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

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