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