fs/bcachefs/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_WRITE|REQ_PREFLUSH,
  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         if (c->opts.journal_flush_disabled)
 178                 return 0;
 179
 180         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
 181                 return -EROFS;
 182
 183         struct bch_inode_unpacked u;
 184         int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
 185                   bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
 186                   bch2_inode_flush_nocow_writes(c, inode);
 187         bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
 188         return ret;
 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         ret = bch2_err_class(ret);
 206         if (ret == -EROFS)
 207                 ret = -EIO;
 208         return ret;
 209 }
 210
 211 /* truncate: */
 212
 213 static inline int range_has_data(struct bch_fs *c, u32 subvol,
 214                                  struct bpos start,
 215                                  struct bpos end)
 216 {
 217         struct btree_trans *trans = bch2_trans_get(c);
 218         struct btree_iter iter;
 219         struct bkey_s_c k;
 220         int ret = 0;
 221 retry:
 222         bch2_trans_begin(trans);
 223
 224         ret = bch2_subvolume_get_snapshot(trans, subvol, &start.snapshot);
 225         if (ret)
 226                 goto err;
 227
 228         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, start, end, 0, k, ret)
 229                 if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) {
 230                         ret = 1;
 231                         break;
 232                 }
 233         start = iter.pos;
 234         bch2_trans_iter_exit(trans, &iter);
 235 err:
 236         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 237                 goto retry;
 238
 239         bch2_trans_put(trans);
 240         return ret;
 241 }
 242
 243 static int __bch2_truncate_folio(struct bch_inode_info *inode,
 244                                  pgoff_t index, loff_t start, loff_t end)
 245 {
 246         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 247         struct address_space *mapping = inode->v.i_mapping;
 248         struct bch_folio *s;
 249         unsigned start_offset;
 250         unsigned end_offset;
 251         unsigned i;
 252         struct folio *folio;
 253         s64 i_sectors_delta = 0;
 254         int ret = 0;
 255         u64 end_pos;
 256
 257         folio = filemap_lock_folio(mapping, index);
 258         if (IS_ERR_OR_NULL(folio)) {
 259                 /*
 260                  * XXX: we're doing two index lookups when we end up reading the
 261                  * folio
 262                  */
 263                 ret = range_has_data(c, inode->ei_subvol,
 264                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
 265                                 POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
 266                 if (ret <= 0)
 267                         return ret;
 268
 269                 folio = __filemap_get_folio(mapping, index,
 270                                             FGP_LOCK|FGP_CREAT, GFP_KERNEL);
 271                 if (IS_ERR_OR_NULL(folio)) {
 272                         ret = -ENOMEM;
 273                         goto out;
 274                 }
 275         }
 276
 277         BUG_ON(start    >= folio_end_pos(folio));
 278         BUG_ON(end      <= folio_pos(folio));
 279
 280         start_offset    = max(start, folio_pos(folio)) - folio_pos(folio);
 281         end_offset      = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
 282
 283         /* Folio boundary? Nothing to do */
 284         if (start_offset == 0 &&
 285             end_offset == folio_size(folio)) {
 286                 ret = 0;
 287                 goto unlock;
 288         }
 289
 290         s = bch2_folio_create(folio, 0);
 291         if (!s) {
 292                 ret = -ENOMEM;
 293                 goto unlock;
 294         }
 295
 296         if (!folio_test_uptodate(folio)) {
 297                 ret = bch2_read_single_folio(folio, mapping);
 298                 if (ret)
 299                         goto unlock;
 300         }
 301
 302         ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
 303         if (ret)
 304                 goto unlock;
 305
 306         for (i = round_up(start_offset, block_bytes(c)) >> 9;
 307              i < round_down(end_offset, block_bytes(c)) >> 9;
 308              i++) {
 309                 s->s[i].nr_replicas     = 0;
 310
 311                 i_sectors_delta -= s->s[i].state == SECTOR_dirty;
 312                 bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
 313         }
 314
 315         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 316
 317         /*
 318          * Caller needs to know whether this folio will be written out by
 319          * writeback - doing an i_size update if necessary - or whether it will
 320          * be responsible for the i_size update.
 321          *
 322          * Note that we shouldn't ever see a folio beyond EOF, but check and
 323          * warn if so. This has been observed by failure to clean up folios
 324          * after a short write and there's still a chance reclaim will fix
 325          * things up.
 326          */
 327         WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
 328         end_pos = folio_end_pos(folio);
 329         if (inode->v.i_size > folio_pos(folio))
 330                 end_pos = min_t(u64, inode->v.i_size, end_pos);
 331         ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
 332
 333         folio_zero_segment(folio, start_offset, end_offset);
 334
 335         /*
 336          * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
 337          *
 338          * XXX: because we aren't currently tracking whether the folio has actual
 339          * data in it (vs. just 0s, or only partially written) this wrong. ick.
 340          */
 341         BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
 342
 343         /*
 344          * This removes any writeable userspace mappings; we need to force
 345          * .page_mkwrite to be called again before any mmapped writes, to
 346          * redirty the full page:
 347          */
 348         folio_mkclean(folio);
 349         filemap_dirty_folio(mapping, folio);
 350 unlock:
 351         folio_unlock(folio);
 352         folio_put(folio);
 353 out:
 354         return ret;
 355 }
 356
 357 static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
 358 {
 359         return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
 360                                      from, ANYSINT_MAX(loff_t));
 361 }
 362
 363 static int bch2_truncate_folios(struct bch_inode_info *inode,
 364                                 loff_t start, loff_t end)
 365 {
 366         int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
 367                                         start, end);
 368
 369         if (ret >= 0 &&
 370             start >> PAGE_SHIFT != end >> PAGE_SHIFT)
 371                 ret = __bch2_truncate_folio(inode,
 372                                         (end - 1) >> PAGE_SHIFT,
 373                                         start, end);
 374         return ret;
 375 }
 376
 377 static int bch2_extend(struct mnt_idmap *idmap,
 378                        struct bch_inode_info *inode,
 379                        struct bch_inode_unpacked *inode_u,
 380                        struct iattr *iattr)
 381 {
 382         struct address_space *mapping = inode->v.i_mapping;
 383         int ret;
 384
 385         /*
 386          * sync appends:
 387          *
 388          * this has to be done _before_ extending i_size:
 389          */
 390         ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
 391         if (ret)
 392                 return ret;
 393
 394         truncate_setsize(&inode->v, iattr->ia_size);
 395
 396         return bch2_setattr_nonsize(idmap, inode, iattr);
 397 }
 398
 399 int bchfs_truncate(struct mnt_idmap *idmap,
 400                   struct bch_inode_info *inode, struct iattr *iattr)
 401 {
 402         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 403         struct address_space *mapping = inode->v.i_mapping;
 404         struct bch_inode_unpacked inode_u;
 405         s64 i_sectors_delta = 0;
 406         int ret = 0;
 407
 408         /*
 409          * If the truncate call with change the size of the file, the
 410          * cmtimes should be updated. If the size will not change, we
 411          * do not need to update the cmtimes.
 412          */
 413         if (iattr->ia_size != inode->v.i_size) {
 414                 if (!(iattr->ia_valid & ATTR_MTIME))
 415                         ktime_get_coarse_real_ts64(&iattr->ia_mtime);
 416                 if (!(iattr->ia_valid & ATTR_CTIME))
 417                         ktime_get_coarse_real_ts64(&iattr->ia_ctime);
 418                 iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
 419         }
 420
 421         inode_dio_wait(&inode->v);
 422         bch2_pagecache_block_get(inode);
 423
 424         ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
 425         if (ret)
 426                 goto err;
 427
 428         /*
 429          * check this before next assertion; on filesystem error our normal
 430          * invariants are a bit broken (truncate has to truncate the page cache
 431          * before the inode).
 432          */
 433         ret = bch2_journal_error(&c->journal);
 434         if (ret)
 435                 goto err;
 436
 437         WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
 438                   inode->v.i_size < inode_u.bi_size,
 439                   "truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
 440                   (u64) inode->v.i_size, inode_u.bi_size);
 441
 442         if (iattr->ia_size > inode->v.i_size) {
 443                 ret = bch2_extend(idmap, inode, &inode_u, iattr);
 444                 goto err;
 445         }
 446
 447         iattr->ia_valid &= ~ATTR_SIZE;
 448
 449         ret = bch2_truncate_folio(inode, iattr->ia_size);
 450         if (unlikely(ret < 0))
 451                 goto err;
 452
 453         truncate_setsize(&inode->v, iattr->ia_size);
 454
 455         /*
 456          * When extending, we're going to write the new i_size to disk
 457          * immediately so we need to flush anything above the current on disk
 458          * i_size first:
 459          *
 460          * Also, when extending we need to flush the page that i_size currently
 461          * straddles - if it's mapped to userspace, we need to ensure that
 462          * userspace has to redirty it and call .mkwrite -> set_page_dirty
 463          * again to allocate the part of the page that was extended.
 464          */
 465         if (iattr->ia_size > inode_u.bi_size)
 466                 ret = filemap_write_and_wait_range(mapping,
 467                                 inode_u.bi_size,
 468                                 iattr->ia_size - 1);
 469         else if (iattr->ia_size & (PAGE_SIZE - 1))
 470                 ret = filemap_write_and_wait_range(mapping,
 471                                 round_down(iattr->ia_size, PAGE_SIZE),
 472                                 iattr->ia_size - 1);
 473         if (ret)
 474                 goto err;
 475
 476         ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta);
 477         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 478
 479         if (unlikely(ret)) {
 480                 /*
 481                  * If we error here, VFS caches are now inconsistent with btree
 482                  */
 483                 set_bit(EI_INODE_ERROR, &inode->ei_flags);
 484                 goto err;
 485         }
 486
 487         bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
 488                                 !bch2_journal_error(&c->journal), c,
 489                                 "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
 490                                 inode->v.i_ino, (u64) inode->v.i_blocks,
 491                                 inode->ei_inode.bi_sectors);
 492
 493         ret = bch2_setattr_nonsize(idmap, inode, iattr);
 494 err:
 495         bch2_pagecache_block_put(inode);
 496         return bch2_err_class(ret);
 497 }
 498
 499 /* fallocate: */
 500
 501 static int inode_update_times_fn(struct btree_trans *trans,
 502                                  struct bch_inode_info *inode,
 503                                  struct bch_inode_unpacked *bi, void *p)
 504 {
 505         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 506
 507         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 508         return 0;
 509 }
 510
 511 static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
 512 {
 513         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 514         u64 end         = offset + len;
 515         u64 block_start = round_up(offset, block_bytes(c));
 516         u64 block_end   = round_down(end, block_bytes(c));
 517         bool truncated_last_page;
 518         int ret = 0;
 519
 520         ret = bch2_truncate_folios(inode, offset, end);
 521         if (unlikely(ret < 0))
 522                 goto err;
 523
 524         truncated_last_page = ret;
 525
 526         truncate_pagecache_range(&inode->v, offset, end - 1);
 527
 528         if (block_start < block_end) {
 529                 s64 i_sectors_delta = 0;
 530
 531                 ret = bch2_fpunch(c, inode_inum(inode),
 532                                   block_start >> 9, block_end >> 9,
 533                                   &i_sectors_delta);
 534                 bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 535         }
 536
 537         mutex_lock(&inode->ei_update_lock);
 538         if (end >= inode->v.i_size && !truncated_last_page) {
 539                 ret = bch2_write_inode_size(c, inode, inode->v.i_size,
 540                                             ATTR_MTIME|ATTR_CTIME);
 541         } else {
 542                 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
 543                                        ATTR_MTIME|ATTR_CTIME);
 544         }
 545         mutex_unlock(&inode->ei_update_lock);
 546 err:
 547         return ret;
 548 }
 549
 550 static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 551                                    loff_t offset, loff_t len,
 552                                    bool insert)
 553 {
 554         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 555         struct address_space *mapping = inode->v.i_mapping;
 556         s64 i_sectors_delta = 0;
 557         int ret = 0;
 558
 559         if ((offset | len) & (block_bytes(c) - 1))
 560                 return -EINVAL;
 561
 562         if (insert) {
 563                 if (offset >= inode->v.i_size)
 564                         return -EINVAL;
 565         } else {
 566                 if (offset + len >= inode->v.i_size)
 567                         return -EINVAL;
 568         }
 569
 570         ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
 571         if (ret)
 572                 return ret;
 573
 574         if (insert)
 575                 i_size_write(&inode->v, inode->v.i_size + len);
 576
 577         ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9,
 578                                      insert, &i_sectors_delta);
 579         if (!ret && !insert)
 580                 i_size_write(&inode->v, inode->v.i_size - len);
 581         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 582
 583         return ret;
 584 }
 585
 586 static int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 587                              u64 start_sector, u64 end_sector)
 588 {
 589         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 590         struct btree_trans *trans = bch2_trans_get(c);
 591         struct btree_iter iter;
 592         struct bpos end_pos = POS(inode->v.i_ino, end_sector);
 593         struct bch_io_opts opts;
 594         int ret = 0;
 595
 596         bch2_inode_opts_get(&opts, c, &inode->ei_inode);
 597
 598         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 599                         POS(inode->v.i_ino, start_sector),
 600                         BTREE_ITER_slots|BTREE_ITER_intent);
 601
 602         while (!ret && bkey_lt(iter.pos, end_pos)) {
 603                 s64 i_sectors_delta = 0;
 604                 struct quota_res quota_res = { 0 };
 605                 struct bkey_s_c k;
 606                 unsigned sectors;
 607                 bool is_allocation;
 608                 u64 hole_start, hole_end;
 609                 u32 snapshot;
 610
 611                 bch2_trans_begin(trans);
 612
 613                 ret = bch2_subvolume_get_snapshot(trans,
 614                                         inode->ei_subvol, &snapshot);
 615                 if (ret)
 616                         goto bkey_err;
 617
 618                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 619
 620                 k = bch2_btree_iter_peek_slot(&iter);
 621                 if ((ret = bkey_err(k)))
 622                         goto bkey_err;
 623
 624                 hole_start      = iter.pos.offset;
 625                 hole_end        = bpos_min(k.k->p, end_pos).offset;
 626                 is_allocation   = bkey_extent_is_allocation(k.k);
 627
 628                 /* already reserved */
 629                 if (bkey_extent_is_reservation(k) &&
 630                     bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
 631                         bch2_btree_iter_advance(&iter);
 632                         continue;
 633                 }
 634
 635                 if (bkey_extent_is_data(k.k) &&
 636                     !(mode & FALLOC_FL_ZERO_RANGE)) {
 637                         bch2_btree_iter_advance(&iter);
 638                         continue;
 639                 }
 640
 641                 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
 642                         /*
 643                          * Lock ordering - can't be holding btree locks while
 644                          * blocking on a folio lock:
 645                          */
 646                         if (bch2_clamp_data_hole(&inode->v,
 647                                                  &hole_start,
 648                                                  &hole_end,
 649                                                  opts.data_replicas, true))
 650                                 ret = drop_locks_do(trans,
 651                                         (bch2_clamp_data_hole(&inode->v,
 652                                                               &hole_start,
 653                                                               &hole_end,
 654                                                               opts.data_replicas, false), 0));
 655                         bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
 656
 657                         if (ret)
 658                                 goto bkey_err;
 659
 660                         if (hole_start == hole_end)
 661                                 continue;
 662                 }
 663
 664                 sectors = hole_end - hole_start;
 665
 666                 if (!is_allocation) {
 667                         ret = bch2_quota_reservation_add(c, inode,
 668                                         &quota_res, sectors, true);
 669                         if (unlikely(ret))
 670                                 goto bkey_err;
 671                 }
 672
 673                 ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter,
 674                                             sectors, opts, &i_sectors_delta,
 675                                             writepoint_hashed((unsigned long) current));
 676                 if (ret)
 677                         goto bkey_err;
 678
 679                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 680
 681                 if (bch2_mark_pagecache_reserved(inode, &hole_start,
 682                                                  iter.pos.offset, true))
 683                         drop_locks_do(trans,
 684                                 bch2_mark_pagecache_reserved(inode, &hole_start,
 685                                                              iter.pos.offset, false));
 686 bkey_err:
 687                 bch2_quota_reservation_put(c, inode, &quota_res);
 688                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 689                         ret = 0;
 690         }
 691
 692         if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
 693                 struct quota_res quota_res = { 0 };
 694                 s64 i_sectors_delta = 0;
 695
 696                 bch2_fpunch_at(trans, &iter, inode_inum(inode),
 697                                end_sector, &i_sectors_delta);
 698                 bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 699                 bch2_quota_reservation_put(c, inode, &quota_res);
 700         }
 701
 702         bch2_trans_iter_exit(trans, &iter);
 703         bch2_trans_put(trans);
 704         return ret;
 705 }
 706
 707 static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 708                             loff_t offset, loff_t len)
 709 {
 710         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 711         u64 end         = offset + len;
 712         u64 block_start = round_down(offset,    block_bytes(c));
 713         u64 block_end   = round_up(end,         block_bytes(c));
 714         bool truncated_last_page = false;
 715         int ret, ret2 = 0;
 716
 717         if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
 718                 ret = inode_newsize_ok(&inode->v, end);
 719                 if (ret)
 720                         return ret;
 721         }
 722
 723         if (mode & FALLOC_FL_ZERO_RANGE) {
 724                 ret = bch2_truncate_folios(inode, offset, end);
 725                 if (unlikely(ret < 0))
 726                         return ret;
 727
 728                 truncated_last_page = ret;
 729
 730                 truncate_pagecache_range(&inode->v, offset, end - 1);
 731
 732                 block_start     = round_up(offset,      block_bytes(c));
 733                 block_end       = round_down(end,       block_bytes(c));
 734         }
 735
 736         ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
 737
 738         /*
 739          * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
 740          * so that the VFS cache i_size is consistent with the btree i_size:
 741          */
 742         if (ret &&
 743             !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
 744                 return ret;
 745
 746         if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
 747                 end = inode->v.i_size;
 748
 749         if (end >= inode->v.i_size &&
 750             (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
 751              !(mode & FALLOC_FL_KEEP_SIZE))) {
 752                 spin_lock(&inode->v.i_lock);
 753                 i_size_write(&inode->v, end);
 754                 spin_unlock(&inode->v.i_lock);
 755
 756                 mutex_lock(&inode->ei_update_lock);
 757                 ret2 = bch2_write_inode_size(c, inode, end, 0);
 758                 mutex_unlock(&inode->ei_update_lock);
 759         }
 760
 761         return ret ?: ret2;
 762 }
 763
 764 long bch2_fallocate_dispatch(struct file *file, int mode,
 765                              loff_t offset, loff_t len)
 766 {
 767         struct bch_inode_info *inode = file_bch_inode(file);
 768         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 769         long ret;
 770
 771         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
 772                 return -EROFS;
 773
 774         inode_lock(&inode->v);
 775         inode_dio_wait(&inode->v);
 776         bch2_pagecache_block_get(inode);
 777
 778         ret = file_modified(file);
 779         if (ret)
 780                 goto err;
 781
 782         if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
 783                 ret = bchfs_fallocate(inode, mode, offset, len);
 784         else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
 785                 ret = bchfs_fpunch(inode, offset, len);
 786         else if (mode == FALLOC_FL_INSERT_RANGE)
 787                 ret = bchfs_fcollapse_finsert(inode, offset, len, true);
 788         else if (mode == FALLOC_FL_COLLAPSE_RANGE)
 789                 ret = bchfs_fcollapse_finsert(inode, offset, len, false);
 790         else
 791                 ret = -EOPNOTSUPP;
 792 err:
 793         bch2_pagecache_block_put(inode);
 794         inode_unlock(&inode->v);
 795         bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
 796
 797         return bch2_err_class(ret);
 798 }
 799
 800 /*
 801  * Take a quota reservation for unallocated blocks in a given file range
 802  * Does not check pagecache
 803  */
 804 static int quota_reserve_range(struct bch_inode_info *inode,
 805                                struct quota_res *res,
 806                                u64 start, u64 end)
 807 {
 808         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 809         struct btree_trans *trans = bch2_trans_get(c);
 810         struct btree_iter iter;
 811         struct bkey_s_c k;
 812         u32 snapshot;
 813         u64 sectors = end - start;
 814         u64 pos = start;
 815         int ret;
 816 retry:
 817         bch2_trans_begin(trans);
 818
 819         ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot);
 820         if (ret)
 821                 goto err;
 822
 823         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 824                              SPOS(inode->v.i_ino, pos, snapshot), 0);
 825
 826         while (!(ret = btree_trans_too_many_iters(trans)) &&
 827                (k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k &&
 828                !(ret = bkey_err(k))) {
 829                 if (bkey_extent_is_allocation(k.k)) {
 830                         u64 s = min(end, k.k->p.offset) -
 831                                 max(start, bkey_start_offset(k.k));
 832                         BUG_ON(s > sectors);
 833                         sectors -= s;
 834                 }
 835                 bch2_btree_iter_advance(&iter);
 836         }
 837         pos = iter.pos.offset;
 838         bch2_trans_iter_exit(trans, &iter);
 839 err:
 840         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 841                 goto retry;
 842
 843         bch2_trans_put(trans);
 844
 845         return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true);
 846 }
 847
 848 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 849                              struct file *file_dst, loff_t pos_dst,
 850                              loff_t len, unsigned remap_flags)
 851 {
 852         struct bch_inode_info *src = file_bch_inode(file_src);
 853         struct bch_inode_info *dst = file_bch_inode(file_dst);
 854         struct bch_fs *c = src->v.i_sb->s_fs_info;
 855         struct quota_res quota_res = { 0 };
 856         s64 i_sectors_delta = 0;
 857         u64 aligned_len;
 858         loff_t ret = 0;
 859
 860         if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
 861                 return -EINVAL;
 862
 863         if (remap_flags & REMAP_FILE_DEDUP)
 864                 return -EOPNOTSUPP;
 865
 866         if ((pos_src & (block_bytes(c) - 1)) ||
 867             (pos_dst & (block_bytes(c) - 1)))
 868                 return -EINVAL;
 869
 870         if (src == dst &&
 871             abs(pos_src - pos_dst) < len)
 872                 return -EINVAL;
 873
 874         lock_two_nondirectories(&src->v, &dst->v);
 875         bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
 876
 877         inode_dio_wait(&src->v);
 878         inode_dio_wait(&dst->v);
 879
 880         ret = generic_remap_file_range_prep(file_src, pos_src,
 881                                             file_dst, pos_dst,
 882                                             &len, remap_flags);
 883         if (ret < 0 || len == 0)
 884                 goto err;
 885
 886         aligned_len = round_up((u64) len, block_bytes(c));
 887
 888         ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
 889                                 pos_dst, pos_dst + len - 1);
 890         if (ret)
 891                 goto err;
 892
 893         ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
 894                                   (pos_dst + aligned_len) >> 9);
 895         if (ret)
 896                 goto err;
 897
 898         file_update_time(file_dst);
 899
 900         bch2_mark_pagecache_unallocated(src, pos_src >> 9,
 901                                    (pos_src + aligned_len) >> 9);
 902
 903         ret = bch2_remap_range(c,
 904                                inode_inum(dst), pos_dst >> 9,
 905                                inode_inum(src), pos_src >> 9,
 906                                aligned_len >> 9,
 907                                pos_dst + len, &i_sectors_delta);
 908         if (ret < 0)
 909                 goto err;
 910
 911         /*
 912          * due to alignment, we might have remapped slightly more than requsted
 913          */
 914         ret = min((u64) ret << 9, (u64) len);
 915
 916         bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
 917
 918         spin_lock(&dst->v.i_lock);
 919         if (pos_dst + ret > dst->v.i_size)
 920                 i_size_write(&dst->v, pos_dst + ret);
 921         spin_unlock(&dst->v.i_lock);
 922
 923         if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
 924             IS_SYNC(file_inode(file_dst)))
 925                 ret = bch2_flush_inode(c, dst);
 926 err:
 927         bch2_quota_reservation_put(c, dst, &quota_res);
 928         bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
 929         unlock_two_nondirectories(&src->v, &dst->v);
 930
 931         return bch2_err_class(ret);
 932 }
 933
 934 /* fseek: */
 935
 936 static loff_t bch2_seek_data(struct file *file, u64 offset)
 937 {
 938         struct bch_inode_info *inode = file_bch_inode(file);
 939         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 940         struct btree_trans *trans;
 941         struct btree_iter iter;
 942         struct bkey_s_c k;
 943         subvol_inum inum = inode_inum(inode);
 944         u64 isize, next_data = MAX_LFS_FILESIZE;
 945         u32 snapshot;
 946         int ret;
 947
 948         isize = i_size_read(&inode->v);
 949         if (offset >= isize)
 950                 return -ENXIO;
 951
 952         trans = bch2_trans_get(c);
 953 retry:
 954         bch2_trans_begin(trans);
 955
 956         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 957         if (ret)
 958                 goto err;
 959
 960         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents,
 961                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
 962                            POS(inode->v.i_ino, U64_MAX),
 963                            0, k, ret) {
 964                 if (bkey_extent_is_data(k.k)) {
 965                         next_data = max(offset, bkey_start_offset(k.k) << 9);
 966                         break;
 967                 } else if (k.k->p.offset >> 9 > isize)
 968                         break;
 969         }
 970         bch2_trans_iter_exit(trans, &iter);
 971 err:
 972         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 973                 goto retry;
 974
 975         bch2_trans_put(trans);
 976         if (ret)
 977                 return ret;
 978
 979         if (next_data > offset)
 980                 next_data = bch2_seek_pagecache_data(&inode->v,
 981                                         offset, next_data, 0, false);
 982
 983         if (next_data >= isize)
 984                 return -ENXIO;
 985
 986         return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
 987 }
 988
 989 static loff_t bch2_seek_hole(struct file *file, u64 offset)
 990 {
 991         struct bch_inode_info *inode = file_bch_inode(file);
 992         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 993         struct btree_trans *trans;
 994         struct btree_iter iter;
 995         struct bkey_s_c k;
 996         subvol_inum inum = inode_inum(inode);
 997         u64 isize, next_hole = MAX_LFS_FILESIZE;
 998         u32 snapshot;
 999         int ret;
1000
1001         isize = i_size_read(&inode->v);
1002         if (offset >= isize)
1003                 return -ENXIO;
1004
1005         trans = bch2_trans_get(c);
1006 retry:
1007         bch2_trans_begin(trans);
1008
1009         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
1010         if (ret)
1011                 goto err;
1012
1013         for_each_btree_key_norestart(trans, iter, BTREE_ID_extents,
1014                            SPOS(inode->v.i_ino, offset >> 9, snapshot),
1015                            BTREE_ITER_slots, k, ret) {
1016                 if (k.k->p.inode != inode->v.i_ino) {
1017                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1018                                         offset, MAX_LFS_FILESIZE, 0, false);
1019                         break;
1020                 } else if (!bkey_extent_is_data(k.k)) {
1021                         next_hole = bch2_seek_pagecache_hole(&inode->v,
1022                                         max(offset, bkey_start_offset(k.k) << 9),
1023                                         k.k->p.offset << 9, 0, false);
1024
1025                         if (next_hole < k.k->p.offset << 9)
1026                                 break;
1027                 } else {
1028                         offset = max(offset, bkey_start_offset(k.k) << 9);
1029                 }
1030         }
1031         bch2_trans_iter_exit(trans, &iter);
1032 err:
1033         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1034                 goto retry;
1035
1036         bch2_trans_put(trans);
1037         if (ret)
1038                 return ret;
1039
1040         if (next_hole > isize)
1041                 next_hole = isize;
1042
1043         return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
1044 }
1045
1046 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
1047 {
1048         loff_t ret;
1049
1050         switch (whence) {
1051         case SEEK_SET:
1052         case SEEK_CUR:
1053         case SEEK_END:
1054                 ret = generic_file_llseek(file, offset, whence);
1055                 break;
1056         case SEEK_DATA:
1057                 ret = bch2_seek_data(file, offset);
1058                 break;
1059         case SEEK_HOLE:
1060                 ret = bch2_seek_hole(file, offset);
1061                 break;
1062         default:
1063                 ret = -EINVAL;
1064                 break;
1065         }
1066
1067         return bch2_err_class(ret);
1068 }
1069
1070 void bch2_fs_fsio_exit(struct bch_fs *c)
1071 {
1072         bioset_exit(&c->nocow_flush_bioset);
1073 }
1074
1075 int bch2_fs_fsio_init(struct bch_fs *c)
1076 {
1077         if (bioset_init(&c->nocow_flush_bioset,
1078                         1, offsetof(struct nocow_flush, bio), 0))
1079                 return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
1080
1081         return 0;
1082 }
1083
1084 #endif /* NO_BCACHEFS_FS */