git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/fs-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "alloc_foreground.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "clock.h"
   9 #include "error.h"
  10 #include "extents.h"
  11 #include "extent_update.h"
  12 #include "fs.h"
  13 #include "fs-io.h"
  14 #include "fs-io-buffered.h"
  15 #include "fs-io-pagecache.h"
  16 #include "fsck.h"
  17 #include "inode.h"
  18 #include "journal.h"
  19 #include "io.h"
  20 #include "keylist.h"
  21 #include "quota.h"
  22 #include "reflink.h"
  23 #include "trace.h"
  24
  25 #include <linux/aio.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/falloc.h>
  28 #include <linux/migrate.h>
  29 #include <linux/mmu_context.h>
  30 #include <linux/pagevec.h>
  31 #include <linux/rmap.h>
  32 #include <linux/sched/signal.h>
  33 #include <linux/task_io_accounting_ops.h>
  34 #include <linux/uio.h>
  35
  36 #include <trace/events/writeback.h>
  37
  38 struct nocow_flush {
  39         struct closure  *cl;
  40         struct bch_dev  *ca;
  41         struct bio      bio;
  42 };
  43
  44 static void nocow_flush_endio(struct bio *_bio)
  45 {
  46
  47         struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio);
  48
  49         closure_put(bio->cl);
  50         percpu_ref_put(&bio->ca->io_ref);
  51         bio_put(&bio->bio);
  52 }
  53
  54 void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
  55                                          struct bch_inode_info *inode,
  56                                          struct closure *cl)
  57 {
  58         struct nocow_flush *bio;
  59         struct bch_dev *ca;
  60         struct bch_devs_mask devs;
  61         unsigned dev;
  62
  63         dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX);
  64         if (dev == BCH_SB_MEMBERS_MAX)
  65                 return;
  66
  67         devs = inode->ei_devs_need_flush;
  68         memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
  69
  70         for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
  71                 rcu_read_lock();
  72                 ca = rcu_dereference(c->devs[dev]);
  73                 if (ca && !percpu_ref_tryget(&ca->io_ref))
  74                         ca = NULL;
  75                 rcu_read_unlock();
  76
  77                 if (!ca)
  78                         continue;
  79
  80                 bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0,
  81                                                     REQ_OP_FLUSH,
  82                                                     GFP_KERNEL,
  83                                                     &c->nocow_flush_bioset),
  84                                    struct nocow_flush, bio);
  85                 bio->cl                 = cl;
  86                 bio->ca                 = ca;
  87                 bio->bio.bi_end_io      = nocow_flush_endio;
  88                 closure_bio_submit(&bio->bio, cl);
  89         }
  90 }
  91
  92 static int bch2_inode_flush_nocow_writes(struct bch_fs *c,
  93                                          struct bch_inode_info *inode)
  94 {
  95         struct closure cl;
  96
  97         closure_init_stack(&cl);
  98         bch2_inode_flush_nocow_writes_async(c, inode, &cl);
  99         closure_sync(&cl);
 100
 101         return 0;
 102 }
 103
 104 /* i_size updates: */
 105
 106 struct inode_new_size {
 107         loff_t          new_size;
 108         u64             now;
 109         unsigned        fields;
 110 };
 111
 112 static int inode_set_size(struct btree_trans *trans,
 113                           struct bch_inode_info *inode,
 114                           struct bch_inode_unpacked *bi,
 115                           void *p)
 116 {
 117         struct inode_new_size *s = p;
 118
 119         bi->bi_size = s->new_size;
 120         if (s->fields & ATTR_ATIME)
 121                 bi->bi_atime = s->now;
 122         if (s->fields & ATTR_MTIME)
 123                 bi->bi_mtime = s->now;
 124         if (s->fields & ATTR_CTIME)
 125                 bi->bi_ctime = s->now;
 126
 127         return 0;
 128 }
 129
 130 int __must_check bch2_write_inode_size(struct bch_fs *c,
 131                                        struct bch_inode_info *inode,
 132                                        loff_t new_size, unsigned fields)
 133 {
 134         struct inode_new_size s = {
 135                 .new_size       = new_size,
 136                 .now            = bch2_current_time(c),
 137                 .fields         = fields,
 138         };
 139
 140         return bch2_write_inode(c, inode, inode_set_size, &s, fields);
 141 }
 142
 143 void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 144                            struct quota_res *quota_res, s64 sectors)
 145 {
 146         bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c,
 147                                 "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
 148                                 inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
 149                                 inode->ei_inode.bi_sectors);
 150         inode->v.i_blocks += sectors;
 151
 152 #ifdef CONFIG_BCACHEFS_QUOTA
 153         if (quota_res &&
 154             !test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) &&
 155             sectors > 0) {
 156                 BUG_ON(sectors > quota_res->sectors);
 157                 BUG_ON(sectors > inode->ei_quota_reserved);
 158
 159                 quota_res->sectors -= sectors;
 160                 inode->ei_quota_reserved -= sectors;
 161         } else {
 162                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
 163         }
 164 #endif
 165 }
 166
 167
 168 /* fsync: */
 169
 170 /*
 171  * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
 172  * insert trigger: look up the btree inode instead
 173  */
 174 static int bch2_flush_inode(struct bch_fs *c,
 175                             struct bch_inode_info *inode)
 176 {
 177         struct bch_inode_unpacked u;
 178         int ret;
 179
 180         if (c->opts.journal_flush_disabled)
 181                 return 0;
 182
 183         ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u);
 184         if (ret)
 185                 return ret;
 186
 187         return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
 188                 bch2_inode_flush_nocow_writes(c, inode);
 189 }
 190
 191 int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 192 {
 193         struct bch_inode_info *inode = file_bch_inode(file);
 194         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 195         int ret, ret2, ret3;
 196
 197         ret = file_write_and_wait_range(file, start, end);
 198         ret2 = sync_inode_metadata(&inode->v, 1);
 199         ret3 = bch2_flush_inode(c, inode);
 200
 201         return bch2_err_class(ret ?: ret2 ?: ret3);
 202 }
 203
 204 /* truncate: */
 205
 206 static inline int range_has_data(struct bch_fs *c, u32 subvol,
 207                                  struct bpos start,
 208                                  struct bpos end)
 209 {
 210         struct btree_trans trans;
 211         struct btree_iter iter;
 212         struct bkey_s_c k;
 213         int ret = 0;
 214
 215         bch2_trans_init(&trans, c, 0, 0);
 216 retry:
 217         bch2_trans_begin(&trans);
 218
 219         ret = bch2_subvolume_get_snapshot(&trans, subvol, &start.snapshot);
 220         if (ret)
 221                 goto err;
 222
 223         for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_extents, start, end, 0, k, ret)
 224                 if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) {
 225                         ret = 1;
 226                         break;
 227                 }
 228         start = iter.pos;
 229         bch2_trans_iter_exit(&trans, &iter);
 230 err:
 231         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 232                 goto retry;
 233
 234         bch2_trans_exit(&trans);
 235         return ret;
 236 }
 237
 238 static int __bch2_truncate_folio(struct bch_inode_info *inode,
 239                                  pgoff_t index, loff_t start, loff_t end)
 240 {
 241         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 242         struct address_space *mapping = inode->v.i_mapping;
 243         struct bch_folio *s;
 244         unsigned start_offset = start & (PAGE_SIZE - 1);
 245         unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1;
 246         unsigned i;
 247         struct folio *folio;
 248         s64 i_sectors_delta = 0;
 249         int ret = 0;
 250         u64 end_pos;
 251
 252         folio = filemap_lock_folio(mapping, index);
 253         if (IS_ERR_OR_NULL(folio)) {
 254                 /*
 255                  * XXX: we're doing two index lookups when we end up reading the
 256                  * folio
 257                  */
 258                 ret = range_has_data(c, inode->ei_subvol,
 259                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
 260                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
 261                 if (ret <= 0)
 262                         return ret;
 263
 264                 folio = __filemap_get_folio(mapping, index,
 265                                             FGP_LOCK|FGP_CREAT, GFP_KERNEL);
 266                 if (IS_ERR_OR_NULL(folio)) {
 267                         ret = -ENOMEM;
 268                         goto out;
 269                 }
 270         }
 271
 272         BUG_ON(start    >= folio_end_pos(folio));
 273         BUG_ON(end      <= folio_pos(folio));
 274
 275         start_offset    = max(start, folio_pos(folio)) - folio_pos(folio);
 276         end_offset      = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
 277
 278         /* Folio boundary? Nothing to do */
 279         if (start_offset == 0 &&
 280             end_offset == folio_size(folio)) {
 281                 ret = 0;
 282                 goto unlock;
 283         }
 284
 285         s = bch2_folio_create(folio, 0);
 286         if (!s) {
 287                 ret = -ENOMEM;
 288                 goto unlock;
 289         }
 290
 291         if (!folio_test_uptodate(folio)) {
 292                 ret = bch2_read_single_folio(folio, mapping);
 293                 if (ret)
 294                         goto unlock;
 295         }
 296
 297         ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
 298         if (ret)
 299                 goto unlock;
 300
 301         for (i = round_up(start_offset, block_bytes(c)) >> 9;
 302              i < round_down(end_offset, block_bytes(c)) >> 9;
 303              i++) {
 304                 s->s[i].nr_replicas     = 0;
 305
 306                 i_sectors_delta -= s->s[i].state == SECTOR_dirty;
 307                 bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
 308         }
 309
 310         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 311
 312         /*
 313          * Caller needs to know whether this folio will be written out by
 314          * writeback - doing an i_size update if necessary - or whether it will
 315          * be responsible for the i_size update.
 316          *
 317          * Note that we shouldn't ever see a folio beyond EOF, but check and
 318          * warn if so. This has been observed by failure to clean up folios
 319          * after a short write and there's still a chance reclaim will fix
 320          * things up.
 321          */
 322         WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
 323         end_pos = folio_end_pos(folio);
 324         if (inode->v.i_size > folio_pos(folio))
 325                 end_pos = min_t(u64, inode->v.i_size, end_pos);
 326         ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
 327
 328         folio_zero_segment(folio, start_offset, end_offset);
 329
 330         /*
 331          * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
 332          *
 333          * XXX: because we aren't currently tracking whether the folio has actual
 334          * data in it (vs. just 0s, or only partially written) this wrong. ick.
 335          */
 336         BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
 337
 338         /*
 339          * This removes any writeable userspace mappings; we need to force
 340          * .page_mkwrite to be called again before any mmapped writes, to
 341          * redirty the full page:
 342          */
 343         folio_mkclean(folio);
 344         filemap_dirty_folio(mapping, folio);
 345 unlock:
 346         folio_unlock(folio);
 347         folio_put(folio);
 348 out:
 349         return ret;
 350 }
 351
 352 static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
 353 {
 354         return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
 355                                      from, ANYSINT_MAX(loff_t));
 356 }
 357
 358 static int bch2_truncate_folios(struct bch_inode_info *inode,
 359                                 loff_t start, loff_t end)
 360 {
 361         int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
 362                                         start, end);
 363
 364         if (ret >= 0 &&
 365             start >> PAGE_SHIFT != end >> PAGE_SHIFT)
 366                 ret = __bch2_truncate_folio(inode,
 367                                         (end - 1) >> PAGE_SHIFT,
 368                                         start, end);
 369         return ret;
 370 }
 371
 372 static int bch2_extend(struct mnt_idmap *idmap,
 373                        struct bch_inode_info *inode,
 374                        struct bch_inode_unpacked *inode_u,
 375                        struct iattr *iattr)
 376 {
 377         struct address_space *mapping = inode->v.i_mapping;
 378         int ret;
 379
 380         /*
 381          * sync appends:
 382          *
 383          * this has to be done _before_ extending i_size:
 384          */
 385         ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
 386         if (ret)
 387                 return ret;
 388
 389         truncate_setsize(&inode->v, iattr->ia_size);
 390
 391         return bch2_setattr_nonsize(idmap, inode, iattr);
 392 }
 393
 394 static int bch2_truncate_finish_fn(struct btree_trans *trans,
 395                                    struct bch_inode_info *inode,
 396                                    struct bch_inode_unpacked *bi,
 397                                    void *p)
 398 {
 399         bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
 400         return 0;
 401 }
 402
 403 static int bch2_truncate_start_fn(struct btree_trans *trans,
 404                                   struct bch_inode_info *inode,
 405                                   struct bch_inode_unpacked *bi, void *p)
 406 {
 407         u64 *new_i_size = p;
 408
 409         bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY;
 410         bi->bi_size = *new_i_size;
 411         return 0;
 412 }
 413
 414 int bch2_truncate(struct mnt_idmap *idmap,
 415                   struct bch_inode_info *inode, struct iattr *iattr)
 416 {
 417         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 418         struct address_space *mapping = inode->v.i_mapping;
 419         struct bch_inode_unpacked inode_u;
 420         u64 new_i_size = iattr->ia_size;
 421         s64 i_sectors_delta = 0;
 422         int ret = 0;
 423
 424         /*
 425          * If the truncate call with change the size of the file, the
 426          * cmtimes should be updated. If the size will not change, we
 427          * do not need to update the cmtimes.
 428          */
 429         if (iattr->ia_size != inode->v.i_size) {
 430                 if (!(iattr->ia_valid & ATTR_MTIME))
 431                         ktime_get_coarse_real_ts64(&iattr->ia_mtime);
 432                 if (!(iattr->ia_valid & ATTR_CTIME))
 433                         ktime_get_coarse_real_ts64(&iattr->ia_ctime);
 434                 iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
 435         }
 436
 437         inode_dio_wait(&inode->v);
 438         bch2_pagecache_block_get(inode);
 439
 440         ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
 441         if (ret)
 442                 goto err;
 443
 444         /*
 445          * check this before next assertion; on filesystem error our normal
 446          * invariants are a bit broken (truncate has to truncate the page cache
 447          * before the inode).
 448          */
 449         ret = bch2_journal_error(&c->journal);
 450         if (ret)
 451                 goto err;
 452
 453         WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
 454                   inode->v.i_size < inode_u.bi_size,
 455                   "truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
 456                   (u64) inode->v.i_size, inode_u.bi_size);
 457
 458         if (iattr->ia_size > inode->v.i_size) {
 459                 ret = bch2_extend(idmap, inode, &inode_u, iattr);
 460                 goto err;
 461         }
 462
 463         iattr->ia_valid &= ~ATTR_SIZE;
 464
 465         ret = bch2_truncate_folio(inode, iattr->ia_size);
 466         if (unlikely(ret < 0))
 467                 goto err;
 468
 469         /*
 470          * When extending, we're going to write the new i_size to disk
 471          * immediately so we need to flush anything above the current on disk
 472          * i_size first:
 473          *
 474          * Also, when extending we need to flush the page that i_size currently
 475          * straddles - if it's mapped to userspace, we need to ensure that
 476          * userspace has to redirty it and call .mkwrite -> set_page_dirty
 477          * again to allocate the part of the page that was extended.
 478          */
 479         if (iattr->ia_size > inode_u.bi_size)
 480                 ret = filemap_write_and_wait_range(mapping,
 481                                 inode_u.bi_size,
 482                                 iattr->ia_size - 1);
 483         else if (iattr->ia_size & (PAGE_SIZE - 1))
 484                 ret = filemap_write_and_wait_range(mapping,
 485                                 round_down(iattr->ia_size, PAGE_SIZE),
 486                                 iattr->ia_size - 1);
 487         if (ret)
 488                 goto err;
 489
 490         mutex_lock(&inode->ei_update_lock);
 491         ret = bch2_write_inode(c, inode, bch2_truncate_start_fn,
 492                                &new_i_size, 0);
 493         mutex_unlock(&inode->ei_update_lock);
 494
 495         if (unlikely(ret))
 496                 goto err;
 497
 498         truncate_setsize(&inode->v, iattr->ia_size);
 499
 500         ret = bch2_fpunch(c, inode_inum(inode),
 501                         round_up(iattr->ia_size, block_bytes(c)) >> 9,
 502                         U64_MAX, &i_sectors_delta);
 503         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 504
 505         bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
 506                                 !bch2_journal_error(&c->journal), c,
 507                                 "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
 508                                 inode->v.i_ino, (u64) inode->v.i_blocks,
 509                                 inode->ei_inode.bi_sectors);
 510         if (unlikely(ret))
 511                 goto err;
 512
 513         mutex_lock(&inode->ei_update_lock);
 514         ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 0);
 515         mutex_unlock(&inode->ei_update_lock);
 516
 517         ret = bch2_setattr_nonsize(idmap, inode, iattr);
 518 err:
 519         bch2_pagecache_block_put(inode);
 520         return bch2_err_class(ret);
 521 }
 522
 523 /* fallocate: */
 524
 525 static int inode_update_times_fn(struct btree_trans *trans,
 526                                  struct bch_inode_info *inode,
 527                                  struct bch_inode_unpacked *bi, void *p)
 528 {
 529         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 530
 531         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 532         return 0;
 533 }
 534
 535 static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
 536 {
 537         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 538         u64 end         = offset + len;
 539         u64 block_start = round_up(offset, block_bytes(c));
 540         u64 block_end   = round_down(end, block_bytes(c));
 541         bool truncated_last_page;
 542         int ret = 0;
 543
 544         ret = bch2_truncate_folios(inode, offset, end);
 545         if (unlikely(ret < 0))
 546                 goto err;
 547
 548         truncated_last_page = ret;
 549
 550         truncate_pagecache_range(&inode->v, offset, end - 1);
 551
 552         if (block_start < block_end) {
 553                 s64 i_sectors_delta = 0;
 554
 555                 ret = bch2_fpunch(c, inode_inum(inode),
 556                                   block_start >> 9, block_end >> 9,
 557                                   &i_sectors_delta);
 558                 bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 559         }
 560
 561         mutex_lock(&inode->ei_update_lock);
 562         if (end >= inode->v.i_size && !truncated_last_page) {
 563                 ret = bch2_write_inode_size(c, inode, inode->v.i_size,
 564                                             ATTR_MTIME|ATTR_CTIME);
 565         } else {
 566                 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
 567                                        ATTR_MTIME|ATTR_CTIME);
 568         }
 569         mutex_unlock(&inode->ei_update_lock);
 570 err:
 571         return ret;
 572 }
 573
 574 static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 575                                    loff_t offset, loff_t len,
 576                                    bool insert)
 577 {
 578         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 579         struct address_space *mapping = inode->v.i_mapping;
 580         struct bkey_buf copy;
 581         struct btree_trans trans;
 582         struct btree_iter src, dst, del;
 583         loff_t shift, new_size;
 584         u64 src_start;
 585         int ret = 0;
 586
 587         if ((offset | len) & (block_bytes(c) - 1))
 588                 return -EINVAL;
 589
 590         if (insert) {
 591                 if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len)
 592                         return -EFBIG;
 593
 594                 if (offset >= inode->v.i_size)
 595                         return -EINVAL;
 596
 597                 src_start       = U64_MAX;
 598                 shift           = len;
 599         } else {
 600                 if (offset + len >= inode->v.i_size)
 601                         return -EINVAL;
 602
 603                 src_start       = offset + len;
 604                 shift           = -len;
 605         }
 606
 607         new_size = inode->v.i_size + shift;
 608
 609         ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
 610         if (ret)
 611                 return ret;
 612
 613         if (insert) {
 614                 i_size_write(&inode->v, new_size);
 615                 mutex_lock(&inode->ei_update_lock);
 616                 ret = bch2_write_inode_size(c, inode, new_size,
 617                                             ATTR_MTIME|ATTR_CTIME);
 618                 mutex_unlock(&inode->ei_update_lock);
 619         } else {
 620                 s64 i_sectors_delta = 0;
 621
 622                 ret = bch2_fpunch(c, inode_inum(inode),
 623                                   offset >> 9, (offset + len) >> 9,
 624                                   &i_sectors_delta);
 625                 bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 626
 627                 if (ret)
 628                         return ret;
 629         }
 630
 631         bch2_bkey_buf_init(&copy);
 632         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
 633         bch2_trans_iter_init(&trans, &src, BTREE_ID_extents,
 634                         POS(inode->v.i_ino, src_start >> 9),
 635                         BTREE_ITER_INTENT);
 636         bch2_trans_copy_iter(&dst, &src);
 637         bch2_trans_copy_iter(&del, &src);
 638
 639         while (ret == 0 ||
 640                bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 641                 struct disk_reservation disk_res =
 642                         bch2_disk_reservation_init(c, 0);
 643                 struct bkey_i delete;
 644                 struct bkey_s_c k;
 645                 struct bpos next_pos;
 646                 struct bpos move_pos = POS(inode->v.i_ino, offset >> 9);
 647                 struct bpos atomic_end;
 648                 unsigned trigger_flags = 0;
 649                 u32 snapshot;
 650
 651                 bch2_trans_begin(&trans);
 652
 653                 ret = bch2_subvolume_get_snapshot(&trans,
 654                                         inode->ei_subvol, &snapshot);
 655                 if (ret)
 656                         continue;
 657
 658                 bch2_btree_iter_set_snapshot(&src, snapshot);
 659                 bch2_btree_iter_set_snapshot(&dst, snapshot);
 660                 bch2_btree_iter_set_snapshot(&del, snapshot);
 661
 662                 bch2_trans_begin(&trans);
 663
 664                 k = insert
 665                         ? bch2_btree_iter_peek_prev(&src)
 666                         : bch2_btree_iter_peek_upto(&src, POS(inode->v.i_ino, U64_MAX));
 667                 if ((ret = bkey_err(k)))
 668                         continue;
 669
 670                 if (!k.k || k.k->p.inode != inode->v.i_ino)
 671                         break;
 672
 673                 if (insert &&
 674                     bkey_le(k.k->p, POS(inode->v.i_ino, offset >> 9)))
 675                         break;
 676 reassemble:
 677                 bch2_bkey_buf_reassemble(&copy, c, k);
 678
 679                 if (insert &&
 680                     bkey_lt(bkey_start_pos(k.k), move_pos))
 681                         bch2_cut_front(move_pos, copy.k);
 682
 683                 copy.k->k.p.offset += shift >> 9;
 684                 bch2_btree_iter_set_pos(&dst, bkey_start_pos(&copy.k->k));
 685
 686                 ret = bch2_extent_atomic_end(&trans, &dst, copy.k, &atomic_end);
 687                 if (ret)
 688                         continue;
 689
 690                 if (!bkey_eq(atomic_end, copy.k->k.p)) {
 691                         if (insert) {
 692                                 move_pos = atomic_end;
 693                                 move_pos.offset -= shift >> 9;
 694                                 goto reassemble;
 695                         } else {
 696                                 bch2_cut_back(atomic_end, copy.k);
 697                         }
 698                 }
 699
 700                 bkey_init(&delete.k);
 701                 delete.k.p = copy.k->k.p;
 702                 delete.k.size = copy.k->k.size;
 703                 delete.k.p.offset -= shift >> 9;
 704                 bch2_btree_iter_set_pos(&del, bkey_start_pos(&delete.k));
 705
 706                 next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p;
 707
 708                 if (copy.k->k.size != k.k->size) {
 709                         /* We might end up splitting compressed extents: */
 710                         unsigned nr_ptrs =
 711                                 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k));
 712
 713                         ret = bch2_disk_reservation_get(c, &disk_res,
 714                                         copy.k->k.size, nr_ptrs,
 715                                         BCH_DISK_RESERVATION_NOFAIL);
 716                         BUG_ON(ret);
 717                 }
 718
 719                 ret =   bch2_btree_iter_traverse(&del) ?:
 720                         bch2_trans_update(&trans, &del, &delete, trigger_flags) ?:
 721                         bch2_trans_update(&trans, &dst, copy.k, trigger_flags) ?:
 722                         bch2_trans_commit(&trans, &disk_res, NULL,
 723                                           BTREE_INSERT_NOFAIL);
 724                 bch2_disk_reservation_put(c, &disk_res);
 725
 726                 if (!ret)
 727                         bch2_btree_iter_set_pos(&src, next_pos);
 728         }
 729         bch2_trans_iter_exit(&trans, &del);
 730         bch2_trans_iter_exit(&trans, &dst);
 731         bch2_trans_iter_exit(&trans, &src);
 732         bch2_trans_exit(&trans);
 733         bch2_bkey_buf_exit(&copy, c);
 734
 735         if (ret)
 736                 return ret;
 737
 738         mutex_lock(&inode->ei_update_lock);
 739         if (!insert) {
 740                 i_size_write(&inode->v, new_size);
 741                 ret = bch2_write_inode_size(c, inode, new_size,
 742                                             ATTR_MTIME|ATTR_CTIME);
 743         } else {
 744                 /* We need an inode update to update bi_journal_seq for fsync: */
 745                 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
 746                                        ATTR_MTIME|ATTR_CTIME);
 747         }
 748         mutex_unlock(&inode->ei_update_lock);
 749         return ret;
 750 }
 751
 752 static int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 753                              u64 start_sector, u64 end_sector)
 754 {
 755         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 756         struct btree_trans trans;
 757         struct btree_iter iter;
 758         struct bpos end_pos = POS(inode->v.i_ino, end_sector);
 759         struct bch_io_opts opts;
 760         int ret = 0;
 761
 762         bch2_inode_opts_get(&opts, c, &inode->ei_inode);
 763         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 512);
 764
 765         bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
 766                         POS(inode->v.i_ino, start_sector),
 767                         BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 768
 769         while (!ret && bkey_lt(iter.pos, end_pos)) {
 770                 s64 i_sectors_delta = 0;
 771                 struct quota_res quota_res = { 0 };
 772                 struct bkey_s_c k;
 773                 unsigned sectors;
 774                 bool is_allocation;
 775                 u64 hole_start, hole_end;
 776                 u32 snapshot;
 777
 778                 bch2_trans_begin(&trans);
 779
 780                 ret = bch2_subvolume_get_snapshot(&trans,
 781                                         inode->ei_subvol, &snapshot);
 782                 if (ret)
 783                         goto bkey_err;
 784
 785                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 786
 787                 k = bch2_btree_iter_peek_slot(&iter);
 788                 if ((ret = bkey_err(k)))
 789                         goto bkey_err;
 790
 791                 hole_start      = iter.pos.offset;
 792                 hole_end        = bpos_min(k.k->p, end_pos).offset;
 793                 is_allocation   = bkey_extent_is_allocation(k.k);
 794
 795                 /* already reserved */
 796                 if (bkey_extent_is_reservation(k) &&
 797                     bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
 798                         bch2_btree_iter_advance(&iter);
 799                         continue;
 800                 }
 801
 802                 if (bkey_extent_is_data(k.k) &&
 803                     !(mode & FALLOC_FL_ZERO_RANGE)) {
 804                         bch2_btree_iter_advance(&iter);
 805                         continue;
 806                 }
 807
 808                 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
 809                         /*
 810                          * Lock ordering - can't be holding btree locks while
 811                          * blocking on a folio lock:
 812                          */
 813                         if (bch2_clamp_data_hole(&inode->v,
 814                                                  &hole_start,
 815                                                  &hole_end,
 816                                                  opts.data_replicas, true))
 817                                 ret = drop_locks_do(&trans,
 818                                         (bch2_clamp_data_hole(&inode->v,
 819                                                               &hole_start,
 820                                                               &hole_end,
 821                                                               opts.data_replicas, false), 0));
 822                         bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
 823
 824                         if (ret)
 825                                 goto bkey_err;
 826
 827                         if (hole_start == hole_end)
 828                                 continue;
 829                 }
 830
 831                 sectors = hole_end - hole_start;
 832
 833                 if (!is_allocation) {
 834                         ret = bch2_quota_reservation_add(c, inode,
 835                                         &quota_res, sectors, true);
 836                         if (unlikely(ret))
 837                                 goto bkey_err;
 838                 }
 839
 840                 ret = bch2_extent_fallocate(&trans, inode_inum(inode), &iter,
 841                                             sectors, opts, &i_sectors_delta,
 842                                             writepoint_hashed((unsigned long) current));
 843                 if (ret)
 844                         goto bkey_err;
 845
 846                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 847
 848                 drop_locks_do(&trans,
 849                         (bch2_mark_pagecache_reserved(inode, hole_start, iter.pos.offset), 0));
 850 bkey_err:
 851                 bch2_quota_reservation_put(c, inode, &quota_res);
 852                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 853                         ret = 0;
 854         }
 855
 856         if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
 857                 struct quota_res quota_res = { 0 };
 858                 s64 i_sectors_delta = 0;
 859
 860                 bch2_fpunch_at(&trans, &iter, inode_inum(inode),
 861                                end_sector, &i_sectors_delta);
 862                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 863                 bch2_quota_reservation_put(c, inode, &quota_res);
 864         }
 865
 866         bch2_trans_iter_exit(&trans, &iter);
 867         bch2_trans_exit(&trans);
 868         return ret;
 869 }
 870
 871 static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 872                             loff_t offset, loff_t len)
 873 {
 874         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 875         u64 end         = offset + len;
 876         u64 block_start = round_down(offset,    block_bytes(c));
 877         u64 block_end   = round_up(end,         block_bytes(c));
 878         bool truncated_last_page = false;
 879         int ret, ret2 = 0;
 880
 881         if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
 882                 ret = inode_newsize_ok(&inode->v, end);
 883                 if (ret)
 884                         return ret;
 885         }
 886
 887         if (mode & FALLOC_FL_ZERO_RANGE) {
 888                 ret = bch2_truncate_folios(inode, offset, end);
 889                 if (unlikely(ret < 0))
 890                         return ret;
 891
 892                 truncated_last_page = ret;
 893
 894                 truncate_pagecache_range(&inode->v, offset, end - 1);
 895
 896                 block_start     = round_up(offset,      block_bytes(c));
 897                 block_end       = round_down(end,       block_bytes(c));
 898         }
 899
 900         ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
 901
 902         /*
 903          * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
 904          * so that the VFS cache i_size is consistent with the btree i_size:
 905          */
 906         if (ret &&
 907             !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
 908                 return ret;
 909
 910         if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
 911                 end = inode->v.i_size;
 912
 913         if (end >= inode->v.i_size &&
 914             (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
 915              !(mode & FALLOC_FL_KEEP_SIZE))) {
 916                 spin_lock(&inode->v.i_lock);
 917                 i_size_write(&inode->v, end);
 918                 spin_unlock(&inode->v.i_lock);
 919
 920                 mutex_lock(&inode->ei_update_lock);
 921                 ret2 = bch2_write_inode_size(c, inode, end, 0);
 922                 mutex_unlock(&inode->ei_update_lock);
 923         }
 924
 925         return ret ?: ret2;
 926 }
 927
 928 long bch2_fallocate_dispatch(struct file *file, int mode,
 929                              loff_t offset, loff_t len)
 930 {
 931         struct bch_inode_info *inode = file_bch_inode(file);
 932         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 933         long ret;
 934
 935         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
 936                 return -EROFS;
 937
 938         inode_lock(&inode->v);
 939         inode_dio_wait(&inode->v);
 940         bch2_pagecache_block_get(inode);
 941
 942         ret = file_modified(file);
 943         if (ret)
 944                 goto err;
 945
 946         if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
 947                 ret = bchfs_fallocate(inode, mode, offset, len);
 948         else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
 949                 ret = bchfs_fpunch(inode, offset, len);
 950         else if (mode == FALLOC_FL_INSERT_RANGE)
 951                 ret = bchfs_fcollapse_finsert(inode, offset, len, true);
 952         else if (mode == FALLOC_FL_COLLAPSE_RANGE)
 953                 ret = bchfs_fcollapse_finsert(inode, offset, len, false);
 954         else
 955                 ret = -EOPNOTSUPP;
 956 err:
 957         bch2_pagecache_block_put(inode);
 958         inode_unlock(&inode->v);
 959         bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
 960
 961         return bch2_err_class(ret);
 962 }
 963
 964 /*
 965  * Take a quota reservation for unallocated blocks in a given file range
 966  * Does not check pagecache
 967  */
 968 static int quota_reserve_range(struct bch_inode_info *inode,
 969                                struct quota_res *res,
 970                                u64 start, u64 end)
 971 {
 972         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 973         struct btree_trans trans;
 974         struct btree_iter iter;
 975         struct bkey_s_c k;
 976         u32 snapshot;
 977         u64 sectors = end - start;
 978         u64 pos = start;
 979         int ret;
 980
 981         bch2_trans_init(&trans, c, 0, 0);
 982 retry:
 983         bch2_trans_begin(&trans);
 984
 985         ret = bch2_subvolume_get_snapshot(&trans, inode->ei_subvol, &snapshot);
 986         if (ret)
 987                 goto err;
 988
 989         bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
 990                              SPOS(inode->v.i_ino, pos, snapshot), 0);
 991
 992         while (!(ret = btree_trans_too_many_iters(&trans)) &&
 993                (k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k &&
 994                !(ret = bkey_err(k))) {
 995                 if (bkey_extent_is_allocation(k.k)) {
 996                         u64 s = min(end, k.k->p.offset) -
 997                                 max(start, bkey_start_offset(k.k));
 998                         BUG_ON(s > sectors);
 999                         sectors -= s;
1000                 }
1001                 bch2_btree_iter_advance(&iter);
1002         }
1003         pos = iter.pos.offset;
1004         bch2_trans_iter_exit(&trans, &iter);
1005 err:
1006         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1007                 goto retry;
1008
1009         bch2_trans_exit(&trans);
1010
1011         if (ret)
1012                 return ret;
1013
1014         return bch2_quota_reservation_add(c, inode, res, sectors, true);
1015 }
1016
1017 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
1018                              struct file *file_dst, loff_t pos_dst,
1019                              loff_t len, unsigned remap_flags)
1020 {
1021         struct bch_inode_info *src = file_bch_inode(file_src);
1022         struct bch_inode_info *dst = file_bch_inode(file_dst);
1023         struct bch_fs *c = src->v.i_sb->s_fs_info;
1024         struct quota_res quota_res = { 0 };
1025         s64 i_sectors_delta = 0;
1026         u64 aligned_len;
1027         loff_t ret = 0;
1028
1029         if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
1030                 return -EINVAL;
1031
1032         if (remap_flags & REMAP_FILE_DEDUP)
1033                 return -EOPNOTSUPP;
1034
1035         if ((pos_src & (block_bytes(c) - 1)) ||
1036             (pos_dst & (block_bytes(c) - 1)))
1037                 return -EINVAL;
1038
1039         if (src == dst &&
1040             abs(pos_src - pos_dst) < len)
1041                 return -EINVAL;
1042
1043         bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);
1044
1045         inode_dio_wait(&src->v);
1046         inode_dio_wait(&dst->v);
1047
1048         ret = generic_remap_file_range_prep(file_src, pos_src,
1049                                             file_dst, pos_dst,
1050                                             &len, remap_flags);
1051         if (ret < 0 || len == 0)
1052                 goto err;
1053
1054         aligned_len = round_up((u64) len, block_bytes(c));
1055
1056         ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
1057                                 pos_dst, pos_dst + len - 1);
1058         if (ret)
1059                 goto err;
1060
1061         ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
1062                                   (pos_dst + aligned_len) >> 9);
1063         if (ret)
1064                 goto err;
1065
1066         file_update_time(file_dst);
1067
1068         bch2_mark_pagecache_unallocated(src, pos_src >> 9,
1069                                    (pos_src + aligned_len) >> 9);
1070
1071         ret = bch2_remap_range(c,
1072                                inode_inum(dst), pos_dst >> 9,
1073                                inode_inum(src), pos_src >> 9,
1074                                aligned_len >> 9,
1075                                pos_dst + len, &i_sectors_delta);
1076         if (ret < 0)
1077                 goto err;
1078
1079         /*
1080          * due to alignment, we might have remapped slightly more than requsted
1081          */
1082         ret = min((u64) ret << 9, (u64) len);
1083
1084         bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
1085
1086         spin_lock(&dst->v.i_lock);
1087         if (pos_dst + ret > dst->v.i_size)
1088                 i_size_write(&dst->v, pos_dst + ret);
1089         spin_unlock(&dst->v.i_lock);
1090
1091         if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
1092             IS_SYNC(file_inode(file_dst)))
1093                 ret = bch2_flush_inode(c, dst);
1094 err:
1095         bch2_quota_reservation_put(c, dst, &quota_res);
1096         bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);
1097
1098         return bch2_err_class(ret);
1099 }
1100
1101 /* fseek: */
1102
1103 static loff_t bch2_seek_data(struct file *file, u64 offset)
1104 {
1105         struct bch_inode_info *inode = file_bch_inode(file);
1106         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1107         struct btree_trans trans;
1108         struct btree_iter iter;
1109         struct bkey_s_c k;
1110         subvol_inum inum = inode_inum(inode);
1111         u64 isize, next_data = MAX_LFS_FILESIZE;
1112         u32 snapshot;
1113         int ret;
1114
1115         isize = i_size_read(&inode->v);
1116         if (offset >= isize)
1117                 return -ENXIO;
1118
1119         bch2_trans_init(&trans, c, 0, 0);
1120 retry:
1121         bch2_trans_begin(&trans);
1122
1123         ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
1124         if (ret)
1125                 goto err;
1126
1127         for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_extents,
1128                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
1129                            POS(inode->v.i_ino, U64_MAX),
1130                            0, k, ret) {
1131                 if (bkey_extent_is_data(k.k)) {
1132                         next_data = max(offset, bkey_start_offset(k.k) << 9);
1133                         break;
1134                 } else if (k.k->p.offset >> 9 > isize)
1135                         break;
1136         }
1137         bch2_trans_iter_exit(&trans, &iter);
1138 err:
1139         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1140                 goto retry;
1141
1142         bch2_trans_exit(&trans);
1143         if (ret)
1144                 return ret;
1145
1146         if (next_data > offset)
1147                 next_data = bch2_seek_pagecache_data(&inode->v,
1148                                         offset, next_data, 0, false);
1149
1150         if (next_data >= isize)
1151                 return -ENXIO;
1152
1153         return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
1154 }
1155
1156 static loff_t bch2_seek_hole(struct file *file, u64 offset)
1157 {
1158         struct bch_inode_info *inode = file_bch_inode(file);
1159         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1160         struct btree_trans trans;
1161         struct btree_iter iter;
1162         struct bkey_s_c k;
1163         subvol_inum inum = inode_inum(inode);
1164         u64 isize, next_hole = MAX_LFS_FILESIZE;
1165         u32 snapshot;
1166         int ret;
1167
1168         isize = i_size_read(&inode->v);
1169         if (offset >= isize)
1170                 return -ENXIO;
1171
1172         bch2_trans_init(&trans, c, 0, 0);
1173 retry:
1174         bch2_trans_begin(&trans);
1175
1176         ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
1177         if (ret)
1178                 goto err;
1179
1180         for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents,
1181                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
1182                            BTREE_ITER_SLOTS, k, ret) {
1183                 if (k.k->p.inode != inode->v.i_ino) {
1184                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1185                                         offset, MAX_LFS_FILESIZE, 0, false);
1186                         break;
1187                 } else if (!bkey_extent_is_data(k.k)) {
1188                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1189                                         max(offset, bkey_start_offset(k.k) << 9),
1190                                         k.k->p.offset << 9, 0, false);
1191
1192                         if (next_hole < k.k->p.offset << 9)
1193                                 break;
1194                 } else {
1195                         offset = max(offset, bkey_start_offset(k.k) << 9);
1196                 }
1197         }
1198         bch2_trans_iter_exit(&trans, &iter);
1199 err:
1200         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1201                 goto retry;
1202
1203         bch2_trans_exit(&trans);
1204         if (ret)
1205                 return ret;
1206
1207         if (next_hole > isize)
1208                 next_hole = isize;
1209
1210         return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
1211 }
1212
1213 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
1214 {
1215         loff_t ret;
1216
1217         switch (whence) {
1218         case SEEK_SET:
1219         case SEEK_CUR:
1220         case SEEK_END:
1221                 ret = generic_file_llseek(file, offset, whence);
1222                 break;
1223         case SEEK_DATA:
1224                 ret = bch2_seek_data(file, offset);
1225                 break;
1226         case SEEK_HOLE:
1227                 ret = bch2_seek_hole(file, offset);
1228                 break;
1229         default:
1230                 ret = -EINVAL;
1231                 break;
1232         }
1233
1234         return bch2_err_class(ret);
1235 }
1236
1237 void bch2_fs_fsio_exit(struct bch_fs *c)
1238 {
1239         bioset_exit(&c->nocow_flush_bioset);
1240 }
1241
1242 int bch2_fs_fsio_init(struct bch_fs *c)
1243 {
1244         if (bioset_init(&c->nocow_flush_bioset,
1245                         1, offsetof(struct nocow_flush, bio), 0))
1246                 return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
1247
1248         return 0;
1249 }
1250
1251 #endif /* NO_BCACHEFS_FS */