fs/bcachefs/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_subvol_is_ro_trans(trans, dir->ei_subvol) ?:
 262                 bch2_create_trans(trans,
 263                                   inode_inum(dir), &dir_u, &inode_u,
 264                                   !(flags & BCH_CREATE_TMPFILE)
 265                                   ? &dentry->d_name : NULL,
 266                                   from_kuid(i_user_ns(&dir->v), current_fsuid()),
 267                                   from_kgid(i_user_ns(&dir->v), current_fsgid()),
 268                                   mode, rdev,
 269                                   default_acl, acl, snapshot_src, flags) ?:
 270                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 271                                 KEY_TYPE_QUOTA_PREALLOC);
 272         if (unlikely(ret))
 273                 goto err_before_quota;
 274
 275         inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
 276         inum.inum = inode_u.bi_inum;
 277
 278         ret   = bch2_subvolume_get(trans, inum.subvol, true,
 279                                    BTREE_ITER_WITH_UPDATES, &subvol) ?:
 280                 bch2_trans_commit(trans, NULL, &journal_seq, 0);
 281         if (unlikely(ret)) {
 282                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 283                                 KEY_TYPE_QUOTA_WARN);
 284 err_before_quota:
 285                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 286                         goto retry;
 287                 goto err_trans;
 288         }
 289
 290         if (!(flags & BCH_CREATE_TMPFILE)) {
 291                 bch2_inode_update_after_write(trans, dir, &dir_u,
 292                                               ATTR_MTIME|ATTR_CTIME);
 293                 mutex_unlock(&dir->ei_update_lock);
 294         }
 295
 296         bch2_iget5_set(&inode->v, &inum);
 297         bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 298
 299         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 300         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 301
 302         /*
 303          * we must insert the new inode into the inode cache before calling
 304          * bch2_trans_exit() and dropping locks, else we could race with another
 305          * thread pulling the inode in and modifying it:
 306          */
 307
 308         inode->v.i_state |= I_CREATING;
 309
 310         old = to_bch_ei(inode_insert5(&inode->v,
 311                                       bch2_inode_hash(inum),
 312                                       bch2_iget5_test,
 313                                       bch2_iget5_set,
 314                                       &inum));
 315         BUG_ON(!old);
 316
 317         if (unlikely(old != inode)) {
 318                 /*
 319                  * We raced, another process pulled the new inode into cache
 320                  * before us:
 321                  */
 322                 make_bad_inode(&inode->v);
 323                 iput(&inode->v);
 324
 325                 inode = old;
 326         } else {
 327                 mutex_lock(&c->vfs_inodes_lock);
 328                 list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
 329                 mutex_unlock(&c->vfs_inodes_lock);
 330                 /*
 331                  * we really don't want insert_inode_locked2() to be setting
 332                  * I_NEW...
 333                  */
 334                 unlock_new_inode(&inode->v);
 335         }
 336
 337         bch2_trans_put(trans);
 338 err:
 339         posix_acl_release(default_acl);
 340         posix_acl_release(acl);
 341         return inode;
 342 err_trans:
 343         if (!(flags & BCH_CREATE_TMPFILE))
 344                 mutex_unlock(&dir->ei_update_lock);
 345
 346         bch2_trans_put(trans);
 347         make_bad_inode(&inode->v);
 348         iput(&inode->v);
 349         inode = ERR_PTR(ret);
 350         goto err;
 351 }
 352
 353 /* methods */
 354
 355 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 356                                   unsigned int flags)
 357 {
 358         struct bch_fs *c = vdir->i_sb->s_fs_info;
 359         struct bch_inode_info *dir = to_bch_ei(vdir);
 360         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 361         struct inode *vinode = NULL;
 362         subvol_inum inum = { .subvol = 1 };
 363         int ret;
 364
 365         ret = bch2_dirent_lookup(c, inode_inum(dir), &hash,
 366                                  &dentry->d_name, &inum);
 367
 368         if (!ret)
 369                 vinode = bch2_vfs_inode_get(c, inum);
 370
 371         return d_splice_alias(vinode, dentry);
 372 }
 373
 374 static int bch2_mknod(struct mnt_idmap *idmap,
 375                       struct inode *vdir, struct dentry *dentry,
 376                       umode_t mode, dev_t rdev)
 377 {
 378         struct bch_inode_info *inode =
 379                 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
 380                               (subvol_inum) { 0 }, 0);
 381
 382         if (IS_ERR(inode))
 383                 return bch2_err_class(PTR_ERR(inode));
 384
 385         d_instantiate(dentry, &inode->v);
 386         return 0;
 387 }
 388
 389 static int bch2_create(struct mnt_idmap *idmap,
 390                        struct inode *vdir, struct dentry *dentry,
 391                        umode_t mode, bool excl)
 392 {
 393         return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
 394 }
 395
 396 static int __bch2_link(struct bch_fs *c,
 397                        struct bch_inode_info *inode,
 398                        struct bch_inode_info *dir,
 399                        struct dentry *dentry)
 400 {
 401         struct btree_trans *trans = bch2_trans_get(c);
 402         struct bch_inode_unpacked dir_u, inode_u;
 403         int ret;
 404
 405         mutex_lock(&inode->ei_update_lock);
 406
 407         ret = commit_do(trans, NULL, NULL, 0,
 408                         bch2_link_trans(trans,
 409                                         inode_inum(dir),   &dir_u,
 410                                         inode_inum(inode), &inode_u,
 411                                         &dentry->d_name));
 412
 413         if (likely(!ret)) {
 414                 bch2_inode_update_after_write(trans, dir, &dir_u,
 415                                               ATTR_MTIME|ATTR_CTIME);
 416                 bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
 417         }
 418
 419         bch2_trans_put(trans);
 420         mutex_unlock(&inode->ei_update_lock);
 421         return ret;
 422 }
 423
 424 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 425                      struct dentry *dentry)
 426 {
 427         struct bch_fs *c = vdir->i_sb->s_fs_info;
 428         struct bch_inode_info *dir = to_bch_ei(vdir);
 429         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 430         int ret;
 431
 432         lockdep_assert_held(&inode->v.i_rwsem);
 433
 434         ret   = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
 435                 bch2_subvol_is_ro(c, inode->ei_subvol) ?:
 436                 __bch2_link(c, inode, dir, dentry);
 437         if (unlikely(ret))
 438                 return bch2_err_class(ret);
 439
 440         ihold(&inode->v);
 441         d_instantiate(dentry, &inode->v);
 442         return 0;
 443 }
 444
 445 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 446                   bool deleting_snapshot)
 447 {
 448         struct bch_fs *c = vdir->i_sb->s_fs_info;
 449         struct bch_inode_info *dir = to_bch_ei(vdir);
 450         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 451         struct bch_inode_unpacked dir_u, inode_u;
 452         struct btree_trans *trans = bch2_trans_get(c);
 453         int ret;
 454
 455         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 456
 457         ret = commit_do(trans, NULL, NULL,
 458                         BCH_TRANS_COMMIT_no_enospc,
 459                 bch2_unlink_trans(trans,
 460                                   inode_inum(dir), &dir_u,
 461                                   &inode_u, &dentry->d_name,
 462                                   deleting_snapshot));
 463         if (unlikely(ret))
 464                 goto err;
 465
 466         bch2_inode_update_after_write(trans, dir, &dir_u,
 467                                       ATTR_MTIME|ATTR_CTIME);
 468         bch2_inode_update_after_write(trans, inode, &inode_u,
 469                                       ATTR_MTIME);
 470
 471         if (inode_u.bi_subvol) {
 472                 /*
 473                  * Subvolume deletion is asynchronous, but we still want to tell
 474                  * the VFS that it's been deleted here:
 475                  */
 476                 set_nlink(&inode->v, 0);
 477         }
 478 err:
 479         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 480         bch2_trans_put(trans);
 481
 482         return ret;
 483 }
 484
 485 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 486 {
 487         struct bch_inode_info *dir= to_bch_ei(vdir);
 488         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 489
 490         int ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
 491                 __bch2_unlink(vdir, dentry, false);
 492         return bch2_err_class(ret);
 493 }
 494
 495 static int bch2_symlink(struct mnt_idmap *idmap,
 496                         struct inode *vdir, struct dentry *dentry,
 497                         const char *symname)
 498 {
 499         struct bch_fs *c = vdir->i_sb->s_fs_info;
 500         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 501         int ret;
 502
 503         inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
 504                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 505         if (IS_ERR(inode))
 506                 return bch2_err_class(PTR_ERR(inode));
 507
 508         inode_lock(&inode->v);
 509         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 510         inode_unlock(&inode->v);
 511
 512         if (unlikely(ret))
 513                 goto err;
 514
 515         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 516         if (unlikely(ret))
 517                 goto err;
 518
 519         ret = __bch2_link(c, inode, dir, dentry);
 520         if (unlikely(ret))
 521                 goto err;
 522
 523         d_instantiate(dentry, &inode->v);
 524         return 0;
 525 err:
 526         iput(&inode->v);
 527         return bch2_err_class(ret);
 528 }
 529
 530 static int bch2_mkdir(struct mnt_idmap *idmap,
 531                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 532 {
 533         return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
 534 }
 535
 536 static int bch2_rename2(struct mnt_idmap *idmap,
 537                         struct inode *src_vdir, struct dentry *src_dentry,
 538                         struct inode *dst_vdir, struct dentry *dst_dentry,
 539                         unsigned flags)
 540 {
 541         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 542         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 543         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 544         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 545         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 546         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 547         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 548         struct btree_trans *trans;
 549         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 550                 ? BCH_RENAME_EXCHANGE
 551                 : dst_dentry->d_inode
 552                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 553         int ret;
 554
 555         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 556                 return -EINVAL;
 557
 558         if (mode == BCH_RENAME_OVERWRITE) {
 559                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 560                                                    0, LLONG_MAX);
 561                 if (ret)
 562                         return ret;
 563         }
 564
 565         trans = bch2_trans_get(c);
 566
 567         bch2_lock_inodes(INODE_UPDATE_LOCK,
 568                          src_dir,
 569                          dst_dir,
 570                          src_inode,
 571                          dst_inode);
 572
 573         ret   = bch2_subvol_is_ro_trans(trans, src_dir->ei_subvol) ?:
 574                 bch2_subvol_is_ro_trans(trans, dst_dir->ei_subvol);
 575         if (ret)
 576                 goto err;
 577
 578         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 579                 ret = bch2_fs_quota_transfer(c, src_inode,
 580                                              dst_dir->ei_qid,
 581                                              1 << QTYP_PRJ,
 582                                              KEY_TYPE_QUOTA_PREALLOC);
 583                 if (ret)
 584                         goto err;
 585         }
 586
 587         if (mode == BCH_RENAME_EXCHANGE &&
 588             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 589                 ret = bch2_fs_quota_transfer(c, dst_inode,
 590                                              src_dir->ei_qid,
 591                                              1 << QTYP_PRJ,
 592                                              KEY_TYPE_QUOTA_PREALLOC);
 593                 if (ret)
 594                         goto err;
 595         }
 596
 597         ret = commit_do(trans, NULL, NULL, 0,
 598                         bch2_rename_trans(trans,
 599                                           inode_inum(src_dir), &src_dir_u,
 600                                           inode_inum(dst_dir), &dst_dir_u,
 601                                           &src_inode_u,
 602                                           &dst_inode_u,
 603                                           &src_dentry->d_name,
 604                                           &dst_dentry->d_name,
 605                                           mode));
 606         if (unlikely(ret))
 607                 goto err;
 608
 609         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 610         BUG_ON(dst_inode &&
 611                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 612
 613         bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
 614                                       ATTR_MTIME|ATTR_CTIME);
 615
 616         if (src_dir != dst_dir)
 617                 bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
 618                                               ATTR_MTIME|ATTR_CTIME);
 619
 620         bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
 621                                       ATTR_CTIME);
 622
 623         if (dst_inode)
 624                 bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
 625                                               ATTR_CTIME);
 626 err:
 627         bch2_trans_put(trans);
 628
 629         bch2_fs_quota_transfer(c, src_inode,
 630                                bch_qid(&src_inode->ei_inode),
 631                                1 << QTYP_PRJ,
 632                                KEY_TYPE_QUOTA_NOCHECK);
 633         if (dst_inode)
 634                 bch2_fs_quota_transfer(c, dst_inode,
 635                                        bch_qid(&dst_inode->ei_inode),
 636                                        1 << QTYP_PRJ,
 637                                        KEY_TYPE_QUOTA_NOCHECK);
 638
 639         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 640                            src_dir,
 641                            dst_dir,
 642                            src_inode,
 643                            dst_inode);
 644
 645         return bch2_err_class(ret);
 646 }
 647
 648 static void bch2_setattr_copy(struct mnt_idmap *idmap,
 649                               struct bch_inode_info *inode,
 650                               struct bch_inode_unpacked *bi,
 651                               struct iattr *attr)
 652 {
 653         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 654         unsigned int ia_valid = attr->ia_valid;
 655
 656         if (ia_valid & ATTR_UID)
 657                 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 658         if (ia_valid & ATTR_GID)
 659                 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 660
 661         if (ia_valid & ATTR_SIZE)
 662                 bi->bi_size = attr->ia_size;
 663
 664         if (ia_valid & ATTR_ATIME)
 665                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 666         if (ia_valid & ATTR_MTIME)
 667                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 668         if (ia_valid & ATTR_CTIME)
 669                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 670
 671         if (ia_valid & ATTR_MODE) {
 672                 umode_t mode = attr->ia_mode;
 673                 kgid_t gid = ia_valid & ATTR_GID
 674                         ? attr->ia_gid
 675                         : inode->v.i_gid;
 676
 677                 if (!in_group_p(gid) &&
 678                     !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
 679                         mode &= ~S_ISGID;
 680                 bi->bi_mode = mode;
 681         }
 682 }
 683
 684 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
 685                          struct bch_inode_info *inode,
 686                          struct iattr *attr)
 687 {
 688         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 689         struct bch_qid qid;
 690         struct btree_trans *trans;
 691         struct btree_iter inode_iter = { NULL };
 692         struct bch_inode_unpacked inode_u;
 693         struct posix_acl *acl = NULL;
 694         int ret;
 695
 696         mutex_lock(&inode->ei_update_lock);
 697
 698         qid = inode->ei_qid;
 699
 700         if (attr->ia_valid & ATTR_UID)
 701                 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 702
 703         if (attr->ia_valid & ATTR_GID)
 704                 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 705
 706         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 707                                      KEY_TYPE_QUOTA_PREALLOC);
 708         if (ret)
 709                 goto err;
 710
 711         trans = bch2_trans_get(c);
 712 retry:
 713         bch2_trans_begin(trans);
 714         kfree(acl);
 715         acl = NULL;
 716
 717         ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
 718                               BTREE_ITER_INTENT);
 719         if (ret)
 720                 goto btree_err;
 721
 722         bch2_setattr_copy(idmap, inode, &inode_u, attr);
 723
 724         if (attr->ia_valid & ATTR_MODE) {
 725                 ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
 726                                      inode_u.bi_mode, &acl);
 727                 if (ret)
 728                         goto btree_err;
 729         }
 730
 731         ret =   bch2_inode_write(trans, &inode_iter, &inode_u) ?:
 732                 bch2_trans_commit(trans, NULL, NULL,
 733                                   BCH_TRANS_COMMIT_no_enospc);
 734 btree_err:
 735         bch2_trans_iter_exit(trans, &inode_iter);
 736
 737         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 738                 goto retry;
 739         if (unlikely(ret))
 740                 goto err_trans;
 741
 742         bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
 743
 744         if (acl)
 745                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 746 err_trans:
 747         bch2_trans_put(trans);
 748 err:
 749         mutex_unlock(&inode->ei_update_lock);
 750
 751         return bch2_err_class(ret);
 752 }
 753
 754 static int bch2_getattr(struct mnt_idmap *idmap,
 755                         const struct path *path, struct kstat *stat,
 756                         u32 request_mask, unsigned query_flags)
 757 {
 758         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 759         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 760
 761         stat->dev       = inode->v.i_sb->s_dev;
 762         stat->ino       = inode->v.i_ino;
 763         stat->mode      = inode->v.i_mode;
 764         stat->nlink     = inode->v.i_nlink;
 765         stat->uid       = inode->v.i_uid;
 766         stat->gid       = inode->v.i_gid;
 767         stat->rdev      = inode->v.i_rdev;
 768         stat->size      = i_size_read(&inode->v);
 769         stat->atime     = inode_get_atime(&inode->v);
 770         stat->mtime     = inode_get_mtime(&inode->v);
 771         stat->ctime     = inode_get_ctime(&inode->v);
 772         stat->blksize   = block_bytes(c);
 773         stat->blocks    = inode->v.i_blocks;
 774
 775         if (request_mask & STATX_BTIME) {
 776                 stat->result_mask |= STATX_BTIME;
 777                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 778         }
 779
 780         if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
 781                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 782         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 783
 784         if (inode->ei_inode.bi_flags & BCH_INODE_append)
 785                 stat->attributes |= STATX_ATTR_APPEND;
 786         stat->attributes_mask    |= STATX_ATTR_APPEND;
 787
 788         if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
 789                 stat->attributes |= STATX_ATTR_NODUMP;
 790         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 791
 792         return 0;
 793 }
 794
 795 static int bch2_setattr(struct mnt_idmap *idmap,
 796                         struct dentry *dentry, struct iattr *iattr)
 797 {
 798         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 799         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 800         int ret;
 801
 802         lockdep_assert_held(&inode->v.i_rwsem);
 803
 804         ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
 805                 setattr_prepare(idmap, dentry, iattr);
 806         if (ret)
 807                 return ret;
 808
 809         return iattr->ia_valid & ATTR_SIZE
 810                 ? bchfs_truncate(idmap, inode, iattr)
 811                 : bch2_setattr_nonsize(idmap, inode, iattr);
 812 }
 813
 814 static int bch2_tmpfile(struct mnt_idmap *idmap,
 815                         struct inode *vdir, struct file *file, umode_t mode)
 816 {
 817         struct bch_inode_info *inode =
 818                 __bch2_create(idmap, to_bch_ei(vdir),
 819                               file->f_path.dentry, mode, 0,
 820                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 821
 822         if (IS_ERR(inode))
 823                 return bch2_err_class(PTR_ERR(inode));
 824
 825         d_mark_tmpfile(file, &inode->v);
 826         d_instantiate(file->f_path.dentry, &inode->v);
 827         return finish_open_simple(file, 0);
 828 }
 829
 830 static int bch2_fill_extent(struct bch_fs *c,
 831                             struct fiemap_extent_info *info,
 832                             struct bkey_s_c k, unsigned flags)
 833 {
 834         if (bkey_extent_is_direct_data(k.k)) {
 835                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 836                 const union bch_extent_entry *entry;
 837                 struct extent_ptr_decoded p;
 838                 int ret;
 839
 840                 if (k.k->type == KEY_TYPE_reflink_v)
 841                         flags |= FIEMAP_EXTENT_SHARED;
 842
 843                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 844                         int flags2 = 0;
 845                         u64 offset = p.ptr.offset;
 846
 847                         if (p.ptr.unwritten)
 848                                 flags2 |= FIEMAP_EXTENT_UNWRITTEN;
 849
 850                         if (p.crc.compression_type)
 851                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 852                         else
 853                                 offset += p.crc.offset;
 854
 855                         if ((offset & (block_sectors(c) - 1)) ||
 856                             (k.k->size & (block_sectors(c) - 1)))
 857                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 858
 859                         ret = fiemap_fill_next_extent(info,
 860                                                 bkey_start_offset(k.k) << 9,
 861                                                 offset << 9,
 862                                                 k.k->size << 9, flags|flags2);
 863                         if (ret)
 864                                 return ret;
 865                 }
 866
 867                 return 0;
 868         } else if (bkey_extent_is_inline_data(k.k)) {
 869                 return fiemap_fill_next_extent(info,
 870                                                bkey_start_offset(k.k) << 9,
 871                                                0, k.k->size << 9,
 872                                                flags|
 873                                                FIEMAP_EXTENT_DATA_INLINE);
 874         } else if (k.k->type == KEY_TYPE_reservation) {
 875                 return fiemap_fill_next_extent(info,
 876                                                bkey_start_offset(k.k) << 9,
 877                                                0, k.k->size << 9,
 878                                                flags|
 879                                                FIEMAP_EXTENT_DELALLOC|
 880                                                FIEMAP_EXTENT_UNWRITTEN);
 881         } else {
 882                 BUG();
 883         }
 884 }
 885
 886 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 887                        u64 start, u64 len)
 888 {
 889         struct bch_fs *c = vinode->i_sb->s_fs_info;
 890         struct bch_inode_info *ei = to_bch_ei(vinode);
 891         struct btree_trans *trans;
 892         struct btree_iter iter;
 893         struct bkey_s_c k;
 894         struct bkey_buf cur, prev;
 895         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 896         unsigned offset_into_extent, sectors;
 897         bool have_extent = false;
 898         u32 snapshot;
 899         int ret = 0;
 900
 901         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 902         if (ret)
 903                 return ret;
 904
 905         if (start + len < start)
 906                 return -EINVAL;
 907
 908         start >>= 9;
 909
 910         bch2_bkey_buf_init(&cur);
 911         bch2_bkey_buf_init(&prev);
 912         trans = bch2_trans_get(c);
 913 retry:
 914         bch2_trans_begin(trans);
 915
 916         ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
 917         if (ret)
 918                 goto err;
 919
 920         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 921                              SPOS(ei->v.i_ino, start, snapshot), 0);
 922
 923         while (!(ret = btree_trans_too_many_iters(trans)) &&
 924                (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
 925                !(ret = bkey_err(k))) {
 926                 enum btree_id data_btree = BTREE_ID_extents;
 927
 928                 if (!bkey_extent_is_data(k.k) &&
 929                     k.k->type != KEY_TYPE_reservation) {
 930                         bch2_btree_iter_advance(&iter);
 931                         continue;
 932                 }
 933
 934                 offset_into_extent      = iter.pos.offset -
 935                         bkey_start_offset(k.k);
 936                 sectors                 = k.k->size - offset_into_extent;
 937
 938                 bch2_bkey_buf_reassemble(&cur, c, k);
 939
 940                 ret = bch2_read_indirect_extent(trans, &data_btree,
 941                                         &offset_into_extent, &cur);
 942                 if (ret)
 943                         break;
 944
 945                 k = bkey_i_to_s_c(cur.k);
 946                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 947
 948                 sectors = min(sectors, k.k->size - offset_into_extent);
 949
 950                 bch2_cut_front(POS(k.k->p.inode,
 951                                    bkey_start_offset(k.k) +
 952                                    offset_into_extent),
 953                                cur.k);
 954                 bch2_key_resize(&cur.k->k, sectors);
 955                 cur.k->k.p = iter.pos;
 956                 cur.k->k.p.offset += cur.k->k.size;
 957
 958                 if (have_extent) {
 959                         bch2_trans_unlock(trans);
 960                         ret = bch2_fill_extent(c, info,
 961                                         bkey_i_to_s_c(prev.k), 0);
 962                         if (ret)
 963                                 break;
 964                 }
 965
 966                 bkey_copy(prev.k, cur.k);
 967                 have_extent = true;
 968
 969                 bch2_btree_iter_set_pos(&iter,
 970                         POS(iter.pos.inode, iter.pos.offset + sectors));
 971         }
 972         start = iter.pos.offset;
 973         bch2_trans_iter_exit(trans, &iter);
 974 err:
 975         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 976                 goto retry;
 977
 978         if (!ret && have_extent) {
 979                 bch2_trans_unlock(trans);
 980                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 981                                        FIEMAP_EXTENT_LAST);
 982         }
 983
 984         bch2_trans_put(trans);
 985         bch2_bkey_buf_exit(&cur, c);
 986         bch2_bkey_buf_exit(&prev, c);
 987         return ret < 0 ? ret : 0;
 988 }
 989
 990 static const struct vm_operations_struct bch_vm_ops = {
 991         .fault          = bch2_page_fault,
 992         .map_pages      = filemap_map_pages,
 993         .page_mkwrite   = bch2_page_mkwrite,
 994 };
 995
 996 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 997 {
 998         file_accessed(file);
 999
1000         vma->vm_ops = &bch_vm_ops;
1001         return 0;
1002 }
1003
1004 /* Directories: */
1005
1006 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1007 {
1008         return generic_file_llseek_size(file, offset, whence,
1009                                         S64_MAX, S64_MAX);
1010 }
1011
1012 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1013 {
1014         struct bch_inode_info *inode = file_bch_inode(file);
1015         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1016
1017         if (!dir_emit_dots(file, ctx))
1018                 return 0;
1019
1020         int ret = bch2_readdir(c, inode_inum(inode), ctx);
1021
1022         bch_err_fn(c, ret);
1023         return bch2_err_class(ret);
1024 }
1025
1026 static int bch2_open(struct inode *vinode, struct file *file)
1027 {
1028         if (file->f_flags & (O_WRONLY|O_RDWR)) {
1029                 struct bch_inode_info *inode = to_bch_ei(vinode);
1030                 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1031
1032                 int ret = bch2_subvol_is_ro(c, inode->ei_subvol);
1033                 if (ret)
1034                         return ret;
1035         }
1036
1037         return generic_file_open(vinode, file);
1038 }
1039
1040 static const struct file_operations bch_file_operations = {
1041         .open           = bch2_open,
1042         .llseek         = bch2_llseek,
1043         .read_iter      = bch2_read_iter,
1044         .write_iter     = bch2_write_iter,
1045         .mmap           = bch2_mmap,
1046         .fsync          = bch2_fsync,
1047         .splice_read    = filemap_splice_read,
1048         .splice_write   = iter_file_splice_write,
1049         .fallocate      = bch2_fallocate_dispatch,
1050         .unlocked_ioctl = bch2_fs_file_ioctl,
1051 #ifdef CONFIG_COMPAT
1052         .compat_ioctl   = bch2_compat_fs_ioctl,
1053 #endif
1054         .remap_file_range = bch2_remap_file_range,
1055 };
1056
1057 static const struct inode_operations bch_file_inode_operations = {
1058         .getattr        = bch2_getattr,
1059         .setattr        = bch2_setattr,
1060         .fiemap         = bch2_fiemap,
1061         .listxattr      = bch2_xattr_list,
1062 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1063         .get_acl        = bch2_get_acl,
1064         .set_acl        = bch2_set_acl,
1065 #endif
1066 };
1067
1068 static const struct inode_operations bch_dir_inode_operations = {
1069         .lookup         = bch2_lookup,
1070         .create         = bch2_create,
1071         .link           = bch2_link,
1072         .unlink         = bch2_unlink,
1073         .symlink        = bch2_symlink,
1074         .mkdir          = bch2_mkdir,
1075         .rmdir          = bch2_unlink,
1076         .mknod          = bch2_mknod,
1077         .rename         = bch2_rename2,
1078         .getattr        = bch2_getattr,
1079         .setattr        = bch2_setattr,
1080         .tmpfile        = bch2_tmpfile,
1081         .listxattr      = bch2_xattr_list,
1082 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1083         .get_acl        = bch2_get_acl,
1084         .set_acl        = bch2_set_acl,
1085 #endif
1086 };
1087
1088 static const struct file_operations bch_dir_file_operations = {
1089         .llseek         = bch2_dir_llseek,
1090         .read           = generic_read_dir,
1091         .iterate_shared = bch2_vfs_readdir,
1092         .fsync          = bch2_fsync,
1093         .unlocked_ioctl = bch2_fs_file_ioctl,
1094 #ifdef CONFIG_COMPAT
1095         .compat_ioctl   = bch2_compat_fs_ioctl,
1096 #endif
1097 };
1098
1099 static const struct inode_operations bch_symlink_inode_operations = {
1100         .get_link       = page_get_link,
1101         .getattr        = bch2_getattr,
1102         .setattr        = bch2_setattr,
1103         .listxattr      = bch2_xattr_list,
1104 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1105         .get_acl        = bch2_get_acl,
1106         .set_acl        = bch2_set_acl,
1107 #endif
1108 };
1109
1110 static const struct inode_operations bch_special_inode_operations = {
1111         .getattr        = bch2_getattr,
1112         .setattr        = bch2_setattr,
1113         .listxattr      = bch2_xattr_list,
1114 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1115         .get_acl        = bch2_get_acl,
1116         .set_acl        = bch2_set_acl,
1117 #endif
1118 };
1119
1120 static const struct address_space_operations bch_address_space_operations = {
1121         .read_folio     = bch2_read_folio,
1122         .writepages     = bch2_writepages,
1123         .readahead      = bch2_readahead,
1124         .dirty_folio    = filemap_dirty_folio,
1125         .write_begin    = bch2_write_begin,
1126         .write_end      = bch2_write_end,
1127         .invalidate_folio = bch2_invalidate_folio,
1128         .release_folio  = bch2_release_folio,
1129         .direct_IO      = noop_direct_IO,
1130 #ifdef CONFIG_MIGRATION
1131         .migrate_folio  = filemap_migrate_folio,
1132 #endif
1133         .error_remove_folio = generic_error_remove_folio,
1134 };
1135
1136 struct bcachefs_fid {
1137         u64             inum;
1138         u32             subvol;
1139         u32             gen;
1140 } __packed;
1141
1142 struct bcachefs_fid_with_parent {
1143         struct bcachefs_fid     fid;
1144         struct bcachefs_fid     dir;
1145 } __packed;
1146
1147 static int bcachefs_fid_valid(int fh_len, int fh_type)
1148 {
1149         switch (fh_type) {
1150         case FILEID_BCACHEFS_WITHOUT_PARENT:
1151                 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1152         case FILEID_BCACHEFS_WITH_PARENT:
1153                 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1154         default:
1155                 return false;
1156         }
1157 }
1158
1159 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1160 {
1161         return (struct bcachefs_fid) {
1162                 .inum   = inode->ei_inode.bi_inum,
1163                 .subvol = inode->ei_subvol,
1164                 .gen    = inode->ei_inode.bi_generation,
1165         };
1166 }
1167
1168 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1169                           struct inode *vdir)
1170 {
1171         struct bch_inode_info *inode    = to_bch_ei(vinode);
1172         struct bch_inode_info *dir      = to_bch_ei(vdir);
1173         int min_len;
1174
1175         if (!S_ISDIR(inode->v.i_mode) && dir) {
1176                 struct bcachefs_fid_with_parent *fid = (void *) fh;
1177
1178                 min_len = sizeof(*fid) / sizeof(u32);
1179                 if (*len < min_len) {
1180                         *len = min_len;
1181                         return FILEID_INVALID;
1182                 }
1183
1184                 fid->fid = bch2_inode_to_fid(inode);
1185                 fid->dir = bch2_inode_to_fid(dir);
1186
1187                 *len = min_len;
1188                 return FILEID_BCACHEFS_WITH_PARENT;
1189         } else {
1190                 struct bcachefs_fid *fid = (void *) fh;
1191
1192                 min_len = sizeof(*fid) / sizeof(u32);
1193                 if (*len < min_len) {
1194                         *len = min_len;
1195                         return FILEID_INVALID;
1196                 }
1197                 *fid = bch2_inode_to_fid(inode);
1198
1199                 *len = min_len;
1200                 return FILEID_BCACHEFS_WITHOUT_PARENT;
1201         }
1202 }
1203
1204 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1205                                         struct bcachefs_fid fid)
1206 {
1207         struct bch_fs *c = sb->s_fs_info;
1208         struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1209                                     .subvol = fid.subvol,
1210                                     .inum = fid.inum,
1211         });
1212         if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1213                 iput(vinode);
1214                 vinode = ERR_PTR(-ESTALE);
1215         }
1216         return vinode;
1217 }
1218
1219 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1220                 int fh_len, int fh_type)
1221 {
1222         struct bcachefs_fid *fid = (void *) _fid;
1223
1224         if (!bcachefs_fid_valid(fh_len, fh_type))
1225                 return NULL;
1226
1227         return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1228 }
1229
1230 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1231                 int fh_len, int fh_type)
1232 {
1233         struct bcachefs_fid_with_parent *fid = (void *) _fid;
1234
1235         if (!bcachefs_fid_valid(fh_len, fh_type) ||
1236             fh_type != FILEID_BCACHEFS_WITH_PARENT)
1237                 return NULL;
1238
1239         return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1240 }
1241
1242 static struct dentry *bch2_get_parent(struct dentry *child)
1243 {
1244         struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1245         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1246         subvol_inum parent_inum = {
1247                 .subvol = inode->ei_inode.bi_parent_subvol ?:
1248                         inode->ei_subvol,
1249                 .inum = inode->ei_inode.bi_dir,
1250         };
1251
1252         return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1253 }
1254
1255 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1256 {
1257         struct bch_inode_info *inode    = to_bch_ei(child->d_inode);
1258         struct bch_inode_info *dir      = to_bch_ei(parent->d_inode);
1259         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1260         struct btree_trans *trans;
1261         struct btree_iter iter1;
1262         struct btree_iter iter2;
1263         struct bkey_s_c k;
1264         struct bkey_s_c_dirent d;
1265         struct bch_inode_unpacked inode_u;
1266         subvol_inum target;
1267         u32 snapshot;
1268         struct qstr dirent_name;
1269         unsigned name_len = 0;
1270         int ret;
1271
1272         if (!S_ISDIR(dir->v.i_mode))
1273                 return -EINVAL;
1274
1275         trans = bch2_trans_get(c);
1276
1277         bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1278                              POS(dir->ei_inode.bi_inum, 0), 0);
1279         bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1280                              POS(dir->ei_inode.bi_inum, 0), 0);
1281 retry:
1282         bch2_trans_begin(trans);
1283
1284         ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1285         if (ret)
1286                 goto err;
1287
1288         bch2_btree_iter_set_snapshot(&iter1, snapshot);
1289         bch2_btree_iter_set_snapshot(&iter2, snapshot);
1290
1291         ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1292         if (ret)
1293                 goto err;
1294
1295         if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1296                 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1297
1298                 k = bch2_btree_iter_peek_slot(&iter1);
1299                 ret = bkey_err(k);
1300                 if (ret)
1301                         goto err;
1302
1303                 if (k.k->type != KEY_TYPE_dirent) {
1304                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1305                         goto err;
1306                 }
1307
1308                 d = bkey_s_c_to_dirent(k);
1309                 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1310                 if (ret > 0)
1311                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1312                 if (ret)
1313                         goto err;
1314
1315                 if (target.subvol       == inode->ei_subvol &&
1316                     target.inum         == inode->ei_inode.bi_inum)
1317                         goto found;
1318         } else {
1319                 /*
1320                  * File with multiple hardlinks and our backref is to the wrong
1321                  * directory - linear search:
1322                  */
1323                 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1324                         if (k.k->p.inode > dir->ei_inode.bi_inum)
1325                                 break;
1326
1327                         if (k.k->type != KEY_TYPE_dirent)
1328                                 continue;
1329
1330                         d = bkey_s_c_to_dirent(k);
1331                         ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1332                         if (ret < 0)
1333                                 break;
1334                         if (ret)
1335                                 continue;
1336
1337                         if (target.subvol       == inode->ei_subvol &&
1338                             target.inum         == inode->ei_inode.bi_inum)
1339                                 goto found;
1340                 }
1341         }
1342
1343         ret = -ENOENT;
1344         goto err;
1345 found:
1346         dirent_name = bch2_dirent_get_name(d);
1347
1348         name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1349         memcpy(name, dirent_name.name, name_len);
1350         name[name_len] = '\0';
1351 err:
1352         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1353                 goto retry;
1354
1355         bch2_trans_iter_exit(trans, &iter1);
1356         bch2_trans_iter_exit(trans, &iter2);
1357         bch2_trans_put(trans);
1358
1359         return ret;
1360 }
1361
1362 static const struct export_operations bch_export_ops = {
1363         .encode_fh      = bch2_encode_fh,
1364         .fh_to_dentry   = bch2_fh_to_dentry,
1365         .fh_to_parent   = bch2_fh_to_parent,
1366         .get_parent     = bch2_get_parent,
1367         .get_name       = bch2_get_name,
1368 };
1369
1370 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1371                                 struct bch_inode_info *inode,
1372                                 struct bch_inode_unpacked *bi,
1373                                 struct bch_subvolume *subvol)
1374 {
1375         bch2_inode_update_after_write(trans, inode, bi, ~0);
1376
1377         if (BCH_SUBVOLUME_SNAP(subvol))
1378                 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1379         else
1380                 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1381
1382         inode->v.i_blocks       = bi->bi_sectors;
1383         inode->v.i_ino          = bi->bi_inum;
1384         inode->v.i_rdev         = bi->bi_dev;
1385         inode->v.i_generation   = bi->bi_generation;
1386         inode->v.i_size         = bi->bi_size;
1387
1388         inode->ei_flags         = 0;
1389         inode->ei_quota_reserved = 0;
1390         inode->ei_qid           = bch_qid(bi);
1391         inode->ei_subvol        = inum.subvol;
1392
1393         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1394
1395         switch (inode->v.i_mode & S_IFMT) {
1396         case S_IFREG:
1397                 inode->v.i_op   = &bch_file_inode_operations;
1398                 inode->v.i_fop  = &bch_file_operations;
1399                 break;
1400         case S_IFDIR:
1401                 inode->v.i_op   = &bch_dir_inode_operations;
1402                 inode->v.i_fop  = &bch_dir_file_operations;
1403                 break;
1404         case S_IFLNK:
1405                 inode_nohighmem(&inode->v);
1406                 inode->v.i_op   = &bch_symlink_inode_operations;
1407                 break;
1408         default:
1409                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1410                 inode->v.i_op   = &bch_special_inode_operations;
1411                 break;
1412         }
1413
1414         mapping_set_large_folios(inode->v.i_mapping);
1415 }
1416
1417 static struct inode *bch2_alloc_inode(struct super_block *sb)
1418 {
1419         struct bch_inode_info *inode;
1420
1421         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1422         if (!inode)
1423                 return NULL;
1424
1425         inode_init_once(&inode->v);
1426         mutex_init(&inode->ei_update_lock);
1427         two_state_lock_init(&inode->ei_pagecache_lock);
1428         INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
1429         mutex_init(&inode->ei_quota_lock);
1430
1431         return &inode->v;
1432 }
1433
1434 static void bch2_i_callback(struct rcu_head *head)
1435 {
1436         struct inode *vinode = container_of(head, struct inode, i_rcu);
1437         struct bch_inode_info *inode = to_bch_ei(vinode);
1438
1439         kmem_cache_free(bch2_inode_cache, inode);
1440 }
1441
1442 static void bch2_destroy_inode(struct inode *vinode)
1443 {
1444         call_rcu(&vinode->i_rcu, bch2_i_callback);
1445 }
1446
1447 static int inode_update_times_fn(struct btree_trans *trans,
1448                                  struct bch_inode_info *inode,
1449                                  struct bch_inode_unpacked *bi,
1450                                  void *p)
1451 {
1452         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1453
1454         bi->bi_atime    = timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1455         bi->bi_mtime    = timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1456         bi->bi_ctime    = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1457
1458         return 0;
1459 }
1460
1461 static int bch2_vfs_write_inode(struct inode *vinode,
1462                                 struct writeback_control *wbc)
1463 {
1464         struct bch_fs *c = vinode->i_sb->s_fs_info;
1465         struct bch_inode_info *inode = to_bch_ei(vinode);
1466         int ret;
1467
1468         mutex_lock(&inode->ei_update_lock);
1469         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1470                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1471         mutex_unlock(&inode->ei_update_lock);
1472
1473         return bch2_err_class(ret);
1474 }
1475
1476 static void bch2_evict_inode(struct inode *vinode)
1477 {
1478         struct bch_fs *c = vinode->i_sb->s_fs_info;
1479         struct bch_inode_info *inode = to_bch_ei(vinode);
1480
1481         truncate_inode_pages_final(&inode->v.i_data);
1482
1483         clear_inode(&inode->v);
1484
1485         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1486
1487         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1488                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1489                                 KEY_TYPE_QUOTA_WARN);
1490                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1491                                 KEY_TYPE_QUOTA_WARN);
1492                 bch2_inode_rm(c, inode_inum(inode));
1493         }
1494
1495         mutex_lock(&c->vfs_inodes_lock);
1496         list_del_init(&inode->ei_vfs_inode_list);
1497         mutex_unlock(&c->vfs_inodes_lock);
1498 }
1499
1500 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1501 {
1502         struct bch_inode_info *inode;
1503         DARRAY(struct bch_inode_info *) grabbed;
1504         bool clean_pass = false, this_pass_clean;
1505
1506         /*
1507          * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1508          * be pruned with d_mark_dontcache().
1509          *
1510          * Once we've had a clean pass where we didn't find any inodes without
1511          * I_DONTCACHE, we wait for them to be freed:
1512          */
1513
1514         darray_init(&grabbed);
1515         darray_make_room(&grabbed, 1024);
1516 again:
1517         cond_resched();
1518         this_pass_clean = true;
1519
1520         mutex_lock(&c->vfs_inodes_lock);
1521         list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1522                 if (!snapshot_list_has_id(s, inode->ei_subvol))
1523                         continue;
1524
1525                 if (!(inode->v.i_state & I_DONTCACHE) &&
1526                     !(inode->v.i_state & I_FREEING) &&
1527                     igrab(&inode->v)) {
1528                         this_pass_clean = false;
1529
1530                         if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1531                                 iput(&inode->v);
1532                                 break;
1533                         }
1534                 } else if (clean_pass && this_pass_clean) {
1535                         wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1536                         DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1537
1538                         prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1539                         mutex_unlock(&c->vfs_inodes_lock);
1540
1541                         schedule();
1542                         finish_wait(wq, &wait.wq_entry);
1543                         goto again;
1544                 }
1545         }
1546         mutex_unlock(&c->vfs_inodes_lock);
1547
1548         darray_for_each(grabbed, i) {
1549                 inode = *i;
1550                 d_mark_dontcache(&inode->v);
1551                 d_prune_aliases(&inode->v);
1552                 iput(&inode->v);
1553         }
1554         grabbed.nr = 0;
1555
1556         if (!clean_pass || !this_pass_clean) {
1557                 clean_pass = this_pass_clean;
1558                 goto again;
1559         }
1560
1561         darray_exit(&grabbed);
1562 }
1563
1564 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1565 {
1566         struct super_block *sb = dentry->d_sb;
1567         struct bch_fs *c = sb->s_fs_info;
1568         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1569         unsigned shift = sb->s_blocksize_bits - 9;
1570         /*
1571          * this assumes inodes take up 64 bytes, which is a decent average
1572          * number:
1573          */
1574         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1575         u64 fsid;
1576
1577         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1578         buf->f_bsize    = sb->s_blocksize;
1579         buf->f_blocks   = usage.capacity >> shift;
1580         buf->f_bfree    = usage.free >> shift;
1581         buf->f_bavail   = avail_factor(usage.free) >> shift;
1582
1583         buf->f_files    = usage.nr_inodes + avail_inodes;
1584         buf->f_ffree    = avail_inodes;
1585
1586         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1587                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1588         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1589         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1590         buf->f_namelen  = BCH_NAME_MAX;
1591
1592         return 0;
1593 }
1594
1595 static int bch2_sync_fs(struct super_block *sb, int wait)
1596 {
1597         struct bch_fs *c = sb->s_fs_info;
1598         int ret;
1599
1600         if (c->opts.journal_flush_disabled)
1601                 return 0;
1602
1603         if (!wait) {
1604                 bch2_journal_flush_async(&c->journal, NULL);
1605                 return 0;
1606         }
1607
1608         ret = bch2_journal_flush(&c->journal);
1609         return bch2_err_class(ret);
1610 }
1611
1612 static struct bch_fs *bch2_path_to_fs(const char *path)
1613 {
1614         struct bch_fs *c;
1615         dev_t dev;
1616         int ret;
1617
1618         ret = lookup_bdev(path, &dev);
1619         if (ret)
1620                 return ERR_PTR(ret);
1621
1622         c = bch2_dev_to_fs(dev);
1623         if (c)
1624                 closure_put(&c->cl);
1625         return c ?: ERR_PTR(-ENOENT);
1626 }
1627
1628 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1629 {
1630         struct bch_fs *c = sb->s_fs_info;
1631         struct bch_opts opts = bch2_opts_empty();
1632         int ret;
1633
1634         ret = bch2_parse_mount_opts(c, &opts, data);
1635         if (ret)
1636                 goto err;
1637
1638         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1639
1640         if (opts.read_only != c->opts.read_only) {
1641                 down_write(&c->state_lock);
1642
1643                 if (opts.read_only) {
1644                         bch2_fs_read_only(c);
1645
1646                         sb->s_flags |= SB_RDONLY;
1647                 } else {
1648                         ret = bch2_fs_read_write(c);
1649                         if (ret) {
1650                                 bch_err(c, "error going rw: %i", ret);
1651                                 up_write(&c->state_lock);
1652                                 ret = -EINVAL;
1653                                 goto err;
1654                         }
1655
1656                         sb->s_flags &= ~SB_RDONLY;
1657                 }
1658
1659                 c->opts.read_only = opts.read_only;
1660
1661                 up_write(&c->state_lock);
1662         }
1663
1664         if (opt_defined(opts, errors))
1665                 c->opts.errors = opts.errors;
1666 err:
1667         return bch2_err_class(ret);
1668 }
1669
1670 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1671 {
1672         struct bch_fs *c = root->d_sb->s_fs_info;
1673         bool first = true;
1674
1675         for_each_online_member(c, ca) {
1676                 if (!first)
1677                         seq_putc(seq, ':');
1678                 first = false;
1679                 seq_puts(seq, ca->disk_sb.sb_name);
1680         }
1681
1682         return 0;
1683 }
1684
1685 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1686 {
1687         struct bch_fs *c = root->d_sb->s_fs_info;
1688         enum bch_opt_id i;
1689         struct printbuf buf = PRINTBUF;
1690         int ret = 0;
1691
1692         for (i = 0; i < bch2_opts_nr; i++) {
1693                 const struct bch_option *opt = &bch2_opt_table[i];
1694                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1695
1696                 if (!(opt->flags & OPT_MOUNT))
1697                         continue;
1698
1699                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1700                         continue;
1701
1702                 printbuf_reset(&buf);
1703                 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1704                                  OPT_SHOW_MOUNT_STYLE);
1705                 seq_putc(seq, ',');
1706                 seq_puts(seq, buf.buf);
1707         }
1708
1709         if (buf.allocation_failure)
1710                 ret = -ENOMEM;
1711         printbuf_exit(&buf);
1712         return ret;
1713 }
1714
1715 static void bch2_put_super(struct super_block *sb)
1716 {
1717         struct bch_fs *c = sb->s_fs_info;
1718
1719         __bch2_fs_stop(c);
1720 }
1721
1722 /*
1723  * bcachefs doesn't currently integrate intwrite freeze protection but the
1724  * internal write references serve the same purpose. Therefore reuse the
1725  * read-only transition code to perform the quiesce. The caveat is that we don't
1726  * currently have the ability to block tasks that want a write reference while
1727  * the superblock is frozen. This is fine for now, but we should either add
1728  * blocking support or find a way to integrate sb_start_intwrite() and friends.
1729  */
1730 static int bch2_freeze(struct super_block *sb)
1731 {
1732         struct bch_fs *c = sb->s_fs_info;
1733
1734         down_write(&c->state_lock);
1735         bch2_fs_read_only(c);
1736         up_write(&c->state_lock);
1737         return 0;
1738 }
1739
1740 static int bch2_unfreeze(struct super_block *sb)
1741 {
1742         struct bch_fs *c = sb->s_fs_info;
1743         int ret;
1744
1745         if (test_bit(BCH_FS_emergency_ro, &c->flags))
1746                 return 0;
1747
1748         down_write(&c->state_lock);
1749         ret = bch2_fs_read_write(c);
1750         up_write(&c->state_lock);
1751         return ret;
1752 }
1753
1754 static const struct super_operations bch_super_operations = {
1755         .alloc_inode    = bch2_alloc_inode,
1756         .destroy_inode  = bch2_destroy_inode,
1757         .write_inode    = bch2_vfs_write_inode,
1758         .evict_inode    = bch2_evict_inode,
1759         .sync_fs        = bch2_sync_fs,
1760         .statfs         = bch2_statfs,
1761         .show_devname   = bch2_show_devname,
1762         .show_options   = bch2_show_options,
1763         .remount_fs     = bch2_remount,
1764         .put_super      = bch2_put_super,
1765         .freeze_fs      = bch2_freeze,
1766         .unfreeze_fs    = bch2_unfreeze,
1767 };
1768
1769 static int bch2_set_super(struct super_block *s, void *data)
1770 {
1771         s->s_fs_info = data;
1772         return 0;
1773 }
1774
1775 static int bch2_noset_super(struct super_block *s, void *data)
1776 {
1777         return -EBUSY;
1778 }
1779
1780 typedef DARRAY(struct bch_fs *) darray_fs;
1781
1782 static int bch2_test_super(struct super_block *s, void *data)
1783 {
1784         struct bch_fs *c = s->s_fs_info;
1785         darray_fs *d = data;
1786
1787         if (!c)
1788                 return false;
1789
1790         darray_for_each(*d, i)
1791                 if (c != *i)
1792                         return false;
1793         return true;
1794 }
1795
1796 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1797                                  int flags, const char *dev_name, void *data)
1798 {
1799         struct bch_fs *c;
1800         struct super_block *sb;
1801         struct inode *vinode;
1802         struct bch_opts opts = bch2_opts_empty();
1803         int ret;
1804
1805         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1806
1807         ret = bch2_parse_mount_opts(NULL, &opts, data);
1808         if (ret)
1809                 return ERR_PTR(ret);
1810
1811         if (!dev_name || strlen(dev_name) == 0)
1812                 return ERR_PTR(-EINVAL);
1813
1814         darray_str devs;
1815         ret = bch2_split_devs(dev_name, &devs);
1816         if (ret)
1817                 return ERR_PTR(ret);
1818
1819         darray_fs devs_to_fs = {};
1820         darray_for_each(devs, i) {
1821                 ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
1822                 if (ret) {
1823                         sb = ERR_PTR(ret);
1824                         goto got_sb;
1825                 }
1826         }
1827
1828         sb = sget(fs_type, bch2_test_super, bch2_noset_super, flags|SB_NOSEC, &devs_to_fs);
1829         if (!IS_ERR(sb))
1830                 goto got_sb;
1831
1832         c = bch2_fs_open(devs.data, devs.nr, opts);
1833         if (IS_ERR(c)) {
1834                 sb = ERR_CAST(c);
1835                 goto got_sb;
1836         }
1837
1838         /* Some options can't be parsed until after the fs is started: */
1839         ret = bch2_parse_mount_opts(c, &opts, data);
1840         if (ret) {
1841                 bch2_fs_stop(c);
1842                 sb = ERR_PTR(ret);
1843                 goto got_sb;
1844         }
1845
1846         bch2_opts_apply(&c->opts, opts);
1847
1848         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1849         if (IS_ERR(sb))
1850                 bch2_fs_stop(c);
1851 got_sb:
1852         darray_exit(&devs_to_fs);
1853         bch2_darray_str_exit(&devs);
1854
1855         if (IS_ERR(sb)) {
1856                 ret = PTR_ERR(sb);
1857                 ret = bch2_err_class(ret);
1858                 return ERR_PTR(ret);
1859         }
1860
1861         c = sb->s_fs_info;
1862
1863         if (sb->s_root) {
1864                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1865                         ret = -EBUSY;
1866                         goto err_put_super;
1867                 }
1868                 goto out;
1869         }
1870
1871         sb->s_blocksize         = block_bytes(c);
1872         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1873         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1874         sb->s_op                = &bch_super_operations;
1875         sb->s_export_op         = &bch_export_ops;
1876 #ifdef CONFIG_BCACHEFS_QUOTA
1877         sb->s_qcop              = &bch2_quotactl_operations;
1878         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1879 #endif
1880         sb->s_xattr             = bch2_xattr_handlers;
1881         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1882         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1883         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1884         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1885         c->vfs_sb               = sb;
1886         strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1887
1888         ret = super_setup_bdi(sb);
1889         if (ret)
1890                 goto err_put_super;
1891
1892         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1893
1894         for_each_online_member(c, ca) {
1895                 struct block_device *bdev = ca->disk_sb.bdev;
1896
1897                 /* XXX: create an anonymous device for multi device filesystems */
1898                 sb->s_bdev      = bdev;
1899                 sb->s_dev       = bdev->bd_dev;
1900                 percpu_ref_put(&ca->io_ref);
1901                 break;
1902         }
1903
1904         c->dev = sb->s_dev;
1905
1906 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1907         if (c->opts.acl)
1908                 sb->s_flags     |= SB_POSIXACL;
1909 #endif
1910
1911         sb->s_shrink->seeks = 0;
1912
1913         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
1914         ret = PTR_ERR_OR_ZERO(vinode);
1915         bch_err_msg(c, ret, "mounting: error getting root inode");
1916         if (ret)
1917                 goto err_put_super;
1918
1919         sb->s_root = d_make_root(vinode);
1920         if (!sb->s_root) {
1921                 bch_err(c, "error mounting: error allocating root dentry");
1922                 ret = -ENOMEM;
1923                 goto err_put_super;
1924         }
1925
1926         sb->s_flags |= SB_ACTIVE;
1927 out:
1928         return dget(sb->s_root);
1929
1930 err_put_super:
1931         deactivate_locked_super(sb);
1932         return ERR_PTR(bch2_err_class(ret));
1933 }
1934
1935 static void bch2_kill_sb(struct super_block *sb)
1936 {
1937         struct bch_fs *c = sb->s_fs_info;
1938
1939         generic_shutdown_super(sb);
1940         bch2_fs_free(c);
1941 }
1942
1943 static struct file_system_type bcache_fs_type = {
1944         .owner          = THIS_MODULE,
1945         .name           = "bcachefs",
1946         .mount          = bch2_mount,
1947         .kill_sb        = bch2_kill_sb,
1948         .fs_flags       = FS_REQUIRES_DEV,
1949 };
1950
1951 MODULE_ALIAS_FS("bcachefs");
1952
1953 void bch2_vfs_exit(void)
1954 {
1955         unregister_filesystem(&bcache_fs_type);
1956         kmem_cache_destroy(bch2_inode_cache);
1957 }
1958
1959 int __init bch2_vfs_init(void)
1960 {
1961         int ret = -ENOMEM;
1962
1963         bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
1964         if (!bch2_inode_cache)
1965                 goto err;
1966
1967         ret = register_filesystem(&bcache_fs_type);
1968         if (ret)
1969                 goto err;
1970
1971         return 0;
1972 err:
1973         bch2_vfs_exit();
1974         return ret;
1975 }
1976
1977 #endif /* NO_BCACHEFS_FS */