git.sesse.net Git - bcachefs-tools-debian/blob - c_src/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 "bkey_buf.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "clock.h"
  10 #include "error.h"
  11 #include "extents.h"
  12 #include "extent_update.h"
  13 #include "fs.h"
  14 #include "fs-io.h"
  15 #include "fs-io-buffered.h"
  16 #include "fs-io-pagecache.h"
  17 #include "fsck.h"
  18 #include "inode.h"
  19 #include "journal.h"
  20 #include "io_misc.h"
  21 #include "keylist.h"
  22 #include "quota.h"
  23 #include "reflink.h"
  24 #include "trace.h"
  25
  26 #include <linux/aio.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/falloc.h>
  29 #include <linux/migrate.h>
  30 #include <linux/mmu_context.h>
  31 #include <linux/pagevec.h>
  32 #include <linux/rmap.h>
  33 #include <linux/sched/signal.h>
  34 #include <linux/task_io_accounting_ops.h>
  35 #include <linux/uio.h>
  36
  37 #include <trace/events/writeback.h>
  38
  39 struct nocow_flush {
  40         struct closure  *cl;
  41         struct bch_dev  *ca;
  42         struct bio      bio;
  43 };
  44
  45 static void nocow_flush_endio(struct bio *_bio)
  46 {
  47
  48         struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio);
  49
  50         closure_put(bio->cl);
  51         percpu_ref_put(&bio->ca->io_ref);
  52         bio_put(&bio->bio);
  53 }
  54
  55 void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
  56                                          struct bch_inode_info *inode,
  57                                          struct closure *cl)
  58 {
  59         struct nocow_flush *bio;
  60         struct bch_dev *ca;
  61         struct bch_devs_mask devs;
  62         unsigned dev;
  63
  64         dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX);
  65         if (dev == BCH_SB_MEMBERS_MAX)
  66                 return;
  67
  68         devs = inode->ei_devs_need_flush;
  69         memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
  70
  71         for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
  72                 rcu_read_lock();
  73                 ca = rcu_dereference(c->devs[dev]);
  74                 if (ca && !percpu_ref_tryget(&ca->io_ref))
  75                         ca = NULL;
  76                 rcu_read_unlock();
  77
  78                 if (!ca)
  79                         continue;
  80
  81                 bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0,
  82                                                     REQ_OP_FLUSH,
  83                                                     GFP_KERNEL,
  84                                                     &c->nocow_flush_bioset),
  85                                    struct nocow_flush, bio);
  86                 bio->cl                 = cl;
  87                 bio->ca                 = ca;
  88                 bio->bio.bi_end_io      = nocow_flush_endio;
  89                 closure_bio_submit(&bio->bio, cl);
  90         }
  91 }
  92
  93 static int bch2_inode_flush_nocow_writes(struct bch_fs *c,
  94                                          struct bch_inode_info *inode)
  95 {
  96         struct closure cl;
  97
  98         closure_init_stack(&cl);
  99         bch2_inode_flush_nocow_writes_async(c, inode, &cl);
 100         closure_sync(&cl);
 101
 102         return 0;
 103 }
 104
 105 /* i_size updates: */
 106
 107 struct inode_new_size {
 108         loff_t          new_size;
 109         u64             now;
 110         unsigned        fields;
 111 };
 112
 113 static int inode_set_size(struct btree_trans *trans,
 114                           struct bch_inode_info *inode,
 115                           struct bch_inode_unpacked *bi,
 116                           void *p)
 117 {
 118         struct inode_new_size *s = p;
 119
 120         bi->bi_size = s->new_size;
 121         if (s->fields & ATTR_ATIME)
 122                 bi->bi_atime = s->now;
 123         if (s->fields & ATTR_MTIME)
 124                 bi->bi_mtime = s->now;
 125         if (s->fields & ATTR_CTIME)
 126                 bi->bi_ctime = s->now;
 127
 128         return 0;
 129 }
 130
 131 int __must_check bch2_write_inode_size(struct bch_fs *c,
 132                                        struct bch_inode_info *inode,
 133                                        loff_t new_size, unsigned fields)
 134 {
 135         struct inode_new_size s = {
 136                 .new_size       = new_size,
 137                 .now            = bch2_current_time(c),
 138                 .fields         = fields,
 139         };
 140
 141         return bch2_write_inode(c, inode, inode_set_size, &s, fields);
 142 }
 143
 144 void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 145                            struct quota_res *quota_res, s64 sectors)
 146 {
 147         bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c,
 148                                 "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
 149                                 inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
 150                                 inode->ei_inode.bi_sectors);
 151         inode->v.i_blocks += sectors;
 152
 153 #ifdef CONFIG_BCACHEFS_QUOTA
 154         if (quota_res &&
 155             !test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) &&
 156             sectors > 0) {
 157                 BUG_ON(sectors > quota_res->sectors);
 158                 BUG_ON(sectors > inode->ei_quota_reserved);
 159
 160                 quota_res->sectors -= sectors;
 161                 inode->ei_quota_reserved -= sectors;
 162         } else {
 163                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
 164         }
 165 #endif
 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;
 196
 197         ret = file_write_and_wait_range(file, start, end);
 198         if (ret)
 199                 goto out;
 200         ret = sync_inode_metadata(&inode->v, 1);
 201         if (ret)
 202                 goto out;
 203         ret = bch2_flush_inode(c, inode);
 204 out:
 205         return bch2_err_class(ret);
 206 }
 207
 208 /* truncate: */
 209
 210 static inline int range_has_data(struct bch_fs *c, u32 subvol,
 211                                  struct bpos start,
 212                                  struct bpos end)
 213 {
 214         struct btree_trans *trans = bch2_trans_get(c);
 215         struct btree_iter iter;
 216         struct bkey_s_c k;
 217         int ret = 0;
 218 retry:
 219         bch2_trans_begin(trans);
 220
 221         ret = bch2_subvolume_get_snapshot(trans, subvol, &start.snapshot);
 222         if (ret)
 223                 goto err;
 224
 225         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, start, end, 0, k, ret)
 226                 if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) {
 227                         ret = 1;
 228                         break;
 229                 }
 230         start = iter.pos;
 231         bch2_trans_iter_exit(trans, &iter);
 232 err:
 233         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 234                 goto retry;
 235
 236         bch2_trans_put(trans);
 237         return ret;
 238 }
 239
 240 static int __bch2_truncate_folio(struct bch_inode_info *inode,
 241                                  pgoff_t index, loff_t start, loff_t end)
 242 {
 243         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 244         struct address_space *mapping = inode->v.i_mapping;
 245         struct bch_folio *s;
 246         unsigned start_offset;
 247         unsigned end_offset;
 248         unsigned i;
 249         struct folio *folio;
 250         s64 i_sectors_delta = 0;
 251         int ret = 0;
 252         u64 end_pos;
 253
 254         folio = filemap_lock_folio(mapping, index);
 255         if (IS_ERR_OR_NULL(folio)) {
 256                 /*
 257                  * XXX: we're doing two index lookups when we end up reading the
 258                  * folio
 259                  */
 260                 ret = range_has_data(c, inode->ei_subvol,
 261                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
 262                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
 263                 if (ret <= 0)
 264                         return ret;
 265
 266                 folio = __filemap_get_folio(mapping, index,
 267                                             FGP_LOCK|FGP_CREAT, GFP_KERNEL);
 268                 if (IS_ERR_OR_NULL(folio)) {
 269                         ret = -ENOMEM;
 270                         goto out;
 271                 }
 272         }
 273
 274         BUG_ON(start    >= folio_end_pos(folio));
 275         BUG_ON(end      <= folio_pos(folio));
 276
 277         start_offset    = max(start, folio_pos(folio)) - folio_pos(folio);
 278         end_offset      = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
 279
 280         /* Folio boundary? Nothing to do */
 281         if (start_offset == 0 &&
 282             end_offset == folio_size(folio)) {
 283                 ret = 0;
 284                 goto unlock;
 285         }
 286
 287         s = bch2_folio_create(folio, 0);
 288         if (!s) {
 289                 ret = -ENOMEM;
 290                 goto unlock;
 291         }
 292
 293         if (!folio_test_uptodate(folio)) {
 294                 ret = bch2_read_single_folio(folio, mapping);
 295                 if (ret)
 296                         goto unlock;
 297         }
 298
 299         ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
 300         if (ret)
 301                 goto unlock;
 302
 303         for (i = round_up(start_offset, block_bytes(c)) >> 9;
 304              i < round_down(end_offset, block_bytes(c)) >> 9;
 305              i++) {
 306                 s->s[i].nr_replicas     = 0;
 307
 308                 i_sectors_delta -= s->s[i].state == SECTOR_dirty;
 309                 bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
 310         }
 311
 312         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 313
 314         /*
 315          * Caller needs to know whether this folio will be written out by
 316          * writeback - doing an i_size update if necessary - or whether it will
 317          * be responsible for the i_size update.
 318          *
 319          * Note that we shouldn't ever see a folio beyond EOF, but check and
 320          * warn if so. This has been observed by failure to clean up folios
 321          * after a short write and there's still a chance reclaim will fix
 322          * things up.
 323          */
 324         WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
 325         end_pos = folio_end_pos(folio);
 326         if (inode->v.i_size > folio_pos(folio))
 327                 end_pos = min_t(u64, inode->v.i_size, end_pos);
 328         ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
 329
 330         folio_zero_segment(folio, start_offset, end_offset);
 331
 332         /*
 333          * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
 334          *
 335          * XXX: because we aren't currently tracking whether the folio has actual
 336          * data in it (vs. just 0s, or only partially written) this wrong. ick.
 337          */
 338         BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
 339
 340         /*
 341          * This removes any writeable userspace mappings; we need to force
 342          * .page_mkwrite to be called again before any mmapped writes, to
 343          * redirty the full page:
 344          */
 345         folio_mkclean(folio);
 346         filemap_dirty_folio(mapping, folio);
 347 unlock:
 348         folio_unlock(folio);
 349         folio_put(folio);
 350 out:
 351         return ret;
 352 }
 353
 354 static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
 355 {
 356         return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
 357                                      from, ANYSINT_MAX(loff_t));
 358 }
 359
 360 static int bch2_truncate_folios(struct bch_inode_info *inode,
 361                                 loff_t start, loff_t end)
 362 {
 363         int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
 364                                         start, end);
 365
 366         if (ret >= 0 &&
 367             start >> PAGE_SHIFT != end >> PAGE_SHIFT)
 368                 ret = __bch2_truncate_folio(inode,
 369                                         (end - 1) >> PAGE_SHIFT,
 370                                         start, end);
 371         return ret;
 372 }
 373
 374 static int bch2_extend(struct mnt_idmap *idmap,
 375                        struct bch_inode_info *inode,
 376                        struct bch_inode_unpacked *inode_u,
 377                        struct iattr *iattr)
 378 {
 379         struct address_space *mapping = inode->v.i_mapping;
 380         int ret;
 381
 382         /*
 383          * sync appends:
 384          *
 385          * this has to be done _before_ extending i_size:
 386          */
 387         ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
 388         if (ret)
 389                 return ret;
 390
 391         truncate_setsize(&inode->v, iattr->ia_size);
 392
 393         return bch2_setattr_nonsize(idmap, inode, iattr);
 394 }
 395
 396 int bchfs_truncate(struct mnt_idmap *idmap,
 397                   struct bch_inode_info *inode, struct iattr *iattr)
 398 {
 399         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 400         struct address_space *mapping = inode->v.i_mapping;
 401         struct bch_inode_unpacked inode_u;
 402         s64 i_sectors_delta = 0;
 403         int ret = 0;
 404
 405         /*
 406          * If the truncate call with change the size of the file, the
 407          * cmtimes should be updated. If the size will not change, we
 408          * do not need to update the cmtimes.
 409          */
 410         if (iattr->ia_size != inode->v.i_size) {
 411                 if (!(iattr->ia_valid & ATTR_MTIME))
 412                         ktime_get_coarse_real_ts64(&iattr->ia_mtime);
 413                 if (!(iattr->ia_valid & ATTR_CTIME))
 414                         ktime_get_coarse_real_ts64(&iattr->ia_ctime);
 415                 iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
 416         }
 417
 418         inode_dio_wait(&inode->v);
 419         bch2_pagecache_block_get(inode);
 420
 421         ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
 422         if (ret)
 423                 goto err;
 424
 425         /*
 426          * check this before next assertion; on filesystem error our normal
 427          * invariants are a bit broken (truncate has to truncate the page cache
 428          * before the inode).
 429          */
 430         ret = bch2_journal_error(&c->journal);
 431         if (ret)
 432                 goto err;
 433
 434         WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
 435                   inode->v.i_size < inode_u.bi_size,
 436                   "truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
 437                   (u64) inode->v.i_size, inode_u.bi_size);
 438
 439         if (iattr->ia_size > inode->v.i_size) {
 440                 ret = bch2_extend(idmap, inode, &inode_u, iattr);
 441                 goto err;
 442         }
 443
 444         iattr->ia_valid &= ~ATTR_SIZE;
 445
 446         ret = bch2_truncate_folio(inode, iattr->ia_size);
 447         if (unlikely(ret < 0))
 448                 goto err;
 449
 450         truncate_setsize(&inode->v, iattr->ia_size);
 451
 452         /*
 453          * When extending, we're going to write the new i_size to disk
 454          * immediately so we need to flush anything above the current on disk
 455          * i_size first:
 456          *
 457          * Also, when extending we need to flush the page that i_size currently
 458          * straddles - if it's mapped to userspace, we need to ensure that
 459          * userspace has to redirty it and call .mkwrite -> set_page_dirty
 460          * again to allocate the part of the page that was extended.
 461          */
 462         if (iattr->ia_size > inode_u.bi_size)
 463                 ret = filemap_write_and_wait_range(mapping,
 464                                 inode_u.bi_size,
 465                                 iattr->ia_size - 1);
 466         else if (iattr->ia_size & (PAGE_SIZE - 1))
 467                 ret = filemap_write_and_wait_range(mapping,
 468                                 round_down(iattr->ia_size, PAGE_SIZE),
 469                                 iattr->ia_size - 1);
 470         if (ret)
 471                 goto err;
 472
 473         ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta);
 474         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 475
 476         if (unlikely(ret)) {
 477                 /*
 478                  * If we error here, VFS caches are now inconsistent with btree
 479                  */
 480                 set_bit(EI_INODE_ERROR, &inode->ei_flags);
 481                 goto err;
 482         }
 483
 484         bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
 485                                 !bch2_journal_error(&c->journal), c,
 486                                 "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
 487                                 inode->v.i_ino, (u64) inode->v.i_blocks,
 488                                 inode->ei_inode.bi_sectors);
 489
 490         ret = bch2_setattr_nonsize(idmap, inode, iattr);
 491 err:
 492         bch2_pagecache_block_put(inode);
 493         return bch2_err_class(ret);
 494 }
 495
 496 /* fallocate: */
 497
 498 static int inode_update_times_fn(struct btree_trans *trans,
 499                                  struct bch_inode_info *inode,
 500                                  struct bch_inode_unpacked *bi, void *p)
 501 {
 502         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 503
 504         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 505         return 0;
 506 }
 507
 508 static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
 509 {
 510         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 511         u64 end         = offset + len;
 512         u64 block_start = round_up(offset, block_bytes(c));
 513         u64 block_end   = round_down(end, block_bytes(c));
 514         bool truncated_last_page;
 515         int ret = 0;
 516
 517         ret = bch2_truncate_folios(inode, offset, end);
 518         if (unlikely(ret < 0))
 519                 goto err;
 520
 521         truncated_last_page = ret;
 522
 523         truncate_pagecache_range(&inode->v, offset, end - 1);
 524
 525         if (block_start < block_end) {
 526                 s64 i_sectors_delta = 0;
 527
 528                 ret = bch2_fpunch(c, inode_inum(inode),
 529                                   block_start >> 9, block_end >> 9,
 530                                   &i_sectors_delta);
 531                 bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 532         }
 533
 534         mutex_lock(&inode->ei_update_lock);
 535         if (end >= inode->v.i_size && !truncated_last_page) {
 536                 ret = bch2_write_inode_size(c, inode, inode->v.i_size,
 537                                             ATTR_MTIME|ATTR_CTIME);
 538         } else {
 539                 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
 540                                        ATTR_MTIME|ATTR_CTIME);
 541         }
 542         mutex_unlock(&inode->ei_update_lock);
 543 err:
 544         return ret;
 545 }
 546
 547 static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 548                                    loff_t offset, loff_t len,
 549                                    bool insert)
 550 {
 551         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 552         struct address_space *mapping = inode->v.i_mapping;
 553         s64 i_sectors_delta = 0;
 554         int ret = 0;
 555
 556         if ((offset | len) & (block_bytes(c) - 1))
 557                 return -EINVAL;
 558
 559         if (insert) {
 560                 if (offset >= inode->v.i_size)
 561                         return -EINVAL;
 562         } else {
 563                 if (offset + len >= inode->v.i_size)
 564                         return -EINVAL;
 565         }
 566
 567         ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
 568         if (ret)
 569                 return ret;
 570
 571         if (insert)
 572                 i_size_write(&inode->v, inode->v.i_size + len);
 573
 574         ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9,
 575                                      insert, &i_sectors_delta);
 576         if (!ret && !insert)
 577                 i_size_write(&inode->v, inode->v.i_size - len);
 578         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 579
 580         return ret;
 581 }
 582
 583 static int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 584                              u64 start_sector, u64 end_sector)
 585 {
 586         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 587         struct btree_trans *trans = bch2_trans_get(c);
 588         struct btree_iter iter;
 589         struct bpos end_pos = POS(inode->v.i_ino, end_sector);
 590         struct bch_io_opts opts;
 591         int ret = 0;
 592
 593         bch2_inode_opts_get(&opts, c, &inode->ei_inode);
 594
 595         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 596                         POS(inode->v.i_ino, start_sector),
 597                         BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 598
 599         while (!ret && bkey_lt(iter.pos, end_pos)) {
 600                 s64 i_sectors_delta = 0;
 601                 struct quota_res quota_res = { 0 };
 602                 struct bkey_s_c k;
 603                 unsigned sectors;
 604                 bool is_allocation;
 605                 u64 hole_start, hole_end;
 606                 u32 snapshot;
 607
 608                 bch2_trans_begin(trans);
 609
 610                 ret = bch2_subvolume_get_snapshot(trans,
 611                                         inode->ei_subvol, &snapshot);
 612                 if (ret)
 613                         goto bkey_err;
 614
 615                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 616
 617                 k = bch2_btree_iter_peek_slot(&iter);
 618                 if ((ret = bkey_err(k)))
 619                         goto bkey_err;
 620
 621                 hole_start      = iter.pos.offset;
 622                 hole_end        = bpos_min(k.k->p, end_pos).offset;
 623                 is_allocation   = bkey_extent_is_allocation(k.k);
 624
 625                 /* already reserved */
 626                 if (bkey_extent_is_reservation(k) &&
 627                     bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
 628                         bch2_btree_iter_advance(&iter);
 629                         continue;
 630                 }
 631
 632                 if (bkey_extent_is_data(k.k) &&
 633                     !(mode & FALLOC_FL_ZERO_RANGE)) {
 634                         bch2_btree_iter_advance(&iter);
 635                         continue;
 636                 }
 637
 638                 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
 639                         /*
 640                          * Lock ordering - can't be holding btree locks while
 641                          * blocking on a folio lock:
 642                          */
 643                         if (bch2_clamp_data_hole(&inode->v,
 644                                                  &hole_start,
 645                                                  &hole_end,
 646                                                  opts.data_replicas, true))
 647                                 ret = drop_locks_do(trans,
 648                                         (bch2_clamp_data_hole(&inode->v,
 649                                                               &hole_start,
 650                                                               &hole_end,
 651                                                               opts.data_replicas, false), 0));
 652                         bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
 653
 654                         if (ret)
 655                                 goto bkey_err;
 656
 657                         if (hole_start == hole_end)
 658                                 continue;
 659                 }
 660
 661                 sectors = hole_end - hole_start;
 662
 663                 if (!is_allocation) {
 664                         ret = bch2_quota_reservation_add(c, inode,
 665                                         &quota_res, sectors, true);
 666                         if (unlikely(ret))
 667                                 goto bkey_err;
 668                 }
 669
 670                 ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter,
 671                                             sectors, opts, &i_sectors_delta,
 672                                             writepoint_hashed((unsigned long) current));
 673                 if (ret)
 674                         goto bkey_err;
 675
 676                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 677
 678                 drop_locks_do(trans,
 679                         (bch2_mark_pagecache_reserved(inode, hole_start, iter.pos.offset), 0));
 680 bkey_err:
 681                 bch2_quota_reservation_put(c, inode, &quota_res);
 682                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 683                         ret = 0;
 684         }
 685
 686         if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
 687                 struct quota_res quota_res = { 0 };
 688                 s64 i_sectors_delta = 0;
 689
 690                 bch2_fpunch_at(trans, &iter, inode_inum(inode),
 691                                end_sector, &i_sectors_delta);
 692                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 693                 bch2_quota_reservation_put(c, inode, &quota_res);
 694         }
 695
 696         bch2_trans_iter_exit(trans, &iter);
 697         bch2_trans_put(trans);
 698         return ret;
 699 }
 700
 701 static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 702                             loff_t offset, loff_t len)
 703 {
 704         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 705         u64 end         = offset + len;
 706         u64 block_start = round_down(offset,    block_bytes(c));
 707         u64 block_end   = round_up(end,         block_bytes(c));
 708         bool truncated_last_page = false;
 709         int ret, ret2 = 0;
 710
 711         if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
 712                 ret = inode_newsize_ok(&inode->v, end);
 713                 if (ret)
 714                         return ret;
 715         }
 716
 717         if (mode & FALLOC_FL_ZERO_RANGE) {
 718                 ret = bch2_truncate_folios(inode, offset, end);
 719                 if (unlikely(ret < 0))
 720                         return ret;
 721
 722                 truncated_last_page = ret;
 723
 724                 truncate_pagecache_range(&inode->v, offset, end - 1);
 725
 726                 block_start     = round_up(offset,      block_bytes(c));
 727                 block_end       = round_down(end,       block_bytes(c));
 728         }
 729
 730         ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
 731
 732         /*
 733          * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
 734          * so that the VFS cache i_size is consistent with the btree i_size:
 735          */
 736         if (ret &&
 737             !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
 738                 return ret;
 739
 740         if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
 741                 end = inode->v.i_size;
 742
 743         if (end >= inode->v.i_size &&
 744             (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
 745              !(mode & FALLOC_FL_KEEP_SIZE))) {
 746                 spin_lock(&inode->v.i_lock);
 747                 i_size_write(&inode->v, end);
 748                 spin_unlock(&inode->v.i_lock);
 749
 750                 mutex_lock(&inode->ei_update_lock);
 751                 ret2 = bch2_write_inode_size(c, inode, end, 0);
 752                 mutex_unlock(&inode->ei_update_lock);
 753         }
 754
 755         return ret ?: ret2;
 756 }
 757
 758 long bch2_fallocate_dispatch(struct file *file, int mode,
 759                              loff_t offset, loff_t len)
 760 {
 761         struct bch_inode_info *inode = file_bch_inode(file);
 762         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 763         long ret;
 764
 765         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
 766                 return -EROFS;
 767
 768         inode_lock(&inode->v);
 769         inode_dio_wait(&inode->v);
 770         bch2_pagecache_block_get(inode);
 771
 772         ret = file_modified(file);
 773         if (ret)
 774                 goto err;
 775
 776         if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
 777                 ret = bchfs_fallocate(inode, mode, offset, len);
 778         else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
 779                 ret = bchfs_fpunch(inode, offset, len);
 780         else if (mode == FALLOC_FL_INSERT_RANGE)
 781                 ret = bchfs_fcollapse_finsert(inode, offset, len, true);
 782         else if (mode == FALLOC_FL_COLLAPSE_RANGE)
 783                 ret = bchfs_fcollapse_finsert(inode, offset, len, false);
 784         else
 785                 ret = -EOPNOTSUPP;
 786 err:
 787         bch2_pagecache_block_put(inode);
 788         inode_unlock(&inode->v);
 789         bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
 790
 791         return bch2_err_class(ret);
 792 }
 793
 794 /*
 795  * Take a quota reservation for unallocated blocks in a given file range
 796  * Does not check pagecache
 797  */
 798 static int quota_reserve_range(struct bch_inode_info *inode,
 799                                struct quota_res *res,
 800                                u64 start, u64 end)
 801 {
 802         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 803         struct btree_trans *trans = bch2_trans_get(c);
 804         struct btree_iter iter;
 805         struct bkey_s_c k;
 806         u32 snapshot;
 807         u64 sectors = end - start;
 808         u64 pos = start;
 809         int ret;
 810 retry:
 811         bch2_trans_begin(trans);
 812
 813         ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot);
 814         if (ret)
 815                 goto err;
 816
 817         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 818                              SPOS(inode->v.i_ino, pos, snapshot), 0);
 819
 820         while (!(ret = btree_trans_too_many_iters(trans)) &&
 821                (k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k &&
 822                !(ret = bkey_err(k))) {
 823                 if (bkey_extent_is_allocation(k.k)) {
 824                         u64 s = min(end, k.k->p.offset) -
 825                                 max(start, bkey_start_offset(k.k));
 826                         BUG_ON(s > sectors);
 827                         sectors -= s;
 828                 }
 829                 bch2_btree_iter_advance(&iter);
 830         }
 831         pos = iter.pos.offset;
 832         bch2_trans_iter_exit(trans, &iter);
 833 err:
 834         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 835                 goto retry;
 836
 837         bch2_trans_put(trans);
 838
 839         return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true);
 840 }
 841
 842 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 843                              struct file *file_dst, loff_t pos_dst,
 844                              loff_t len, unsigned remap_flags)
 845 {
 846         struct bch_inode_info *src = file_bch_inode(file_src);
 847         struct bch_inode_info *dst = file_bch_inode(file_dst);
 848         struct bch_fs *c = src->v.i_sb->s_fs_info;
 849         struct quota_res quota_res = { 0 };
 850         s64 i_sectors_delta = 0;
 851         u64 aligned_len;
 852         loff_t ret = 0;
 853
 854         if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
 855                 return -EINVAL;
 856
 857         if (remap_flags & REMAP_FILE_DEDUP)
 858                 return -EOPNOTSUPP;
 859
 860         if ((pos_src & (block_bytes(c) - 1)) ||
 861             (pos_dst & (block_bytes(c) - 1)))
 862                 return -EINVAL;
 863
 864         if (src == dst &&
 865             abs(pos_src - pos_dst) < len)
 866                 return -EINVAL;
 867
 868         lock_two_nondirectories(&src->v, &dst->v);
 869         bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
 870
 871         inode_dio_wait(&src->v);
 872         inode_dio_wait(&dst->v);
 873
 874         ret = generic_remap_file_range_prep(file_src, pos_src,
 875                                             file_dst, pos_dst,
 876                                             &len, remap_flags);
 877         if (ret < 0 || len == 0)
 878                 goto err;
 879
 880         aligned_len = round_up((u64) len, block_bytes(c));
 881
 882         ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
 883                                 pos_dst, pos_dst + len - 1);
 884         if (ret)
 885                 goto err;
 886
 887         ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
 888                                   (pos_dst + aligned_len) >> 9);
 889         if (ret)
 890                 goto err;
 891
 892         file_update_time(file_dst);
 893
 894         bch2_mark_pagecache_unallocated(src, pos_src >> 9,
 895                                    (pos_src + aligned_len) >> 9);
 896
 897         ret = bch2_remap_range(c,
 898                                inode_inum(dst), pos_dst >> 9,
 899                                inode_inum(src), pos_src >> 9,
 900                                aligned_len >> 9,
 901                                pos_dst + len, &i_sectors_delta);
 902         if (ret < 0)
 903                 goto err;
 904
 905         /*
 906          * due to alignment, we might have remapped slightly more than requsted
 907          */
 908         ret = min((u64) ret << 9, (u64) len);
 909
 910         bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
 911
 912         spin_lock(&dst->v.i_lock);
 913         if (pos_dst + ret > dst->v.i_size)
 914                 i_size_write(&dst->v, pos_dst + ret);
 915         spin_unlock(&dst->v.i_lock);
 916
 917         if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
 918             IS_SYNC(file_inode(file_dst)))
 919                 ret = bch2_flush_inode(c, dst);
 920 err:
 921         bch2_quota_reservation_put(c, dst, &quota_res);
 922         bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
 923         unlock_two_nondirectories(&src->v, &dst->v);
 924
 925         return bch2_err_class(ret);
 926 }
 927
 928 /* fseek: */
 929
 930 static loff_t bch2_seek_data(struct file *file, u64 offset)
 931 {
 932         struct bch_inode_info *inode = file_bch_inode(file);
 933         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 934         struct btree_trans *trans;
 935         struct btree_iter iter;
 936         struct bkey_s_c k;
 937         subvol_inum inum = inode_inum(inode);
 938         u64 isize, next_data = MAX_LFS_FILESIZE;
 939         u32 snapshot;
 940         int ret;
 941
 942         isize = i_size_read(&inode->v);
 943         if (offset >= isize)
 944                 return -ENXIO;
 945
 946         trans = bch2_trans_get(c);
 947 retry:
 948         bch2_trans_begin(trans);
 949
 950         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 951         if (ret)
 952                 goto err;
 953
 954         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents,
 955                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
 956                            POS(inode->v.i_ino, U64_MAX),
 957                            0, k, ret) {
 958                 if (bkey_extent_is_data(k.k)) {
 959                         next_data = max(offset, bkey_start_offset(k.k) << 9);
 960                         break;
 961                 } else if (k.k->p.offset >> 9 > isize)
 962                         break;
 963         }
 964         bch2_trans_iter_exit(trans, &iter);
 965 err:
 966         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 967                 goto retry;
 968
 969         bch2_trans_put(trans);
 970         if (ret)
 971                 return ret;
 972
 973         if (next_data > offset)
 974                 next_data = bch2_seek_pagecache_data(&inode->v,
 975                                         offset, next_data, 0, false);
 976
 977         if (next_data >= isize)
 978                 return -ENXIO;
 979
 980         return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
 981 }
 982
 983 static loff_t bch2_seek_hole(struct file *file, u64 offset)
 984 {
 985         struct bch_inode_info *inode = file_bch_inode(file);
 986         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 987         struct btree_trans *trans;
 988         struct btree_iter iter;
 989         struct bkey_s_c k;
 990         subvol_inum inum = inode_inum(inode);
 991         u64 isize, next_hole = MAX_LFS_FILESIZE;
 992         u32 snapshot;
 993         int ret;
 994
 995         isize = i_size_read(&inode->v);
 996         if (offset >= isize)
 997                 return -ENXIO;
 998
 999         trans = bch2_trans_get(c);
1000 retry:
1001         bch2_trans_begin(trans);
1002
1003         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
1004         if (ret)
1005                 goto err;
1006
1007         for_each_btree_key_norestart(trans, iter, BTREE_ID_extents,
1008                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
1009                            BTREE_ITER_SLOTS, k, ret) {
1010                 if (k.k->p.inode != inode->v.i_ino) {
1011                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1012                                         offset, MAX_LFS_FILESIZE, 0, false);
1013                         break;
1014                 } else if (!bkey_extent_is_data(k.k)) {
1015                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1016                                         max(offset, bkey_start_offset(k.k) << 9),
1017                                         k.k->p.offset << 9, 0, false);
1018
1019                         if (next_hole < k.k->p.offset << 9)
1020                                 break;
1021                 } else {
1022                         offset = max(offset, bkey_start_offset(k.k) << 9);
1023                 }
1024         }
1025         bch2_trans_iter_exit(trans, &iter);
1026 err:
1027         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1028                 goto retry;
1029
1030         bch2_trans_put(trans);
1031         if (ret)
1032                 return ret;
1033
1034         if (next_hole > isize)
1035                 next_hole = isize;
1036
1037         return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
1038 }
1039
1040 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
1041 {
1042         loff_t ret;
1043
1044         switch (whence) {
1045         case SEEK_SET:
1046         case SEEK_CUR:
1047         case SEEK_END:
1048                 ret = generic_file_llseek(file, offset, whence);
1049                 break;
1050         case SEEK_DATA:
1051                 ret = bch2_seek_data(file, offset);
1052                 break;
1053         case SEEK_HOLE:
1054                 ret = bch2_seek_hole(file, offset);
1055                 break;
1056         default:
1057                 ret = -EINVAL;
1058                 break;
1059         }
1060
1061         return bch2_err_class(ret);
1062 }
1063
1064 void bch2_fs_fsio_exit(struct bch_fs *c)
1065 {
1066         bioset_exit(&c->nocow_flush_bioset);
1067 }
1068
1069 int bch2_fs_fsio_init(struct bch_fs *c)
1070 {
1071         if (bioset_init(&c->nocow_flush_bioset,
1072                         1, offsetof(struct nocow_flush, bio), 0))
1073                 return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
1074
1075         return 0;
1076 }
1077
1078 #endif /* NO_BCACHEFS_FS */