fs/btrfs/space-info.c

   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
 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  */
 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  */
 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  */
 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
 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
 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
 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
 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
 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
 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
 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  */
 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
 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
 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
 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
 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
 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
 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  */
 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  */
 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
 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
 837 static bool need_preemptive_reclaim(struct btrfs_fs_info *fs_info,
 838                                     struct btrfs_space_info *space_info)
 839 {
 840         u64 global_rsv_size = fs_info->global_block_rsv.reserved;
 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 += fs_info->delayed_refs_rsv.reserved +
 941                         fs_info->delayed_block_rsv.reserved;
 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
 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  */
 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  */
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  */
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                 u64 global_rsv_size = global_rsv->reserved;
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                         delayed_block_rsv->reserved +
1169                         delayed_refs_rsv->reserved +
1170                         trans_rsv->reserved;
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                            (delayed_block_rsv->reserved +
1191                             delayed_refs_rsv->reserved)) {
1192                         to_reclaim = space_info->bytes_pinned;
1193                         flush = COMMIT_TRANS;
1194                 } else if (delayed_block_rsv->reserved >
1195                            delayed_refs_rsv->reserved) {
1196                         to_reclaim = delayed_block_rsv->reserved;
1197                         flush = FLUSH_DELAYED_ITEMS_NR;
1198                 } else {
1199                         to_reclaim = delayed_refs_rsv->reserved;
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
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
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
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
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
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  */
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  */
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
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
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  */
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  */
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  */
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  */
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. */
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  */
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 }