Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
dc17ff8f CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
dc17ff8f CM |
4 | */ |
5 | ||
dc17ff8f | 6 | #include <linux/slab.h> |
d6bfde87 | 7 | #include <linux/blkdev.h> |
f421950f | 8 | #include <linux/writeback.h> |
a3d46aea | 9 | #include <linux/sched/mm.h> |
9b569ea0 | 10 | #include "messages.h" |
602cbe91 | 11 | #include "misc.h" |
dc17ff8f CM |
12 | #include "ctree.h" |
13 | #include "transaction.h" | |
14 | #include "btrfs_inode.h" | |
e6dcd2dc | 15 | #include "extent_io.h" |
199c2a9c | 16 | #include "disk-io.h" |
ebb8765b | 17 | #include "compression.h" |
86736342 | 18 | #include "delalloc-space.h" |
7dbeaad0 | 19 | #include "qgroup.h" |
b945a463 | 20 | #include "subpage.h" |
af142b6f | 21 | #include "file.h" |
7f0add25 | 22 | #include "super.h" |
dc17ff8f | 23 | |
6352b91d MX |
24 | static struct kmem_cache *btrfs_ordered_extent_cache; |
25 | ||
e6dcd2dc | 26 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 27 | { |
bffe633e | 28 | if (entry->file_offset + entry->num_bytes < entry->file_offset) |
e6dcd2dc | 29 | return (u64)-1; |
bffe633e | 30 | return entry->file_offset + entry->num_bytes; |
dc17ff8f CM |
31 | } |
32 | ||
d352ac68 CM |
33 | /* returns NULL if the insertion worked, or it returns the node it did find |
34 | * in the tree | |
35 | */ | |
e6dcd2dc CM |
36 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
37 | struct rb_node *node) | |
dc17ff8f | 38 | { |
d397712b CM |
39 | struct rb_node **p = &root->rb_node; |
40 | struct rb_node *parent = NULL; | |
e6dcd2dc | 41 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 42 | |
d397712b | 43 | while (*p) { |
dc17ff8f | 44 | parent = *p; |
e6dcd2dc | 45 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 46 | |
e6dcd2dc | 47 | if (file_offset < entry->file_offset) |
dc17ff8f | 48 | p = &(*p)->rb_left; |
e6dcd2dc | 49 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
50 | p = &(*p)->rb_right; |
51 | else | |
52 | return parent; | |
53 | } | |
54 | ||
55 | rb_link_node(node, parent, p); | |
56 | rb_insert_color(node, root); | |
57 | return NULL; | |
58 | } | |
59 | ||
d352ac68 CM |
60 | /* |
61 | * look for a given offset in the tree, and if it can't be found return the | |
62 | * first lesser offset | |
63 | */ | |
e6dcd2dc CM |
64 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
65 | struct rb_node **prev_ret) | |
dc17ff8f | 66 | { |
d397712b | 67 | struct rb_node *n = root->rb_node; |
dc17ff8f | 68 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
69 | struct rb_node *test; |
70 | struct btrfs_ordered_extent *entry; | |
71 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 72 | |
d397712b | 73 | while (n) { |
e6dcd2dc | 74 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
75 | prev = n; |
76 | prev_entry = entry; | |
dc17ff8f | 77 | |
e6dcd2dc | 78 | if (file_offset < entry->file_offset) |
dc17ff8f | 79 | n = n->rb_left; |
e6dcd2dc | 80 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
81 | n = n->rb_right; |
82 | else | |
83 | return n; | |
84 | } | |
85 | if (!prev_ret) | |
86 | return NULL; | |
87 | ||
d397712b | 88 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
89 | test = rb_next(prev); |
90 | if (!test) | |
91 | break; | |
92 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
93 | rb_node); | |
94 | if (file_offset < entry_end(prev_entry)) | |
95 | break; | |
96 | ||
97 | prev = test; | |
98 | } | |
99 | if (prev) | |
100 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
101 | rb_node); | |
d397712b | 102 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
103 | test = rb_prev(prev); |
104 | if (!test) | |
105 | break; | |
106 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
107 | rb_node); | |
108 | prev = test; | |
dc17ff8f CM |
109 | } |
110 | *prev_ret = prev; | |
111 | return NULL; | |
112 | } | |
113 | ||
4b46fce2 JB |
114 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
115 | u64 len) | |
116 | { | |
117 | if (file_offset + len <= entry->file_offset || | |
bffe633e | 118 | entry->file_offset + entry->num_bytes <= file_offset) |
4b46fce2 JB |
119 | return 0; |
120 | return 1; | |
121 | } | |
122 | ||
d352ac68 CM |
123 | /* |
124 | * look find the first ordered struct that has this offset, otherwise | |
125 | * the first one less than this offset | |
126 | */ | |
e6dcd2dc CM |
127 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
128 | u64 file_offset) | |
dc17ff8f | 129 | { |
e6dcd2dc | 130 | struct rb_root *root = &tree->tree; |
c87fb6fd | 131 | struct rb_node *prev = NULL; |
dc17ff8f | 132 | struct rb_node *ret; |
e6dcd2dc CM |
133 | struct btrfs_ordered_extent *entry; |
134 | ||
135 | if (tree->last) { | |
136 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
137 | rb_node); | |
20bbf20e | 138 | if (in_range(file_offset, entry->file_offset, entry->num_bytes)) |
e6dcd2dc CM |
139 | return tree->last; |
140 | } | |
141 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 142 | if (!ret) |
e6dcd2dc CM |
143 | ret = prev; |
144 | if (ret) | |
145 | tree->last = ret; | |
dc17ff8f CM |
146 | return ret; |
147 | } | |
148 | ||
53d9981c CH |
149 | static struct btrfs_ordered_extent *alloc_ordered_extent( |
150 | struct btrfs_inode *inode, u64 file_offset, u64 num_bytes, | |
151 | u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes, | |
152 | u64 offset, unsigned long flags, int compress_type) | |
dc17ff8f | 153 | { |
e6dcd2dc | 154 | struct btrfs_ordered_extent *entry; |
7dbeaad0 QW |
155 | int ret; |
156 | ||
cb36a9bb OS |
157 | if (flags & |
158 | ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) { | |
7dbeaad0 | 159 | /* For nocow write, we can release the qgroup rsv right now */ |
8b8a979f | 160 | ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes); |
7dbeaad0 | 161 | if (ret < 0) |
cf6d1aa4 | 162 | return ERR_PTR(ret); |
7dbeaad0 QW |
163 | } else { |
164 | /* | |
165 | * The ordered extent has reserved qgroup space, release now | |
166 | * and pass the reserved number for qgroup_record to free. | |
167 | */ | |
72b7d15b | 168 | ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes); |
7dbeaad0 | 169 | if (ret < 0) |
cf6d1aa4 | 170 | return ERR_PTR(ret); |
7dbeaad0 | 171 | } |
6352b91d | 172 | entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); |
dc17ff8f | 173 | if (!entry) |
cf6d1aa4 | 174 | return ERR_PTR(-ENOMEM); |
dc17ff8f | 175 | |
e6dcd2dc | 176 | entry->file_offset = file_offset; |
bffe633e | 177 | entry->num_bytes = num_bytes; |
cb36a9bb OS |
178 | entry->ram_bytes = ram_bytes; |
179 | entry->disk_bytenr = disk_bytenr; | |
bffe633e | 180 | entry->disk_num_bytes = disk_num_bytes; |
cb36a9bb | 181 | entry->offset = offset; |
bffe633e | 182 | entry->bytes_left = num_bytes; |
da69fea9 | 183 | entry->inode = igrab(&inode->vfs_inode); |
261507a0 | 184 | entry->compress_type = compress_type; |
77cef2ec | 185 | entry->truncated_len = (u64)-1; |
7dbeaad0 | 186 | entry->qgroup_rsv = ret; |
cb36a9bb | 187 | entry->flags = flags; |
e76edab7 | 188 | refcount_set(&entry->refs, 1); |
e6dcd2dc CM |
189 | init_waitqueue_head(&entry->wait); |
190 | INIT_LIST_HEAD(&entry->list); | |
48778179 | 191 | INIT_LIST_HEAD(&entry->log_list); |
3eaa2885 | 192 | INIT_LIST_HEAD(&entry->root_extent_list); |
9afab882 MX |
193 | INIT_LIST_HEAD(&entry->work_list); |
194 | init_completion(&entry->completion); | |
dc17ff8f | 195 | |
53d9981c CH |
196 | /* |
197 | * We don't need the count_max_extents here, we can assume that all of | |
198 | * that work has been done at higher layers, so this is truly the | |
199 | * smallest the extent is going to get. | |
200 | */ | |
201 | spin_lock(&inode->lock); | |
202 | btrfs_mod_outstanding_extents(inode, 1); | |
203 | spin_unlock(&inode->lock); | |
204 | ||
205 | return entry; | |
206 | } | |
207 | ||
208 | static void insert_ordered_extent(struct btrfs_ordered_extent *entry) | |
209 | { | |
210 | struct btrfs_inode *inode = BTRFS_I(entry->inode); | |
211 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; | |
212 | struct btrfs_root *root = inode->root; | |
213 | struct btrfs_fs_info *fs_info = root->fs_info; | |
214 | struct rb_node *node; | |
215 | ||
acbf1dd0 | 216 | trace_btrfs_ordered_extent_add(inode, entry); |
1abe9b8a | 217 | |
53d9981c CH |
218 | percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes, |
219 | fs_info->delalloc_batch); | |
220 | ||
221 | /* One ref for the tree. */ | |
222 | refcount_inc(&entry->refs); | |
223 | ||
5fd02043 | 224 | spin_lock_irq(&tree->lock); |
53d9981c | 225 | node = tree_insert(&tree->tree, entry->file_offset, &entry->rb_node); |
43c04fb1 | 226 | if (node) |
511a32b5 NB |
227 | btrfs_panic(fs_info, -EEXIST, |
228 | "inconsistency in ordered tree at offset %llu", | |
53d9981c | 229 | entry->file_offset); |
5fd02043 | 230 | spin_unlock_irq(&tree->lock); |
d397712b | 231 | |
199c2a9c | 232 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 233 | list_add_tail(&entry->root_extent_list, |
199c2a9c MX |
234 | &root->ordered_extents); |
235 | root->nr_ordered_extents++; | |
236 | if (root->nr_ordered_extents == 1) { | |
0b246afa | 237 | spin_lock(&fs_info->ordered_root_lock); |
199c2a9c | 238 | BUG_ON(!list_empty(&root->ordered_root)); |
0b246afa JM |
239 | list_add_tail(&root->ordered_root, &fs_info->ordered_roots); |
240 | spin_unlock(&fs_info->ordered_root_lock); | |
199c2a9c MX |
241 | } |
242 | spin_unlock(&root->ordered_extent_lock); | |
53d9981c | 243 | } |
3eaa2885 | 244 | |
53d9981c CH |
245 | /* |
246 | * Add an ordered extent to the per-inode tree. | |
247 | * | |
248 | * @inode: Inode that this extent is for. | |
249 | * @file_offset: Logical offset in file where the extent starts. | |
250 | * @num_bytes: Logical length of extent in file. | |
251 | * @ram_bytes: Full length of unencoded data. | |
252 | * @disk_bytenr: Offset of extent on disk. | |
253 | * @disk_num_bytes: Size of extent on disk. | |
254 | * @offset: Offset into unencoded data where file data starts. | |
255 | * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*). | |
256 | * @compress_type: Compression algorithm used for data. | |
257 | * | |
258 | * Most of these parameters correspond to &struct btrfs_file_extent_item. The | |
259 | * tree is given a single reference on the ordered extent that was inserted, and | |
260 | * the returned pointer is given a second reference. | |
261 | * | |
262 | * Return: the new ordered extent or error pointer. | |
263 | */ | |
264 | struct btrfs_ordered_extent *btrfs_alloc_ordered_extent( | |
265 | struct btrfs_inode *inode, u64 file_offset, | |
266 | u64 num_bytes, u64 ram_bytes, u64 disk_bytenr, | |
267 | u64 disk_num_bytes, u64 offset, unsigned long flags, | |
268 | int compress_type) | |
269 | { | |
270 | struct btrfs_ordered_extent *entry; | |
8b62f87b | 271 | |
53d9981c | 272 | ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0); |
cf6d1aa4 | 273 | |
53d9981c CH |
274 | entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes, |
275 | disk_bytenr, disk_num_bytes, offset, flags, | |
276 | compress_type); | |
277 | if (!IS_ERR(entry)) | |
278 | insert_ordered_extent(entry); | |
cf6d1aa4 BB |
279 | return entry; |
280 | } | |
281 | ||
eb84ae03 CM |
282 | /* |
283 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
284 | * when an ordered extent is finished. If the list covers more than one |
285 | * ordered extent, it is split across multiples. | |
eb84ae03 | 286 | */ |
f9756261 | 287 | void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry, |
143bede5 | 288 | struct btrfs_ordered_sum *sum) |
dc17ff8f | 289 | { |
e6dcd2dc | 290 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 291 | |
f9756261 | 292 | tree = &BTRFS_I(entry->inode)->ordered_tree; |
5fd02043 | 293 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 294 | list_add_tail(&sum->list, &entry->list); |
5fd02043 | 295 | spin_unlock_irq(&tree->lock); |
dc17ff8f CM |
296 | } |
297 | ||
711f447b CH |
298 | static void finish_ordered_fn(struct btrfs_work *work) |
299 | { | |
300 | struct btrfs_ordered_extent *ordered_extent; | |
301 | ||
302 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
303 | btrfs_finish_ordered_io(ordered_extent); | |
304 | } | |
305 | ||
53df2586 CH |
306 | static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered, |
307 | struct page *page, u64 file_offset, | |
308 | u64 len, bool uptodate) | |
309 | { | |
310 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
311 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
312 | ||
313 | lockdep_assert_held(&inode->ordered_tree.lock); | |
314 | ||
315 | if (page) { | |
316 | ASSERT(page->mapping); | |
317 | ASSERT(page_offset(page) <= file_offset); | |
318 | ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE); | |
319 | ||
320 | /* | |
321 | * Ordered (Private2) bit indicates whether we still have | |
322 | * pending io unfinished for the ordered extent. | |
323 | * | |
324 | * If there's no such bit, we need to skip to next range. | |
325 | */ | |
326 | if (!btrfs_page_test_ordered(fs_info, page, file_offset, len)) | |
327 | return false; | |
328 | btrfs_page_clear_ordered(fs_info, page, file_offset, len); | |
329 | } | |
330 | ||
331 | /* Now we're fine to update the accounting. */ | |
332 | if (WARN_ON_ONCE(len > ordered->bytes_left)) { | |
333 | btrfs_crit(fs_info, | |
334 | "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu", | |
335 | inode->root->root_key.objectid, btrfs_ino(inode), | |
336 | ordered->file_offset, ordered->num_bytes, | |
337 | len, ordered->bytes_left); | |
338 | ordered->bytes_left = 0; | |
339 | } else { | |
340 | ordered->bytes_left -= len; | |
341 | } | |
342 | ||
343 | if (!uptodate) | |
344 | set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); | |
345 | ||
346 | if (ordered->bytes_left) | |
347 | return false; | |
348 | ||
349 | /* | |
350 | * All the IO of the ordered extent is finished, we need to queue | |
351 | * the finish_func to be executed. | |
352 | */ | |
353 | set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags); | |
354 | cond_wake_up(&ordered->wait); | |
355 | refcount_inc(&ordered->refs); | |
356 | trace_btrfs_ordered_extent_mark_finished(inode, ordered); | |
357 | return true; | |
358 | } | |
359 | ||
2d6f107e CH |
360 | static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered) |
361 | { | |
362 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
363 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
364 | struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ? | |
365 | fs_info->endio_freespace_worker : fs_info->endio_write_workers; | |
366 | ||
367 | btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL); | |
368 | btrfs_queue_work(wq, &ordered->work); | |
369 | } | |
370 | ||
122e9ede CH |
371 | bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered, |
372 | struct page *page, u64 file_offset, u64 len, | |
373 | bool uptodate) | |
374 | { | |
375 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
376 | unsigned long flags; | |
377 | bool ret; | |
378 | ||
379 | trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate); | |
380 | ||
381 | spin_lock_irqsave(&inode->ordered_tree.lock, flags); | |
382 | ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate); | |
383 | spin_unlock_irqrestore(&inode->ordered_tree.lock, flags); | |
384 | ||
385 | if (ret) | |
386 | btrfs_queue_ordered_fn(ordered); | |
387 | return ret; | |
388 | } | |
389 | ||
163cf09c | 390 | /* |
e65f152e | 391 | * Mark all ordered extents io inside the specified range finished. |
163cf09c | 392 | * |
143823cf | 393 | * @page: The involved page for the operation. |
e65f152e QW |
394 | * For uncompressed buffered IO, the page status also needs to be |
395 | * updated to indicate whether the pending ordered io is finished. | |
396 | * Can be NULL for direct IO and compressed write. | |
397 | * For these cases, callers are ensured they won't execute the | |
398 | * endio function twice. | |
163cf09c | 399 | * |
e65f152e | 400 | * This function is called for endio, thus the range must have ordered |
143823cf | 401 | * extent(s) covering it. |
163cf09c | 402 | */ |
e65f152e | 403 | void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode, |
711f447b CH |
404 | struct page *page, u64 file_offset, |
405 | u64 num_bytes, bool uptodate) | |
163cf09c | 406 | { |
7095821e | 407 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; |
163cf09c CM |
408 | struct rb_node *node; |
409 | struct btrfs_ordered_extent *entry = NULL; | |
5fd02043 | 410 | unsigned long flags; |
e65f152e QW |
411 | u64 cur = file_offset; |
412 | ||
6648cedd CH |
413 | trace_btrfs_writepage_end_io_hook(inode, file_offset, |
414 | file_offset + num_bytes - 1, | |
415 | uptodate); | |
416 | ||
5fd02043 | 417 | spin_lock_irqsave(&tree->lock, flags); |
e65f152e QW |
418 | while (cur < file_offset + num_bytes) { |
419 | u64 entry_end; | |
420 | u64 end; | |
421 | u32 len; | |
422 | ||
423 | node = tree_search(tree, cur); | |
424 | /* No ordered extents at all */ | |
425 | if (!node) | |
426 | break; | |
5fd02043 | 427 | |
e65f152e QW |
428 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
429 | entry_end = entry->file_offset + entry->num_bytes; | |
58f74b22 | 430 | /* |
e65f152e QW |
431 | * |<-- OE --->| | |
432 | * cur | |
433 | * Go to next OE. | |
58f74b22 | 434 | */ |
e65f152e QW |
435 | if (cur >= entry_end) { |
436 | node = rb_next(node); | |
437 | /* No more ordered extents, exit */ | |
438 | if (!node) | |
439 | break; | |
440 | entry = rb_entry(node, struct btrfs_ordered_extent, | |
441 | rb_node); | |
442 | ||
443 | /* Go to next ordered extent and continue */ | |
444 | cur = entry->file_offset; | |
445 | continue; | |
446 | } | |
447 | /* | |
448 | * | |<--- OE --->| | |
449 | * cur | |
450 | * Go to the start of OE. | |
451 | */ | |
452 | if (cur < entry->file_offset) { | |
453 | cur = entry->file_offset; | |
454 | continue; | |
455 | } | |
456 | ||
457 | /* | |
458 | * Now we are definitely inside one ordered extent. | |
459 | * | |
460 | * |<--- OE --->| | |
461 | * | | |
462 | * cur | |
463 | */ | |
464 | end = min(entry->file_offset + entry->num_bytes, | |
465 | file_offset + num_bytes) - 1; | |
466 | ASSERT(end + 1 - cur < U32_MAX); | |
467 | len = end + 1 - cur; | |
468 | ||
53df2586 | 469 | if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) { |
e65f152e | 470 | spin_unlock_irqrestore(&tree->lock, flags); |
2d6f107e | 471 | btrfs_queue_ordered_fn(entry); |
e65f152e QW |
472 | spin_lock_irqsave(&tree->lock, flags); |
473 | } | |
474 | cur += len; | |
163cf09c | 475 | } |
5fd02043 | 476 | spin_unlock_irqrestore(&tree->lock, flags); |
163cf09c CM |
477 | } |
478 | ||
eb84ae03 | 479 | /* |
58f74b22 QW |
480 | * Finish IO for one ordered extent across a given range. The range can only |
481 | * contain one ordered extent. | |
482 | * | |
483 | * @cached: The cached ordered extent. If not NULL, we can skip the tree | |
484 | * search and use the ordered extent directly. | |
485 | * Will be also used to store the finished ordered extent. | |
486 | * @file_offset: File offset for the finished IO | |
487 | * @io_size: Length of the finish IO range | |
eb84ae03 | 488 | * |
58f74b22 QW |
489 | * Return true if the ordered extent is finished in the range, and update |
490 | * @cached. | |
491 | * Return false otherwise. | |
492 | * | |
493 | * NOTE: The range can NOT cross multiple ordered extents. | |
494 | * Thus caller should ensure the range doesn't cross ordered extents. | |
eb84ae03 | 495 | */ |
58f74b22 QW |
496 | bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode, |
497 | struct btrfs_ordered_extent **cached, | |
f41b6ba9 | 498 | u64 file_offset, u64 io_size) |
dc17ff8f | 499 | { |
90c0304c | 500 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; |
dc17ff8f | 501 | struct rb_node *node; |
5a1a3df1 | 502 | struct btrfs_ordered_extent *entry = NULL; |
5fd02043 | 503 | unsigned long flags; |
58f74b22 | 504 | bool finished = false; |
e6dcd2dc | 505 | |
5fd02043 JB |
506 | spin_lock_irqsave(&tree->lock, flags); |
507 | if (cached && *cached) { | |
508 | entry = *cached; | |
509 | goto have_entry; | |
510 | } | |
511 | ||
e6dcd2dc | 512 | node = tree_search(tree, file_offset); |
58f74b22 | 513 | if (!node) |
e6dcd2dc | 514 | goto out; |
dc17ff8f | 515 | |
e6dcd2dc | 516 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 517 | have_entry: |
20bbf20e | 518 | if (!in_range(file_offset, entry->file_offset, entry->num_bytes)) |
e6dcd2dc | 519 | goto out; |
e6dcd2dc | 520 | |
58f74b22 | 521 | if (io_size > entry->bytes_left) |
90c0304c | 522 | btrfs_crit(inode->root->fs_info, |
efe120a0 | 523 | "bad ordered accounting left %llu size %llu", |
c1c9ff7c | 524 | entry->bytes_left, io_size); |
58f74b22 | 525 | |
8b62b72b | 526 | entry->bytes_left -= io_size; |
5fd02043 | 527 | |
af7a6509 | 528 | if (entry->bytes_left == 0) { |
58f74b22 QW |
529 | /* |
530 | * Ensure only one caller can set the flag and finished_ret | |
531 | * accordingly | |
532 | */ | |
533 | finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
093258e6 DS |
534 | /* test_and_set_bit implies a barrier */ |
535 | cond_wake_up_nomb(&entry->wait); | |
af7a6509 | 536 | } |
e6dcd2dc | 537 | out: |
58f74b22 | 538 | if (finished && cached && entry) { |
5a1a3df1 | 539 | *cached = entry; |
e76edab7 | 540 | refcount_inc(&entry->refs); |
5bea2508 | 541 | trace_btrfs_ordered_extent_dec_test_pending(inode, entry); |
5a1a3df1 | 542 | } |
5fd02043 | 543 | spin_unlock_irqrestore(&tree->lock, flags); |
58f74b22 | 544 | return finished; |
e6dcd2dc | 545 | } |
dc17ff8f | 546 | |
eb84ae03 CM |
547 | /* |
548 | * used to drop a reference on an ordered extent. This will free | |
549 | * the extent if the last reference is dropped | |
550 | */ | |
143bede5 | 551 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc | 552 | { |
ba1da2f4 CM |
553 | struct list_head *cur; |
554 | struct btrfs_ordered_sum *sum; | |
555 | ||
acbf1dd0 | 556 | trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry); |
1abe9b8a | 557 | |
e76edab7 | 558 | if (refcount_dec_and_test(&entry->refs)) { |
61de718f | 559 | ASSERT(list_empty(&entry->root_extent_list)); |
48778179 | 560 | ASSERT(list_empty(&entry->log_list)); |
61de718f | 561 | ASSERT(RB_EMPTY_NODE(&entry->rb_node)); |
5fd02043 | 562 | if (entry->inode) |
e55cf7ca | 563 | btrfs_add_delayed_iput(BTRFS_I(entry->inode)); |
d397712b | 564 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
565 | cur = entry->list.next; |
566 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
567 | list_del(&sum->list); | |
a3d46aea | 568 | kvfree(sum); |
ba1da2f4 | 569 | } |
6352b91d | 570 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
ba1da2f4 | 571 | } |
dc17ff8f | 572 | } |
cee36a03 | 573 | |
eb84ae03 CM |
574 | /* |
575 | * remove an ordered extent from the tree. No references are dropped | |
5fd02043 | 576 | * and waiters are woken up. |
eb84ae03 | 577 | */ |
71fe0a55 | 578 | void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode, |
5fd02043 | 579 | struct btrfs_ordered_extent *entry) |
cee36a03 | 580 | { |
e6dcd2dc | 581 | struct btrfs_ordered_inode_tree *tree; |
8b62f87b | 582 | struct btrfs_root *root = btrfs_inode->root; |
71fe0a55 | 583 | struct btrfs_fs_info *fs_info = root->fs_info; |
cee36a03 | 584 | struct rb_node *node; |
48778179 | 585 | bool pending; |
5f4403e1 IA |
586 | bool freespace_inode; |
587 | ||
588 | /* | |
589 | * If this is a free space inode the thread has not acquired the ordered | |
590 | * extents lockdep map. | |
591 | */ | |
592 | freespace_inode = btrfs_is_free_space_inode(btrfs_inode); | |
cee36a03 | 593 | |
8b53779e | 594 | btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered); |
ebfe4d4e | 595 | /* This is paired with btrfs_alloc_ordered_extent. */ |
8b62f87b JB |
596 | spin_lock(&btrfs_inode->lock); |
597 | btrfs_mod_outstanding_extents(btrfs_inode, -1); | |
598 | spin_unlock(&btrfs_inode->lock); | |
7c0c7269 OS |
599 | if (root != fs_info->tree_root) { |
600 | u64 release; | |
601 | ||
602 | if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags)) | |
603 | release = entry->disk_num_bytes; | |
604 | else | |
605 | release = entry->num_bytes; | |
606 | btrfs_delalloc_release_metadata(btrfs_inode, release, false); | |
607 | } | |
8b62f87b | 608 | |
5deb17e1 JB |
609 | percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes, |
610 | fs_info->delalloc_batch); | |
4297ff84 | 611 | |
8b62f87b | 612 | tree = &btrfs_inode->ordered_tree; |
5fd02043 | 613 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 614 | node = &entry->rb_node; |
cee36a03 | 615 | rb_erase(node, &tree->tree); |
61de718f | 616 | RB_CLEAR_NODE(node); |
1b8e7e45 FDBM |
617 | if (tree->last == node) |
618 | tree->last = NULL; | |
e6dcd2dc | 619 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); |
48778179 | 620 | pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags); |
5fd02043 | 621 | spin_unlock_irq(&tree->lock); |
3eaa2885 | 622 | |
48778179 FM |
623 | /* |
624 | * The current running transaction is waiting on us, we need to let it | |
625 | * know that we're complete and wake it up. | |
626 | */ | |
627 | if (pending) { | |
628 | struct btrfs_transaction *trans; | |
629 | ||
630 | /* | |
631 | * The checks for trans are just a formality, it should be set, | |
632 | * but if it isn't we don't want to deref/assert under the spin | |
633 | * lock, so be nice and check if trans is set, but ASSERT() so | |
634 | * if it isn't set a developer will notice. | |
635 | */ | |
636 | spin_lock(&fs_info->trans_lock); | |
637 | trans = fs_info->running_transaction; | |
638 | if (trans) | |
639 | refcount_inc(&trans->use_count); | |
640 | spin_unlock(&fs_info->trans_lock); | |
641 | ||
642 | ASSERT(trans); | |
643 | if (trans) { | |
644 | if (atomic_dec_and_test(&trans->pending_ordered)) | |
645 | wake_up(&trans->pending_wait); | |
646 | btrfs_put_transaction(trans); | |
647 | } | |
648 | } | |
649 | ||
8b53779e IA |
650 | btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered); |
651 | ||
199c2a9c | 652 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 653 | list_del_init(&entry->root_extent_list); |
199c2a9c | 654 | root->nr_ordered_extents--; |
5a3f23d5 | 655 | |
71fe0a55 | 656 | trace_btrfs_ordered_extent_remove(btrfs_inode, entry); |
1abe9b8a | 657 | |
199c2a9c | 658 | if (!root->nr_ordered_extents) { |
0b246afa | 659 | spin_lock(&fs_info->ordered_root_lock); |
199c2a9c MX |
660 | BUG_ON(list_empty(&root->ordered_root)); |
661 | list_del_init(&root->ordered_root); | |
0b246afa | 662 | spin_unlock(&fs_info->ordered_root_lock); |
199c2a9c MX |
663 | } |
664 | spin_unlock(&root->ordered_extent_lock); | |
e6dcd2dc | 665 | wake_up(&entry->wait); |
5f4403e1 IA |
666 | if (!freespace_inode) |
667 | btrfs_lockdep_release(fs_info, btrfs_ordered_extent); | |
cee36a03 CM |
668 | } |
669 | ||
d458b054 | 670 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
9afab882 MX |
671 | { |
672 | struct btrfs_ordered_extent *ordered; | |
673 | ||
674 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); | |
36d45567 | 675 | btrfs_start_ordered_extent(ordered); |
9afab882 MX |
676 | complete(&ordered->completion); |
677 | } | |
678 | ||
d352ac68 CM |
679 | /* |
680 | * wait for all the ordered extents in a root. This is done when balancing | |
681 | * space between drives. | |
682 | */ | |
6374e57a | 683 | u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr, |
578def7c | 684 | const u64 range_start, const u64 range_len) |
3eaa2885 | 685 | { |
0b246afa | 686 | struct btrfs_fs_info *fs_info = root->fs_info; |
578def7c FM |
687 | LIST_HEAD(splice); |
688 | LIST_HEAD(skipped); | |
689 | LIST_HEAD(works); | |
9afab882 | 690 | struct btrfs_ordered_extent *ordered, *next; |
6374e57a | 691 | u64 count = 0; |
578def7c | 692 | const u64 range_end = range_start + range_len; |
3eaa2885 | 693 | |
31f3d255 | 694 | mutex_lock(&root->ordered_extent_mutex); |
199c2a9c MX |
695 | spin_lock(&root->ordered_extent_lock); |
696 | list_splice_init(&root->ordered_extents, &splice); | |
b0244199 | 697 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
698 | ordered = list_first_entry(&splice, struct btrfs_ordered_extent, |
699 | root_extent_list); | |
578def7c | 700 | |
bffe633e OS |
701 | if (range_end <= ordered->disk_bytenr || |
702 | ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) { | |
578def7c FM |
703 | list_move_tail(&ordered->root_extent_list, &skipped); |
704 | cond_resched_lock(&root->ordered_extent_lock); | |
705 | continue; | |
706 | } | |
707 | ||
199c2a9c MX |
708 | list_move_tail(&ordered->root_extent_list, |
709 | &root->ordered_extents); | |
e76edab7 | 710 | refcount_inc(&ordered->refs); |
199c2a9c | 711 | spin_unlock(&root->ordered_extent_lock); |
3eaa2885 | 712 | |
a44903ab QW |
713 | btrfs_init_work(&ordered->flush_work, |
714 | btrfs_run_ordered_extent_work, NULL, NULL); | |
199c2a9c | 715 | list_add_tail(&ordered->work_list, &works); |
0b246afa | 716 | btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work); |
3eaa2885 | 717 | |
9afab882 | 718 | cond_resched(); |
199c2a9c | 719 | spin_lock(&root->ordered_extent_lock); |
6374e57a | 720 | if (nr != U64_MAX) |
b0244199 MX |
721 | nr--; |
722 | count++; | |
3eaa2885 | 723 | } |
578def7c | 724 | list_splice_tail(&skipped, &root->ordered_extents); |
b0244199 | 725 | list_splice_tail(&splice, &root->ordered_extents); |
199c2a9c | 726 | spin_unlock(&root->ordered_extent_lock); |
9afab882 MX |
727 | |
728 | list_for_each_entry_safe(ordered, next, &works, work_list) { | |
729 | list_del_init(&ordered->work_list); | |
730 | wait_for_completion(&ordered->completion); | |
9afab882 | 731 | btrfs_put_ordered_extent(ordered); |
9afab882 MX |
732 | cond_resched(); |
733 | } | |
31f3d255 | 734 | mutex_unlock(&root->ordered_extent_mutex); |
b0244199 MX |
735 | |
736 | return count; | |
3eaa2885 CM |
737 | } |
738 | ||
042528f8 | 739 | void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr, |
6374e57a | 740 | const u64 range_start, const u64 range_len) |
199c2a9c MX |
741 | { |
742 | struct btrfs_root *root; | |
84af994b | 743 | LIST_HEAD(splice); |
6374e57a | 744 | u64 done; |
199c2a9c | 745 | |
8b9d83cd | 746 | mutex_lock(&fs_info->ordered_operations_mutex); |
199c2a9c MX |
747 | spin_lock(&fs_info->ordered_root_lock); |
748 | list_splice_init(&fs_info->ordered_roots, &splice); | |
b0244199 | 749 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
750 | root = list_first_entry(&splice, struct btrfs_root, |
751 | ordered_root); | |
00246528 | 752 | root = btrfs_grab_root(root); |
199c2a9c MX |
753 | BUG_ON(!root); |
754 | list_move_tail(&root->ordered_root, | |
755 | &fs_info->ordered_roots); | |
756 | spin_unlock(&fs_info->ordered_root_lock); | |
757 | ||
578def7c FM |
758 | done = btrfs_wait_ordered_extents(root, nr, |
759 | range_start, range_len); | |
00246528 | 760 | btrfs_put_root(root); |
199c2a9c MX |
761 | |
762 | spin_lock(&fs_info->ordered_root_lock); | |
6374e57a | 763 | if (nr != U64_MAX) { |
b0244199 | 764 | nr -= done; |
b0244199 | 765 | } |
199c2a9c | 766 | } |
931aa877 | 767 | list_splice_tail(&splice, &fs_info->ordered_roots); |
199c2a9c | 768 | spin_unlock(&fs_info->ordered_root_lock); |
8b9d83cd | 769 | mutex_unlock(&fs_info->ordered_operations_mutex); |
199c2a9c MX |
770 | } |
771 | ||
eb84ae03 | 772 | /* |
36d45567 | 773 | * Start IO and wait for a given ordered extent to finish. |
eb84ae03 | 774 | * |
36d45567 CH |
775 | * Wait on page writeback for all the pages in the extent and the IO completion |
776 | * code to insert metadata into the btree corresponding to the extent. | |
eb84ae03 | 777 | */ |
36d45567 | 778 | void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc CM |
779 | { |
780 | u64 start = entry->file_offset; | |
bffe633e | 781 | u64 end = start + entry->num_bytes - 1; |
c0a43603 | 782 | struct btrfs_inode *inode = BTRFS_I(entry->inode); |
5f4403e1 | 783 | bool freespace_inode; |
e1b81e67 | 784 | |
c0a43603 | 785 | trace_btrfs_ordered_extent_start(inode, entry); |
1abe9b8a | 786 | |
5f4403e1 IA |
787 | /* |
788 | * If this is a free space inode do not take the ordered extents lockdep | |
789 | * map. | |
790 | */ | |
791 | freespace_inode = btrfs_is_free_space_inode(inode); | |
792 | ||
eb84ae03 CM |
793 | /* |
794 | * pages in the range can be dirty, clean or writeback. We | |
795 | * start IO on any dirty ones so the wait doesn't stall waiting | |
b2570314 | 796 | * for the flusher thread to find them |
eb84ae03 | 797 | */ |
4b46fce2 | 798 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
c0a43603 | 799 | filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end); |
36d45567 CH |
800 | |
801 | if (!freespace_inode) | |
802 | btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent); | |
803 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags)); | |
e6dcd2dc | 804 | } |
cee36a03 | 805 | |
eb84ae03 CM |
806 | /* |
807 | * Used to wait on ordered extents across a large range of bytes. | |
808 | */ | |
0ef8b726 | 809 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc | 810 | { |
0ef8b726 | 811 | int ret = 0; |
28aeeac1 | 812 | int ret_wb = 0; |
e6dcd2dc | 813 | u64 end; |
e5a2217e | 814 | u64 orig_end; |
e6dcd2dc | 815 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
816 | |
817 | if (start + len < start) { | |
cf260db4 | 818 | orig_end = OFFSET_MAX; |
e5a2217e CM |
819 | } else { |
820 | orig_end = start + len - 1; | |
cf260db4 ZL |
821 | if (orig_end > OFFSET_MAX) |
822 | orig_end = OFFSET_MAX; | |
e5a2217e | 823 | } |
551ebb2d | 824 | |
e5a2217e CM |
825 | /* start IO across the range first to instantiate any delalloc |
826 | * extents | |
827 | */ | |
728404da | 828 | ret = btrfs_fdatawrite_range(inode, start, orig_end); |
0ef8b726 JB |
829 | if (ret) |
830 | return ret; | |
728404da | 831 | |
28aeeac1 FM |
832 | /* |
833 | * If we have a writeback error don't return immediately. Wait first | |
834 | * for any ordered extents that haven't completed yet. This is to make | |
835 | * sure no one can dirty the same page ranges and call writepages() | |
836 | * before the ordered extents complete - to avoid failures (-EEXIST) | |
837 | * when adding the new ordered extents to the ordered tree. | |
838 | */ | |
839 | ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end); | |
e5a2217e | 840 | |
f421950f | 841 | end = orig_end; |
d397712b | 842 | while (1) { |
6d072c8e | 843 | ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end); |
d397712b | 844 | if (!ordered) |
e6dcd2dc | 845 | break; |
e5a2217e | 846 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
847 | btrfs_put_ordered_extent(ordered); |
848 | break; | |
849 | } | |
bffe633e | 850 | if (ordered->file_offset + ordered->num_bytes <= start) { |
e6dcd2dc CM |
851 | btrfs_put_ordered_extent(ordered); |
852 | break; | |
853 | } | |
36d45567 | 854 | btrfs_start_ordered_extent(ordered); |
e6dcd2dc | 855 | end = ordered->file_offset; |
e75fd33b FM |
856 | /* |
857 | * If the ordered extent had an error save the error but don't | |
858 | * exit without waiting first for all other ordered extents in | |
859 | * the range to complete. | |
860 | */ | |
0ef8b726 JB |
861 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) |
862 | ret = -EIO; | |
e6dcd2dc | 863 | btrfs_put_ordered_extent(ordered); |
e75fd33b | 864 | if (end == 0 || end == start) |
e6dcd2dc CM |
865 | break; |
866 | end--; | |
867 | } | |
28aeeac1 | 868 | return ret_wb ? ret_wb : ret; |
cee36a03 CM |
869 | } |
870 | ||
eb84ae03 CM |
871 | /* |
872 | * find an ordered extent corresponding to file_offset. return NULL if | |
873 | * nothing is found, otherwise take a reference on the extent and return it | |
874 | */ | |
c3504372 | 875 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode, |
e6dcd2dc CM |
876 | u64 file_offset) |
877 | { | |
878 | struct btrfs_ordered_inode_tree *tree; | |
879 | struct rb_node *node; | |
880 | struct btrfs_ordered_extent *entry = NULL; | |
24533f6a | 881 | unsigned long flags; |
e6dcd2dc | 882 | |
c3504372 | 883 | tree = &inode->ordered_tree; |
24533f6a | 884 | spin_lock_irqsave(&tree->lock, flags); |
e6dcd2dc CM |
885 | node = tree_search(tree, file_offset); |
886 | if (!node) | |
887 | goto out; | |
888 | ||
889 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
20bbf20e | 890 | if (!in_range(file_offset, entry->file_offset, entry->num_bytes)) |
e6dcd2dc | 891 | entry = NULL; |
5bea2508 | 892 | if (entry) { |
e76edab7 | 893 | refcount_inc(&entry->refs); |
5bea2508 JT |
894 | trace_btrfs_ordered_extent_lookup(inode, entry); |
895 | } | |
e6dcd2dc | 896 | out: |
24533f6a | 897 | spin_unlock_irqrestore(&tree->lock, flags); |
e6dcd2dc CM |
898 | return entry; |
899 | } | |
900 | ||
4b46fce2 JB |
901 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
902 | * extents that exist in the range, rather than just the start of the range. | |
903 | */ | |
a776c6fa NB |
904 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range( |
905 | struct btrfs_inode *inode, u64 file_offset, u64 len) | |
4b46fce2 JB |
906 | { |
907 | struct btrfs_ordered_inode_tree *tree; | |
908 | struct rb_node *node; | |
909 | struct btrfs_ordered_extent *entry = NULL; | |
910 | ||
a776c6fa | 911 | tree = &inode->ordered_tree; |
5fd02043 | 912 | spin_lock_irq(&tree->lock); |
4b46fce2 JB |
913 | node = tree_search(tree, file_offset); |
914 | if (!node) { | |
915 | node = tree_search(tree, file_offset + len); | |
916 | if (!node) | |
917 | goto out; | |
918 | } | |
919 | ||
920 | while (1) { | |
921 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
922 | if (range_overlaps(entry, file_offset, len)) | |
923 | break; | |
924 | ||
925 | if (entry->file_offset >= file_offset + len) { | |
926 | entry = NULL; | |
927 | break; | |
928 | } | |
929 | entry = NULL; | |
930 | node = rb_next(node); | |
931 | if (!node) | |
932 | break; | |
933 | } | |
934 | out: | |
5bea2508 | 935 | if (entry) { |
e76edab7 | 936 | refcount_inc(&entry->refs); |
5bea2508 JT |
937 | trace_btrfs_ordered_extent_lookup_range(inode, entry); |
938 | } | |
5fd02043 | 939 | spin_unlock_irq(&tree->lock); |
4b46fce2 JB |
940 | return entry; |
941 | } | |
942 | ||
48778179 FM |
943 | /* |
944 | * Adds all ordered extents to the given list. The list ends up sorted by the | |
945 | * file_offset of the ordered extents. | |
946 | */ | |
947 | void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode, | |
948 | struct list_head *list) | |
949 | { | |
950 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; | |
951 | struct rb_node *n; | |
952 | ||
953 | ASSERT(inode_is_locked(&inode->vfs_inode)); | |
954 | ||
955 | spin_lock_irq(&tree->lock); | |
956 | for (n = rb_first(&tree->tree); n; n = rb_next(n)) { | |
957 | struct btrfs_ordered_extent *ordered; | |
958 | ||
959 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); | |
960 | ||
961 | if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags)) | |
962 | continue; | |
963 | ||
964 | ASSERT(list_empty(&ordered->log_list)); | |
965 | list_add_tail(&ordered->log_list, list); | |
966 | refcount_inc(&ordered->refs); | |
5bea2508 | 967 | trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered); |
48778179 FM |
968 | } |
969 | spin_unlock_irq(&tree->lock); | |
970 | } | |
971 | ||
eb84ae03 CM |
972 | /* |
973 | * lookup and return any extent before 'file_offset'. NULL is returned | |
974 | * if none is found | |
975 | */ | |
e6dcd2dc | 976 | struct btrfs_ordered_extent * |
6d072c8e | 977 | btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset) |
e6dcd2dc CM |
978 | { |
979 | struct btrfs_ordered_inode_tree *tree; | |
980 | struct rb_node *node; | |
981 | struct btrfs_ordered_extent *entry = NULL; | |
982 | ||
6d072c8e | 983 | tree = &inode->ordered_tree; |
5fd02043 | 984 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
985 | node = tree_search(tree, file_offset); |
986 | if (!node) | |
987 | goto out; | |
988 | ||
989 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
e76edab7 | 990 | refcount_inc(&entry->refs); |
5bea2508 | 991 | trace_btrfs_ordered_extent_lookup_first(inode, entry); |
e6dcd2dc | 992 | out: |
5fd02043 | 993 | spin_unlock_irq(&tree->lock); |
e6dcd2dc | 994 | return entry; |
81d7ed29 | 995 | } |
dbe674a9 | 996 | |
c095f333 QW |
997 | /* |
998 | * Lookup the first ordered extent that overlaps the range | |
999 | * [@file_offset, @file_offset + @len). | |
1000 | * | |
1001 | * The difference between this and btrfs_lookup_first_ordered_extent() is | |
1002 | * that this one won't return any ordered extent that does not overlap the range. | |
1003 | * And the difference against btrfs_lookup_ordered_extent() is, this function | |
1004 | * ensures the first ordered extent gets returned. | |
1005 | */ | |
1006 | struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range( | |
1007 | struct btrfs_inode *inode, u64 file_offset, u64 len) | |
1008 | { | |
1009 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; | |
1010 | struct rb_node *node; | |
1011 | struct rb_node *cur; | |
1012 | struct rb_node *prev; | |
1013 | struct rb_node *next; | |
1014 | struct btrfs_ordered_extent *entry = NULL; | |
1015 | ||
1016 | spin_lock_irq(&tree->lock); | |
1017 | node = tree->tree.rb_node; | |
1018 | /* | |
1019 | * Here we don't want to use tree_search() which will use tree->last | |
1020 | * and screw up the search order. | |
1021 | * And __tree_search() can't return the adjacent ordered extents | |
1022 | * either, thus here we do our own search. | |
1023 | */ | |
1024 | while (node) { | |
1025 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
1026 | ||
1027 | if (file_offset < entry->file_offset) { | |
1028 | node = node->rb_left; | |
1029 | } else if (file_offset >= entry_end(entry)) { | |
1030 | node = node->rb_right; | |
1031 | } else { | |
1032 | /* | |
1033 | * Direct hit, got an ordered extent that starts at | |
1034 | * @file_offset | |
1035 | */ | |
1036 | goto out; | |
1037 | } | |
1038 | } | |
1039 | if (!entry) { | |
1040 | /* Empty tree */ | |
1041 | goto out; | |
1042 | } | |
1043 | ||
1044 | cur = &entry->rb_node; | |
1045 | /* We got an entry around @file_offset, check adjacent entries */ | |
1046 | if (entry->file_offset < file_offset) { | |
1047 | prev = cur; | |
1048 | next = rb_next(cur); | |
1049 | } else { | |
1050 | prev = rb_prev(cur); | |
1051 | next = cur; | |
1052 | } | |
1053 | if (prev) { | |
1054 | entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node); | |
1055 | if (range_overlaps(entry, file_offset, len)) | |
1056 | goto out; | |
1057 | } | |
1058 | if (next) { | |
1059 | entry = rb_entry(next, struct btrfs_ordered_extent, rb_node); | |
1060 | if (range_overlaps(entry, file_offset, len)) | |
1061 | goto out; | |
1062 | } | |
1063 | /* No ordered extent in the range */ | |
1064 | entry = NULL; | |
1065 | out: | |
5bea2508 | 1066 | if (entry) { |
c095f333 | 1067 | refcount_inc(&entry->refs); |
5bea2508 JT |
1068 | trace_btrfs_ordered_extent_lookup_first_range(inode, entry); |
1069 | } | |
1070 | ||
c095f333 QW |
1071 | spin_unlock_irq(&tree->lock); |
1072 | return entry; | |
1073 | } | |
1074 | ||
ffa87214 | 1075 | /* |
43dd529a DS |
1076 | * Lock the passed range and ensures all pending ordered extents in it are run |
1077 | * to completion. | |
ffa87214 | 1078 | * |
ffa87214 NB |
1079 | * @inode: Inode whose ordered tree is to be searched |
1080 | * @start: Beginning of range to flush | |
1081 | * @end: Last byte of range to lock | |
1082 | * @cached_state: If passed, will return the extent state responsible for the | |
43dd529a DS |
1083 | * locked range. It's the caller's responsibility to free the |
1084 | * cached state. | |
ffa87214 | 1085 | * |
43dd529a DS |
1086 | * Always return with the given range locked, ensuring after it's called no |
1087 | * order extent can be pending. | |
ffa87214 | 1088 | */ |
b272ae22 | 1089 | void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start, |
ffa87214 NB |
1090 | u64 end, |
1091 | struct extent_state **cached_state) | |
1092 | { | |
1093 | struct btrfs_ordered_extent *ordered; | |
a3b46b86 NA |
1094 | struct extent_state *cache = NULL; |
1095 | struct extent_state **cachedp = &cache; | |
bd80d94e NB |
1096 | |
1097 | if (cached_state) | |
a3b46b86 | 1098 | cachedp = cached_state; |
ffa87214 NB |
1099 | |
1100 | while (1) { | |
570eb97b | 1101 | lock_extent(&inode->io_tree, start, end, cachedp); |
ffa87214 NB |
1102 | ordered = btrfs_lookup_ordered_range(inode, start, |
1103 | end - start + 1); | |
bd80d94e NB |
1104 | if (!ordered) { |
1105 | /* | |
1106 | * If no external cached_state has been passed then | |
1107 | * decrement the extra ref taken for cachedp since we | |
1108 | * aren't exposing it outside of this function | |
1109 | */ | |
1110 | if (!cached_state) | |
a3b46b86 | 1111 | refcount_dec(&cache->refs); |
ffa87214 | 1112 | break; |
bd80d94e | 1113 | } |
570eb97b | 1114 | unlock_extent(&inode->io_tree, start, end, cachedp); |
36d45567 | 1115 | btrfs_start_ordered_extent(ordered); |
ffa87214 NB |
1116 | btrfs_put_ordered_extent(ordered); |
1117 | } | |
1118 | } | |
1119 | ||
d2c7a19f JB |
1120 | /* |
1121 | * Lock the passed range and ensure all pending ordered extents in it are run | |
1122 | * to completion in nowait mode. | |
1123 | * | |
1124 | * Return true if btrfs_lock_ordered_range does not return any extents, | |
1125 | * otherwise false. | |
1126 | */ | |
632ddfa2 JB |
1127 | bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end, |
1128 | struct extent_state **cached_state) | |
d2c7a19f JB |
1129 | { |
1130 | struct btrfs_ordered_extent *ordered; | |
1131 | ||
632ddfa2 | 1132 | if (!try_lock_extent(&inode->io_tree, start, end, cached_state)) |
d2c7a19f JB |
1133 | return false; |
1134 | ||
1135 | ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1); | |
1136 | if (!ordered) | |
1137 | return true; | |
1138 | ||
1139 | btrfs_put_ordered_extent(ordered); | |
632ddfa2 | 1140 | unlock_extent(&inode->io_tree, start, end, cached_state); |
d2c7a19f JB |
1141 | |
1142 | return false; | |
1143 | } | |
1144 | ||
8f4af4b8 | 1145 | /* Split out a new ordered extent for this first @len bytes of @ordered. */ |
b0307e28 CH |
1146 | struct btrfs_ordered_extent *btrfs_split_ordered_extent( |
1147 | struct btrfs_ordered_extent *ordered, u64 len) | |
d22002fd | 1148 | { |
816f589b CH |
1149 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); |
1150 | struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree; | |
1151 | struct btrfs_root *root = inode->root; | |
1152 | struct btrfs_fs_info *fs_info = root->fs_info; | |
f0792b79 CH |
1153 | u64 file_offset = ordered->file_offset; |
1154 | u64 disk_bytenr = ordered->disk_bytenr; | |
52b1fdca CH |
1155 | unsigned long flags = ordered->flags; |
1156 | struct btrfs_ordered_sum *sum, *tmpsum; | |
816f589b | 1157 | struct btrfs_ordered_extent *new; |
8f4af4b8 | 1158 | struct rb_node *node; |
52b1fdca | 1159 | u64 offset = 0; |
d22002fd | 1160 | |
816f589b | 1161 | trace_btrfs_ordered_extent_split(inode, ordered); |
5bea2508 | 1162 | |
f0792b79 CH |
1163 | ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED))); |
1164 | ||
8f4af4b8 CH |
1165 | /* |
1166 | * The entire bio must be covered by the ordered extent, but we can't | |
1167 | * reduce the original extent to a zero length either. | |
1168 | */ | |
1169 | if (WARN_ON_ONCE(len >= ordered->num_bytes)) | |
b0307e28 | 1170 | return ERR_PTR(-EINVAL); |
52b1fdca CH |
1171 | /* We cannot split partially completed ordered extents. */ |
1172 | if (ordered->bytes_left) { | |
1173 | ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS)); | |
1174 | if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) | |
1175 | return ERR_PTR(-EINVAL); | |
1176 | } | |
e44ca71c CH |
1177 | /* We cannot split a compressed ordered extent. */ |
1178 | if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) | |
b0307e28 | 1179 | return ERR_PTR(-EINVAL); |
e44ca71c | 1180 | |
816f589b CH |
1181 | new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr, |
1182 | len, 0, flags, ordered->compress_type); | |
1183 | if (IS_ERR(new)) | |
1184 | return new; | |
1185 | ||
1186 | /* One ref for the tree. */ | |
1187 | refcount_inc(&new->refs); | |
1188 | ||
1189 | spin_lock_irq(&root->ordered_extent_lock); | |
1190 | spin_lock(&tree->lock); | |
d22002fd NA |
1191 | /* Remove from tree once */ |
1192 | node = &ordered->rb_node; | |
1193 | rb_erase(node, &tree->tree); | |
1194 | RB_CLEAR_NODE(node); | |
1195 | if (tree->last == node) | |
1196 | tree->last = NULL; | |
1197 | ||
8f4af4b8 CH |
1198 | ordered->file_offset += len; |
1199 | ordered->disk_bytenr += len; | |
1200 | ordered->num_bytes -= len; | |
1201 | ordered->disk_num_bytes -= len; | |
52b1fdca CH |
1202 | |
1203 | if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) { | |
1204 | ASSERT(ordered->bytes_left == 0); | |
1205 | new->bytes_left = 0; | |
1206 | } else { | |
1207 | ordered->bytes_left -= len; | |
1208 | } | |
1209 | ||
1210 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) { | |
1211 | if (ordered->truncated_len > len) { | |
1212 | ordered->truncated_len -= len; | |
1213 | } else { | |
1214 | new->truncated_len = ordered->truncated_len; | |
1215 | ordered->truncated_len = 0; | |
1216 | } | |
1217 | } | |
1218 | ||
1219 | list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) { | |
1220 | if (offset == len) | |
1221 | break; | |
1222 | list_move_tail(&sum->list, &new->list); | |
1223 | offset += sum->len; | |
1224 | } | |
d22002fd NA |
1225 | |
1226 | /* Re-insert the node */ | |
1227 | node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node); | |
1228 | if (node) | |
1229 | btrfs_panic(fs_info, -EEXIST, | |
1230 | "zoned: inconsistency in ordered tree at offset %llu", | |
816f589b | 1231 | ordered->file_offset); |
d22002fd | 1232 | |
816f589b CH |
1233 | node = tree_insert(&tree->tree, new->file_offset, &new->rb_node); |
1234 | if (node) | |
1235 | btrfs_panic(fs_info, -EEXIST, | |
1236 | "zoned: inconsistency in ordered tree at offset %llu", | |
1237 | new->file_offset); | |
1238 | spin_unlock(&tree->lock); | |
f0792b79 | 1239 | |
816f589b CH |
1240 | list_add_tail(&new->root_extent_list, &root->ordered_extents); |
1241 | root->nr_ordered_extents++; | |
1242 | spin_unlock_irq(&root->ordered_extent_lock); | |
1243 | return new; | |
d22002fd NA |
1244 | } |
1245 | ||
6352b91d MX |
1246 | int __init ordered_data_init(void) |
1247 | { | |
1248 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", | |
1249 | sizeof(struct btrfs_ordered_extent), 0, | |
fba4b697 | 1250 | SLAB_MEM_SPREAD, |
6352b91d MX |
1251 | NULL); |
1252 | if (!btrfs_ordered_extent_cache) | |
1253 | return -ENOMEM; | |
25287e0a | 1254 | |
6352b91d MX |
1255 | return 0; |
1256 | } | |
1257 | ||
e67c718b | 1258 | void __cold ordered_data_exit(void) |
6352b91d | 1259 | { |
5598e900 | 1260 | kmem_cache_destroy(btrfs_ordered_extent_cache); |
6352b91d | 1261 | } |