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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "bkey_buf.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "chardev.h"
  10 #include "dirent.h"
  11 #include "errcode.h"
  12 #include "extents.h"
  13 #include "fs.h"
  14 #include "fs-common.h"
  15 #include "fs-io.h"
  16 #include "fs-ioctl.h"
  17 #include "fs-io-buffered.h"
  18 #include "fs-io-direct.h"
  19 #include "fs-io-pagecache.h"
  20 #include "fsck.h"
  21 #include "inode.h"
  22 #include "io_read.h"
  23 #include "journal.h"
  24 #include "keylist.h"
  25 #include "quota.h"
  26 #include "snapshot.h"
  27 #include "super.h"
  28 #include "xattr.h"
  29
  30 #include <linux/aio.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/fiemap.h>
  34 #include <linux/module.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/posix_acl.h>
  37 #include <linux/random.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/statfs.h>
  40 #include <linux/string.h>
  41 #include <linux/xattr.h>
  42
  43 static struct kmem_cache *bch2_inode_cache;
  44
  45 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
  46                                 struct bch_inode_info *,
  47                                 struct bch_inode_unpacked *,
  48                                 struct bch_subvolume *);
  49
  50 void bch2_inode_update_after_write(struct btree_trans *trans,
  51                                    struct bch_inode_info *inode,
  52                                    struct bch_inode_unpacked *bi,
  53                                    unsigned fields)
  54 {
  55         struct bch_fs *c = trans->c;
  56
  57         BUG_ON(bi->bi_inum != inode->v.i_ino);
  58
  59         bch2_assert_pos_locked(trans, BTREE_ID_inodes,
  60                                POS(0, bi->bi_inum),
  61                                c->opts.inodes_use_key_cache);
  62
  63         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
  64         i_uid_write(&inode->v, bi->bi_uid);
  65         i_gid_write(&inode->v, bi->bi_gid);
  66         inode->v.i_mode = bi->bi_mode;
  67
  68         if (fields & ATTR_ATIME)
  69                 inode_set_atime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_atime));
  70         if (fields & ATTR_MTIME)
  71                 inode_set_mtime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_mtime));
  72         if (fields & ATTR_CTIME)
  73                 inode_set_ctime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_ctime));
  74
  75         inode->ei_inode         = *bi;
  76
  77         bch2_inode_flags_to_vfs(inode);
  78 }
  79
  80 int __must_check bch2_write_inode(struct bch_fs *c,
  81                                   struct bch_inode_info *inode,
  82                                   inode_set_fn set,
  83                                   void *p, unsigned fields)
  84 {
  85         struct btree_trans *trans = bch2_trans_get(c);
  86         struct btree_iter iter = { NULL };
  87         struct bch_inode_unpacked inode_u;
  88         int ret;
  89 retry:
  90         bch2_trans_begin(trans);
  91
  92         ret   = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
  93                                 BTREE_ITER_INTENT) ?:
  94                 (set ? set(trans, inode, &inode_u, p) : 0) ?:
  95                 bch2_inode_write(trans, &iter, &inode_u) ?:
  96                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
  97
  98         /*
  99          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 100          * this is important for inode updates via bchfs_write_index_update
 101          */
 102         if (!ret)
 103                 bch2_inode_update_after_write(trans, inode, &inode_u, fields);
 104
 105         bch2_trans_iter_exit(trans, &iter);
 106
 107         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 108                 goto retry;
 109
 110         bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
 111                              "inode %u:%llu not found when updating",
 112                              inode_inum(inode).subvol,
 113                              inode_inum(inode).inum);
 114
 115         bch2_trans_put(trans);
 116         return ret < 0 ? ret : 0;
 117 }
 118
 119 int bch2_fs_quota_transfer(struct bch_fs *c,
 120                            struct bch_inode_info *inode,
 121                            struct bch_qid new_qid,
 122                            unsigned qtypes,
 123                            enum quota_acct_mode mode)
 124 {
 125         unsigned i;
 126         int ret;
 127
 128         qtypes &= enabled_qtypes(c);
 129
 130         for (i = 0; i < QTYP_NR; i++)
 131                 if (new_qid.q[i] == inode->ei_qid.q[i])
 132                         qtypes &= ~(1U << i);
 133
 134         if (!qtypes)
 135                 return 0;
 136
 137         mutex_lock(&inode->ei_quota_lock);
 138
 139         ret = bch2_quota_transfer(c, qtypes, new_qid,
 140                                   inode->ei_qid,
 141                                   inode->v.i_blocks +
 142                                   inode->ei_quota_reserved,
 143                                   mode);
 144         if (!ret)
 145                 for (i = 0; i < QTYP_NR; i++)
 146                         if (qtypes & (1 << i))
 147                                 inode->ei_qid.q[i] = new_qid.q[i];
 148
 149         mutex_unlock(&inode->ei_quota_lock);
 150
 151         return ret;
 152 }
 153
 154 static int bch2_iget5_test(struct inode *vinode, void *p)
 155 {
 156         struct bch_inode_info *inode = to_bch_ei(vinode);
 157         subvol_inum *inum = p;
 158
 159         return inode->ei_subvol == inum->subvol &&
 160                 inode->ei_inode.bi_inum == inum->inum;
 161 }
 162
 163 static int bch2_iget5_set(struct inode *vinode, void *p)
 164 {
 165         struct bch_inode_info *inode = to_bch_ei(vinode);
 166         subvol_inum *inum = p;
 167
 168         inode->v.i_ino          = inum->inum;
 169         inode->ei_subvol        = inum->subvol;
 170         inode->ei_inode.bi_inum = inum->inum;
 171         return 0;
 172 }
 173
 174 static unsigned bch2_inode_hash(subvol_inum inum)
 175 {
 176         return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
 177 }
 178
 179 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 180 {
 181         struct bch_inode_unpacked inode_u;
 182         struct bch_inode_info *inode;
 183         struct btree_trans *trans;
 184         struct bch_subvolume subvol;
 185         int ret;
 186
 187         inode = to_bch_ei(iget5_locked(c->vfs_sb,
 188                                        bch2_inode_hash(inum),
 189                                        bch2_iget5_test,
 190                                        bch2_iget5_set,
 191                                        &inum));
 192         if (unlikely(!inode))
 193                 return ERR_PTR(-ENOMEM);
 194         if (!(inode->v.i_state & I_NEW))
 195                 return &inode->v;
 196
 197         trans = bch2_trans_get(c);
 198         ret = lockrestart_do(trans,
 199                 bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
 200                 bch2_inode_find_by_inum_trans(trans, inum, &inode_u));
 201
 202         if (!ret)
 203                 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 204         bch2_trans_put(trans);
 205
 206         if (ret) {
 207                 iget_failed(&inode->v);
 208                 return ERR_PTR(bch2_err_class(ret));
 209         }
 210
 211         mutex_lock(&c->vfs_inodes_lock);
 212         list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
 213         mutex_unlock(&c->vfs_inodes_lock);
 214
 215         unlock_new_inode(&inode->v);
 216
 217         return &inode->v;
 218 }
 219
 220 struct bch_inode_info *
 221 __bch2_create(struct mnt_idmap *idmap,
 222               struct bch_inode_info *dir, struct dentry *dentry,
 223               umode_t mode, dev_t rdev, subvol_inum snapshot_src,
 224               unsigned flags)
 225 {
 226         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 227         struct btree_trans *trans;
 228         struct bch_inode_unpacked dir_u;
 229         struct bch_inode_info *inode, *old;
 230         struct bch_inode_unpacked inode_u;
 231         struct posix_acl *default_acl = NULL, *acl = NULL;
 232         subvol_inum inum;
 233         struct bch_subvolume subvol;
 234         u64 journal_seq = 0;
 235         int ret;
 236
 237         /*
 238          * preallocate acls + vfs inode before btree transaction, so that
 239          * nothing can fail after the transaction succeeds:
 240          */
 241 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 242         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 243         if (ret)
 244                 return ERR_PTR(ret);
 245 #endif
 246         inode = to_bch_ei(new_inode(c->vfs_sb));
 247         if (unlikely(!inode)) {
 248                 inode = ERR_PTR(-ENOMEM);
 249                 goto err;
 250         }
 251
 252         bch2_inode_init_early(c, &inode_u);
 253
 254         if (!(flags & BCH_CREATE_TMPFILE))
 255                 mutex_lock(&dir->ei_update_lock);
 256
 257         trans = bch2_trans_get(c);
 258 retry:
 259         bch2_trans_begin(trans);
 260
 261         ret   = bch2_create_trans(trans,
 262                                   inode_inum(dir), &dir_u, &inode_u,
 263                                   !(flags & BCH_CREATE_TMPFILE)
 264                                   ? &dentry->d_name : NULL,
 265                                   from_kuid(i_user_ns(&dir->v), current_fsuid()),
 266                                   from_kgid(i_user_ns(&dir->v), current_fsgid()),
 267                                   mode, rdev,
 268                                   default_acl, acl, snapshot_src, flags) ?:
 269                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 270                                 KEY_TYPE_QUOTA_PREALLOC);
 271         if (unlikely(ret))
 272                 goto err_before_quota;
 273
 274         inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
 275         inum.inum = inode_u.bi_inum;
 276
 277         ret   = bch2_subvolume_get(trans, inum.subvol, true,
 278                                    BTREE_ITER_WITH_UPDATES, &subvol) ?:
 279                 bch2_trans_commit(trans, NULL, &journal_seq, 0);
 280         if (unlikely(ret)) {
 281                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 282                                 KEY_TYPE_QUOTA_WARN);
 283 err_before_quota:
 284                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 285                         goto retry;
 286                 goto err_trans;
 287         }
 288
 289         if (!(flags & BCH_CREATE_TMPFILE)) {
 290                 bch2_inode_update_after_write(trans, dir, &dir_u,
 291                                               ATTR_MTIME|ATTR_CTIME);
 292                 mutex_unlock(&dir->ei_update_lock);
 293         }
 294
 295         bch2_iget5_set(&inode->v, &inum);
 296         bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 297
 298         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 299         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 300
 301         /*
 302          * we must insert the new inode into the inode cache before calling
 303          * bch2_trans_exit() and dropping locks, else we could race with another
 304          * thread pulling the inode in and modifying it:
 305          */
 306
 307         inode->v.i_state |= I_CREATING;
 308
 309         old = to_bch_ei(inode_insert5(&inode->v,
 310                                       bch2_inode_hash(inum),
 311                                       bch2_iget5_test,
 312                                       bch2_iget5_set,
 313                                       &inum));
 314         BUG_ON(!old);
 315
 316         if (unlikely(old != inode)) {
 317                 /*
 318                  * We raced, another process pulled the new inode into cache
 319                  * before us:
 320                  */
 321                 make_bad_inode(&inode->v);
 322                 iput(&inode->v);
 323
 324                 inode = old;
 325         } else {
 326                 mutex_lock(&c->vfs_inodes_lock);
 327                 list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
 328                 mutex_unlock(&c->vfs_inodes_lock);
 329                 /*
 330                  * we really don't want insert_inode_locked2() to be setting
 331                  * I_NEW...
 332                  */
 333                 unlock_new_inode(&inode->v);
 334         }
 335
 336         bch2_trans_put(trans);
 337 err:
 338         posix_acl_release(default_acl);
 339         posix_acl_release(acl);
 340         return inode;
 341 err_trans:
 342         if (!(flags & BCH_CREATE_TMPFILE))
 343                 mutex_unlock(&dir->ei_update_lock);
 344
 345         bch2_trans_put(trans);
 346         make_bad_inode(&inode->v);
 347         iput(&inode->v);
 348         inode = ERR_PTR(ret);
 349         goto err;
 350 }
 351
 352 /* methods */
 353
 354 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 355                                   unsigned int flags)
 356 {
 357         struct bch_fs *c = vdir->i_sb->s_fs_info;
 358         struct bch_inode_info *dir = to_bch_ei(vdir);
 359         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 360         struct inode *vinode = NULL;
 361         subvol_inum inum = { .subvol = 1 };
 362         int ret;
 363
 364         ret = bch2_dirent_lookup(c, inode_inum(dir), &hash,
 365                                  &dentry->d_name, &inum);
 366
 367         if (!ret)
 368                 vinode = bch2_vfs_inode_get(c, inum);
 369
 370         return d_splice_alias(vinode, dentry);
 371 }
 372
 373 static int bch2_mknod(struct mnt_idmap *idmap,
 374                       struct inode *vdir, struct dentry *dentry,
 375                       umode_t mode, dev_t rdev)
 376 {
 377         struct bch_inode_info *inode =
 378                 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
 379                               (subvol_inum) { 0 }, 0);
 380
 381         if (IS_ERR(inode))
 382                 return bch2_err_class(PTR_ERR(inode));
 383
 384         d_instantiate(dentry, &inode->v);
 385         return 0;
 386 }
 387
 388 static int bch2_create(struct mnt_idmap *idmap,
 389                        struct inode *vdir, struct dentry *dentry,
 390                        umode_t mode, bool excl)
 391 {
 392         return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
 393 }
 394
 395 static int __bch2_link(struct bch_fs *c,
 396                        struct bch_inode_info *inode,
 397                        struct bch_inode_info *dir,
 398                        struct dentry *dentry)
 399 {
 400         struct btree_trans *trans = bch2_trans_get(c);
 401         struct bch_inode_unpacked dir_u, inode_u;
 402         int ret;
 403
 404         mutex_lock(&inode->ei_update_lock);
 405
 406         ret = commit_do(trans, NULL, NULL, 0,
 407                         bch2_link_trans(trans,
 408                                         inode_inum(dir),   &dir_u,
 409                                         inode_inum(inode), &inode_u,
 410                                         &dentry->d_name));
 411
 412         if (likely(!ret)) {
 413                 bch2_inode_update_after_write(trans, dir, &dir_u,
 414                                               ATTR_MTIME|ATTR_CTIME);
 415                 bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
 416         }
 417
 418         bch2_trans_put(trans);
 419         mutex_unlock(&inode->ei_update_lock);
 420         return ret;
 421 }
 422
 423 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 424                      struct dentry *dentry)
 425 {
 426         struct bch_fs *c = vdir->i_sb->s_fs_info;
 427         struct bch_inode_info *dir = to_bch_ei(vdir);
 428         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 429         int ret;
 430
 431         lockdep_assert_held(&inode->v.i_rwsem);
 432
 433         ret = __bch2_link(c, inode, dir, dentry);
 434         if (unlikely(ret))
 435                 return ret;
 436
 437         ihold(&inode->v);
 438         d_instantiate(dentry, &inode->v);
 439         return 0;
 440 }
 441
 442 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 443                   bool deleting_snapshot)
 444 {
 445         struct bch_fs *c = vdir->i_sb->s_fs_info;
 446         struct bch_inode_info *dir = to_bch_ei(vdir);
 447         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 448         struct bch_inode_unpacked dir_u, inode_u;
 449         struct btree_trans *trans = bch2_trans_get(c);
 450         int ret;
 451
 452         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 453
 454         ret = commit_do(trans, NULL, NULL,
 455                         BCH_TRANS_COMMIT_no_enospc,
 456                 bch2_unlink_trans(trans,
 457                                   inode_inum(dir), &dir_u,
 458                                   &inode_u, &dentry->d_name,
 459                                   deleting_snapshot));
 460         if (unlikely(ret))
 461                 goto err;
 462
 463         bch2_inode_update_after_write(trans, dir, &dir_u,
 464                                       ATTR_MTIME|ATTR_CTIME);
 465         bch2_inode_update_after_write(trans, inode, &inode_u,
 466                                       ATTR_MTIME);
 467
 468         if (inode_u.bi_subvol) {
 469                 /*
 470                  * Subvolume deletion is asynchronous, but we still want to tell
 471                  * the VFS that it's been deleted here:
 472                  */
 473                 set_nlink(&inode->v, 0);
 474         }
 475 err:
 476         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 477         bch2_trans_put(trans);
 478
 479         return ret;
 480 }
 481
 482 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 483 {
 484         return __bch2_unlink(vdir, dentry, false);
 485 }
 486
 487 static int bch2_symlink(struct mnt_idmap *idmap,
 488                         struct inode *vdir, struct dentry *dentry,
 489                         const char *symname)
 490 {
 491         struct bch_fs *c = vdir->i_sb->s_fs_info;
 492         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 493         int ret;
 494
 495         inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
 496                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 497         if (IS_ERR(inode))
 498                 return bch2_err_class(PTR_ERR(inode));
 499
 500         inode_lock(&inode->v);
 501         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 502         inode_unlock(&inode->v);
 503
 504         if (unlikely(ret))
 505                 goto err;
 506
 507         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 508         if (unlikely(ret))
 509                 goto err;
 510
 511         ret = __bch2_link(c, inode, dir, dentry);
 512         if (unlikely(ret))
 513                 goto err;
 514
 515         d_instantiate(dentry, &inode->v);
 516         return 0;
 517 err:
 518         iput(&inode->v);
 519         return ret;
 520 }
 521
 522 static int bch2_mkdir(struct mnt_idmap *idmap,
 523                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 524 {
 525         return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
 526 }
 527
 528 static int bch2_rename2(struct mnt_idmap *idmap,
 529                         struct inode *src_vdir, struct dentry *src_dentry,
 530                         struct inode *dst_vdir, struct dentry *dst_dentry,
 531                         unsigned flags)
 532 {
 533         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 534         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 535         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 536         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 537         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 538         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 539         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 540         struct btree_trans *trans;
 541         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 542                 ? BCH_RENAME_EXCHANGE
 543                 : dst_dentry->d_inode
 544                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 545         int ret;
 546
 547         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 548                 return -EINVAL;
 549
 550         if (mode == BCH_RENAME_OVERWRITE) {
 551                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 552                                                    0, LLONG_MAX);
 553                 if (ret)
 554                         return ret;
 555         }
 556
 557         trans = bch2_trans_get(c);
 558
 559         bch2_lock_inodes(INODE_UPDATE_LOCK,
 560                          src_dir,
 561                          dst_dir,
 562                          src_inode,
 563                          dst_inode);
 564
 565         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 566                 ret = bch2_fs_quota_transfer(c, src_inode,
 567                                              dst_dir->ei_qid,
 568                                              1 << QTYP_PRJ,
 569                                              KEY_TYPE_QUOTA_PREALLOC);
 570                 if (ret)
 571                         goto err;
 572         }
 573
 574         if (mode == BCH_RENAME_EXCHANGE &&
 575             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 576                 ret = bch2_fs_quota_transfer(c, dst_inode,
 577                                              src_dir->ei_qid,
 578                                              1 << QTYP_PRJ,
 579                                              KEY_TYPE_QUOTA_PREALLOC);
 580                 if (ret)
 581                         goto err;
 582         }
 583
 584         ret = commit_do(trans, NULL, NULL, 0,
 585                         bch2_rename_trans(trans,
 586                                           inode_inum(src_dir), &src_dir_u,
 587                                           inode_inum(dst_dir), &dst_dir_u,
 588                                           &src_inode_u,
 589                                           &dst_inode_u,
 590                                           &src_dentry->d_name,
 591                                           &dst_dentry->d_name,
 592                                           mode));
 593         if (unlikely(ret))
 594                 goto err;
 595
 596         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 597         BUG_ON(dst_inode &&
 598                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 599
 600         bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
 601                                       ATTR_MTIME|ATTR_CTIME);
 602
 603         if (src_dir != dst_dir)
 604                 bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
 605                                               ATTR_MTIME|ATTR_CTIME);
 606
 607         bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
 608                                       ATTR_CTIME);
 609
 610         if (dst_inode)
 611                 bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
 612                                               ATTR_CTIME);
 613 err:
 614         bch2_trans_put(trans);
 615
 616         bch2_fs_quota_transfer(c, src_inode,
 617                                bch_qid(&src_inode->ei_inode),
 618                                1 << QTYP_PRJ,
 619                                KEY_TYPE_QUOTA_NOCHECK);
 620         if (dst_inode)
 621                 bch2_fs_quota_transfer(c, dst_inode,
 622                                        bch_qid(&dst_inode->ei_inode),
 623                                        1 << QTYP_PRJ,
 624                                        KEY_TYPE_QUOTA_NOCHECK);
 625
 626         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 627                            src_dir,
 628                            dst_dir,
 629                            src_inode,
 630                            dst_inode);
 631
 632         return ret;
 633 }
 634
 635 static void bch2_setattr_copy(struct mnt_idmap *idmap,
 636                               struct bch_inode_info *inode,
 637                               struct bch_inode_unpacked *bi,
 638                               struct iattr *attr)
 639 {
 640         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 641         unsigned int ia_valid = attr->ia_valid;
 642
 643         if (ia_valid & ATTR_UID)
 644                 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 645         if (ia_valid & ATTR_GID)
 646                 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 647
 648         if (ia_valid & ATTR_SIZE)
 649                 bi->bi_size = attr->ia_size;
 650
 651         if (ia_valid & ATTR_ATIME)
 652                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 653         if (ia_valid & ATTR_MTIME)
 654                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 655         if (ia_valid & ATTR_CTIME)
 656                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 657
 658         if (ia_valid & ATTR_MODE) {
 659                 umode_t mode = attr->ia_mode;
 660                 kgid_t gid = ia_valid & ATTR_GID
 661                         ? attr->ia_gid
 662                         : inode->v.i_gid;
 663
 664                 if (!in_group_p(gid) &&
 665                     !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
 666                         mode &= ~S_ISGID;
 667                 bi->bi_mode = mode;
 668         }
 669 }
 670
 671 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
 672                          struct bch_inode_info *inode,
 673                          struct iattr *attr)
 674 {
 675         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 676         struct bch_qid qid;
 677         struct btree_trans *trans;
 678         struct btree_iter inode_iter = { NULL };
 679         struct bch_inode_unpacked inode_u;
 680         struct posix_acl *acl = NULL;
 681         int ret;
 682
 683         mutex_lock(&inode->ei_update_lock);
 684
 685         qid = inode->ei_qid;
 686
 687         if (attr->ia_valid & ATTR_UID)
 688                 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 689
 690         if (attr->ia_valid & ATTR_GID)
 691                 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 692
 693         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 694                                      KEY_TYPE_QUOTA_PREALLOC);
 695         if (ret)
 696                 goto err;
 697
 698         trans = bch2_trans_get(c);
 699 retry:
 700         bch2_trans_begin(trans);
 701         kfree(acl);
 702         acl = NULL;
 703
 704         ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
 705                               BTREE_ITER_INTENT);
 706         if (ret)
 707                 goto btree_err;
 708
 709         bch2_setattr_copy(idmap, inode, &inode_u, attr);
 710
 711         if (attr->ia_valid & ATTR_MODE) {
 712                 ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
 713                                      inode_u.bi_mode, &acl);
 714                 if (ret)
 715                         goto btree_err;
 716         }
 717
 718         ret =   bch2_inode_write(trans, &inode_iter, &inode_u) ?:
 719                 bch2_trans_commit(trans, NULL, NULL,
 720                                   BCH_TRANS_COMMIT_no_enospc);
 721 btree_err:
 722         bch2_trans_iter_exit(trans, &inode_iter);
 723
 724         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 725                 goto retry;
 726         if (unlikely(ret))
 727                 goto err_trans;
 728
 729         bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
 730
 731         if (acl)
 732                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 733 err_trans:
 734         bch2_trans_put(trans);
 735 err:
 736         mutex_unlock(&inode->ei_update_lock);
 737
 738         return bch2_err_class(ret);
 739 }
 740
 741 static int bch2_getattr(struct mnt_idmap *idmap,
 742                         const struct path *path, struct kstat *stat,
 743                         u32 request_mask, unsigned query_flags)
 744 {
 745         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 746         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 747
 748         stat->dev       = inode->v.i_sb->s_dev;
 749         stat->ino       = inode->v.i_ino;
 750         stat->mode      = inode->v.i_mode;
 751         stat->nlink     = inode->v.i_nlink;
 752         stat->uid       = inode->v.i_uid;
 753         stat->gid       = inode->v.i_gid;
 754         stat->rdev      = inode->v.i_rdev;
 755         stat->size      = i_size_read(&inode->v);
 756         stat->atime     = inode_get_atime(&inode->v);
 757         stat->mtime     = inode_get_mtime(&inode->v);
 758         stat->ctime     = inode_get_ctime(&inode->v);
 759         stat->blksize   = block_bytes(c);
 760         stat->blocks    = inode->v.i_blocks;
 761
 762         if (request_mask & STATX_BTIME) {
 763                 stat->result_mask |= STATX_BTIME;
 764                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 765         }
 766
 767         if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
 768                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 769         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 770
 771         if (inode->ei_inode.bi_flags & BCH_INODE_append)
 772                 stat->attributes |= STATX_ATTR_APPEND;
 773         stat->attributes_mask    |= STATX_ATTR_APPEND;
 774
 775         if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
 776                 stat->attributes |= STATX_ATTR_NODUMP;
 777         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 778
 779         return 0;
 780 }
 781
 782 static int bch2_setattr(struct mnt_idmap *idmap,
 783                         struct dentry *dentry, struct iattr *iattr)
 784 {
 785         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 786         int ret;
 787
 788         lockdep_assert_held(&inode->v.i_rwsem);
 789
 790         ret = setattr_prepare(idmap, dentry, iattr);
 791         if (ret)
 792                 return ret;
 793
 794         return iattr->ia_valid & ATTR_SIZE
 795                 ? bchfs_truncate(idmap, inode, iattr)
 796                 : bch2_setattr_nonsize(idmap, inode, iattr);
 797 }
 798
 799 static int bch2_tmpfile(struct mnt_idmap *idmap,
 800                         struct inode *vdir, struct file *file, umode_t mode)
 801 {
 802         struct bch_inode_info *inode =
 803                 __bch2_create(idmap, to_bch_ei(vdir),
 804                               file->f_path.dentry, mode, 0,
 805                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 806
 807         if (IS_ERR(inode))
 808                 return bch2_err_class(PTR_ERR(inode));
 809
 810         d_mark_tmpfile(file, &inode->v);
 811         d_instantiate(file->f_path.dentry, &inode->v);
 812         return finish_open_simple(file, 0);
 813 }
 814
 815 static int bch2_fill_extent(struct bch_fs *c,
 816                             struct fiemap_extent_info *info,
 817                             struct bkey_s_c k, unsigned flags)
 818 {
 819         if (bkey_extent_is_direct_data(k.k)) {
 820                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 821                 const union bch_extent_entry *entry;
 822                 struct extent_ptr_decoded p;
 823                 int ret;
 824
 825                 if (k.k->type == KEY_TYPE_reflink_v)
 826                         flags |= FIEMAP_EXTENT_SHARED;
 827
 828                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 829                         int flags2 = 0;
 830                         u64 offset = p.ptr.offset;
 831
 832                         if (p.ptr.unwritten)
 833                                 flags2 |= FIEMAP_EXTENT_UNWRITTEN;
 834
 835                         if (p.crc.compression_type)
 836                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 837                         else
 838                                 offset += p.crc.offset;
 839
 840                         if ((offset & (block_sectors(c) - 1)) ||
 841                             (k.k->size & (block_sectors(c) - 1)))
 842                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 843
 844                         ret = fiemap_fill_next_extent(info,
 845                                                 bkey_start_offset(k.k) << 9,
 846                                                 offset << 9,
 847                                                 k.k->size << 9, flags|flags2);
 848                         if (ret)
 849                                 return ret;
 850                 }
 851
 852                 return 0;
 853         } else if (bkey_extent_is_inline_data(k.k)) {
 854                 return fiemap_fill_next_extent(info,
 855                                                bkey_start_offset(k.k) << 9,
 856                                                0, k.k->size << 9,
 857                                                flags|
 858                                                FIEMAP_EXTENT_DATA_INLINE);
 859         } else if (k.k->type == KEY_TYPE_reservation) {
 860                 return fiemap_fill_next_extent(info,
 861                                                bkey_start_offset(k.k) << 9,
 862                                                0, k.k->size << 9,
 863                                                flags|
 864                                                FIEMAP_EXTENT_DELALLOC|
 865                                                FIEMAP_EXTENT_UNWRITTEN);
 866         } else {
 867                 BUG();
 868         }
 869 }
 870
 871 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 872                        u64 start, u64 len)
 873 {
 874         struct bch_fs *c = vinode->i_sb->s_fs_info;
 875         struct bch_inode_info *ei = to_bch_ei(vinode);
 876         struct btree_trans *trans;
 877         struct btree_iter iter;
 878         struct bkey_s_c k;
 879         struct bkey_buf cur, prev;
 880         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 881         unsigned offset_into_extent, sectors;
 882         bool have_extent = false;
 883         u32 snapshot;
 884         int ret = 0;
 885
 886         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 887         if (ret)
 888                 return ret;
 889
 890         if (start + len < start)
 891                 return -EINVAL;
 892
 893         start >>= 9;
 894
 895         bch2_bkey_buf_init(&cur);
 896         bch2_bkey_buf_init(&prev);
 897         trans = bch2_trans_get(c);
 898 retry:
 899         bch2_trans_begin(trans);
 900
 901         ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
 902         if (ret)
 903                 goto err;
 904
 905         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 906                              SPOS(ei->v.i_ino, start, snapshot), 0);
 907
 908         while (!(ret = btree_trans_too_many_iters(trans)) &&
 909                (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
 910                !(ret = bkey_err(k))) {
 911                 enum btree_id data_btree = BTREE_ID_extents;
 912
 913                 if (!bkey_extent_is_data(k.k) &&
 914                     k.k->type != KEY_TYPE_reservation) {
 915                         bch2_btree_iter_advance(&iter);
 916                         continue;
 917                 }
 918
 919                 offset_into_extent      = iter.pos.offset -
 920                         bkey_start_offset(k.k);
 921                 sectors                 = k.k->size - offset_into_extent;
 922
 923                 bch2_bkey_buf_reassemble(&cur, c, k);
 924
 925                 ret = bch2_read_indirect_extent(trans, &data_btree,
 926                                         &offset_into_extent, &cur);
 927                 if (ret)
 928                         break;
 929
 930                 k = bkey_i_to_s_c(cur.k);
 931                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 932
 933                 sectors = min(sectors, k.k->size - offset_into_extent);
 934
 935                 bch2_cut_front(POS(k.k->p.inode,
 936                                    bkey_start_offset(k.k) +
 937                                    offset_into_extent),
 938                                cur.k);
 939                 bch2_key_resize(&cur.k->k, sectors);
 940                 cur.k->k.p = iter.pos;
 941                 cur.k->k.p.offset += cur.k->k.size;
 942
 943                 if (have_extent) {
 944                         bch2_trans_unlock(trans);
 945                         ret = bch2_fill_extent(c, info,
 946                                         bkey_i_to_s_c(prev.k), 0);
 947                         if (ret)
 948                                 break;
 949                 }
 950
 951                 bkey_copy(prev.k, cur.k);
 952                 have_extent = true;
 953
 954                 bch2_btree_iter_set_pos(&iter,
 955                         POS(iter.pos.inode, iter.pos.offset + sectors));
 956         }
 957         start = iter.pos.offset;
 958         bch2_trans_iter_exit(trans, &iter);
 959 err:
 960         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 961                 goto retry;
 962
 963         if (!ret && have_extent) {
 964                 bch2_trans_unlock(trans);
 965                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 966                                        FIEMAP_EXTENT_LAST);
 967         }
 968
 969         bch2_trans_put(trans);
 970         bch2_bkey_buf_exit(&cur, c);
 971         bch2_bkey_buf_exit(&prev, c);
 972         return ret < 0 ? ret : 0;
 973 }
 974
 975 static const struct vm_operations_struct bch_vm_ops = {
 976         .fault          = bch2_page_fault,
 977         .map_pages      = filemap_map_pages,
 978         .page_mkwrite   = bch2_page_mkwrite,
 979 };
 980
 981 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 982 {
 983         file_accessed(file);
 984
 985         vma->vm_ops = &bch_vm_ops;
 986         return 0;
 987 }
 988
 989 /* Directories: */
 990
 991 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 992 {
 993         return generic_file_llseek_size(file, offset, whence,
 994                                         S64_MAX, S64_MAX);
 995 }
 996
 997 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 998 {
 999         struct bch_inode_info *inode = file_bch_inode(file);
1000         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1001         int ret;
1002
1003         if (!dir_emit_dots(file, ctx))
1004                 return 0;
1005
1006         ret = bch2_readdir(c, inode_inum(inode), ctx);
1007         if (ret)
1008                 bch_err_fn(c, ret);
1009
1010         return bch2_err_class(ret);
1011 }
1012
1013 static const struct file_operations bch_file_operations = {
1014         .llseek         = bch2_llseek,
1015         .read_iter      = bch2_read_iter,
1016         .write_iter     = bch2_write_iter,
1017         .mmap           = bch2_mmap,
1018         .open           = generic_file_open,
1019         .fsync          = bch2_fsync,
1020         .splice_read    = filemap_splice_read,
1021         .splice_write   = iter_file_splice_write,
1022         .fallocate      = bch2_fallocate_dispatch,
1023         .unlocked_ioctl = bch2_fs_file_ioctl,
1024 #ifdef CONFIG_COMPAT
1025         .compat_ioctl   = bch2_compat_fs_ioctl,
1026 #endif
1027         .remap_file_range = bch2_remap_file_range,
1028 };
1029
1030 static const struct inode_operations bch_file_inode_operations = {
1031         .getattr        = bch2_getattr,
1032         .setattr        = bch2_setattr,
1033         .fiemap         = bch2_fiemap,
1034         .listxattr      = bch2_xattr_list,
1035 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1036         .get_acl        = bch2_get_acl,
1037         .set_acl        = bch2_set_acl,
1038 #endif
1039 };
1040
1041 static const struct inode_operations bch_dir_inode_operations = {
1042         .lookup         = bch2_lookup,
1043         .create         = bch2_create,
1044         .link           = bch2_link,
1045         .unlink         = bch2_unlink,
1046         .symlink        = bch2_symlink,
1047         .mkdir          = bch2_mkdir,
1048         .rmdir          = bch2_unlink,
1049         .mknod          = bch2_mknod,
1050         .rename         = bch2_rename2,
1051         .getattr        = bch2_getattr,
1052         .setattr        = bch2_setattr,
1053         .tmpfile        = bch2_tmpfile,
1054         .listxattr      = bch2_xattr_list,
1055 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1056         .get_acl        = bch2_get_acl,
1057         .set_acl        = bch2_set_acl,
1058 #endif
1059 };
1060
1061 static const struct file_operations bch_dir_file_operations = {
1062         .llseek         = bch2_dir_llseek,
1063         .read           = generic_read_dir,
1064         .iterate_shared = bch2_vfs_readdir,
1065         .fsync          = bch2_fsync,
1066         .unlocked_ioctl = bch2_fs_file_ioctl,
1067 #ifdef CONFIG_COMPAT
1068         .compat_ioctl   = bch2_compat_fs_ioctl,
1069 #endif
1070 };
1071
1072 static const struct inode_operations bch_symlink_inode_operations = {
1073         .get_link       = page_get_link,
1074         .getattr        = bch2_getattr,
1075         .setattr        = bch2_setattr,
1076         .listxattr      = bch2_xattr_list,
1077 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1078         .get_acl        = bch2_get_acl,
1079         .set_acl        = bch2_set_acl,
1080 #endif
1081 };
1082
1083 static const struct inode_operations bch_special_inode_operations = {
1084         .getattr        = bch2_getattr,
1085         .setattr        = bch2_setattr,
1086         .listxattr      = bch2_xattr_list,
1087 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1088         .get_acl        = bch2_get_acl,
1089         .set_acl        = bch2_set_acl,
1090 #endif
1091 };
1092
1093 static const struct address_space_operations bch_address_space_operations = {
1094         .read_folio     = bch2_read_folio,
1095         .writepages     = bch2_writepages,
1096         .readahead      = bch2_readahead,
1097         .dirty_folio    = filemap_dirty_folio,
1098         .write_begin    = bch2_write_begin,
1099         .write_end      = bch2_write_end,
1100         .invalidate_folio = bch2_invalidate_folio,
1101         .release_folio  = bch2_release_folio,
1102         .direct_IO      = noop_direct_IO,
1103 #ifdef CONFIG_MIGRATION
1104         .migrate_folio  = filemap_migrate_folio,
1105 #endif
1106         .error_remove_page = generic_error_remove_page,
1107 };
1108
1109 struct bcachefs_fid {
1110         u64             inum;
1111         u32             subvol;
1112         u32             gen;
1113 } __packed;
1114
1115 struct bcachefs_fid_with_parent {
1116         struct bcachefs_fid     fid;
1117         struct bcachefs_fid     dir;
1118 } __packed;
1119
1120 static int bcachefs_fid_valid(int fh_len, int fh_type)
1121 {
1122         switch (fh_type) {
1123         case FILEID_BCACHEFS_WITHOUT_PARENT:
1124                 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1125         case FILEID_BCACHEFS_WITH_PARENT:
1126                 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1127         default:
1128                 return false;
1129         }
1130 }
1131
1132 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1133 {
1134         return (struct bcachefs_fid) {
1135                 .inum   = inode->ei_inode.bi_inum,
1136                 .subvol = inode->ei_subvol,
1137                 .gen    = inode->ei_inode.bi_generation,
1138         };
1139 }
1140
1141 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1142                           struct inode *vdir)
1143 {
1144         struct bch_inode_info *inode    = to_bch_ei(vinode);
1145         struct bch_inode_info *dir      = to_bch_ei(vdir);
1146
1147         if (*len < sizeof(struct bcachefs_fid_with_parent) / sizeof(u32))
1148                 return FILEID_INVALID;
1149
1150         if (!S_ISDIR(inode->v.i_mode) && dir) {
1151                 struct bcachefs_fid_with_parent *fid = (void *) fh;
1152
1153                 fid->fid = bch2_inode_to_fid(inode);
1154                 fid->dir = bch2_inode_to_fid(dir);
1155
1156                 *len = sizeof(*fid) / sizeof(u32);
1157                 return FILEID_BCACHEFS_WITH_PARENT;
1158         } else {
1159                 struct bcachefs_fid *fid = (void *) fh;
1160
1161                 *fid = bch2_inode_to_fid(inode);
1162
1163                 *len = sizeof(*fid) / sizeof(u32);
1164                 return FILEID_BCACHEFS_WITHOUT_PARENT;
1165         }
1166 }
1167
1168 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1169                                         struct bcachefs_fid fid)
1170 {
1171         struct bch_fs *c = sb->s_fs_info;
1172         struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1173                                     .subvol = fid.subvol,
1174                                     .inum = fid.inum,
1175         });
1176         if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1177                 iput(vinode);
1178                 vinode = ERR_PTR(-ESTALE);
1179         }
1180         return vinode;
1181 }
1182
1183 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1184                 int fh_len, int fh_type)
1185 {
1186         struct bcachefs_fid *fid = (void *) _fid;
1187
1188         if (!bcachefs_fid_valid(fh_len, fh_type))
1189                 return NULL;
1190
1191         return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1192 }
1193
1194 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1195                 int fh_len, int fh_type)
1196 {
1197         struct bcachefs_fid_with_parent *fid = (void *) _fid;
1198
1199         if (!bcachefs_fid_valid(fh_len, fh_type) ||
1200             fh_type != FILEID_BCACHEFS_WITH_PARENT)
1201                 return NULL;
1202
1203         return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1204 }
1205
1206 static struct dentry *bch2_get_parent(struct dentry *child)
1207 {
1208         struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1209         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1210         subvol_inum parent_inum = {
1211                 .subvol = inode->ei_inode.bi_parent_subvol ?:
1212                         inode->ei_subvol,
1213                 .inum = inode->ei_inode.bi_dir,
1214         };
1215
1216         return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1217 }
1218
1219 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1220 {
1221         struct bch_inode_info *inode    = to_bch_ei(child->d_inode);
1222         struct bch_inode_info *dir      = to_bch_ei(parent->d_inode);
1223         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1224         struct btree_trans *trans;
1225         struct btree_iter iter1;
1226         struct btree_iter iter2;
1227         struct bkey_s_c k;
1228         struct bkey_s_c_dirent d;
1229         struct bch_inode_unpacked inode_u;
1230         subvol_inum target;
1231         u32 snapshot;
1232         struct qstr dirent_name;
1233         unsigned name_len = 0;
1234         int ret;
1235
1236         if (!S_ISDIR(dir->v.i_mode))
1237                 return -EINVAL;
1238
1239         trans = bch2_trans_get(c);
1240
1241         bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1242                              POS(dir->ei_inode.bi_inum, 0), 0);
1243         bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1244                              POS(dir->ei_inode.bi_inum, 0), 0);
1245 retry:
1246         bch2_trans_begin(trans);
1247
1248         ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1249         if (ret)
1250                 goto err;
1251
1252         bch2_btree_iter_set_snapshot(&iter1, snapshot);
1253         bch2_btree_iter_set_snapshot(&iter2, snapshot);
1254
1255         ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1256         if (ret)
1257                 goto err;
1258
1259         if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1260                 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1261
1262                 k = bch2_btree_iter_peek_slot(&iter1);
1263                 ret = bkey_err(k);
1264                 if (ret)
1265                         goto err;
1266
1267                 if (k.k->type != KEY_TYPE_dirent) {
1268                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1269                         goto err;
1270                 }
1271
1272                 d = bkey_s_c_to_dirent(k);
1273                 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1274                 if (ret > 0)
1275                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1276                 if (ret)
1277                         goto err;
1278
1279                 if (target.subvol       == inode->ei_subvol &&
1280                     target.inum         == inode->ei_inode.bi_inum)
1281                         goto found;
1282         } else {
1283                 /*
1284                  * File with multiple hardlinks and our backref is to the wrong
1285                  * directory - linear search:
1286                  */
1287                 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1288                         if (k.k->p.inode > dir->ei_inode.bi_inum)
1289                                 break;
1290
1291                         if (k.k->type != KEY_TYPE_dirent)
1292                                 continue;
1293
1294                         d = bkey_s_c_to_dirent(k);
1295                         ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1296                         if (ret < 0)
1297                                 break;
1298                         if (ret)
1299                                 continue;
1300
1301                         if (target.subvol       == inode->ei_subvol &&
1302                             target.inum         == inode->ei_inode.bi_inum)
1303                                 goto found;
1304                 }
1305         }
1306
1307         ret = -ENOENT;
1308         goto err;
1309 found:
1310         dirent_name = bch2_dirent_get_name(d);
1311
1312         name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1313         memcpy(name, dirent_name.name, name_len);
1314         name[name_len] = '\0';
1315 err:
1316         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1317                 goto retry;
1318
1319         bch2_trans_iter_exit(trans, &iter1);
1320         bch2_trans_iter_exit(trans, &iter2);
1321         bch2_trans_put(trans);
1322
1323         return ret;
1324 }
1325
1326 static const struct export_operations bch_export_ops = {
1327         .encode_fh      = bch2_encode_fh,
1328         .fh_to_dentry   = bch2_fh_to_dentry,
1329         .fh_to_parent   = bch2_fh_to_parent,
1330         .get_parent     = bch2_get_parent,
1331         .get_name       = bch2_get_name,
1332 };
1333
1334 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1335                                 struct bch_inode_info *inode,
1336                                 struct bch_inode_unpacked *bi,
1337                                 struct bch_subvolume *subvol)
1338 {
1339         bch2_inode_update_after_write(trans, inode, bi, ~0);
1340
1341         if (BCH_SUBVOLUME_SNAP(subvol))
1342                 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1343         else
1344                 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1345
1346         inode->v.i_blocks       = bi->bi_sectors;
1347         inode->v.i_ino          = bi->bi_inum;
1348         inode->v.i_rdev         = bi->bi_dev;
1349         inode->v.i_generation   = bi->bi_generation;
1350         inode->v.i_size         = bi->bi_size;
1351
1352         inode->ei_flags         = 0;
1353         inode->ei_quota_reserved = 0;
1354         inode->ei_qid           = bch_qid(bi);
1355         inode->ei_subvol        = inum.subvol;
1356
1357         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1358
1359         switch (inode->v.i_mode & S_IFMT) {
1360         case S_IFREG:
1361                 inode->v.i_op   = &bch_file_inode_operations;
1362                 inode->v.i_fop  = &bch_file_operations;
1363                 break;
1364         case S_IFDIR:
1365                 inode->v.i_op   = &bch_dir_inode_operations;
1366                 inode->v.i_fop  = &bch_dir_file_operations;
1367                 break;
1368         case S_IFLNK:
1369                 inode_nohighmem(&inode->v);
1370                 inode->v.i_op   = &bch_symlink_inode_operations;
1371                 break;
1372         default:
1373                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1374                 inode->v.i_op   = &bch_special_inode_operations;
1375                 break;
1376         }
1377
1378         mapping_set_large_folios(inode->v.i_mapping);
1379 }
1380
1381 static struct inode *bch2_alloc_inode(struct super_block *sb)
1382 {
1383         struct bch_inode_info *inode;
1384
1385         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1386         if (!inode)
1387                 return NULL;
1388
1389         inode_init_once(&inode->v);
1390         mutex_init(&inode->ei_update_lock);
1391         two_state_lock_init(&inode->ei_pagecache_lock);
1392         INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
1393         mutex_init(&inode->ei_quota_lock);
1394
1395         return &inode->v;
1396 }
1397
1398 static void bch2_i_callback(struct rcu_head *head)
1399 {
1400         struct inode *vinode = container_of(head, struct inode, i_rcu);
1401         struct bch_inode_info *inode = to_bch_ei(vinode);
1402
1403         kmem_cache_free(bch2_inode_cache, inode);
1404 }
1405
1406 static void bch2_destroy_inode(struct inode *vinode)
1407 {
1408         call_rcu(&vinode->i_rcu, bch2_i_callback);
1409 }
1410
1411 static int inode_update_times_fn(struct btree_trans *trans,
1412                                  struct bch_inode_info *inode,
1413                                  struct bch_inode_unpacked *bi,
1414                                  void *p)
1415 {
1416         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1417
1418         bi->bi_atime    = timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1419         bi->bi_mtime    = timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1420         bi->bi_ctime    = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1421
1422         return 0;
1423 }
1424
1425 static int bch2_vfs_write_inode(struct inode *vinode,
1426                                 struct writeback_control *wbc)
1427 {
1428         struct bch_fs *c = vinode->i_sb->s_fs_info;
1429         struct bch_inode_info *inode = to_bch_ei(vinode);
1430         int ret;
1431
1432         mutex_lock(&inode->ei_update_lock);
1433         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1434                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1435         mutex_unlock(&inode->ei_update_lock);
1436
1437         return bch2_err_class(ret);
1438 }
1439
1440 static void bch2_evict_inode(struct inode *vinode)
1441 {
1442         struct bch_fs *c = vinode->i_sb->s_fs_info;
1443         struct bch_inode_info *inode = to_bch_ei(vinode);
1444
1445         truncate_inode_pages_final(&inode->v.i_data);
1446
1447         clear_inode(&inode->v);
1448
1449         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1450
1451         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1452                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1453                                 KEY_TYPE_QUOTA_WARN);
1454                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1455                                 KEY_TYPE_QUOTA_WARN);
1456                 bch2_inode_rm(c, inode_inum(inode));
1457         }
1458
1459         mutex_lock(&c->vfs_inodes_lock);
1460         list_del_init(&inode->ei_vfs_inode_list);
1461         mutex_unlock(&c->vfs_inodes_lock);
1462 }
1463
1464 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1465 {
1466         struct bch_inode_info *inode, **i;
1467         DARRAY(struct bch_inode_info *) grabbed;
1468         bool clean_pass = false, this_pass_clean;
1469
1470         /*
1471          * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1472          * be pruned with d_mark_dontcache().
1473          *
1474          * Once we've had a clean pass where we didn't find any inodes without
1475          * I_DONTCACHE, we wait for them to be freed:
1476          */
1477
1478         darray_init(&grabbed);
1479         darray_make_room(&grabbed, 1024);
1480 again:
1481         cond_resched();
1482         this_pass_clean = true;
1483
1484         mutex_lock(&c->vfs_inodes_lock);
1485         list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1486                 if (!snapshot_list_has_id(s, inode->ei_subvol))
1487                         continue;
1488
1489                 if (!(inode->v.i_state & I_DONTCACHE) &&
1490                     !(inode->v.i_state & I_FREEING) &&
1491                     igrab(&inode->v)) {
1492                         this_pass_clean = false;
1493
1494                         if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1495                                 iput(&inode->v);
1496                                 break;
1497                         }
1498                 } else if (clean_pass && this_pass_clean) {
1499                         wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1500                         DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1501
1502                         prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1503                         mutex_unlock(&c->vfs_inodes_lock);
1504
1505                         schedule();
1506                         finish_wait(wq, &wait.wq_entry);
1507                         goto again;
1508                 }
1509         }
1510         mutex_unlock(&c->vfs_inodes_lock);
1511
1512         darray_for_each(grabbed, i) {
1513                 inode = *i;
1514                 d_mark_dontcache(&inode->v);
1515                 d_prune_aliases(&inode->v);
1516                 iput(&inode->v);
1517         }
1518         grabbed.nr = 0;
1519
1520         if (!clean_pass || !this_pass_clean) {
1521                 clean_pass = this_pass_clean;
1522                 goto again;
1523         }
1524
1525         darray_exit(&grabbed);
1526 }
1527
1528 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1529 {
1530         struct super_block *sb = dentry->d_sb;
1531         struct bch_fs *c = sb->s_fs_info;
1532         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1533         unsigned shift = sb->s_blocksize_bits - 9;
1534         /*
1535          * this assumes inodes take up 64 bytes, which is a decent average
1536          * number:
1537          */
1538         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1539         u64 fsid;
1540
1541         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1542         buf->f_bsize    = sb->s_blocksize;
1543         buf->f_blocks   = usage.capacity >> shift;
1544         buf->f_bfree    = usage.free >> shift;
1545         buf->f_bavail   = avail_factor(usage.free) >> shift;
1546
1547         buf->f_files    = usage.nr_inodes + avail_inodes;
1548         buf->f_ffree    = avail_inodes;
1549
1550         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1551                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1552         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1553         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1554         buf->f_namelen  = BCH_NAME_MAX;
1555
1556         return 0;
1557 }
1558
1559 static int bch2_sync_fs(struct super_block *sb, int wait)
1560 {
1561         struct bch_fs *c = sb->s_fs_info;
1562         int ret;
1563
1564         if (c->opts.journal_flush_disabled)
1565                 return 0;
1566
1567         if (!wait) {
1568                 bch2_journal_flush_async(&c->journal, NULL);
1569                 return 0;
1570         }
1571
1572         ret = bch2_journal_flush(&c->journal);
1573         return bch2_err_class(ret);
1574 }
1575
1576 static struct bch_fs *bch2_path_to_fs(const char *path)
1577 {
1578         struct bch_fs *c;
1579         dev_t dev;
1580         int ret;
1581
1582         ret = lookup_bdev(path, &dev);
1583         if (ret)
1584                 return ERR_PTR(ret);
1585
1586         c = bch2_dev_to_fs(dev);
1587         if (c)
1588                 closure_put(&c->cl);
1589         return c ?: ERR_PTR(-ENOENT);
1590 }
1591
1592 static char **split_devs(const char *_dev_name, unsigned *nr)
1593 {
1594         char *dev_name = NULL, **devs = NULL, *s;
1595         size_t i = 0, nr_devs = 0;
1596
1597         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1598         if (!dev_name)
1599                 return NULL;
1600
1601         for (s = dev_name; s; s = strchr(s + 1, ':'))
1602                 nr_devs++;
1603
1604         devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1605         if (!devs) {
1606                 kfree(dev_name);
1607                 return NULL;
1608         }
1609
1610         while ((s = strsep(&dev_name, ":")))
1611                 devs[i++] = s;
1612
1613         *nr = nr_devs;
1614         return devs;
1615 }
1616
1617 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1618 {
1619         struct bch_fs *c = sb->s_fs_info;
1620         struct bch_opts opts = bch2_opts_empty();
1621         int ret;
1622
1623         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1624
1625         ret = bch2_parse_mount_opts(c, &opts, data);
1626         if (ret)
1627                 goto err;
1628
1629         if (opts.read_only != c->opts.read_only) {
1630                 down_write(&c->state_lock);
1631
1632                 if (opts.read_only) {
1633                         bch2_fs_read_only(c);
1634
1635                         sb->s_flags |= SB_RDONLY;
1636                 } else {
1637                         ret = bch2_fs_read_write(c);
1638                         if (ret) {
1639                                 bch_err(c, "error going rw: %i", ret);
1640                                 up_write(&c->state_lock);
1641                                 ret = -EINVAL;
1642                                 goto err;
1643                         }
1644
1645                         sb->s_flags &= ~SB_RDONLY;
1646                 }
1647
1648                 c->opts.read_only = opts.read_only;
1649
1650                 up_write(&c->state_lock);
1651         }
1652
1653         if (opt_defined(opts, errors))
1654                 c->opts.errors = opts.errors;
1655 err:
1656         return bch2_err_class(ret);
1657 }
1658
1659 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1660 {
1661         struct bch_fs *c = root->d_sb->s_fs_info;
1662         struct bch_dev *ca;
1663         unsigned i;
1664         bool first = true;
1665
1666         for_each_online_member(ca, c, i) {
1667                 if (!first)
1668                         seq_putc(seq, ':');
1669                 first = false;
1670                 seq_puts(seq, ca->disk_sb.sb_name);
1671         }
1672
1673         return 0;
1674 }
1675
1676 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1677 {
1678         struct bch_fs *c = root->d_sb->s_fs_info;
1679         enum bch_opt_id i;
1680         struct printbuf buf = PRINTBUF;
1681         int ret = 0;
1682
1683         for (i = 0; i < bch2_opts_nr; i++) {
1684                 const struct bch_option *opt = &bch2_opt_table[i];
1685                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1686
1687                 if (!(opt->flags & OPT_MOUNT))
1688                         continue;
1689
1690                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1691                         continue;
1692
1693                 printbuf_reset(&buf);
1694                 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1695                                  OPT_SHOW_MOUNT_STYLE);
1696                 seq_putc(seq, ',');
1697                 seq_puts(seq, buf.buf);
1698         }
1699
1700         if (buf.allocation_failure)
1701                 ret = -ENOMEM;
1702         printbuf_exit(&buf);
1703         return ret;
1704 }
1705
1706 static void bch2_put_super(struct super_block *sb)
1707 {
1708         struct bch_fs *c = sb->s_fs_info;
1709
1710         __bch2_fs_stop(c);
1711 }
1712
1713 /*
1714  * bcachefs doesn't currently integrate intwrite freeze protection but the
1715  * internal write references serve the same purpose. Therefore reuse the
1716  * read-only transition code to perform the quiesce. The caveat is that we don't
1717  * currently have the ability to block tasks that want a write reference while
1718  * the superblock is frozen. This is fine for now, but we should either add
1719  * blocking support or find a way to integrate sb_start_intwrite() and friends.
1720  */
1721 static int bch2_freeze(struct super_block *sb)
1722 {
1723         struct bch_fs *c = sb->s_fs_info;
1724
1725         down_write(&c->state_lock);
1726         bch2_fs_read_only(c);
1727         up_write(&c->state_lock);
1728         return 0;
1729 }
1730
1731 static int bch2_unfreeze(struct super_block *sb)
1732 {
1733         struct bch_fs *c = sb->s_fs_info;
1734         int ret;
1735
1736         down_write(&c->state_lock);
1737         ret = bch2_fs_read_write(c);
1738         up_write(&c->state_lock);
1739         return ret;
1740 }
1741
1742 static const struct super_operations bch_super_operations = {
1743         .alloc_inode    = bch2_alloc_inode,
1744         .destroy_inode  = bch2_destroy_inode,
1745         .write_inode    = bch2_vfs_write_inode,
1746         .evict_inode    = bch2_evict_inode,
1747         .sync_fs        = bch2_sync_fs,
1748         .statfs         = bch2_statfs,
1749         .show_devname   = bch2_show_devname,
1750         .show_options   = bch2_show_options,
1751         .remount_fs     = bch2_remount,
1752         .put_super      = bch2_put_super,
1753         .freeze_fs      = bch2_freeze,
1754         .unfreeze_fs    = bch2_unfreeze,
1755 };
1756
1757 static int bch2_set_super(struct super_block *s, void *data)
1758 {
1759         s->s_fs_info = data;
1760         return 0;
1761 }
1762
1763 static int bch2_noset_super(struct super_block *s, void *data)
1764 {
1765         return -EBUSY;
1766 }
1767
1768 static int bch2_test_super(struct super_block *s, void *data)
1769 {
1770         struct bch_fs *c = s->s_fs_info;
1771         struct bch_fs **devs = data;
1772         unsigned i;
1773
1774         if (!c)
1775                 return false;
1776
1777         for (i = 0; devs[i]; i++)
1778                 if (c != devs[i])
1779                         return false;
1780         return true;
1781 }
1782
1783 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1784                                  int flags, const char *dev_name, void *data)
1785 {
1786         struct bch_fs *c;
1787         struct bch_dev *ca;
1788         struct super_block *sb;
1789         struct inode *vinode;
1790         struct bch_opts opts = bch2_opts_empty();
1791         char **devs;
1792         struct bch_fs **devs_to_fs = NULL;
1793         unsigned i, nr_devs;
1794         int ret;
1795
1796         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1797
1798         ret = bch2_parse_mount_opts(NULL, &opts, data);
1799         if (ret)
1800                 return ERR_PTR(ret);
1801
1802         if (!dev_name || strlen(dev_name) == 0)
1803                 return ERR_PTR(-EINVAL);
1804
1805         devs = split_devs(dev_name, &nr_devs);
1806         if (!devs)
1807                 return ERR_PTR(-ENOMEM);
1808
1809         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1810         if (!devs_to_fs) {
1811                 sb = ERR_PTR(-ENOMEM);
1812                 goto got_sb;
1813         }
1814
1815         for (i = 0; i < nr_devs; i++)
1816                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1817
1818         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1819                   flags|SB_NOSEC, devs_to_fs);
1820         if (!IS_ERR(sb))
1821                 goto got_sb;
1822
1823         c = bch2_fs_open(devs, nr_devs, opts);
1824         if (IS_ERR(c)) {
1825                 sb = ERR_CAST(c);
1826                 goto got_sb;
1827         }
1828
1829         /* Some options can't be parsed until after the fs is started: */
1830         ret = bch2_parse_mount_opts(c, &opts, data);
1831         if (ret) {
1832                 bch2_fs_stop(c);
1833                 sb = ERR_PTR(ret);
1834                 goto got_sb;
1835         }
1836
1837         bch2_opts_apply(&c->opts, opts);
1838
1839         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1840         if (IS_ERR(sb))
1841                 bch2_fs_stop(c);
1842 got_sb:
1843         kfree(devs_to_fs);
1844         kfree(devs[0]);
1845         kfree(devs);
1846
1847         if (IS_ERR(sb)) {
1848                 ret = PTR_ERR(sb);
1849                 ret = bch2_err_class(ret);
1850                 return ERR_PTR(ret);
1851         }
1852
1853         c = sb->s_fs_info;
1854
1855         if (sb->s_root) {
1856                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1857                         ret = -EBUSY;
1858                         goto err_put_super;
1859                 }
1860                 goto out;
1861         }
1862
1863         sb->s_blocksize         = block_bytes(c);
1864         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1865         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1866         sb->s_op                = &bch_super_operations;
1867         sb->s_export_op         = &bch_export_ops;
1868 #ifdef CONFIG_BCACHEFS_QUOTA
1869         sb->s_qcop              = &bch2_quotactl_operations;
1870         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1871 #endif
1872         sb->s_xattr             = bch2_xattr_handlers;
1873         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1874         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1875         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1876         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1877         c->vfs_sb               = sb;
1878         strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1879
1880         ret = super_setup_bdi(sb);
1881         if (ret)
1882                 goto err_put_super;
1883
1884         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1885
1886         for_each_online_member(ca, c, i) {
1887                 struct block_device *bdev = ca->disk_sb.bdev;
1888
1889                 /* XXX: create an anonymous device for multi device filesystems */
1890                 sb->s_bdev      = bdev;
1891                 sb->s_dev       = bdev->bd_dev;
1892                 percpu_ref_put(&ca->io_ref);
1893                 break;
1894         }
1895
1896         c->dev = sb->s_dev;
1897
1898 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1899         if (c->opts.acl)
1900                 sb->s_flags     |= SB_POSIXACL;
1901 #endif
1902
1903         sb->s_shrink->seeks = 0;
1904
1905         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
1906         ret = PTR_ERR_OR_ZERO(vinode);
1907         if (ret) {
1908                 bch_err_msg(c, ret, "mounting: error getting root inode");
1909                 goto err_put_super;
1910         }
1911
1912         sb->s_root = d_make_root(vinode);
1913         if (!sb->s_root) {
1914                 bch_err(c, "error mounting: error allocating root dentry");
1915                 ret = -ENOMEM;
1916                 goto err_put_super;
1917         }
1918
1919         sb->s_flags |= SB_ACTIVE;
1920 out:
1921         return dget(sb->s_root);
1922
1923 err_put_super:
1924         deactivate_locked_super(sb);
1925         return ERR_PTR(bch2_err_class(ret));
1926 }
1927
1928 static void bch2_kill_sb(struct super_block *sb)
1929 {
1930         struct bch_fs *c = sb->s_fs_info;
1931
1932         generic_shutdown_super(sb);
1933         bch2_fs_free(c);
1934 }
1935
1936 static struct file_system_type bcache_fs_type = {
1937         .owner          = THIS_MODULE,
1938         .name           = "bcachefs",
1939         .mount          = bch2_mount,
1940         .kill_sb        = bch2_kill_sb,
1941         .fs_flags       = FS_REQUIRES_DEV,
1942 };
1943
1944 MODULE_ALIAS_FS("bcachefs");
1945
1946 void bch2_vfs_exit(void)
1947 {
1948         unregister_filesystem(&bcache_fs_type);
1949         kmem_cache_destroy(bch2_inode_cache);
1950 }
1951
1952 int __init bch2_vfs_init(void)
1953 {
1954         int ret = -ENOMEM;
1955
1956         bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
1957         if (!bch2_inode_cache)
1958                 goto err;
1959
1960         ret = register_filesystem(&bcache_fs_type);
1961         if (ret)
1962                 goto err;
1963
1964         return 0;
1965 err:
1966         bch2_vfs_exit();
1967         return ret;
1968 }
1969
1970 #endif /* NO_BCACHEFS_FS */