail_len, |
2692 | 0, 1); | |
d7781546 QW |
2693 | if (ret) |
2694 | goto out_only_mutex; | |
51f395ad | 2695 | } |
0061280d | 2696 | } |
2aaa6655 JB |
2697 | } |
2698 | ||
2699 | if (lockend < lockstart) { | |
e8c1c76e FM |
2700 | ret = 0; |
2701 | goto out_only_mutex; | |
2aaa6655 JB |
2702 | } |
2703 | ||
55961c8a | 2704 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state); |
2aaa6655 JB |
2705 | |
2706 | path = btrfs_alloc_path(); | |
2707 | if (!path) { | |
2708 | ret = -ENOMEM; | |
2709 | goto out; | |
2710 | } | |
2711 | ||
bfc78479 NB |
2712 | ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart, |
2713 | lockend, NULL, &trans); | |
9cba40a6 FM |
2714 | btrfs_free_path(path); |
2715 | if (ret) | |
2716 | goto out; | |
2aaa6655 | 2717 | |
9cba40a6 | 2718 | ASSERT(trans != NULL); |
e1f5790e | 2719 | inode_inc_iversion(inode); |
2a9462de | 2720 | inode->i_mtime = inode_set_ctime_current(inode); |
8b9d0322 | 2721 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
e8c1c76e | 2722 | updated_inode = true; |
3a45bb20 | 2723 | btrfs_end_transaction(trans); |
2ff7e61e | 2724 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2725 | out: |
570eb97b JB |
2726 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2727 | &cached_state); | |
d7781546 | 2728 | out_only_mutex: |
9cba40a6 | 2729 | if (!updated_inode && truncated_block && !ret) { |
e8c1c76e FM |
2730 | /* |
2731 | * If we only end up zeroing part of a page, we still need to | |
2732 | * update the inode item, so that all the time fields are | |
2733 | * updated as well as the necessary btrfs inode in memory fields | |
2734 | * for detecting, at fsync time, if the inode isn't yet in the | |
2735 | * log tree or it's there but not up to date. | |
2736 | */ | |
2a9462de | 2737 | struct timespec64 now = inode_set_ctime_current(inode); |
17900668 FM |
2738 | |
2739 | inode_inc_iversion(inode); | |
2740 | inode->i_mtime = now; | |
e8c1c76e FM |
2741 | trans = btrfs_start_transaction(root, 1); |
2742 | if (IS_ERR(trans)) { | |
9cba40a6 | 2743 | ret = PTR_ERR(trans); |
e8c1c76e | 2744 | } else { |
9cba40a6 FM |
2745 | int ret2; |
2746 | ||
8b9d0322 | 2747 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
9cba40a6 FM |
2748 | ret2 = btrfs_end_transaction(trans); |
2749 | if (!ret) | |
2750 | ret = ret2; | |
e8c1c76e FM |
2751 | } |
2752 | } | |
e5d4d75b | 2753 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
9cba40a6 | 2754 | return ret; |
2aaa6655 JB |
2755 | } |
2756 | ||
14524a84 QW |
2757 | /* Helper structure to record which range is already reserved */ |
2758 | struct falloc_range { | |
2759 | struct list_head list; | |
2760 | u64 start; | |
2761 | u64 len; | |
2762 | }; | |
2763 | ||
2764 | /* | |
2765 | * Helper function to add falloc range | |
2766 | * | |
2767 | * Caller should have locked the larger range of extent containing | |
2768 | * [start, len) | |
2769 | */ | |
2770 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2771 | { | |
14524a84 QW |
2772 | struct falloc_range *range = NULL; |
2773 | ||
77d25534 NB |
2774 | if (!list_empty(head)) { |
2775 | /* | |
2776 | * As fallocate iterates by bytenr order, we only need to check | |
2777 | * the last range. | |
2778 | */ | |
2779 | range = list_last_entry(head, struct falloc_range, list); | |
2780 | if (range->start + range->len == start) { | |
2781 | range->len += len; | |
2782 | return 0; | |
2783 | } | |
14524a84 | 2784 | } |
77d25534 | 2785 | |
32fc932e | 2786 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2787 | if (!range) |
2788 | return -ENOMEM; | |
2789 | range->start = start; | |
2790 | range->len = len; | |
2791 | list_add_tail(&range->list, head); | |
2792 | return 0; | |
2793 | } | |
2794 | ||
f27451f2 FM |
2795 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2796 | const u64 end, | |
2797 | const int mode) | |
2798 | { | |
2799 | struct btrfs_trans_handle *trans; | |
2800 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2801 | int ret; | |
2802 | int ret2; | |
2803 | ||
2804 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2805 | return 0; | |
2806 | ||
2807 | trans = btrfs_start_transaction(root, 1); | |
2808 | if (IS_ERR(trans)) | |
2809 | return PTR_ERR(trans); | |
2810 | ||
2a9462de | 2811 | inode_set_ctime_current(inode); |
f27451f2 | 2812 | i_size_write(inode, end); |
76aea537 | 2813 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
8b9d0322 | 2814 | ret = btrfs_update_inode(trans, BTRFS_I(inode)); |
f27451f2 FM |
2815 | ret2 = btrfs_end_transaction(trans); |
2816 | ||
2817 | return ret ? ret : ret2; | |
2818 | } | |
2819 | ||
81fdf638 | 2820 | enum { |
f262fa8d DS |
2821 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
2822 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
2823 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
2824 | }; |
2825 | ||
948dfeb8 | 2826 | static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode, |
f27451f2 FM |
2827 | u64 offset) |
2828 | { | |
ee8ba05c | 2829 | const u64 sectorsize = inode->root->fs_info->sectorsize; |
f27451f2 | 2830 | struct extent_map *em; |
81fdf638 | 2831 | int ret; |
f27451f2 FM |
2832 | |
2833 | offset = round_down(offset, sectorsize); | |
948dfeb8 | 2834 | em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize); |
f27451f2 FM |
2835 | if (IS_ERR(em)) |
2836 | return PTR_ERR(em); | |
2837 | ||
2838 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
2839 | ret = RANGE_BOUNDARY_HOLE; |
2840 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2841 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
2842 | else | |
2843 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
2844 | |
2845 | free_extent_map(em); | |
2846 | return ret; | |
2847 | } | |
2848 | ||
2849 | static int btrfs_zero_range(struct inode *inode, | |
2850 | loff_t offset, | |
2851 | loff_t len, | |
2852 | const int mode) | |
2853 | { | |
2854 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2855 | struct extent_map *em; | |
2856 | struct extent_changeset *data_reserved = NULL; | |
2857 | int ret; | |
2858 | u64 alloc_hint = 0; | |
ee8ba05c | 2859 | const u64 sectorsize = fs_info->sectorsize; |
f27451f2 FM |
2860 | u64 alloc_start = round_down(offset, sectorsize); |
2861 | u64 alloc_end = round_up(offset + len, sectorsize); | |
2862 | u64 bytes_to_reserve = 0; | |
2863 | bool space_reserved = false; | |
2864 | ||
39b07b5d OS |
2865 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2866 | alloc_end - alloc_start); | |
f27451f2 FM |
2867 | if (IS_ERR(em)) { |
2868 | ret = PTR_ERR(em); | |
2869 | goto out; | |
2870 | } | |
2871 | ||
2872 | /* | |
2873 | * Avoid hole punching and extent allocation for some cases. More cases | |
2874 | * could be considered, but these are unlikely common and we keep things | |
2875 | * as simple as possible for now. Also, intentionally, if the target | |
2876 | * range contains one or more prealloc extents together with regular | |
2877 | * extents and holes, we drop all the existing extents and allocate a | |
2878 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
2879 | */ | |
2880 | if (em->start <= alloc_start && | |
2881 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2882 | const u64 em_end = em->start + em->len; | |
2883 | ||
2884 | if (em_end >= offset + len) { | |
2885 | /* | |
2886 | * The whole range is already a prealloc extent, | |
2887 | * do nothing except updating the inode's i_size if | |
2888 | * needed. | |
2889 | */ | |
2890 | free_extent_map(em); | |
2891 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2892 | mode); | |
2893 | goto out; | |
2894 | } | |
2895 | /* | |
2896 | * Part of the range is already a prealloc extent, so operate | |
2897 | * only on the remaining part of the range. | |
2898 | */ | |
2899 | alloc_start = em_end; | |
2900 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
2901 | len = offset + len - alloc_start; | |
2902 | offset = alloc_start; | |
2903 | alloc_hint = em->block_start + em->len; | |
2904 | } | |
2905 | free_extent_map(em); | |
2906 | ||
2907 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
2908 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
39b07b5d OS |
2909 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
2910 | sectorsize); | |
f27451f2 FM |
2911 | if (IS_ERR(em)) { |
2912 | ret = PTR_ERR(em); | |
2913 | goto out; | |
2914 | } | |
2915 | ||
2916 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2917 | free_extent_map(em); | |
2918 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2919 | mode); | |
2920 | goto out; | |
2921 | } | |
2922 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
2923 | free_extent_map(em); | |
217f42eb NB |
2924 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
2925 | 0); | |
f27451f2 FM |
2926 | if (!ret) |
2927 | ret = btrfs_fallocate_update_isize(inode, | |
2928 | offset + len, | |
2929 | mode); | |
2930 | return ret; | |
2931 | } | |
2932 | free_extent_map(em); | |
2933 | alloc_start = round_down(offset, sectorsize); | |
2934 | alloc_end = alloc_start + sectorsize; | |
2935 | goto reserve_space; | |
2936 | } | |
2937 | ||
2938 | alloc_start = round_up(offset, sectorsize); | |
2939 | alloc_end = round_down(offset + len, sectorsize); | |
2940 | ||
2941 | /* | |
2942 | * For unaligned ranges, check the pages at the boundaries, they might | |
2943 | * map to an extent, in which case we need to partially zero them, or | |
2944 | * they might map to a hole, in which case we need our allocation range | |
2945 | * to cover them. | |
2946 | */ | |
2947 | if (!IS_ALIGNED(offset, sectorsize)) { | |
948dfeb8 NB |
2948 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
2949 | offset); | |
f27451f2 FM |
2950 | if (ret < 0) |
2951 | goto out; | |
81fdf638 | 2952 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2953 | alloc_start = round_down(offset, sectorsize); |
2954 | ret = 0; | |
81fdf638 | 2955 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb | 2956 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
f27451f2 FM |
2957 | if (ret) |
2958 | goto out; | |
81fdf638 FM |
2959 | } else { |
2960 | ret = 0; | |
f27451f2 FM |
2961 | } |
2962 | } | |
2963 | ||
2964 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
948dfeb8 | 2965 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
f27451f2 FM |
2966 | offset + len); |
2967 | if (ret < 0) | |
2968 | goto out; | |
81fdf638 | 2969 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2970 | alloc_end = round_up(offset + len, sectorsize); |
2971 | ret = 0; | |
81fdf638 | 2972 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb NB |
2973 | ret = btrfs_truncate_block(BTRFS_I(inode), offset + len, |
2974 | 0, 1); | |
f27451f2 FM |
2975 | if (ret) |
2976 | goto out; | |
81fdf638 FM |
2977 | } else { |
2978 | ret = 0; | |
f27451f2 FM |
2979 | } |
2980 | } | |
2981 | ||
2982 | reserve_space: | |
2983 | if (alloc_start < alloc_end) { | |
2984 | struct extent_state *cached_state = NULL; | |
2985 | const u64 lockstart = alloc_start; | |
2986 | const u64 lockend = alloc_end - 1; | |
2987 | ||
2988 | bytes_to_reserve = alloc_end - alloc_start; | |
2989 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
2990 | bytes_to_reserve); | |
2991 | if (ret < 0) | |
2992 | goto out; | |
2993 | space_reserved = true; | |
55961c8a FM |
2994 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
2995 | &cached_state); | |
7661a3e0 | 2996 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved, |
a7f8b1c2 | 2997 | alloc_start, bytes_to_reserve); |
4f6a49de | 2998 | if (ret) { |
570eb97b JB |
2999 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
3000 | lockend, &cached_state); | |
a7f8b1c2 | 3001 | goto out; |
4f6a49de | 3002 | } |
f27451f2 FM |
3003 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, |
3004 | alloc_end - alloc_start, | |
3005 | i_blocksize(inode), | |
3006 | offset + len, &alloc_hint); | |
570eb97b JB |
3007 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3008 | &cached_state); | |
f27451f2 | 3009 | /* btrfs_prealloc_file_range releases reserved space on error */ |
9f13ce74 | 3010 | if (ret) { |
f27451f2 | 3011 | space_reserved = false; |
9f13ce74 FM |
3012 | goto out; |
3013 | } | |
f27451f2 | 3014 | } |
9f13ce74 | 3015 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3016 | out: |
3017 | if (ret && space_reserved) | |
25ce28ca | 3018 | btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved, |
f27451f2 FM |
3019 | alloc_start, bytes_to_reserve); |
3020 | extent_changeset_free(data_reserved); | |
3021 | ||
3022 | return ret; | |
3023 | } | |
3024 | ||
2fe17c10 CH |
3025 | static long btrfs_fallocate(struct file *file, int mode, |
3026 | loff_t offset, loff_t len) | |
3027 | { | |
496ad9aa | 3028 | struct inode *inode = file_inode(file); |
2fe17c10 | 3029 | struct extent_state *cached_state = NULL; |
364ecf36 | 3030 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3031 | struct falloc_range *range; |
3032 | struct falloc_range *tmp; | |
84af994b | 3033 | LIST_HEAD(reserve_list); |
2fe17c10 CH |
3034 | u64 cur_offset; |
3035 | u64 last_byte; | |
3036 | u64 alloc_start; | |
3037 | u64 alloc_end; | |
3038 | u64 alloc_hint = 0; | |
3039 | u64 locked_end; | |
14524a84 | 3040 | u64 actual_end = 0; |
47e1d1c7 FM |
3041 | u64 data_space_needed = 0; |
3042 | u64 data_space_reserved = 0; | |
3043 | u64 qgroup_reserved = 0; | |
2fe17c10 | 3044 | struct extent_map *em; |
ee8ba05c | 3045 | int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize; |
2fe17c10 CH |
3046 | int ret; |
3047 | ||
f1569c4c NA |
3048 | /* Do not allow fallocate in ZONED mode */ |
3049 | if (btrfs_is_zoned(btrfs_sb(inode->i_sb))) | |
3050 | return -EOPNOTSUPP; | |
3051 | ||
797f4277 MX |
3052 | alloc_start = round_down(offset, blocksize); |
3053 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3054 | cur_offset = alloc_start; |
2fe17c10 | 3055 | |
2aaa6655 | 3056 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3057 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3058 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3059 | return -EOPNOTSUPP; |
3060 | ||
2aaa6655 | 3061 | if (mode & FALLOC_FL_PUNCH_HOLE) |
05fd9564 | 3062 | return btrfs_punch_hole(file, offset, len); |
2aaa6655 | 3063 | |
29b6352b | 3064 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
2a162ce9 DI |
3065 | |
3066 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3067 | ret = inode_newsize_ok(inode, offset + len); | |
3068 | if (ret) | |
3069 | goto out; | |
3070 | } | |
2fe17c10 | 3071 | |
05fd9564 DW |
3072 | ret = file_modified(file); |
3073 | if (ret) | |
3074 | goto out; | |
3075 | ||
14524a84 QW |
3076 | /* |
3077 | * TODO: Move these two operations after we have checked | |
3078 | * accurate reserved space, or fallocate can still fail but | |
3079 | * with page truncated or size expanded. | |
3080 | * | |
3081 | * But that's a minor problem and won't do much harm BTW. | |
3082 | */ | |
2fe17c10 | 3083 | if (alloc_start > inode->i_size) { |
b06359a3 | 3084 | ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode), |
a41ad394 | 3085 | alloc_start); |
2fe17c10 CH |
3086 | if (ret) |
3087 | goto out; | |
0f6925fa | 3088 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3089 | /* |
3090 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3091 | * need to zero out the end of the block if i_size lands in the |
3092 | * middle of a block. | |
a71754fc | 3093 | */ |
217f42eb | 3094 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
a71754fc JB |
3095 | if (ret) |
3096 | goto out; | |
2fe17c10 CH |
3097 | } |
3098 | ||
a71754fc | 3099 | /* |
ffa8fc60 FM |
3100 | * We have locked the inode at the VFS level (in exclusive mode) and we |
3101 | * have locked the i_mmap_lock lock (in exclusive mode). Now before | |
3102 | * locking the file range, flush all dealloc in the range and wait for | |
3103 | * all ordered extents in the range to complete. After this we can lock | |
3104 | * the file range and, due to the previous locking we did, we know there | |
3105 | * can't be more delalloc or ordered extents in the range. | |
a71754fc | 3106 | */ |
0ef8b726 JB |
3107 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3108 | alloc_end - alloc_start); | |
3109 | if (ret) | |
3110 | goto out; | |
a71754fc | 3111 | |
f27451f2 FM |
3112 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3113 | ret = btrfs_zero_range(inode, offset, len, mode); | |
e5d4d75b | 3114 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
f27451f2 FM |
3115 | return ret; |
3116 | } | |
3117 | ||
2fe17c10 | 3118 | locked_end = alloc_end - 1; |
570eb97b JB |
3119 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3120 | &cached_state); | |
2fe17c10 | 3121 | |
63c34cb4 FM |
3122 | btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end); |
3123 | ||
14524a84 | 3124 | /* First, check if we exceed the qgroup limit */ |
6b7d6e93 | 3125 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3126 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
39b07b5d | 3127 | alloc_end - cur_offset); |
9986277e DC |
3128 | if (IS_ERR(em)) { |
3129 | ret = PTR_ERR(em); | |
79787eaa JM |
3130 | break; |
3131 | } | |
2fe17c10 | 3132 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3133 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3134 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3135 | if (em->block_start == EXTENT_MAP_HOLE || |
3136 | (cur_offset >= inode->i_size && | |
3137 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
47e1d1c7 FM |
3138 | const u64 range_len = last_byte - cur_offset; |
3139 | ||
3140 | ret = add_falloc_range(&reserve_list, cur_offset, range_len); | |
14524a84 QW |
3141 | if (ret < 0) { |
3142 | free_extent_map(em); | |
3143 | break; | |
3d850dd4 | 3144 | } |
7661a3e0 | 3145 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), |
47e1d1c7 | 3146 | &data_reserved, cur_offset, range_len); |
be2d253c FM |
3147 | if (ret < 0) { |
3148 | free_extent_map(em); | |
14524a84 | 3149 | break; |
be2d253c | 3150 | } |
47e1d1c7 FM |
3151 | qgroup_reserved += range_len; |
3152 | data_space_needed += range_len; | |
2fe17c10 CH |
3153 | } |
3154 | free_extent_map(em); | |
2fe17c10 | 3155 | cur_offset = last_byte; |
14524a84 QW |
3156 | } |
3157 | ||
47e1d1c7 FM |
3158 | if (!ret && data_space_needed > 0) { |
3159 | /* | |
3160 | * We are safe to reserve space here as we can't have delalloc | |
3161 | * in the range, see above. | |
3162 | */ | |
3163 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3164 | data_space_needed); | |
3165 | if (!ret) | |
3166 | data_space_reserved = data_space_needed; | |
3167 | } | |
3168 | ||
14524a84 QW |
3169 | /* |
3170 | * If ret is still 0, means we're OK to fallocate. | |
3171 | * Or just cleanup the list and exit. | |
3172 | */ | |
3173 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
47e1d1c7 | 3174 | if (!ret) { |
14524a84 QW |
3175 | ret = btrfs_prealloc_file_range(inode, mode, |
3176 | range->start, | |
93407472 | 3177 | range->len, i_blocksize(inode), |
14524a84 | 3178 | offset + len, &alloc_hint); |
47e1d1c7 FM |
3179 | /* |
3180 | * btrfs_prealloc_file_range() releases space even | |
3181 | * if it returns an error. | |
3182 | */ | |
3183 | data_space_reserved -= range->len; | |
3184 | qgroup_reserved -= range->len; | |
3185 | } else if (data_space_reserved > 0) { | |
25ce28ca | 3186 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
47e1d1c7 FM |
3187 | data_reserved, range->start, |
3188 | range->len); | |
3189 | data_space_reserved -= range->len; | |
3190 | qgroup_reserved -= range->len; | |
3191 | } else if (qgroup_reserved > 0) { | |
3192 | btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved, | |
3193 | range->start, range->len); | |
3194 | qgroup_reserved -= range->len; | |
3195 | } | |
14524a84 QW |
3196 | list_del(&range->list); |
3197 | kfree(range); | |
3198 | } | |
3199 | if (ret < 0) | |
3200 | goto out_unlock; | |
3201 | ||
f27451f2 FM |
3202 | /* |
3203 | * We didn't need to allocate any more space, but we still extended the | |
3204 | * size of the file so we need to update i_size and the inode item. | |
3205 | */ | |
3206 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3207 | out_unlock: |
570eb97b JB |
3208 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3209 | &cached_state); | |
2fe17c10 | 3210 | out: |
e5d4d75b | 3211 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP); |
364ecf36 | 3212 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3213 | return ret; |
3214 | } | |
3215 | ||
b6e83356 | 3216 | /* |
ac3c0d36 FM |
3217 | * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range |
3218 | * that has unflushed and/or flushing delalloc. There might be other adjacent | |
3219 | * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps | |
3220 | * looping while it gets adjacent subranges, and merging them together. | |
b6e83356 FM |
3221 | */ |
3222 | static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end, | |
b3e744fe | 3223 | struct extent_state **cached_state, |
af979fd6 | 3224 | bool *search_io_tree, |
b6e83356 FM |
3225 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
3226 | { | |
40daf3e0 | 3227 | u64 len = end + 1 - start; |
8ddc8274 FM |
3228 | u64 delalloc_len = 0; |
3229 | struct btrfs_ordered_extent *oe; | |
3230 | u64 oe_start; | |
3231 | u64 oe_end; | |
b6e83356 FM |
3232 | |
3233 | /* | |
3234 | * Search the io tree first for EXTENT_DELALLOC. If we find any, it | |
3235 | * means we have delalloc (dirty pages) for which writeback has not | |
3236 | * started yet. | |
3237 | */ | |
8ddc8274 FM |
3238 | if (*search_io_tree) { |
3239 | spin_lock(&inode->lock); | |
3240 | if (inode->delalloc_bytes > 0) { | |
3241 | spin_unlock(&inode->lock); | |
3242 | *delalloc_start_ret = start; | |
3243 | delalloc_len = count_range_bits(&inode->io_tree, | |
3244 | delalloc_start_ret, end, | |
8c6e53a7 | 3245 | len, EXTENT_DELALLOC, 1, |
b3e744fe | 3246 | cached_state); |
8ddc8274 FM |
3247 | } else { |
3248 | spin_unlock(&inode->lock); | |
3249 | } | |
a2853ffc FM |
3250 | } |
3251 | ||
40daf3e0 FM |
3252 | if (delalloc_len > 0) { |
3253 | /* | |
3254 | * If delalloc was found then *delalloc_start_ret has a sector size | |
3255 | * aligned value (rounded down). | |
3256 | */ | |
b6e83356 FM |
3257 | *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1; |
3258 | ||
40daf3e0 FM |
3259 | if (*delalloc_start_ret == start) { |
3260 | /* Delalloc for the whole range, nothing more to do. */ | |
3261 | if (*delalloc_end_ret == end) | |
3262 | return true; | |
8ddc8274 | 3263 | /* Else trim our search range for ordered extents. */ |
40daf3e0 FM |
3264 | start = *delalloc_end_ret + 1; |
3265 | len = end + 1 - start; | |
3266 | } | |
af979fd6 FM |
3267 | } else { |
3268 | /* No delalloc, future calls don't need to search again. */ | |
3269 | *search_io_tree = false; | |
40daf3e0 FM |
3270 | } |
3271 | ||
a2853ffc | 3272 | /* |
8ddc8274 FM |
3273 | * Now also check if there's any ordered extent in the range. |
3274 | * We do this because: | |
b6e83356 FM |
3275 | * |
3276 | * 1) When delalloc is flushed, the file range is locked, we clear the | |
8ddc8274 FM |
3277 | * EXTENT_DELALLOC bit from the io tree and create an extent map and |
3278 | * an ordered extent for the write. So we might just have been called | |
3279 | * after delalloc is flushed and before the ordered extent completes | |
3280 | * and inserts the new file extent item in the subvolume's btree; | |
b6e83356 | 3281 | * |
8ddc8274 | 3282 | * 2) We may have an ordered extent created by flushing delalloc for a |
b6e83356 FM |
3283 | * subrange that starts before the subrange we found marked with |
3284 | * EXTENT_DELALLOC in the io tree. | |
8ddc8274 FM |
3285 | * |
3286 | * We could also use the extent map tree to find such delalloc that is | |
3287 | * being flushed, but using the ordered extents tree is more efficient | |
3288 | * because it's usually much smaller as ordered extents are removed from | |
3289 | * the tree once they complete. With the extent maps, we mau have them | |
3290 | * in the extent map tree for a very long time, and they were either | |
3291 | * created by previous writes or loaded by read operations. | |
b6e83356 | 3292 | */ |
8ddc8274 FM |
3293 | oe = btrfs_lookup_first_ordered_range(inode, start, len); |
3294 | if (!oe) | |
d47704bd | 3295 | return (delalloc_len > 0); |
d47704bd | 3296 | |
8ddc8274 FM |
3297 | /* The ordered extent may span beyond our search range. */ |
3298 | oe_start = max(oe->file_offset, start); | |
3299 | oe_end = min(oe->file_offset + oe->num_bytes - 1, end); | |
b6e83356 | 3300 | |
8ddc8274 | 3301 | btrfs_put_ordered_extent(oe); |
b6e83356 | 3302 | |
8ddc8274 | 3303 | /* Don't have unflushed delalloc, return the ordered extent range. */ |
b6e83356 | 3304 | if (delalloc_len == 0) { |
8ddc8274 FM |
3305 | *delalloc_start_ret = oe_start; |
3306 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3307 | return true; |
3308 | } | |
3309 | ||
3310 | /* | |
8ddc8274 FM |
3311 | * We have both unflushed delalloc (io_tree) and an ordered extent. |
3312 | * If the ranges are adjacent returned a combined range, otherwise | |
3313 | * return the leftmost range. | |
b6e83356 | 3314 | */ |
8ddc8274 FM |
3315 | if (oe_start < *delalloc_start_ret) { |
3316 | if (oe_end < *delalloc_start_ret) | |
3317 | *delalloc_end_ret = oe_end; | |
3318 | *delalloc_start_ret = oe_start; | |
3319 | } else if (*delalloc_end_ret + 1 == oe_start) { | |
3320 | *delalloc_end_ret = oe_end; | |
b6e83356 FM |
3321 | } |
3322 | ||
b6e83356 FM |
3323 | return true; |
3324 | } | |
3325 | ||
3326 | /* | |
3327 | * Check if there's delalloc in a given range. | |
3328 | * | |
3329 | * @inode: The inode. | |
3330 | * @start: The start offset of the range. It does not need to be | |
3331 | * sector size aligned. | |
3332 | * @end: The end offset (inclusive value) of the search range. | |
3333 | * It does not need to be sector size aligned. | |
b3e744fe FM |
3334 | * @cached_state: Extent state record used for speeding up delalloc |
3335 | * searches in the inode's io_tree. Can be NULL. | |
b6e83356 FM |
3336 | * @delalloc_start_ret: Output argument, set to the start offset of the |
3337 | * subrange found with delalloc (may not be sector size | |
3338 | * aligned). | |
3339 | * @delalloc_end_ret: Output argument, set to he end offset (inclusive value) | |
3340 | * of the subrange found with delalloc. | |
3341 | * | |
3342 | * Returns true if a subrange with delalloc is found within the given range, and | |
3343 | * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and | |
3344 | * end offsets of the subrange. | |
3345 | */ | |
ac3c0d36 | 3346 | bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end, |
b3e744fe | 3347 | struct extent_state **cached_state, |
ac3c0d36 | 3348 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) |
b6e83356 FM |
3349 | { |
3350 | u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize); | |
3351 | u64 prev_delalloc_end = 0; | |
af979fd6 | 3352 | bool search_io_tree = true; |
b6e83356 FM |
3353 | bool ret = false; |
3354 | ||
2f2e84ca | 3355 | while (cur_offset <= end) { |
b6e83356 FM |
3356 | u64 delalloc_start; |
3357 | u64 delalloc_end; | |
3358 | bool delalloc; | |
3359 | ||
3360 | delalloc = find_delalloc_subrange(inode, cur_offset, end, | |
b3e744fe | 3361 | cached_state, &search_io_tree, |
b6e83356 FM |
3362 | &delalloc_start, |
3363 | &delalloc_end); | |
3364 | if (!delalloc) | |
3365 | break; | |
3366 | ||
3367 | if (prev_delalloc_end == 0) { | |
3368 | /* First subrange found. */ | |
3369 | *delalloc_start_ret = max(delalloc_start, start); | |
3370 | *delalloc_end_ret = delalloc_end; | |
3371 | ret = true; | |
3372 | } else if (delalloc_start == prev_delalloc_end + 1) { | |
3373 | /* Subrange adjacent to the previous one, merge them. */ | |
3374 | *delalloc_end_ret = delalloc_end; | |
3375 | } else { | |
3376 | /* Subrange not adjacent to the previous one, exit. */ | |
3377 | break; | |
3378 | } | |
3379 | ||
3380 | prev_delalloc_end = delalloc_end; | |
3381 | cur_offset = delalloc_end + 1; | |
3382 | cond_resched(); | |
3383 | } | |
3384 | ||
3385 | return ret; | |
3386 | } | |
3387 | ||
3388 | /* | |
3389 | * Check if there's a hole or delalloc range in a range representing a hole (or | |
3390 | * prealloc extent) found in the inode's subvolume btree. | |
3391 | * | |
3392 | * @inode: The inode. | |
3393 | * @whence: Seek mode (SEEK_DATA or SEEK_HOLE). | |
3394 | * @start: Start offset of the hole region. It does not need to be sector | |
3395 | * size aligned. | |
3396 | * @end: End offset (inclusive value) of the hole region. It does not | |
3397 | * need to be sector size aligned. | |
3398 | * @start_ret: Return parameter, used to set the start of the subrange in the | |
3399 | * hole that matches the search criteria (seek mode), if such | |
3400 | * subrange is found (return value of the function is true). | |
3401 | * The value returned here may not be sector size aligned. | |
3402 | * | |
3403 | * Returns true if a subrange matching the given seek mode is found, and if one | |
3404 | * is found, it updates @start_ret with the start of the subrange. | |
3405 | */ | |
3406 | static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence, | |
3c32c721 | 3407 | struct extent_state **cached_state, |
b6e83356 FM |
3408 | u64 start, u64 end, u64 *start_ret) |
3409 | { | |
3410 | u64 delalloc_start; | |
3411 | u64 delalloc_end; | |
3412 | bool delalloc; | |
3413 | ||
3c32c721 | 3414 | delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state, |
ac3c0d36 | 3415 | &delalloc_start, &delalloc_end); |
b6e83356 FM |
3416 | if (delalloc && whence == SEEK_DATA) { |
3417 | *start_ret = delalloc_start; | |
3418 | return true; | |
3419 | } | |
3420 | ||
3421 | if (delalloc && whence == SEEK_HOLE) { | |
3422 | /* | |
3423 | * We found delalloc but it starts after out start offset. So we | |
3424 | * have a hole between our start offset and the delalloc start. | |
3425 | */ | |
3426 | if (start < delalloc_start) { | |
3427 | *start_ret = start; | |
3428 | return true; | |
3429 | } | |
3430 | /* | |
3431 | * Delalloc range starts at our start offset. | |
3432 | * If the delalloc range's length is smaller than our range, | |
3433 | * then it means we have a hole that starts where the delalloc | |
3434 | * subrange ends. | |
3435 | */ | |
3436 | if (delalloc_end < end) { | |
3437 | *start_ret = delalloc_end + 1; | |
3438 | return true; | |
3439 | } | |
3440 | ||
3441 | /* There's delalloc for the whole range. */ | |
3442 | return false; | |
3443 | } | |
3444 | ||
3445 | if (!delalloc && whence == SEEK_HOLE) { | |
3446 | *start_ret = start; | |
3447 | return true; | |
3448 | } | |
3449 | ||
3450 | /* | |
3451 | * No delalloc in the range and we are seeking for data. The caller has | |
3452 | * to iterate to the next extent item in the subvolume btree. | |
3453 | */ | |
3454 | return false; | |
3455 | } | |
3456 | ||
3c32c721 | 3457 | static loff_t find_desired_extent(struct file *file, loff_t offset, int whence) |
b2675157 | 3458 | { |
3c32c721 FM |
3459 | struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host); |
3460 | struct btrfs_file_private *private = file->private_data; | |
cca5de97 | 3461 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
b2675157 | 3462 | struct extent_state *cached_state = NULL; |
3c32c721 | 3463 | struct extent_state **delalloc_cached_state; |
b6e83356 FM |
3464 | const loff_t i_size = i_size_read(&inode->vfs_inode); |
3465 | const u64 ino = btrfs_ino(inode); | |
3466 | struct btrfs_root *root = inode->root; | |
3467 | struct btrfs_path *path; | |
3468 | struct btrfs_key key; | |
3469 | u64 last_extent_end; | |
4d1a40c6 LB |
3470 | u64 lockstart; |
3471 | u64 lockend; | |
3472 | u64 start; | |
b6e83356 FM |
3473 | int ret; |
3474 | bool found = false; | |
b2675157 | 3475 | |
bc80230e | 3476 | if (i_size == 0 || offset >= i_size) |
4d1a40c6 LB |
3477 | return -ENXIO; |
3478 | ||
b6e83356 FM |
3479 | /* |
3480 | * Quick path. If the inode has no prealloc extents and its number of | |
3481 | * bytes used matches its i_size, then it can not have holes. | |
3482 | */ | |
3483 | if (whence == SEEK_HOLE && | |
3484 | !(inode->flags & BTRFS_INODE_PREALLOC) && | |
3485 | inode_get_bytes(&inode->vfs_inode) == i_size) | |
3486 | return i_size; | |
3487 | ||
3c32c721 FM |
3488 | if (!private) { |
3489 | private = kzalloc(sizeof(*private), GFP_KERNEL); | |
3490 | /* | |
3491 | * No worries if memory allocation failed. | |
3492 | * The private structure is used only for speeding up multiple | |
3493 | * lseek SEEK_HOLE/DATA calls to a file when there's delalloc, | |
3494 | * so everything will still be correct. | |
3495 | */ | |
3496 | file->private_data = private; | |
3497 | } | |
3498 | ||
3499 | if (private) | |
3500 | delalloc_cached_state = &private->llseek_cached_state; | |
3501 | else | |
3502 | delalloc_cached_state = NULL; | |
3503 | ||
4d1a40c6 | 3504 | /* |
bc80230e | 3505 | * offset can be negative, in this case we start finding DATA/HOLE from |
4d1a40c6 LB |
3506 | * the very start of the file. |
3507 | */ | |
bc80230e | 3508 | start = max_t(loff_t, 0, offset); |
4d1a40c6 | 3509 | |
0b246afa | 3510 | lockstart = round_down(start, fs_info->sectorsize); |
d79b7c26 | 3511 | lockend = round_up(i_size, fs_info->sectorsize); |
b2675157 | 3512 | if (lockend <= lockstart) |
0b246afa | 3513 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3514 | lockend--; |
b6e83356 FM |
3515 | |
3516 | path = btrfs_alloc_path(); | |
3517 | if (!path) | |
3518 | return -ENOMEM; | |
3519 | path->reada = READA_FORWARD; | |
3520 | ||
3521 | key.objectid = ino; | |
3522 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3523 | key.offset = start; | |
3524 | ||
3525 | last_extent_end = lockstart; | |
b2675157 | 3526 | |
570eb97b | 3527 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b2675157 | 3528 | |
b6e83356 FM |
3529 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3530 | if (ret < 0) { | |
3531 | goto out; | |
3532 | } else if (ret > 0 && path->slots[0] > 0) { | |
3533 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3534 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3535 | path->slots[0]--; | |
3536 | } | |
3537 | ||
d79b7c26 | 3538 | while (start < i_size) { |
b6e83356 FM |
3539 | struct extent_buffer *leaf = path->nodes[0]; |
3540 | struct btrfs_file_extent_item *extent; | |
3541 | u64 extent_end; | |
1f55ee6d | 3542 | u8 type; |
b6e83356 FM |
3543 | |
3544 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3545 | ret = btrfs_next_leaf(root, path); | |
3546 | if (ret < 0) | |
3547 | goto out; | |
3548 | else if (ret > 0) | |
3549 | break; | |
3550 | ||
3551 | leaf = path->nodes[0]; | |
b2675157 JB |
3552 | } |
3553 | ||
b6e83356 FM |
3554 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3555 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
7f4ca37c | 3556 | break; |
b2675157 | 3557 | |
b6e83356 FM |
3558 | extent_end = btrfs_file_extent_end(path); |
3559 | ||
3560 | /* | |
3561 | * In the first iteration we may have a slot that points to an | |
3562 | * extent that ends before our start offset, so skip it. | |
3563 | */ | |
3564 | if (extent_end <= start) { | |
3565 | path->slots[0]++; | |
3566 | continue; | |
3567 | } | |
3568 | ||
3569 | /* We have an implicit hole, NO_HOLES feature is likely set. */ | |
3570 | if (last_extent_end < key.offset) { | |
3571 | u64 search_start = last_extent_end; | |
3572 | u64 found_start; | |
3573 | ||
3574 | /* | |
3575 | * First iteration, @start matches @offset and it's | |
3576 | * within the hole. | |
3577 | */ | |
3578 | if (start == offset) | |
3579 | search_start = offset; | |
3580 | ||
3581 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3582 | delalloc_cached_state, |
b6e83356 FM |
3583 | search_start, |
3584 | key.offset - 1, | |
3585 | &found_start); | |
3586 | if (found) { | |
3587 | start = found_start; | |
3588 | break; | |
3589 | } | |
3590 | /* | |
3591 | * Didn't find data or a hole (due to delalloc) in the | |
3592 | * implicit hole range, so need to analyze the extent. | |
3593 | */ | |
3594 | } | |
3595 | ||
3596 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
3597 | struct btrfs_file_extent_item); | |
1f55ee6d | 3598 | type = btrfs_file_extent_type(leaf, extent); |
b6e83356 | 3599 | |
1f55ee6d FM |
3600 | /* |
3601 | * Can't access the extent's disk_bytenr field if this is an | |
3602 | * inline extent, since at that offset, it's where the extent | |
3603 | * data starts. | |
3604 | */ | |
3605 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
3606 | (type == BTRFS_FILE_EXTENT_REG && | |
3607 | btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) { | |
b6e83356 FM |
3608 | /* |
3609 | * Explicit hole or prealloc extent, search for delalloc. | |
3610 | * A prealloc extent is treated like a hole. | |
3611 | */ | |
3612 | u64 search_start = key.offset; | |
3613 | u64 found_start; | |
3614 | ||
3615 | /* | |
3616 | * First iteration, @start matches @offset and it's | |
3617 | * within the hole. | |
3618 | */ | |
3619 | if (start == offset) | |
3620 | search_start = offset; | |
3621 | ||
3622 | found = find_desired_extent_in_hole(inode, whence, | |
3c32c721 | 3623 | delalloc_cached_state, |
b6e83356 FM |
3624 | search_start, |
3625 | extent_end - 1, | |
3626 | &found_start); | |
3627 | if (found) { | |
3628 | start = found_start; | |
3629 | break; | |
3630 | } | |
3631 | /* | |
3632 | * Didn't find data or a hole (due to delalloc) in the | |
3633 | * implicit hole range, so need to analyze the next | |
3634 | * extent item. | |
3635 | */ | |
3636 | } else { | |
3637 | /* | |
3638 | * Found a regular or inline extent. | |
3639 | * If we are seeking for data, adjust the start offset | |
3640 | * and stop, we're done. | |
3641 | */ | |
3642 | if (whence == SEEK_DATA) { | |
3643 | start = max_t(u64, key.offset, offset); | |
3644 | found = true; | |
3645 | break; | |
3646 | } | |
3647 | /* | |
3648 | * Else, we are seeking for a hole, check the next file | |
3649 | * extent item. | |
3650 | */ | |
3651 | } | |
3652 | ||
3653 | start = extent_end; | |
3654 | last_extent_end = extent_end; | |
3655 | path->slots[0]++; | |
aed0ca18 FM |
3656 | if (fatal_signal_pending(current)) { |
3657 | ret = -EINTR; | |
b6e83356 | 3658 | goto out; |
aed0ca18 | 3659 | } |
b2675157 JB |
3660 | cond_resched(); |
3661 | } | |
b6e83356 FM |
3662 | |
3663 | /* We have an implicit hole from the last extent found up to i_size. */ | |
3664 | if (!found && start < i_size) { | |
3c32c721 FM |
3665 | found = find_desired_extent_in_hole(inode, whence, |
3666 | delalloc_cached_state, start, | |
b6e83356 FM |
3667 | i_size - 1, &start); |
3668 | if (!found) | |
3669 | start = i_size; | |
3670 | } | |
3671 | ||
3672 | out: | |
570eb97b | 3673 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b6e83356 FM |
3674 | btrfs_free_path(path); |
3675 | ||
3676 | if (ret < 0) | |
3677 | return ret; | |
3678 | ||
3679 | if (whence == SEEK_DATA && start >= i_size) | |
3680 | return -ENXIO; | |
bc80230e | 3681 | |
b6e83356 | 3682 | return min_t(loff_t, start, i_size); |
b2675157 JB |
3683 | } |
3684 | ||
965c8e59 | 3685 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3686 | { |
3687 | struct inode *inode = file->f_mapping->host; | |
b2675157 | 3688 | |
965c8e59 | 3689 | switch (whence) { |
2034f3b4 NB |
3690 | default: |
3691 | return generic_file_llseek(file, offset, whence); | |
b2675157 JB |
3692 | case SEEK_DATA: |
3693 | case SEEK_HOLE: | |
29b6352b | 3694 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
3c32c721 | 3695 | offset = find_desired_extent(file, offset, whence); |
e5d4d75b | 3696 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
bc80230e | 3697 | break; |
b2675157 JB |
3698 | } |
3699 | ||
bc80230e NB |
3700 | if (offset < 0) |
3701 | return offset; | |
3702 | ||
2034f3b4 | 3703 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
3704 | } |
3705 | ||
edf064e7 GR |
3706 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3707 | { | |
14605409 BB |
3708 | int ret; |
3709 | ||
f02c75e6 CH |
3710 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC | |
3711 | FMODE_CAN_ODIRECT; | |
14605409 BB |
3712 | |
3713 | ret = fsverity_file_open(inode, filp); | |
3714 | if (ret) | |
3715 | return ret; | |
edf064e7 GR |
3716 | return generic_file_open(inode, filp); |
3717 | } | |
3718 | ||
4e4cabec GR |
3719 | static int check_direct_read(struct btrfs_fs_info *fs_info, |
3720 | const struct iov_iter *iter, loff_t offset) | |
3721 | { | |
3722 | int ret; | |
3723 | int i, seg; | |
3724 | ||
3725 | ret = check_direct_IO(fs_info, iter, offset); | |
3726 | if (ret < 0) | |
3727 | return ret; | |
3728 | ||
3729 | if (!iter_is_iovec(iter)) | |
3730 | return 0; | |
3731 | ||
de4f5fed JA |
3732 | for (seg = 0; seg < iter->nr_segs; seg++) { |
3733 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
3734 | const struct iovec *iov1 = iter_iov(iter) + seg; | |
3735 | const struct iovec *iov2 = iter_iov(iter) + i; | |
3736 | ||
3737 | if (iov1->iov_base == iov2->iov_base) | |
4e4cabec | 3738 | return -EINVAL; |
de4f5fed JA |
3739 | } |
3740 | } | |
4e4cabec GR |
3741 | return 0; |
3742 | } | |
3743 | ||
3744 | static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to) | |
3745 | { | |
3746 | struct inode *inode = file_inode(iocb->ki_filp); | |
51bd9563 FM |
3747 | size_t prev_left = 0; |
3748 | ssize_t read = 0; | |
4e4cabec GR |
3749 | ssize_t ret; |
3750 | ||
14605409 BB |
3751 | if (fsverity_active(inode)) |
3752 | return 0; | |
3753 | ||
4e4cabec GR |
3754 | if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos)) |
3755 | return 0; | |
3756 | ||
29b6352b | 3757 | btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 FM |
3758 | again: |
3759 | /* | |
3760 | * This is similar to what we do for direct IO writes, see the comment | |
3761 | * at btrfs_direct_write(), but we also disable page faults in addition | |
3762 | * to disabling them only at the iov_iter level. This is because when | |
3763 | * reading from a hole or prealloc extent, iomap calls iov_iter_zero(), | |
3764 | * which can still trigger page fault ins despite having set ->nofault | |
3765 | * to true of our 'to' iov_iter. | |
3766 | * | |
3767 | * The difference to direct IO writes is that we deadlock when trying | |
3768 | * to lock the extent range in the inode's tree during he page reads | |
3769 | * triggered by the fault in (while for writes it is due to waiting for | |
3770 | * our own ordered extent). This is because for direct IO reads, | |
3771 | * btrfs_dio_iomap_begin() returns with the extent range locked, which | |
3772 | * is only unlocked in the endio callback (end_bio_extent_readpage()). | |
3773 | */ | |
3774 | pagefault_disable(); | |
3775 | to->nofault = true; | |
8184620a | 3776 | ret = btrfs_dio_read(iocb, to, read); |
51bd9563 FM |
3777 | to->nofault = false; |
3778 | pagefault_enable(); | |
3779 | ||
3780 | /* No increment (+=) because iomap returns a cumulative value. */ | |
3781 | if (ret > 0) | |
3782 | read = ret; | |
3783 | ||
3784 | if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) { | |
3785 | const size_t left = iov_iter_count(to); | |
3786 | ||
3787 | if (left == prev_left) { | |
3788 | /* | |
3789 | * We didn't make any progress since the last attempt, | |
3790 | * fallback to a buffered read for the remainder of the | |
3791 | * range. This is just to avoid any possibility of looping | |
3792 | * for too long. | |
3793 | */ | |
3794 | ret = read; | |
3795 | } else { | |
3796 | /* | |
3797 | * We made some progress since the last retry or this is | |
3798 | * the first time we are retrying. Fault in as many pages | |
3799 | * as possible and retry. | |
3800 | */ | |
3801 | fault_in_iov_iter_writeable(to, left); | |
3802 | prev_left = left; | |
3803 | goto again; | |
3804 | } | |
3805 | } | |
e5d4d75b | 3806 | btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED); |
51bd9563 | 3807 | return ret < 0 ? ret : read; |
4e4cabec GR |
3808 | } |
3809 | ||
f85781fb GR |
3810 | static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
3811 | { | |
3812 | ssize_t ret = 0; | |
3813 | ||
3814 | if (iocb->ki_flags & IOCB_DIRECT) { | |
4e4cabec | 3815 | ret = btrfs_direct_read(iocb, to); |
0425e7ba JT |
3816 | if (ret < 0 || !iov_iter_count(to) || |
3817 | iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp))) | |
f85781fb GR |
3818 | return ret; |
3819 | } | |
3820 | ||
87fa0f3e | 3821 | return filemap_read(iocb, to, ret); |
f85781fb GR |
3822 | } |
3823 | ||
828c0950 | 3824 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3825 | .llseek = btrfs_file_llseek, |
f85781fb | 3826 | .read_iter = btrfs_file_read_iter, |
2cb1e089 | 3827 | .splice_read = filemap_splice_read, |
b30ac0fc | 3828 | .write_iter = btrfs_file_write_iter, |
d7776591 | 3829 | .splice_write = iter_file_splice_write, |
9ebefb18 | 3830 | .mmap = btrfs_file_mmap, |
edf064e7 | 3831 | .open = btrfs_file_open, |
e1b81e67 | 3832 | .release = btrfs_release_file, |
b0c58223 | 3833 | .get_unmapped_area = thp_get_unmapped_area, |
39279cc3 | 3834 | .fsync = btrfs_sync_file, |
2fe17c10 | 3835 | .fallocate = btrfs_fallocate, |
34287aa3 | 3836 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3837 | #ifdef CONFIG_COMPAT |
4c63c245 | 3838 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3839 | #endif |
2e5dfc99 | 3840 | .remap_file_range = btrfs_remap_file_range, |
39279cc3 | 3841 | }; |
9247f317 | 3842 | |
728404da FM |
3843 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) |
3844 | { | |
3845 | int ret; | |
3846 | ||
3847 | /* | |
3848 | * So with compression we will find and lock a dirty page and clear the | |
3849 | * first one as dirty, setup an async extent, and immediately return | |
3850 | * with the entire range locked but with nobody actually marked with | |
3851 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3852 | * expect it to work since it will just kick off a thread to do the | |
3853 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3854 | * since it will wait on the page lock, which won't be unlocked until | |
3855 | * after the pages have been marked as writeback and so we're good to go | |
3856 | * from there. We have to do this otherwise we'll miss the ordered | |
3857 | * extents and that results in badness. Please Josef, do not think you | |
3858 | * know better and pull this out at some point in the future, it is | |
3859 | * right and you are wrong. | |
3860 | */ | |
3861 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3862 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3863 | &BTRFS_I(inode)->runtime_flags)) | |
3864 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3865 | ||
3866 | return ret; | |
3867 | } |