xref: /openbmc/linux/fs/btrfs/space-info.c (revision 82220b18)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "misc.h"
4 #include "ctree.h"
5 #include "space-info.h"
6 #include "sysfs.h"
7 #include "volumes.h"
8 #include "free-space-cache.h"
9 #include "ordered-data.h"
10 #include "transaction.h"
11 #include "block-group.h"
12 #include "zoned.h"
13 #include "fs.h"
14 #include "accessors.h"
15 #include "extent-tree.h"
16 
17 /*
18  * HOW DOES SPACE RESERVATION WORK
19  *
20  * If you want to know about delalloc specifically, there is a separate comment
21  * for that with the delalloc code.  This comment is about how the whole system
22  * works generally.
23  *
24  * BASIC CONCEPTS
25  *
26  *   1) space_info.  This is the ultimate arbiter of how much space we can use.
27  *   There's a description of the bytes_ fields with the struct declaration,
28  *   refer to that for specifics on each field.  Suffice it to say that for
29  *   reservations we care about total_bytes - SUM(space_info->bytes_) when
30  *   determining if there is space to make an allocation.  There is a space_info
31  *   for METADATA, SYSTEM, and DATA areas.
32  *
33  *   2) block_rsv's.  These are basically buckets for every different type of
34  *   metadata reservation we have.  You can see the comment in the block_rsv
35  *   code on the rules for each type, but generally block_rsv->reserved is how
36  *   much space is accounted for in space_info->bytes_may_use.
37  *
38  *   3) btrfs_calc*_size.  These are the worst case calculations we used based
39  *   on the number of items we will want to modify.  We have one for changing
40  *   items, and one for inserting new items.  Generally we use these helpers to
41  *   determine the size of the block reserves, and then use the actual bytes
42  *   values to adjust the space_info counters.
43  *
44  * MAKING RESERVATIONS, THE NORMAL CASE
45  *
46  *   We call into either btrfs_reserve_data_bytes() or
47  *   btrfs_reserve_metadata_bytes(), depending on which we're looking for, with
48  *   num_bytes we want to reserve.
49  *
50  *   ->reserve
51  *     space_info->bytes_may_reserve += num_bytes
52  *
53  *   ->extent allocation
54  *     Call btrfs_add_reserved_bytes() which does
55  *     space_info->bytes_may_reserve -= num_bytes
56  *     space_info->bytes_reserved += extent_bytes
57  *
58  *   ->insert reference
59  *     Call btrfs_update_block_group() which does
60  *     space_info->bytes_reserved -= extent_bytes
61  *     space_info->bytes_used += extent_bytes
62  *
63  * MAKING RESERVATIONS, FLUSHING NORMALLY (non-priority)
64  *
65  *   Assume we are unable to simply make the reservation because we do not have
66  *   enough space
67  *
68  *   -> __reserve_bytes
69  *     create a reserve_ticket with ->bytes set to our reservation, add it to
70  *     the tail of space_info->tickets, kick async flush thread
71  *
72  *   ->handle_reserve_ticket
73  *     wait on ticket->wait for ->bytes to be reduced to 0, or ->error to be set
74  *     on the ticket.
75  *
76  *   -> btrfs_async_reclaim_metadata_space/btrfs_async_reclaim_data_space
77  *     Flushes various things attempting to free up space.
78  *
79  *   -> btrfs_try_granting_tickets()
80  *     This is called by anything that either subtracts space from
81  *     space_info->bytes_may_use, ->bytes_pinned, etc, or adds to the
82  *     space_info->total_bytes.  This loops through the ->priority_tickets and
83  *     then the ->tickets list checking to see if the reservation can be
84  *     completed.  If it can the space is added to space_info->bytes_may_use and
85  *     the ticket is woken up.
86  *
87  *   -> ticket wakeup
88  *     Check if ->bytes == 0, if it does we got our reservation and we can carry
89  *     on, if not return the appropriate error (ENOSPC, but can be EINTR if we
90  *     were interrupted.)
91  *
92  * MAKING RESERVATIONS, FLUSHING HIGH PRIORITY
93  *
94  *   Same as the above, except we add ourselves to the
95  *   space_info->priority_tickets, and we do not use ticket->wait, we simply
96  *   call flush_space() ourselves for the states that are safe for us to call
97  *   without deadlocking and hope for the best.
98  *
99  * THE FLUSHING STATES
100  *
101  *   Generally speaking we will have two cases for each state, a "nice" state
102  *   and a "ALL THE THINGS" state.  In btrfs we delay a lot of work in order to
103  *   reduce the locking over head on the various trees, and even to keep from
104  *   doing any work at all in the case of delayed refs.  Each of these delayed
105  *   things however hold reservations, and so letting them run allows us to
106  *   reclaim space so we can make new reservations.
107  *
108  *   FLUSH_DELAYED_ITEMS
109  *     Every inode has a delayed item to update the inode.  Take a simple write
110  *     for example, we would update the inode item at write time to update the
111  *     mtime, and then again at finish_ordered_io() time in order to update the
112  *     isize or bytes.  We keep these delayed items to coalesce these operations
113  *     into a single operation done on demand.  These are an easy way to reclaim
114  *     metadata space.
115  *
116  *   FLUSH_DELALLOC
117  *     Look at the delalloc comment to get an idea of how much space is reserved
118  *     for delayed allocation.  We can reclaim some of this space simply by
119  *     running delalloc, but usually we need to wait for ordered extents to
120  *     reclaim the bulk of this space.
121  *
122  *   FLUSH_DELAYED_REFS
123  *     We have a block reserve for the outstanding delayed refs space, and every
124  *     delayed ref operation holds a reservation.  Running these is a quick way
125  *     to reclaim space, but we want to hold this until the end because COW can
126  *     churn a lot and we can avoid making some extent tree modifications if we
127  *     are able to delay for as long as possible.
128  *
129  *   ALLOC_CHUNK
130  *     We will skip this the first time through space reservation, because of
131  *     overcommit and we don't want to have a lot of useless metadata space when
132  *     our worst case reservations will likely never come true.
133  *
134  *   RUN_DELAYED_IPUTS
135  *     If we're freeing inodes we're likely freeing checksums, file extent
136  *     items, and extent tree items.  Loads of space could be freed up by these
137  *     operations, however they won't be usable until the transaction commits.
138  *
139  *   COMMIT_TRANS
140  *     This will commit the transaction.  Historically we had a lot of logic
141  *     surrounding whether or not we'd commit the transaction, but this waits born
142  *     out of a pre-tickets era where we could end up committing the transaction
143  *     thousands of times in a row without making progress.  Now thanks to our
144  *     ticketing system we know if we're not making progress and can error
145  *     everybody out after a few commits rather than burning the disk hoping for
146  *     a different answer.
147  *
148  * OVERCOMMIT
149  *
150  *   Because we hold so many reservations for metadata we will allow you to
151  *   reserve more space than is currently free in the currently allocate
152  *   metadata space.  This only happens with metadata, data does not allow
153  *   overcommitting.
154  *
155  *   You can see the current logic for when we allow overcommit in
156  *   btrfs_can_overcommit(), but it only applies to unallocated space.  If there
157  *   is no unallocated space to be had, all reservations are kept within the
158  *   free space in the allocated metadata chunks.
159  *
160  *   Because of overcommitting, you generally want to use the
161  *   btrfs_can_overcommit() logic for metadata allocations, as it does the right
162  *   thing with or without extra unallocated space.
163  */
164 
btrfs_space_info_used(struct btrfs_space_info * s_info,bool may_use_included)165 u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info,
166 			  bool may_use_included)
167 {
168 	ASSERT(s_info);
169 	return s_info->bytes_used + s_info->bytes_reserved +
170 		s_info->bytes_pinned + s_info->bytes_readonly +
171 		s_info->bytes_zone_unusable +
172 		(may_use_included ? s_info->bytes_may_use : 0);
173 }
174 
175 /*
176  * after adding space to the filesystem, we need to clear the full flags
177  * on all the space infos.
178  */
btrfs_clear_space_info_full(struct btrfs_fs_info * info)179 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
180 {
181 	struct list_head *head = &info->space_info;
182 	struct btrfs_space_info *found;
183 
184 	list_for_each_entry(found, head, list)
185 		found->full = 0;
186 }
187 
188 /*
189  * Block groups with more than this value (percents) of unusable space will be
190  * scheduled for background reclaim.
191  */
192 #define BTRFS_DEFAULT_ZONED_RECLAIM_THRESH			(75)
193 
194 /*
195  * Calculate chunk size depending on volume type (regular or zoned).
196  */
calc_chunk_size(const struct btrfs_fs_info * fs_info,u64 flags)197 static u64 calc_chunk_size(const struct btrfs_fs_info *fs_info, u64 flags)
198 {
199 	if (btrfs_is_zoned(fs_info))
200 		return fs_info->zone_size;
201 
202 	ASSERT(flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
203 
204 	if (flags & BTRFS_BLOCK_GROUP_DATA)
205 		return BTRFS_MAX_DATA_CHUNK_SIZE;
206 	else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
207 		return SZ_32M;
208 
209 	/* Handle BTRFS_BLOCK_GROUP_METADATA */
210 	if (fs_info->fs_devices->total_rw_bytes > 50ULL * SZ_1G)
211 		return SZ_1G;
212 
213 	return SZ_256M;
214 }
215 
216 /*
217  * Update default chunk size.
218  */
btrfs_update_space_info_chunk_size(struct btrfs_space_info * space_info,u64 chunk_size)219 void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
220 					u64 chunk_size)
221 {
222 	WRITE_ONCE(space_info->chunk_size, chunk_size);
223 }
224 
create_space_info(struct btrfs_fs_info * info,u64 flags)225 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
226 {
227 
228 	struct btrfs_space_info *space_info;
229 	int i;
230 	int ret;
231 
232 	space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
233 	if (!space_info)
234 		return -ENOMEM;
235 
236 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
237 		INIT_LIST_HEAD(&space_info->block_groups[i]);
238 	init_rwsem(&space_info->groups_sem);
239 	spin_lock_init(&space_info->lock);
240 	space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
241 	space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
242 	INIT_LIST_HEAD(&space_info->ro_bgs);
243 	INIT_LIST_HEAD(&space_info->tickets);
244 	INIT_LIST_HEAD(&space_info->priority_tickets);
245 	space_info->clamp = 1;
246 	btrfs_update_space_info_chunk_size(space_info, calc_chunk_size(info, flags));
247 
248 	if (btrfs_is_zoned(info))
249 		space_info->bg_reclaim_threshold = BTRFS_DEFAULT_ZONED_RECLAIM_THRESH;
250 
251 	ret = btrfs_sysfs_add_space_info_type(info, space_info);
252 	if (ret)
253 		return ret;
254 
255 	list_add(&space_info->list, &info->space_info);
256 	if (flags & BTRFS_BLOCK_GROUP_DATA)
257 		info->data_sinfo = space_info;
258 
259 	return ret;
260 }
261 
btrfs_init_space_info(struct btrfs_fs_info * fs_info)262 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
263 {
264 	struct btrfs_super_block *disk_super;
265 	u64 features;
266 	u64 flags;
267 	int mixed = 0;
268 	int ret;
269 
270 	disk_super = fs_info->super_copy;
271 	if (!btrfs_super_root(disk_super))
272 		return -EINVAL;
273 
274 	features = btrfs_super_incompat_flags(disk_super);
275 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
276 		mixed = 1;
277 
278 	flags = BTRFS_BLOCK_GROUP_SYSTEM;
279 	ret = create_space_info(fs_info, flags);
280 	if (ret)
281 		goto out;
282 
283 	if (mixed) {
284 		flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
285 		ret = create_space_info(fs_info, flags);
286 	} else {
287 		flags = BTRFS_BLOCK_GROUP_METADATA;
288 		ret = create_space_info(fs_info, flags);
289 		if (ret)
290 			goto out;
291 
292 		flags = BTRFS_BLOCK_GROUP_DATA;
293 		ret = create_space_info(fs_info, flags);
294 	}
295 out:
296 	return ret;
297 }
298 
btrfs_add_bg_to_space_info(struct btrfs_fs_info * info,struct btrfs_block_group * block_group)299 void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
300 				struct btrfs_block_group *block_group)
301 {
302 	struct btrfs_space_info *found;
303 	int factor, index;
304 
305 	factor = btrfs_bg_type_to_factor(block_group->flags);
306 
307 	found = btrfs_find_space_info(info, block_group->flags);
308 	ASSERT(found);
309 	spin_lock(&found->lock);
310 	found->total_bytes += block_group->length;
311 	found->disk_total += block_group->length * factor;
312 	found->bytes_used += block_group->used;
313 	found->disk_used += block_group->used * factor;
314 	found->bytes_readonly += block_group->bytes_super;
315 	found->bytes_zone_unusable += block_group->zone_unusable;
316 	if (block_group->length > 0)
317 		found->full = 0;
318 	btrfs_try_granting_tickets(info, found);
319 	spin_unlock(&found->lock);
320 
321 	block_group->space_info = found;
322 
323 	index = btrfs_bg_flags_to_raid_index(block_group->flags);
324 	down_write(&found->groups_sem);
325 	list_add_tail(&block_group->list, &found->block_groups[index]);
326 	up_write(&found->groups_sem);
327 }
328 
btrfs_find_space_info(struct btrfs_fs_info * info,u64 flags)329 struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
330 					       u64 flags)
331 {
332 	struct list_head *head = &info->space_info;
333 	struct btrfs_space_info *found;
334 
335 	flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
336 
337 	list_for_each_entry(found, head, list) {
338 		if (found->flags & flags)
339 			return found;
340 	}
341 	return NULL;
342 }
343 
calc_available_free_space(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,enum btrfs_reserve_flush_enum flush)344 static u64 calc_available_free_space(struct btrfs_fs_info *fs_info,
345 			  struct btrfs_space_info *space_info,
346 			  enum btrfs_reserve_flush_enum flush)
347 {
348 	u64 profile;
349 	u64 avail;
350 	int factor;
351 
352 	if (space_info->flags & BTRFS_BLOCK_GROUP_SYSTEM)
353 		profile = btrfs_system_alloc_profile(fs_info);
354 	else
355 		profile = btrfs_metadata_alloc_profile(fs_info);
356 
357 	avail = atomic64_read(&fs_info->free_chunk_space);
358 
359 	/*
360 	 * If we have dup, raid1 or raid10 then only half of the free
361 	 * space is actually usable.  For raid56, the space info used
362 	 * doesn't include the parity drive, so we don't have to
363 	 * change the math
364 	 */
365 	factor = btrfs_bg_type_to_factor(profile);
366 	avail = div_u64(avail, factor);
367 
368 	/*
369 	 * If we aren't flushing all things, let us overcommit up to
370 	 * 1/2th of the space. If we can flush, don't let us overcommit
371 	 * too much, let it overcommit up to 1/8 of the space.
372 	 */
373 	if (flush == BTRFS_RESERVE_FLUSH_ALL)
374 		avail >>= 3;
375 	else
376 		avail >>= 1;
377 	return avail;
378 }
379 
btrfs_can_overcommit(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,u64 bytes,enum btrfs_reserve_flush_enum flush)380 int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
381 			 struct btrfs_space_info *space_info, u64 bytes,
382 			 enum btrfs_reserve_flush_enum flush)
383 {
384 	u64 avail;
385 	u64 used;
386 
387 	/* Don't overcommit when in mixed mode */
388 	if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
389 		return 0;
390 
391 	used = btrfs_space_info_used(space_info, true);
392 	avail = calc_available_free_space(fs_info, space_info, flush);
393 
394 	if (used + bytes < space_info->total_bytes + avail)
395 		return 1;
396 	return 0;
397 }
398 
remove_ticket(struct btrfs_space_info * space_info,struct reserve_ticket * ticket)399 static void remove_ticket(struct btrfs_space_info *space_info,
400 			  struct reserve_ticket *ticket)
401 {
402 	if (!list_empty(&ticket->list)) {
403 		list_del_init(&ticket->list);
404 		ASSERT(space_info->reclaim_size >= ticket->bytes);
405 		space_info->reclaim_size -= ticket->bytes;
406 	}
407 }
408 
409 /*
410  * This is for space we already have accounted in space_info->bytes_may_use, so
411  * basically when we're returning space from block_rsv's.
412  */
btrfs_try_granting_tickets(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info)413 void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
414 				struct btrfs_space_info *space_info)
415 {
416 	struct list_head *head;
417 	enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
418 
419 	lockdep_assert_held(&space_info->lock);
420 
421 	head = &space_info->priority_tickets;
422 again:
423 	while (!list_empty(head)) {
424 		struct reserve_ticket *ticket;
425 		u64 used = btrfs_space_info_used(space_info, true);
426 
427 		ticket = list_first_entry(head, struct reserve_ticket, list);
428 
429 		/* Check and see if our ticket can be satisfied now. */
430 		if ((used + ticket->bytes <= space_info->total_bytes) ||
431 		    btrfs_can_overcommit(fs_info, space_info, ticket->bytes,
432 					 flush)) {
433 			btrfs_space_info_update_bytes_may_use(fs_info,
434 							      space_info,
435 							      ticket->bytes);
436 			remove_ticket(space_info, ticket);
437 			ticket->bytes = 0;
438 			space_info->tickets_id++;
439 			wake_up(&ticket->wait);
440 		} else {
441 			break;
442 		}
443 	}
444 
445 	if (head == &space_info->priority_tickets) {
446 		head = &space_info->tickets;
447 		flush = BTRFS_RESERVE_FLUSH_ALL;
448 		goto again;
449 	}
450 }
451 
452 #define DUMP_BLOCK_RSV(fs_info, rsv_name)				\
453 do {									\
454 	struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name;		\
455 	spin_lock(&__rsv->lock);					\
456 	btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu",	\
457 		   __rsv->size, __rsv->reserved);			\
458 	spin_unlock(&__rsv->lock);					\
459 } while (0)
460 
space_info_flag_to_str(const struct btrfs_space_info * space_info)461 static const char *space_info_flag_to_str(const struct btrfs_space_info *space_info)
462 {
463 	switch (space_info->flags) {
464 	case BTRFS_BLOCK_GROUP_SYSTEM:
465 		return "SYSTEM";
466 	case BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA:
467 		return "DATA+METADATA";
468 	case BTRFS_BLOCK_GROUP_DATA:
469 		return "DATA";
470 	case BTRFS_BLOCK_GROUP_METADATA:
471 		return "METADATA";
472 	default:
473 		return "UNKNOWN";
474 	}
475 }
476 
dump_global_block_rsv(struct btrfs_fs_info * fs_info)477 static void dump_global_block_rsv(struct btrfs_fs_info *fs_info)
478 {
479 	DUMP_BLOCK_RSV(fs_info, global_block_rsv);
480 	DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
481 	DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
482 	DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
483 	DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
484 }
485 
__btrfs_dump_space_info(struct btrfs_fs_info * fs_info,struct btrfs_space_info * info)486 static void __btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
487 				    struct btrfs_space_info *info)
488 {
489 	const char *flag_str = space_info_flag_to_str(info);
490 	lockdep_assert_held(&info->lock);
491 
492 	/* The free space could be negative in case of overcommit */
493 	btrfs_info(fs_info, "space_info %s has %lld free, is %sfull",
494 		   flag_str,
495 		   (s64)(info->total_bytes - btrfs_space_info_used(info, true)),
496 		   info->full ? "" : "not ");
497 	btrfs_info(fs_info,
498 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu zone_unusable=%llu",
499 		info->total_bytes, info->bytes_used, info->bytes_pinned,
500 		info->bytes_reserved, info->bytes_may_use,
501 		info->bytes_readonly, info->bytes_zone_unusable);
502 }
503 
btrfs_dump_space_info(struct btrfs_fs_info * fs_info,struct btrfs_space_info * info,u64 bytes,int dump_block_groups)504 void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
505 			   struct btrfs_space_info *info, u64 bytes,
506 			   int dump_block_groups)
507 {
508 	struct btrfs_block_group *cache;
509 	u64 total_avail = 0;
510 	int index = 0;
511 
512 	spin_lock(&info->lock);
513 	__btrfs_dump_space_info(fs_info, info);
514 	dump_global_block_rsv(fs_info);
515 	spin_unlock(&info->lock);
516 
517 	if (!dump_block_groups)
518 		return;
519 
520 	down_read(&info->groups_sem);
521 again:
522 	list_for_each_entry(cache, &info->block_groups[index], list) {
523 		u64 avail;
524 
525 		spin_lock(&cache->lock);
526 		avail = cache->length - cache->used - cache->pinned -
527 			cache->reserved - cache->delalloc_bytes -
528 			cache->bytes_super - cache->zone_unusable;
529 		btrfs_info(fs_info,
530 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %llu delalloc %llu super %llu zone_unusable (%llu bytes available) %s",
531 			   cache->start, cache->length, cache->used, cache->pinned,
532 			   cache->reserved, cache->delalloc_bytes,
533 			   cache->bytes_super, cache->zone_unusable,
534 			   avail, cache->ro ? "[readonly]" : "");
535 		spin_unlock(&cache->lock);
536 		btrfs_dump_free_space(cache, bytes);
537 		total_avail += avail;
538 	}
539 	if (++index < BTRFS_NR_RAID_TYPES)
540 		goto again;
541 	up_read(&info->groups_sem);
542 
543 	btrfs_info(fs_info, "%llu bytes available across all block groups", total_avail);
544 }
545 
calc_reclaim_items_nr(const struct btrfs_fs_info * fs_info,u64 to_reclaim)546 static inline u64 calc_reclaim_items_nr(const struct btrfs_fs_info *fs_info,
547 					u64 to_reclaim)
548 {
549 	u64 bytes;
550 	u64 nr;
551 
552 	bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
553 	nr = div64_u64(to_reclaim, bytes);
554 	if (!nr)
555 		nr = 1;
556 	return nr;
557 }
558 
calc_delayed_refs_nr(const struct btrfs_fs_info * fs_info,u64 to_reclaim)559 static inline u64 calc_delayed_refs_nr(const struct btrfs_fs_info *fs_info,
560 				       u64 to_reclaim)
561 {
562 	const u64 bytes = btrfs_calc_delayed_ref_bytes(fs_info, 1);
563 	u64 nr;
564 
565 	nr = div64_u64(to_reclaim, bytes);
566 	if (!nr)
567 		nr = 1;
568 	return nr;
569 }
570 
571 #define EXTENT_SIZE_PER_ITEM	SZ_256K
572 
573 /*
574  * shrink metadata reservation for delalloc
575  */
shrink_delalloc(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,u64 to_reclaim,bool wait_ordered,bool for_preempt)576 static void shrink_delalloc(struct btrfs_fs_info *fs_info,
577 			    struct btrfs_space_info *space_info,
578 			    u64 to_reclaim, bool wait_ordered,
579 			    bool for_preempt)
580 {
581 	struct btrfs_trans_handle *trans;
582 	u64 delalloc_bytes;
583 	u64 ordered_bytes;
584 	u64 items;
585 	long time_left;
586 	int loops;
587 
588 	delalloc_bytes = percpu_counter_sum_positive(&fs_info->delalloc_bytes);
589 	ordered_bytes = percpu_counter_sum_positive(&fs_info->ordered_bytes);
590 	if (delalloc_bytes == 0 && ordered_bytes == 0)
591 		return;
592 
593 	/* Calc the number of the pages we need flush for space reservation */
594 	if (to_reclaim == U64_MAX) {
595 		items = U64_MAX;
596 	} else {
597 		/*
598 		 * to_reclaim is set to however much metadata we need to
599 		 * reclaim, but reclaiming that much data doesn't really track
600 		 * exactly.  What we really want to do is reclaim full inode's
601 		 * worth of reservations, however that's not available to us
602 		 * here.  We will take a fraction of the delalloc bytes for our
603 		 * flushing loops and hope for the best.  Delalloc will expand
604 		 * the amount we write to cover an entire dirty extent, which
605 		 * will reclaim the metadata reservation for that range.  If
606 		 * it's not enough subsequent flush stages will be more
607 		 * aggressive.
608 		 */
609 		to_reclaim = max(to_reclaim, delalloc_bytes >> 3);
610 		items = calc_reclaim_items_nr(fs_info, to_reclaim) * 2;
611 	}
612 
613 	trans = current->journal_info;
614 
615 	/*
616 	 * If we are doing more ordered than delalloc we need to just wait on
617 	 * ordered extents, otherwise we'll waste time trying to flush delalloc
618 	 * that likely won't give us the space back we need.
619 	 */
620 	if (ordered_bytes > delalloc_bytes && !for_preempt)
621 		wait_ordered = true;
622 
623 	loops = 0;
624 	while ((delalloc_bytes || ordered_bytes) && loops < 3) {
625 		u64 temp = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
626 		long nr_pages = min_t(u64, temp, LONG_MAX);
627 		int async_pages;
628 
629 		btrfs_start_delalloc_roots(fs_info, nr_pages, true);
630 
631 		/*
632 		 * We need to make sure any outstanding async pages are now
633 		 * processed before we continue.  This is because things like
634 		 * sync_inode() try to be smart and skip writing if the inode is
635 		 * marked clean.  We don't use filemap_fwrite for flushing
636 		 * because we want to control how many pages we write out at a
637 		 * time, thus this is the only safe way to make sure we've
638 		 * waited for outstanding compressed workers to have started
639 		 * their jobs and thus have ordered extents set up properly.
640 		 *
641 		 * This exists because we do not want to wait for each
642 		 * individual inode to finish its async work, we simply want to
643 		 * start the IO on everybody, and then come back here and wait
644 		 * for all of the async work to catch up.  Once we're done with
645 		 * that we know we'll have ordered extents for everything and we
646 		 * can decide if we wait for that or not.
647 		 *
648 		 * If we choose to replace this in the future, make absolutely
649 		 * sure that the proper waiting is being done in the async case,
650 		 * as there have been bugs in that area before.
651 		 */
652 		async_pages = atomic_read(&fs_info->async_delalloc_pages);
653 		if (!async_pages)
654 			goto skip_async;
655 
656 		/*
657 		 * We don't want to wait forever, if we wrote less pages in this
658 		 * loop than we have outstanding, only wait for that number of
659 		 * pages, otherwise we can wait for all async pages to finish
660 		 * before continuing.
661 		 */
662 		if (async_pages > nr_pages)
663 			async_pages -= nr_pages;
664 		else
665 			async_pages = 0;
666 		wait_event(fs_info->async_submit_wait,
667 			   atomic_read(&fs_info->async_delalloc_pages) <=
668 			   async_pages);
669 skip_async:
670 		loops++;
671 		if (wait_ordered && !trans) {
672 			btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
673 		} else {
674 			time_left = schedule_timeout_killable(1);
675 			if (time_left)
676 				break;
677 		}
678 
679 		/*
680 		 * If we are for preemption we just want a one-shot of delalloc
681 		 * flushing so we can stop flushing if we decide we don't need
682 		 * to anymore.
683 		 */
684 		if (for_preempt)
685 			break;
686 
687 		spin_lock(&space_info->lock);
688 		if (list_empty(&space_info->tickets) &&
689 		    list_empty(&space_info->priority_tickets)) {
690 			spin_unlock(&space_info->lock);
691 			break;
692 		}
693 		spin_unlock(&space_info->lock);
694 
695 		delalloc_bytes = percpu_counter_sum_positive(
696 						&fs_info->delalloc_bytes);
697 		ordered_bytes = percpu_counter_sum_positive(
698 						&fs_info->ordered_bytes);
699 	}
700 }
701 
702 /*
703  * Try to flush some data based on policy set by @state. This is only advisory
704  * and may fail for various reasons. The caller is supposed to examine the
705  * state of @space_info to detect the outcome.
706  */
flush_space(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,u64 num_bytes,enum btrfs_flush_state state,bool for_preempt)707 static void flush_space(struct btrfs_fs_info *fs_info,
708 		       struct btrfs_space_info *space_info, u64 num_bytes,
709 		       enum btrfs_flush_state state, bool for_preempt)
710 {
711 	struct btrfs_root *root = fs_info->tree_root;
712 	struct btrfs_trans_handle *trans;
713 	int nr;
714 	int ret = 0;
715 
716 	switch (state) {
717 	case FLUSH_DELAYED_ITEMS_NR:
718 	case FLUSH_DELAYED_ITEMS:
719 		if (state == FLUSH_DELAYED_ITEMS_NR)
720 			nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
721 		else
722 			nr = -1;
723 
724 		trans = btrfs_join_transaction_nostart(root);
725 		if (IS_ERR(trans)) {
726 			ret = PTR_ERR(trans);
727 			if (ret == -ENOENT)
728 				ret = 0;
729 			break;
730 		}
731 		ret = btrfs_run_delayed_items_nr(trans, nr);
732 		btrfs_end_transaction(trans);
733 		break;
734 	case FLUSH_DELALLOC:
735 	case FLUSH_DELALLOC_WAIT:
736 	case FLUSH_DELALLOC_FULL:
737 		if (state == FLUSH_DELALLOC_FULL)
738 			num_bytes = U64_MAX;
739 		shrink_delalloc(fs_info, space_info, num_bytes,
740 				state != FLUSH_DELALLOC, for_preempt);
741 		break;
742 	case FLUSH_DELAYED_REFS_NR:
743 	case FLUSH_DELAYED_REFS:
744 		trans = btrfs_join_transaction_nostart(root);
745 		if (IS_ERR(trans)) {
746 			ret = PTR_ERR(trans);
747 			if (ret == -ENOENT)
748 				ret = 0;
749 			break;
750 		}
751 		if (state == FLUSH_DELAYED_REFS_NR)
752 			nr = calc_delayed_refs_nr(fs_info, num_bytes);
753 		else
754 			nr = 0;
755 		btrfs_run_delayed_refs(trans, nr);
756 		btrfs_end_transaction(trans);
757 		break;
758 	case ALLOC_CHUNK:
759 	case ALLOC_CHUNK_FORCE:
760 		trans = btrfs_join_transaction(root);
761 		if (IS_ERR(trans)) {
762 			ret = PTR_ERR(trans);
763 			break;
764 		}
765 		ret = btrfs_chunk_alloc(trans,
766 				btrfs_get_alloc_profile(fs_info, space_info->flags),
767 				(state == ALLOC_CHUNK) ? CHUNK_ALLOC_NO_FORCE :
768 					CHUNK_ALLOC_FORCE);
769 		btrfs_end_transaction(trans);
770 
771 		if (ret > 0 || ret == -ENOSPC)
772 			ret = 0;
773 		break;
774 	case RUN_DELAYED_IPUTS:
775 		/*
776 		 * If we have pending delayed iputs then we could free up a
777 		 * bunch of pinned space, so make sure we run the iputs before
778 		 * we do our pinned bytes check below.
779 		 */
780 		btrfs_run_delayed_iputs(fs_info);
781 		btrfs_wait_on_delayed_iputs(fs_info);
782 		break;
783 	case COMMIT_TRANS:
784 		ASSERT(current->journal_info == NULL);
785 		/*
786 		 * We don't want to start a new transaction, just attach to the
787 		 * current one or wait it fully commits in case its commit is
788 		 * happening at the moment. Note: we don't use a nostart join
789 		 * because that does not wait for a transaction to fully commit
790 		 * (only for it to be unblocked, state TRANS_STATE_UNBLOCKED).
791 		 */
792 		trans = btrfs_attach_transaction_barrier(root);
793 		if (IS_ERR(trans)) {
794 			ret = PTR_ERR(trans);
795 			if (ret == -ENOENT)
796 				ret = 0;
797 			break;
798 		}
799 		ret = btrfs_commit_transaction(trans);
800 		break;
801 	default:
802 		ret = -ENOSPC;
803 		break;
804 	}
805 
806 	trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
807 				ret, for_preempt);
808 	return;
809 }
810 
811 static inline u64
btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info)812 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
813 				 struct btrfs_space_info *space_info)
814 {
815 	u64 used;
816 	u64 avail;
817 	u64 to_reclaim = space_info->reclaim_size;
818 
819 	lockdep_assert_held(&space_info->lock);
820 
821 	avail = calc_available_free_space(fs_info, space_info,
822 					  BTRFS_RESERVE_FLUSH_ALL);
823 	used = btrfs_space_info_used(space_info, true);
824 
825 	/*
826 	 * We may be flushing because suddenly we have less space than we had
827 	 * before, and now we're well over-committed based on our current free
828 	 * space.  If that's the case add in our overage so we make sure to put
829 	 * appropriate pressure on the flushing state machine.
830 	 */
831 	if (space_info->total_bytes + avail < used)
832 		to_reclaim += used - (space_info->total_bytes + avail);
833 
834 	return to_reclaim;
835 }
836 
need_preemptive_reclaim(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info)837 static bool need_preemptive_reclaim(struct btrfs_fs_info *fs_info,
838 				    struct btrfs_space_info *space_info)
839 {
840 	const u64 global_rsv_size = btrfs_block_rsv_reserved(&fs_info->global_block_rsv);
841 	u64 ordered, delalloc;
842 	u64 thresh;
843 	u64 used;
844 
845 	thresh = mult_perc(space_info->total_bytes, 90);
846 
847 	lockdep_assert_held(&space_info->lock);
848 
849 	/* If we're just plain full then async reclaim just slows us down. */
850 	if ((space_info->bytes_used + space_info->bytes_reserved +
851 	     global_rsv_size) >= thresh)
852 		return false;
853 
854 	used = space_info->bytes_may_use + space_info->bytes_pinned;
855 
856 	/* The total flushable belongs to the global rsv, don't flush. */
857 	if (global_rsv_size >= used)
858 		return false;
859 
860 	/*
861 	 * 128MiB is 1/4 of the maximum global rsv size.  If we have less than
862 	 * that devoted to other reservations then there's no sense in flushing,
863 	 * we don't have a lot of things that need flushing.
864 	 */
865 	if (used - global_rsv_size <= SZ_128M)
866 		return false;
867 
868 	/*
869 	 * We have tickets queued, bail so we don't compete with the async
870 	 * flushers.
871 	 */
872 	if (space_info->reclaim_size)
873 		return false;
874 
875 	/*
876 	 * If we have over half of the free space occupied by reservations or
877 	 * pinned then we want to start flushing.
878 	 *
879 	 * We do not do the traditional thing here, which is to say
880 	 *
881 	 *   if (used >= ((total_bytes + avail) / 2))
882 	 *     return 1;
883 	 *
884 	 * because this doesn't quite work how we want.  If we had more than 50%
885 	 * of the space_info used by bytes_used and we had 0 available we'd just
886 	 * constantly run the background flusher.  Instead we want it to kick in
887 	 * if our reclaimable space exceeds our clamped free space.
888 	 *
889 	 * Our clamping range is 2^1 -> 2^8.  Practically speaking that means
890 	 * the following:
891 	 *
892 	 * Amount of RAM        Minimum threshold       Maximum threshold
893 	 *
894 	 *        256GiB                     1GiB                  128GiB
895 	 *        128GiB                   512MiB                   64GiB
896 	 *         64GiB                   256MiB                   32GiB
897 	 *         32GiB                   128MiB                   16GiB
898 	 *         16GiB                    64MiB                    8GiB
899 	 *
900 	 * These are the range our thresholds will fall in, corresponding to how
901 	 * much delalloc we need for the background flusher to kick in.
902 	 */
903 
904 	thresh = calc_available_free_space(fs_info, space_info,
905 					   BTRFS_RESERVE_FLUSH_ALL);
906 	used = space_info->bytes_used + space_info->bytes_reserved +
907 	       space_info->bytes_readonly + global_rsv_size;
908 	if (used < space_info->total_bytes)
909 		thresh += space_info->total_bytes - used;
910 	thresh >>= space_info->clamp;
911 
912 	used = space_info->bytes_pinned;
913 
914 	/*
915 	 * If we have more ordered bytes than delalloc bytes then we're either
916 	 * doing a lot of DIO, or we simply don't have a lot of delalloc waiting
917 	 * around.  Preemptive flushing is only useful in that it can free up
918 	 * space before tickets need to wait for things to finish.  In the case
919 	 * of ordered extents, preemptively waiting on ordered extents gets us
920 	 * nothing, if our reservations are tied up in ordered extents we'll
921 	 * simply have to slow down writers by forcing them to wait on ordered
922 	 * extents.
923 	 *
924 	 * In the case that ordered is larger than delalloc, only include the
925 	 * block reserves that we would actually be able to directly reclaim
926 	 * from.  In this case if we're heavy on metadata operations this will
927 	 * clearly be heavy enough to warrant preemptive flushing.  In the case
928 	 * of heavy DIO or ordered reservations, preemptive flushing will just
929 	 * waste time and cause us to slow down.
930 	 *
931 	 * We want to make sure we truly are maxed out on ordered however, so
932 	 * cut ordered in half, and if it's still higher than delalloc then we
933 	 * can keep flushing.  This is to avoid the case where we start
934 	 * flushing, and now delalloc == ordered and we stop preemptively
935 	 * flushing when we could still have several gigs of delalloc to flush.
936 	 */
937 	ordered = percpu_counter_read_positive(&fs_info->ordered_bytes) >> 1;
938 	delalloc = percpu_counter_read_positive(&fs_info->delalloc_bytes);
939 	if (ordered >= delalloc)
940 		used += btrfs_block_rsv_reserved(&fs_info->delayed_refs_rsv) +
941 			btrfs_block_rsv_reserved(&fs_info->delayed_block_rsv);
942 	else
943 		used += space_info->bytes_may_use - global_rsv_size;
944 
945 	return (used >= thresh && !btrfs_fs_closing(fs_info) &&
946 		!test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
947 }
948 
steal_from_global_rsv(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,struct reserve_ticket * ticket)949 static bool steal_from_global_rsv(struct btrfs_fs_info *fs_info,
950 				  struct btrfs_space_info *space_info,
951 				  struct reserve_ticket *ticket)
952 {
953 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
954 	u64 min_bytes;
955 
956 	if (!ticket->steal)
957 		return false;
958 
959 	if (global_rsv->space_info != space_info)
960 		return false;
961 
962 	spin_lock(&global_rsv->lock);
963 	min_bytes = mult_perc(global_rsv->size, 10);
964 	if (global_rsv->reserved < min_bytes + ticket->bytes) {
965 		spin_unlock(&global_rsv->lock);
966 		return false;
967 	}
968 	global_rsv->reserved -= ticket->bytes;
969 	remove_ticket(space_info, ticket);
970 	ticket->bytes = 0;
971 	wake_up(&ticket->wait);
972 	space_info->tickets_id++;
973 	if (global_rsv->reserved < global_rsv->size)
974 		global_rsv->full = 0;
975 	spin_unlock(&global_rsv->lock);
976 
977 	return true;
978 }
979 
980 /*
981  * maybe_fail_all_tickets - we've exhausted our flushing, start failing tickets
982  * @fs_info - fs_info for this fs
983  * @space_info - the space info we were flushing
984  *
985  * We call this when we've exhausted our flushing ability and haven't made
986  * progress in satisfying tickets.  The reservation code handles tickets in
987  * order, so if there is a large ticket first and then smaller ones we could
988  * very well satisfy the smaller tickets.  This will attempt to wake up any
989  * tickets in the list to catch this case.
990  *
991  * This function returns true if it was able to make progress by clearing out
992  * other tickets, or if it stumbles across a ticket that was smaller than the
993  * first ticket.
994  */
maybe_fail_all_tickets(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info)995 static bool maybe_fail_all_tickets(struct btrfs_fs_info *fs_info,
996 				   struct btrfs_space_info *space_info)
997 {
998 	struct reserve_ticket *ticket;
999 	u64 tickets_id = space_info->tickets_id;
1000 	const bool aborted = BTRFS_FS_ERROR(fs_info);
1001 
1002 	trace_btrfs_fail_all_tickets(fs_info, space_info);
1003 
1004 	if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
1005 		btrfs_info(fs_info, "cannot satisfy tickets, dumping space info");
1006 		__btrfs_dump_space_info(fs_info, space_info);
1007 	}
1008 
1009 	while (!list_empty(&space_info->tickets) &&
1010 	       tickets_id == space_info->tickets_id) {
1011 		ticket = list_first_entry(&space_info->tickets,
1012 					  struct reserve_ticket, list);
1013 
1014 		if (!aborted && steal_from_global_rsv(fs_info, space_info, ticket))
1015 			return true;
1016 
1017 		if (!aborted && btrfs_test_opt(fs_info, ENOSPC_DEBUG))
1018 			btrfs_info(fs_info, "failing ticket with %llu bytes",
1019 				   ticket->bytes);
1020 
1021 		remove_ticket(space_info, ticket);
1022 		if (aborted)
1023 			ticket->error = -EIO;
1024 		else
1025 			ticket->error = -ENOSPC;
1026 		wake_up(&ticket->wait);
1027 
1028 		/*
1029 		 * We're just throwing tickets away, so more flushing may not
1030 		 * trip over btrfs_try_granting_tickets, so we need to call it
1031 		 * here to see if we can make progress with the next ticket in
1032 		 * the list.
1033 		 */
1034 		if (!aborted)
1035 			btrfs_try_granting_tickets(fs_info, space_info);
1036 	}
1037 	return (tickets_id != space_info->tickets_id);
1038 }
1039 
1040 /*
1041  * This is for normal flushers, we can wait all goddamned day if we want to.  We
1042  * will loop and continuously try to flush as long as we are making progress.
1043  * We count progress as clearing off tickets each time we have to loop.
1044  */
btrfs_async_reclaim_metadata_space(struct work_struct * work)1045 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
1046 {
1047 	struct btrfs_fs_info *fs_info;
1048 	struct btrfs_space_info *space_info;
1049 	u64 to_reclaim;
1050 	enum btrfs_flush_state flush_state;
1051 	int commit_cycles = 0;
1052 	u64 last_tickets_id;
1053 
1054 	fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
1055 	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
1056 
1057 	spin_lock(&space_info->lock);
1058 	to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info);
1059 	if (!to_reclaim) {
1060 		space_info->flush = 0;
1061 		spin_unlock(&space_info->lock);
1062 		return;
1063 	}
1064 	last_tickets_id = space_info->tickets_id;
1065 	spin_unlock(&space_info->lock);
1066 
1067 	flush_state = FLUSH_DELAYED_ITEMS_NR;
1068 	do {
1069 		flush_space(fs_info, space_info, to_reclaim, flush_state, false);
1070 		spin_lock(&space_info->lock);
1071 		if (list_empty(&space_info->tickets)) {
1072 			space_info->flush = 0;
1073 			spin_unlock(&space_info->lock);
1074 			return;
1075 		}
1076 		to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
1077 							      space_info);
1078 		if (last_tickets_id == space_info->tickets_id) {
1079 			flush_state++;
1080 		} else {
1081 			last_tickets_id = space_info->tickets_id;
1082 			flush_state = FLUSH_DELAYED_ITEMS_NR;
1083 			if (commit_cycles)
1084 				commit_cycles--;
1085 		}
1086 
1087 		/*
1088 		 * We do not want to empty the system of delalloc unless we're
1089 		 * under heavy pressure, so allow one trip through the flushing
1090 		 * logic before we start doing a FLUSH_DELALLOC_FULL.
1091 		 */
1092 		if (flush_state == FLUSH_DELALLOC_FULL && !commit_cycles)
1093 			flush_state++;
1094 
1095 		/*
1096 		 * We don't want to force a chunk allocation until we've tried
1097 		 * pretty hard to reclaim space.  Think of the case where we
1098 		 * freed up a bunch of space and so have a lot of pinned space
1099 		 * to reclaim.  We would rather use that than possibly create a
1100 		 * underutilized metadata chunk.  So if this is our first run
1101 		 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
1102 		 * commit the transaction.  If nothing has changed the next go
1103 		 * around then we can force a chunk allocation.
1104 		 */
1105 		if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
1106 			flush_state++;
1107 
1108 		if (flush_state > COMMIT_TRANS) {
1109 			commit_cycles++;
1110 			if (commit_cycles > 2) {
1111 				if (maybe_fail_all_tickets(fs_info, space_info)) {
1112 					flush_state = FLUSH_DELAYED_ITEMS_NR;
1113 					commit_cycles--;
1114 				} else {
1115 					space_info->flush = 0;
1116 				}
1117 			} else {
1118 				flush_state = FLUSH_DELAYED_ITEMS_NR;
1119 			}
1120 		}
1121 		spin_unlock(&space_info->lock);
1122 	} while (flush_state <= COMMIT_TRANS);
1123 }
1124 
1125 /*
1126  * This handles pre-flushing of metadata space before we get to the point that
1127  * we need to start blocking threads on tickets.  The logic here is different
1128  * from the other flush paths because it doesn't rely on tickets to tell us how
1129  * much we need to flush, instead it attempts to keep us below the 80% full
1130  * watermark of space by flushing whichever reservation pool is currently the
1131  * largest.
1132  */
btrfs_preempt_reclaim_metadata_space(struct work_struct * work)1133 static void btrfs_preempt_reclaim_metadata_space(struct work_struct *work)
1134 {
1135 	struct btrfs_fs_info *fs_info;
1136 	struct btrfs_space_info *space_info;
1137 	struct btrfs_block_rsv *delayed_block_rsv;
1138 	struct btrfs_block_rsv *delayed_refs_rsv;
1139 	struct btrfs_block_rsv *global_rsv;
1140 	struct btrfs_block_rsv *trans_rsv;
1141 	int loops = 0;
1142 
1143 	fs_info = container_of(work, struct btrfs_fs_info,
1144 			       preempt_reclaim_work);
1145 	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
1146 	delayed_block_rsv = &fs_info->delayed_block_rsv;
1147 	delayed_refs_rsv = &fs_info->delayed_refs_rsv;
1148 	global_rsv = &fs_info->global_block_rsv;
1149 	trans_rsv = &fs_info->trans_block_rsv;
1150 
1151 	spin_lock(&space_info->lock);
1152 	while (need_preemptive_reclaim(fs_info, space_info)) {
1153 		enum btrfs_flush_state flush;
1154 		u64 delalloc_size = 0;
1155 		u64 to_reclaim, block_rsv_size;
1156 		const u64 global_rsv_size = btrfs_block_rsv_reserved(global_rsv);
1157 
1158 		loops++;
1159 
1160 		/*
1161 		 * We don't have a precise counter for the metadata being
1162 		 * reserved for delalloc, so we'll approximate it by subtracting
1163 		 * out the block rsv's space from the bytes_may_use.  If that
1164 		 * amount is higher than the individual reserves, then we can
1165 		 * assume it's tied up in delalloc reservations.
1166 		 */
1167 		block_rsv_size = global_rsv_size +
1168 			btrfs_block_rsv_reserved(delayed_block_rsv) +
1169 			btrfs_block_rsv_reserved(delayed_refs_rsv) +
1170 			btrfs_block_rsv_reserved(trans_rsv);
1171 		if (block_rsv_size < space_info->bytes_may_use)
1172 			delalloc_size = space_info->bytes_may_use - block_rsv_size;
1173 
1174 		/*
1175 		 * We don't want to include the global_rsv in our calculation,
1176 		 * because that's space we can't touch.  Subtract it from the
1177 		 * block_rsv_size for the next checks.
1178 		 */
1179 		block_rsv_size -= global_rsv_size;
1180 
1181 		/*
1182 		 * We really want to avoid flushing delalloc too much, as it
1183 		 * could result in poor allocation patterns, so only flush it if
1184 		 * it's larger than the rest of the pools combined.
1185 		 */
1186 		if (delalloc_size > block_rsv_size) {
1187 			to_reclaim = delalloc_size;
1188 			flush = FLUSH_DELALLOC;
1189 		} else if (space_info->bytes_pinned >
1190 			   (btrfs_block_rsv_reserved(delayed_block_rsv) +
1191 			    btrfs_block_rsv_reserved(delayed_refs_rsv))) {
1192 			to_reclaim = space_info->bytes_pinned;
1193 			flush = COMMIT_TRANS;
1194 		} else if (btrfs_block_rsv_reserved(delayed_block_rsv) >
1195 			   btrfs_block_rsv_reserved(delayed_refs_rsv)) {
1196 			to_reclaim = btrfs_block_rsv_reserved(delayed_block_rsv);
1197 			flush = FLUSH_DELAYED_ITEMS_NR;
1198 		} else {
1199 			to_reclaim = btrfs_block_rsv_reserved(delayed_refs_rsv);
1200 			flush = FLUSH_DELAYED_REFS_NR;
1201 		}
1202 
1203 		spin_unlock(&space_info->lock);
1204 
1205 		/*
1206 		 * We don't want to reclaim everything, just a portion, so scale
1207 		 * down the to_reclaim by 1/4.  If it takes us down to 0,
1208 		 * reclaim 1 items worth.
1209 		 */
1210 		to_reclaim >>= 2;
1211 		if (!to_reclaim)
1212 			to_reclaim = btrfs_calc_insert_metadata_size(fs_info, 1);
1213 		flush_space(fs_info, space_info, to_reclaim, flush, true);
1214 		cond_resched();
1215 		spin_lock(&space_info->lock);
1216 	}
1217 
1218 	/* We only went through once, back off our clamping. */
1219 	if (loops == 1 && !space_info->reclaim_size)
1220 		space_info->clamp = max(1, space_info->clamp - 1);
1221 	trace_btrfs_done_preemptive_reclaim(fs_info, space_info);
1222 	spin_unlock(&space_info->lock);
1223 }
1224 
1225 /*
1226  * FLUSH_DELALLOC_WAIT:
1227  *   Space is freed from flushing delalloc in one of two ways.
1228  *
1229  *   1) compression is on and we allocate less space than we reserved
1230  *   2) we are overwriting existing space
1231  *
1232  *   For #1 that extra space is reclaimed as soon as the delalloc pages are
1233  *   COWed, by way of btrfs_add_reserved_bytes() which adds the actual extent
1234  *   length to ->bytes_reserved, and subtracts the reserved space from
1235  *   ->bytes_may_use.
1236  *
1237  *   For #2 this is trickier.  Once the ordered extent runs we will drop the
1238  *   extent in the range we are overwriting, which creates a delayed ref for
1239  *   that freed extent.  This however is not reclaimed until the transaction
1240  *   commits, thus the next stages.
1241  *
1242  * RUN_DELAYED_IPUTS
1243  *   If we are freeing inodes, we want to make sure all delayed iputs have
1244  *   completed, because they could have been on an inode with i_nlink == 0, and
1245  *   thus have been truncated and freed up space.  But again this space is not
1246  *   immediately re-usable, it comes in the form of a delayed ref, which must be
1247  *   run and then the transaction must be committed.
1248  *
1249  * COMMIT_TRANS
1250  *   This is where we reclaim all of the pinned space generated by running the
1251  *   iputs
1252  *
1253  * ALLOC_CHUNK_FORCE
1254  *   For data we start with alloc chunk force, however we could have been full
1255  *   before, and then the transaction commit could have freed new block groups,
1256  *   so if we now have space to allocate do the force chunk allocation.
1257  */
1258 static const enum btrfs_flush_state data_flush_states[] = {
1259 	FLUSH_DELALLOC_FULL,
1260 	RUN_DELAYED_IPUTS,
1261 	COMMIT_TRANS,
1262 	ALLOC_CHUNK_FORCE,
1263 };
1264 
btrfs_async_reclaim_data_space(struct work_struct * work)1265 static void btrfs_async_reclaim_data_space(struct work_struct *work)
1266 {
1267 	struct btrfs_fs_info *fs_info;
1268 	struct btrfs_space_info *space_info;
1269 	u64 last_tickets_id;
1270 	enum btrfs_flush_state flush_state = 0;
1271 
1272 	fs_info = container_of(work, struct btrfs_fs_info, async_data_reclaim_work);
1273 	space_info = fs_info->data_sinfo;
1274 
1275 	spin_lock(&space_info->lock);
1276 	if (list_empty(&space_info->tickets)) {
1277 		space_info->flush = 0;
1278 		spin_unlock(&space_info->lock);
1279 		return;
1280 	}
1281 	last_tickets_id = space_info->tickets_id;
1282 	spin_unlock(&space_info->lock);
1283 
1284 	while (!space_info->full) {
1285 		flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE, false);
1286 		spin_lock(&space_info->lock);
1287 		if (list_empty(&space_info->tickets)) {
1288 			space_info->flush = 0;
1289 			spin_unlock(&space_info->lock);
1290 			return;
1291 		}
1292 
1293 		/* Something happened, fail everything and bail. */
1294 		if (BTRFS_FS_ERROR(fs_info))
1295 			goto aborted_fs;
1296 		last_tickets_id = space_info->tickets_id;
1297 		spin_unlock(&space_info->lock);
1298 	}
1299 
1300 	while (flush_state < ARRAY_SIZE(data_flush_states)) {
1301 		flush_space(fs_info, space_info, U64_MAX,
1302 			    data_flush_states[flush_state], false);
1303 		spin_lock(&space_info->lock);
1304 		if (list_empty(&space_info->tickets)) {
1305 			space_info->flush = 0;
1306 			spin_unlock(&space_info->lock);
1307 			return;
1308 		}
1309 
1310 		if (last_tickets_id == space_info->tickets_id) {
1311 			flush_state++;
1312 		} else {
1313 			last_tickets_id = space_info->tickets_id;
1314 			flush_state = 0;
1315 		}
1316 
1317 		if (flush_state >= ARRAY_SIZE(data_flush_states)) {
1318 			if (space_info->full) {
1319 				if (maybe_fail_all_tickets(fs_info, space_info))
1320 					flush_state = 0;
1321 				else
1322 					space_info->flush = 0;
1323 			} else {
1324 				flush_state = 0;
1325 			}
1326 
1327 			/* Something happened, fail everything and bail. */
1328 			if (BTRFS_FS_ERROR(fs_info))
1329 				goto aborted_fs;
1330 
1331 		}
1332 		spin_unlock(&space_info->lock);
1333 	}
1334 	return;
1335 
1336 aborted_fs:
1337 	maybe_fail_all_tickets(fs_info, space_info);
1338 	space_info->flush = 0;
1339 	spin_unlock(&space_info->lock);
1340 }
1341 
btrfs_init_async_reclaim_work(struct btrfs_fs_info * fs_info)1342 void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info)
1343 {
1344 	INIT_WORK(&fs_info->async_reclaim_work, btrfs_async_reclaim_metadata_space);
1345 	INIT_WORK(&fs_info->async_data_reclaim_work, btrfs_async_reclaim_data_space);
1346 	INIT_WORK(&fs_info->preempt_reclaim_work,
1347 		  btrfs_preempt_reclaim_metadata_space);
1348 }
1349 
1350 static const enum btrfs_flush_state priority_flush_states[] = {
1351 	FLUSH_DELAYED_ITEMS_NR,
1352 	FLUSH_DELAYED_ITEMS,
1353 	ALLOC_CHUNK,
1354 };
1355 
1356 static const enum btrfs_flush_state evict_flush_states[] = {
1357 	FLUSH_DELAYED_ITEMS_NR,
1358 	FLUSH_DELAYED_ITEMS,
1359 	FLUSH_DELAYED_REFS_NR,
1360 	FLUSH_DELAYED_REFS,
1361 	FLUSH_DELALLOC,
1362 	FLUSH_DELALLOC_WAIT,
1363 	FLUSH_DELALLOC_FULL,
1364 	ALLOC_CHUNK,
1365 	COMMIT_TRANS,
1366 };
1367 
priority_reclaim_metadata_space(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,struct reserve_ticket * ticket,const enum btrfs_flush_state * states,int states_nr)1368 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
1369 				struct btrfs_space_info *space_info,
1370 				struct reserve_ticket *ticket,
1371 				const enum btrfs_flush_state *states,
1372 				int states_nr)
1373 {
1374 	u64 to_reclaim;
1375 	int flush_state = 0;
1376 
1377 	spin_lock(&space_info->lock);
1378 	to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info);
1379 	/*
1380 	 * This is the priority reclaim path, so to_reclaim could be >0 still
1381 	 * because we may have only satisfied the priority tickets and still
1382 	 * left non priority tickets on the list.  We would then have
1383 	 * to_reclaim but ->bytes == 0.
1384 	 */
1385 	if (ticket->bytes == 0) {
1386 		spin_unlock(&space_info->lock);
1387 		return;
1388 	}
1389 
1390 	while (flush_state < states_nr) {
1391 		spin_unlock(&space_info->lock);
1392 		flush_space(fs_info, space_info, to_reclaim, states[flush_state],
1393 			    false);
1394 		flush_state++;
1395 		spin_lock(&space_info->lock);
1396 		if (ticket->bytes == 0) {
1397 			spin_unlock(&space_info->lock);
1398 			return;
1399 		}
1400 	}
1401 
1402 	/*
1403 	 * Attempt to steal from the global rsv if we can, except if the fs was
1404 	 * turned into error mode due to a transaction abort when flushing space
1405 	 * above, in that case fail with the abort error instead of returning
1406 	 * success to the caller if we can steal from the global rsv - this is
1407 	 * just to have caller fail immeditelly instead of later when trying to
1408 	 * modify the fs, making it easier to debug -ENOSPC problems.
1409 	 */
1410 	if (BTRFS_FS_ERROR(fs_info)) {
1411 		ticket->error = BTRFS_FS_ERROR(fs_info);
1412 		remove_ticket(space_info, ticket);
1413 	} else if (!steal_from_global_rsv(fs_info, space_info, ticket)) {
1414 		ticket->error = -ENOSPC;
1415 		remove_ticket(space_info, ticket);
1416 	}
1417 
1418 	/*
1419 	 * We must run try_granting_tickets here because we could be a large
1420 	 * ticket in front of a smaller ticket that can now be satisfied with
1421 	 * the available space.
1422 	 */
1423 	btrfs_try_granting_tickets(fs_info, space_info);
1424 	spin_unlock(&space_info->lock);
1425 }
1426 
priority_reclaim_data_space(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,struct reserve_ticket * ticket)1427 static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
1428 					struct btrfs_space_info *space_info,
1429 					struct reserve_ticket *ticket)
1430 {
1431 	spin_lock(&space_info->lock);
1432 
1433 	/* We could have been granted before we got here. */
1434 	if (ticket->bytes == 0) {
1435 		spin_unlock(&space_info->lock);
1436 		return;
1437 	}
1438 
1439 	while (!space_info->full) {
1440 		spin_unlock(&space_info->lock);
1441 		flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE, false);
1442 		spin_lock(&space_info->lock);
1443 		if (ticket->bytes == 0) {
1444 			spin_unlock(&space_info->lock);
1445 			return;
1446 		}
1447 	}
1448 
1449 	ticket->error = -ENOSPC;
1450 	remove_ticket(space_info, ticket);
1451 	btrfs_try_granting_tickets(fs_info, space_info);
1452 	spin_unlock(&space_info->lock);
1453 }
1454 
wait_reserve_ticket(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,struct reserve_ticket * ticket)1455 static void wait_reserve_ticket(struct btrfs_fs_info *fs_info,
1456 				struct btrfs_space_info *space_info,
1457 				struct reserve_ticket *ticket)
1458 
1459 {
1460 	DEFINE_WAIT(wait);
1461 	int ret = 0;
1462 
1463 	spin_lock(&space_info->lock);
1464 	while (ticket->bytes > 0 && ticket->error == 0) {
1465 		ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
1466 		if (ret) {
1467 			/*
1468 			 * Delete us from the list. After we unlock the space
1469 			 * info, we don't want the async reclaim job to reserve
1470 			 * space for this ticket. If that would happen, then the
1471 			 * ticket's task would not known that space was reserved
1472 			 * despite getting an error, resulting in a space leak
1473 			 * (bytes_may_use counter of our space_info).
1474 			 */
1475 			remove_ticket(space_info, ticket);
1476 			ticket->error = -EINTR;
1477 			break;
1478 		}
1479 		spin_unlock(&space_info->lock);
1480 
1481 		schedule();
1482 
1483 		finish_wait(&ticket->wait, &wait);
1484 		spin_lock(&space_info->lock);
1485 	}
1486 	spin_unlock(&space_info->lock);
1487 }
1488 
1489 /*
1490  * Do the appropriate flushing and waiting for a ticket.
1491  *
1492  * @fs_info:    the filesystem
1493  * @space_info: space info for the reservation
1494  * @ticket:     ticket for the reservation
1495  * @start_ns:   timestamp when the reservation started
1496  * @orig_bytes: amount of bytes originally reserved
1497  * @flush:      how much we can flush
1498  *
1499  * This does the work of figuring out how to flush for the ticket, waiting for
1500  * the reservation, and returning the appropriate error if there is one.
1501  */
handle_reserve_ticket(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,struct reserve_ticket * ticket,u64 start_ns,u64 orig_bytes,enum btrfs_reserve_flush_enum flush)1502 static int handle_reserve_ticket(struct btrfs_fs_info *fs_info,
1503 				 struct btrfs_space_info *space_info,
1504 				 struct reserve_ticket *ticket,
1505 				 u64 start_ns, u64 orig_bytes,
1506 				 enum btrfs_reserve_flush_enum flush)
1507 {
1508 	int ret;
1509 
1510 	switch (flush) {
1511 	case BTRFS_RESERVE_FLUSH_DATA:
1512 	case BTRFS_RESERVE_FLUSH_ALL:
1513 	case BTRFS_RESERVE_FLUSH_ALL_STEAL:
1514 		wait_reserve_ticket(fs_info, space_info, ticket);
1515 		break;
1516 	case BTRFS_RESERVE_FLUSH_LIMIT:
1517 		priority_reclaim_metadata_space(fs_info, space_info, ticket,
1518 						priority_flush_states,
1519 						ARRAY_SIZE(priority_flush_states));
1520 		break;
1521 	case BTRFS_RESERVE_FLUSH_EVICT:
1522 		priority_reclaim_metadata_space(fs_info, space_info, ticket,
1523 						evict_flush_states,
1524 						ARRAY_SIZE(evict_flush_states));
1525 		break;
1526 	case BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE:
1527 		priority_reclaim_data_space(fs_info, space_info, ticket);
1528 		break;
1529 	default:
1530 		ASSERT(0);
1531 		break;
1532 	}
1533 
1534 	ret = ticket->error;
1535 	ASSERT(list_empty(&ticket->list));
1536 	/*
1537 	 * Check that we can't have an error set if the reservation succeeded,
1538 	 * as that would confuse tasks and lead them to error out without
1539 	 * releasing reserved space (if an error happens the expectation is that
1540 	 * space wasn't reserved at all).
1541 	 */
1542 	ASSERT(!(ticket->bytes == 0 && ticket->error));
1543 	trace_btrfs_reserve_ticket(fs_info, space_info->flags, orig_bytes,
1544 				   start_ns, flush, ticket->error);
1545 	return ret;
1546 }
1547 
1548 /*
1549  * This returns true if this flush state will go through the ordinary flushing
1550  * code.
1551  */
is_normal_flushing(enum btrfs_reserve_flush_enum flush)1552 static inline bool is_normal_flushing(enum btrfs_reserve_flush_enum flush)
1553 {
1554 	return	(flush == BTRFS_RESERVE_FLUSH_ALL) ||
1555 		(flush == BTRFS_RESERVE_FLUSH_ALL_STEAL);
1556 }
1557 
maybe_clamp_preempt(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info)1558 static inline void maybe_clamp_preempt(struct btrfs_fs_info *fs_info,
1559 				       struct btrfs_space_info *space_info)
1560 {
1561 	u64 ordered = percpu_counter_sum_positive(&fs_info->ordered_bytes);
1562 	u64 delalloc = percpu_counter_sum_positive(&fs_info->delalloc_bytes);
1563 
1564 	/*
1565 	 * If we're heavy on ordered operations then clamping won't help us.  We
1566 	 * need to clamp specifically to keep up with dirty'ing buffered
1567 	 * writers, because there's not a 1:1 correlation of writing delalloc
1568 	 * and freeing space, like there is with flushing delayed refs or
1569 	 * delayed nodes.  If we're already more ordered than delalloc then
1570 	 * we're keeping up, otherwise we aren't and should probably clamp.
1571 	 */
1572 	if (ordered < delalloc)
1573 		space_info->clamp = min(space_info->clamp + 1, 8);
1574 }
1575 
can_steal(enum btrfs_reserve_flush_enum flush)1576 static inline bool can_steal(enum btrfs_reserve_flush_enum flush)
1577 {
1578 	return (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL ||
1579 		flush == BTRFS_RESERVE_FLUSH_EVICT);
1580 }
1581 
1582 /*
1583  * NO_FLUSH and FLUSH_EMERGENCY don't want to create a ticket, they just want to
1584  * fail as quickly as possible.
1585  */
can_ticket(enum btrfs_reserve_flush_enum flush)1586 static inline bool can_ticket(enum btrfs_reserve_flush_enum flush)
1587 {
1588 	return (flush != BTRFS_RESERVE_NO_FLUSH &&
1589 		flush != BTRFS_RESERVE_FLUSH_EMERGENCY);
1590 }
1591 
1592 /*
1593  * Try to reserve bytes from the block_rsv's space.
1594  *
1595  * @fs_info:    the filesystem
1596  * @space_info: space info we want to allocate from
1597  * @orig_bytes: number of bytes we want
1598  * @flush:      whether or not we can flush to make our reservation
1599  *
1600  * This will reserve orig_bytes number of bytes from the space info associated
1601  * with the block_rsv.  If there is not enough space it will make an attempt to
1602  * flush out space to make room.  It will do this by flushing delalloc if
1603  * possible or committing the transaction.  If flush is 0 then no attempts to
1604  * regain reservations will be made and this will fail if there is not enough
1605  * space already.
1606  */
__reserve_bytes(struct btrfs_fs_info * fs_info,struct btrfs_space_info * space_info,u64 orig_bytes,enum btrfs_reserve_flush_enum flush)1607 static int __reserve_bytes(struct btrfs_fs_info *fs_info,
1608 			   struct btrfs_space_info *space_info, u64 orig_bytes,
1609 			   enum btrfs_reserve_flush_enum flush)
1610 {
1611 	struct work_struct *async_work;
1612 	struct reserve_ticket ticket;
1613 	u64 start_ns = 0;
1614 	u64 used;
1615 	int ret = -ENOSPC;
1616 	bool pending_tickets;
1617 
1618 	ASSERT(orig_bytes);
1619 	/*
1620 	 * If have a transaction handle (current->journal_info != NULL), then
1621 	 * the flush method can not be neither BTRFS_RESERVE_FLUSH_ALL* nor
1622 	 * BTRFS_RESERVE_FLUSH_EVICT, as we could deadlock because those
1623 	 * flushing methods can trigger transaction commits.
1624 	 */
1625 	if (current->journal_info) {
1626 		/* One assert per line for easier debugging. */
1627 		ASSERT(flush != BTRFS_RESERVE_FLUSH_ALL);
1628 		ASSERT(flush != BTRFS_RESERVE_FLUSH_ALL_STEAL);
1629 		ASSERT(flush != BTRFS_RESERVE_FLUSH_EVICT);
1630 	}
1631 
1632 	if (flush == BTRFS_RESERVE_FLUSH_DATA)
1633 		async_work = &fs_info->async_data_reclaim_work;
1634 	else
1635 		async_work = &fs_info->async_reclaim_work;
1636 
1637 	spin_lock(&space_info->lock);
1638 	used = btrfs_space_info_used(space_info, true);
1639 
1640 	/*
1641 	 * We don't want NO_FLUSH allocations to jump everybody, they can
1642 	 * generally handle ENOSPC in a different way, so treat them the same as
1643 	 * normal flushers when it comes to skipping pending tickets.
1644 	 */
1645 	if (is_normal_flushing(flush) || (flush == BTRFS_RESERVE_NO_FLUSH))
1646 		pending_tickets = !list_empty(&space_info->tickets) ||
1647 			!list_empty(&space_info->priority_tickets);
1648 	else
1649 		pending_tickets = !list_empty(&space_info->priority_tickets);
1650 
1651 	/*
1652 	 * Carry on if we have enough space (short-circuit) OR call
1653 	 * can_overcommit() to ensure we can overcommit to continue.
1654 	 */
1655 	if (!pending_tickets &&
1656 	    ((used + orig_bytes <= space_info->total_bytes) ||
1657 	     btrfs_can_overcommit(fs_info, space_info, orig_bytes, flush))) {
1658 		btrfs_space_info_update_bytes_may_use(fs_info, space_info,
1659 						      orig_bytes);
1660 		ret = 0;
1661 	}
1662 
1663 	/*
1664 	 * Things are dire, we need to make a reservation so we don't abort.  We
1665 	 * will let this reservation go through as long as we have actual space
1666 	 * left to allocate for the block.
1667 	 */
1668 	if (ret && unlikely(flush == BTRFS_RESERVE_FLUSH_EMERGENCY)) {
1669 		used = btrfs_space_info_used(space_info, false);
1670 		if (used + orig_bytes <= space_info->total_bytes) {
1671 			btrfs_space_info_update_bytes_may_use(fs_info, space_info,
1672 							      orig_bytes);
1673 			ret = 0;
1674 		}
1675 	}
1676 
1677 	/*
1678 	 * If we couldn't make a reservation then setup our reservation ticket
1679 	 * and kick the async worker if it's not already running.
1680 	 *
1681 	 * If we are a priority flusher then we just need to add our ticket to
1682 	 * the list and we will do our own flushing further down.
1683 	 */
1684 	if (ret && can_ticket(flush)) {
1685 		ticket.bytes = orig_bytes;
1686 		ticket.error = 0;
1687 		space_info->reclaim_size += ticket.bytes;
1688 		init_waitqueue_head(&ticket.wait);
1689 		ticket.steal = can_steal(flush);
1690 		if (trace_btrfs_reserve_ticket_enabled())
1691 			start_ns = ktime_get_ns();
1692 
1693 		if (flush == BTRFS_RESERVE_FLUSH_ALL ||
1694 		    flush == BTRFS_RESERVE_FLUSH_ALL_STEAL ||
1695 		    flush == BTRFS_RESERVE_FLUSH_DATA) {
1696 			list_add_tail(&ticket.list, &space_info->tickets);
1697 			if (!space_info->flush) {
1698 				/*
1699 				 * We were forced to add a reserve ticket, so
1700 				 * our preemptive flushing is unable to keep
1701 				 * up.  Clamp down on the threshold for the
1702 				 * preemptive flushing in order to keep up with
1703 				 * the workload.
1704 				 */
1705 				maybe_clamp_preempt(fs_info, space_info);
1706 
1707 				space_info->flush = 1;
1708 				trace_btrfs_trigger_flush(fs_info,
1709 							  space_info->flags,
1710 							  orig_bytes, flush,
1711 							  "enospc");
1712 				queue_work(system_unbound_wq, async_work);
1713 			}
1714 		} else {
1715 			list_add_tail(&ticket.list,
1716 				      &space_info->priority_tickets);
1717 		}
1718 	} else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
1719 		/*
1720 		 * We will do the space reservation dance during log replay,
1721 		 * which means we won't have fs_info->fs_root set, so don't do
1722 		 * the async reclaim as we will panic.
1723 		 */
1724 		if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
1725 		    !work_busy(&fs_info->preempt_reclaim_work) &&
1726 		    need_preemptive_reclaim(fs_info, space_info)) {
1727 			trace_btrfs_trigger_flush(fs_info, space_info->flags,
1728 						  orig_bytes, flush, "preempt");
1729 			queue_work(system_unbound_wq,
1730 				   &fs_info->preempt_reclaim_work);
1731 		}
1732 	}
1733 	spin_unlock(&space_info->lock);
1734 	if (!ret || !can_ticket(flush))
1735 		return ret;
1736 
1737 	return handle_reserve_ticket(fs_info, space_info, &ticket, start_ns,
1738 				     orig_bytes, flush);
1739 }
1740 
1741 /*
1742  * Try to reserve metadata bytes from the block_rsv's space.
1743  *
1744  * @fs_info:    the filesystem
1745  * @block_rsv:  block_rsv we're allocating for
1746  * @orig_bytes: number of bytes we want
1747  * @flush:      whether or not we can flush to make our reservation
1748  *
1749  * This will reserve orig_bytes number of bytes from the space info associated
1750  * with the block_rsv.  If there is not enough space it will make an attempt to
1751  * flush out space to make room.  It will do this by flushing delalloc if
1752  * possible or committing the transaction.  If flush is 0 then no attempts to
1753  * regain reservations will be made and this will fail if there is not enough
1754  * space already.
1755  */
btrfs_reserve_metadata_bytes(struct btrfs_fs_info * fs_info,struct btrfs_block_rsv * block_rsv,u64 orig_bytes,enum btrfs_reserve_flush_enum flush)1756 int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
1757 				 struct btrfs_block_rsv *block_rsv,
1758 				 u64 orig_bytes,
1759 				 enum btrfs_reserve_flush_enum flush)
1760 {
1761 	int ret;
1762 
1763 	ret = __reserve_bytes(fs_info, block_rsv->space_info, orig_bytes, flush);
1764 	if (ret == -ENOSPC) {
1765 		trace_btrfs_space_reservation(fs_info, "space_info:enospc",
1766 					      block_rsv->space_info->flags,
1767 					      orig_bytes, 1);
1768 
1769 		if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
1770 			btrfs_dump_space_info(fs_info, block_rsv->space_info,
1771 					      orig_bytes, 0);
1772 	}
1773 	return ret;
1774 }
1775 
1776 /*
1777  * Try to reserve data bytes for an allocation.
1778  *
1779  * @fs_info: the filesystem
1780  * @bytes:   number of bytes we need
1781  * @flush:   how we are allowed to flush
1782  *
1783  * This will reserve bytes from the data space info.  If there is not enough
1784  * space then we will attempt to flush space as specified by flush.
1785  */
btrfs_reserve_data_bytes(struct btrfs_fs_info * fs_info,u64 bytes,enum btrfs_reserve_flush_enum flush)1786 int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
1787 			     enum btrfs_reserve_flush_enum flush)
1788 {
1789 	struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
1790 	int ret;
1791 
1792 	ASSERT(flush == BTRFS_RESERVE_FLUSH_DATA ||
1793 	       flush == BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE ||
1794 	       flush == BTRFS_RESERVE_NO_FLUSH);
1795 	ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA);
1796 
1797 	ret = __reserve_bytes(fs_info, data_sinfo, bytes, flush);
1798 	if (ret == -ENOSPC) {
1799 		trace_btrfs_space_reservation(fs_info, "space_info:enospc",
1800 					      data_sinfo->flags, bytes, 1);
1801 		if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
1802 			btrfs_dump_space_info(fs_info, data_sinfo, bytes, 0);
1803 	}
1804 	return ret;
1805 }
1806 
1807 /* Dump all the space infos when we abort a transaction due to ENOSPC. */
btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info * fs_info)1808 __cold void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info)
1809 {
1810 	struct btrfs_space_info *space_info;
1811 
1812 	btrfs_info(fs_info, "dumping space info:");
1813 	list_for_each_entry(space_info, &fs_info->space_info, list) {
1814 		spin_lock(&space_info->lock);
1815 		__btrfs_dump_space_info(fs_info, space_info);
1816 		spin_unlock(&space_info->lock);
1817 	}
1818 	dump_global_block_rsv(fs_info);
1819 }
1820 
1821 /*
1822  * Account the unused space of all the readonly block group in the space_info.
1823  * takes mirrors into account.
1824  */
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info * sinfo)1825 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
1826 {
1827 	struct btrfs_block_group *block_group;
1828 	u64 free_bytes = 0;
1829 	int factor;
1830 
1831 	/* It's df, we don't care if it's racy */
1832 	if (list_empty(&sinfo->ro_bgs))
1833 		return 0;
1834 
1835 	spin_lock(&sinfo->lock);
1836 	list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
1837 		spin_lock(&block_group->lock);
1838 
1839 		if (!block_group->ro) {
1840 			spin_unlock(&block_group->lock);
1841 			continue;
1842 		}
1843 
1844 		factor = btrfs_bg_type_to_factor(block_group->flags);
1845 		free_bytes += (block_group->length -
1846 			       block_group->used) * factor;
1847 
1848 		spin_unlock(&block_group->lock);
1849 	}
1850 	spin_unlock(&sinfo->lock);
1851 
1852 	return free_bytes;
1853 }
1854