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 "extents.h"
  12 #include "fs.h"
  13 #include "fs-common.h"
  14 #include "fs-io.h"
  15 #include "fs-ioctl.h"
  16 #include "fsck.h"
  17 #include "inode.h"
  18 #include "io.h"
  19 #include "journal.h"
  20 #include "keylist.h"
  21 #include "quota.h"
  22 #include "super.h"
  23 #include "xattr.h"
  24
  25 #include <linux/aio.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/exportfs.h>
  28 #include <linux/fiemap.h>
  29 #include <linux/module.h>
  30 #include <linux/pagemap.h>
  31 #include <linux/posix_acl.h>
  32 #include <linux/random.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/statfs.h>
  35 #include <linux/string.h>
  36 #include <linux/xattr.h>
  37
  38 static struct kmem_cache *bch2_inode_cache;
  39
  40 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
  41                                 struct bch_inode_info *,
  42                                 struct bch_inode_unpacked *,
  43                                 struct bch_subvolume *);
  44
  45 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
  46 {
  47         BUG_ON(atomic_long_read(&lock->v) == 0);
  48
  49         if (atomic_long_sub_return_release(i, &lock->v) == 0)
  50                 wake_up_all(&lock->wait);
  51 }
  52
  53 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
  54 {
  55         long v = atomic_long_read(&lock->v), old;
  56
  57         do {
  58                 old = v;
  59
  60                 if (i > 0 ? v < 0 : v > 0)
  61                         return false;
  62         } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
  63                                         old, old + i)) != old);
  64         return true;
  65 }
  66
  67 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
  68 {
  69         wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
  70 }
  71
  72 void bch2_pagecache_add_put(struct pagecache_lock *lock)
  73 {
  74         __pagecache_lock_put(lock, 1);
  75 }
  76
  77 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
  78 {
  79         return __pagecache_lock_tryget(lock, 1);
  80 }
  81
  82 void bch2_pagecache_add_get(struct pagecache_lock *lock)
  83 {
  84         __pagecache_lock_get(lock, 1);
  85 }
  86
  87 void bch2_pagecache_block_put(struct pagecache_lock *lock)
  88 {
  89         __pagecache_lock_put(lock, -1);
  90 }
  91
  92 void bch2_pagecache_block_get(struct pagecache_lock *lock)
  93 {
  94         __pagecache_lock_get(lock, -1);
  95 }
  96
  97 void bch2_inode_update_after_write(struct btree_trans *trans,
  98                                    struct bch_inode_info *inode,
  99                                    struct bch_inode_unpacked *bi,
 100                                    unsigned fields)
 101 {
 102         struct bch_fs *c = trans->c;
 103
 104         BUG_ON(bi->bi_inum != inode->v.i_ino);
 105
 106         bch2_assert_pos_locked(trans, BTREE_ID_inodes,
 107                                POS(0, bi->bi_inum),
 108                                c->opts.inodes_use_key_cache);
 109
 110         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
 111         i_uid_write(&inode->v, bi->bi_uid);
 112         i_gid_write(&inode->v, bi->bi_gid);
 113         inode->v.i_mode = bi->bi_mode;
 114
 115         if (fields & ATTR_ATIME)
 116                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
 117         if (fields & ATTR_MTIME)
 118                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
 119         if (fields & ATTR_CTIME)
 120                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
 121
 122         inode->ei_inode         = *bi;
 123
 124         bch2_inode_flags_to_vfs(inode);
 125 }
 126
 127 int __must_check bch2_write_inode(struct bch_fs *c,
 128                                   struct bch_inode_info *inode,
 129                                   inode_set_fn set,
 130                                   void *p, unsigned fields)
 131 {
 132         struct btree_trans trans;
 133         struct btree_iter iter = { NULL };
 134         struct bch_inode_unpacked inode_u;
 135         int ret;
 136
 137         bch2_trans_init(&trans, c, 0, 512);
 138 retry:
 139         bch2_trans_begin(&trans);
 140
 141         ret   = bch2_inode_peek(&trans, &iter, &inode_u, inode_inum(inode),
 142                                 BTREE_ITER_INTENT) ?:
 143                 (set ? set(inode, &inode_u, p) : 0) ?:
 144                 bch2_inode_write(&trans, &iter, &inode_u) ?:
 145                 bch2_trans_commit(&trans, NULL, NULL, BTREE_INSERT_NOFAIL);
 146
 147         /*
 148          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 149          * this is important for inode updates via bchfs_write_index_update
 150          */
 151         if (!ret)
 152                 bch2_inode_update_after_write(&trans, inode, &inode_u, fields);
 153
 154         bch2_trans_iter_exit(&trans, &iter);
 155
 156         if (ret == -EINTR)
 157                 goto retry;
 158
 159         bch2_trans_exit(&trans);
 160         return ret < 0 ? ret : 0;
 161 }
 162
 163 int bch2_fs_quota_transfer(struct bch_fs *c,
 164                            struct bch_inode_info *inode,
 165                            struct bch_qid new_qid,
 166                            unsigned qtypes,
 167                            enum quota_acct_mode mode)
 168 {
 169         unsigned i;
 170         int ret;
 171
 172         qtypes &= enabled_qtypes(c);
 173
 174         for (i = 0; i < QTYP_NR; i++)
 175                 if (new_qid.q[i] == inode->ei_qid.q[i])
 176                         qtypes &= ~(1U << i);
 177
 178         if (!qtypes)
 179                 return 0;
 180
 181         mutex_lock(&inode->ei_quota_lock);
 182
 183         ret = bch2_quota_transfer(c, qtypes, new_qid,
 184                                   inode->ei_qid,
 185                                   inode->v.i_blocks +
 186                                   inode->ei_quota_reserved,
 187                                   mode);
 188         if (!ret)
 189                 for (i = 0; i < QTYP_NR; i++)
 190                         if (qtypes & (1 << i))
 191                                 inode->ei_qid.q[i] = new_qid.q[i];
 192
 193         mutex_unlock(&inode->ei_quota_lock);
 194
 195         return ret;
 196 }
 197
 198 static int bch2_iget5_test(struct inode *vinode, void *p)
 199 {
 200         struct bch_inode_info *inode = to_bch_ei(vinode);
 201         subvol_inum *inum = p;
 202
 203         return inode->ei_subvol == inum->subvol &&
 204                 inode->ei_inode.bi_inum == inum->inum;
 205 }
 206
 207 static int bch2_iget5_set(struct inode *vinode, void *p)
 208 {
 209         struct bch_inode_info *inode = to_bch_ei(vinode);
 210         subvol_inum *inum = p;
 211
 212         inode->v.i_ino          = inum->inum;
 213         inode->ei_subvol        = inum->subvol;
 214         inode->ei_inode.bi_inum = inum->inum;
 215         return 0;
 216 }
 217
 218 static unsigned bch2_inode_hash(subvol_inum inum)
 219 {
 220         return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
 221 }
 222
 223 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 224 {
 225         struct bch_inode_unpacked inode_u;
 226         struct bch_inode_info *inode;
 227         struct btree_trans trans;
 228         struct bch_subvolume subvol;
 229         int ret;
 230
 231         inode = to_bch_ei(iget5_locked(c->vfs_sb,
 232                                        bch2_inode_hash(inum),
 233                                        bch2_iget5_test,
 234                                        bch2_iget5_set,
 235                                        &inum));
 236         if (unlikely(!inode))
 237                 return ERR_PTR(-ENOMEM);
 238         if (!(inode->v.i_state & I_NEW))
 239                 return &inode->v;
 240
 241         bch2_trans_init(&trans, c, 8, 0);
 242         ret = lockrestart_do(&trans,
 243                 bch2_subvolume_get(&trans, inum.subvol, true, 0, &subvol) ?:
 244                 bch2_inode_find_by_inum_trans(&trans, inum, &inode_u));
 245
 246         if (!ret)
 247                 bch2_vfs_inode_init(&trans, inum, inode, &inode_u, &subvol);
 248         bch2_trans_exit(&trans);
 249
 250         if (ret) {
 251                 iget_failed(&inode->v);
 252                 return ERR_PTR(ret);
 253         }
 254
 255         unlock_new_inode(&inode->v);
 256
 257         return &inode->v;
 258 }
 259
 260 struct bch_inode_info *
 261 __bch2_create(struct user_namespace *mnt_userns,
 262               struct bch_inode_info *dir, struct dentry *dentry,
 263               umode_t mode, dev_t rdev, subvol_inum snapshot_src,
 264               unsigned flags)
 265 {
 266         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 267         struct btree_trans trans;
 268         struct bch_inode_unpacked dir_u;
 269         struct bch_inode_info *inode, *old;
 270         struct bch_inode_unpacked inode_u;
 271         struct posix_acl *default_acl = NULL, *acl = NULL;
 272         subvol_inum inum;
 273         struct bch_subvolume subvol;
 274         u64 journal_seq = 0;
 275         int ret;
 276
 277         /*
 278          * preallocate acls + vfs inode before btree transaction, so that
 279          * nothing can fail after the transaction succeeds:
 280          */
 281 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 282         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 283         if (ret)
 284                 return ERR_PTR(ret);
 285 #endif
 286         inode = to_bch_ei(new_inode(c->vfs_sb));
 287         if (unlikely(!inode)) {
 288                 inode = ERR_PTR(-ENOMEM);
 289                 goto err;
 290         }
 291
 292         bch2_inode_init_early(c, &inode_u);
 293
 294         if (!(flags & BCH_CREATE_TMPFILE))
 295                 mutex_lock(&dir->ei_update_lock);
 296
 297         bch2_trans_init(&trans, c, 8,
 298                         2048 + (!(flags & BCH_CREATE_TMPFILE)
 299                                 ? dentry->d_name.len : 0));
 300 retry:
 301         bch2_trans_begin(&trans);
 302
 303         ret   = bch2_create_trans(&trans,
 304                                   inode_inum(dir), &dir_u, &inode_u,
 305                                   !(flags & BCH_CREATE_TMPFILE)
 306                                   ? &dentry->d_name : NULL,
 307                                   from_kuid(mnt_userns, current_fsuid()),
 308                                   from_kgid(mnt_userns, current_fsgid()),
 309                                   mode, rdev,
 310                                   default_acl, acl, snapshot_src, flags) ?:
 311                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 312                                 KEY_TYPE_QUOTA_PREALLOC);
 313         if (unlikely(ret))
 314                 goto err_before_quota;
 315
 316         inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
 317         inum.inum = inode_u.bi_inum;
 318
 319         ret   = bch2_subvolume_get(&trans, inum.subvol, true,
 320                                    BTREE_ITER_WITH_UPDATES, &subvol) ?:
 321                 bch2_trans_commit(&trans, NULL, &journal_seq, 0);
 322         if (unlikely(ret)) {
 323                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 324                                 KEY_TYPE_QUOTA_WARN);
 325 err_before_quota:
 326                 if (ret == -EINTR)
 327                         goto retry;
 328                 goto err_trans;
 329         }
 330
 331         if (!(flags & BCH_CREATE_TMPFILE)) {
 332                 bch2_inode_update_after_write(&trans, dir, &dir_u,
 333                                               ATTR_MTIME|ATTR_CTIME);
 334                 mutex_unlock(&dir->ei_update_lock);
 335         }
 336
 337         bch2_iget5_set(&inode->v, &inum);
 338         bch2_vfs_inode_init(&trans, inum, inode, &inode_u, &subvol);
 339
 340         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 341         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 342
 343         /*
 344          * we must insert the new inode into the inode cache before calling
 345          * bch2_trans_exit() and dropping locks, else we could race with another
 346          * thread pulling the inode in and modifying it:
 347          */
 348
 349         inode->v.i_state |= I_CREATING;
 350
 351         old = to_bch_ei(inode_insert5(&inode->v,
 352                                       bch2_inode_hash(inum),
 353                                       bch2_iget5_test,
 354                                       bch2_iget5_set,
 355                                       &inum));
 356         BUG_ON(!old);
 357
 358         if (unlikely(old != inode)) {
 359                 /*
 360                  * We raced, another process pulled the new inode into cache
 361                  * before us:
 362                  */
 363                 make_bad_inode(&inode->v);
 364                 iput(&inode->v);
 365
 366                 inode = old;
 367         } else {
 368                 /*
 369                  * we really don't want insert_inode_locked2() to be setting
 370                  * I_NEW...
 371                  */
 372                 unlock_new_inode(&inode->v);
 373         }
 374
 375         bch2_trans_exit(&trans);
 376 err:
 377         posix_acl_release(default_acl);
 378         posix_acl_release(acl);
 379         return inode;
 380 err_trans:
 381         if (!(flags & BCH_CREATE_TMPFILE))
 382                 mutex_unlock(&dir->ei_update_lock);
 383
 384         bch2_trans_exit(&trans);
 385         make_bad_inode(&inode->v);
 386         iput(&inode->v);
 387         inode = ERR_PTR(ret);
 388         goto err;
 389 }
 390
 391 /* methods */
 392
 393 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 394                                   unsigned int flags)
 395 {
 396         struct bch_fs *c = vdir->i_sb->s_fs_info;
 397         struct bch_inode_info *dir = to_bch_ei(vdir);
 398         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 399         struct inode *vinode = NULL;
 400         subvol_inum inum = { .subvol = 1 };
 401         int ret;
 402
 403         ret = bch2_dirent_lookup(c, inode_inum(dir), &hash,
 404                                  &dentry->d_name, &inum);
 405
 406         if (!ret)
 407                 vinode = bch2_vfs_inode_get(c, inum);
 408
 409         return d_splice_alias(vinode, dentry);
 410 }
 411
 412 static int bch2_mknod(struct user_namespace *mnt_userns,
 413                       struct inode *vdir, struct dentry *dentry,
 414                       umode_t mode, dev_t rdev)
 415 {
 416         struct bch_inode_info *inode =
 417                 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, rdev,
 418                               (subvol_inum) { 0 }, 0);
 419
 420         if (IS_ERR(inode))
 421                 return PTR_ERR(inode);
 422
 423         d_instantiate(dentry, &inode->v);
 424         return 0;
 425 }
 426
 427 static int bch2_create(struct user_namespace *mnt_userns,
 428                        struct inode *vdir, struct dentry *dentry,
 429                        umode_t mode, bool excl)
 430 {
 431         return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFREG, 0);
 432 }
 433
 434 static int __bch2_link(struct bch_fs *c,
 435                        struct bch_inode_info *inode,
 436                        struct bch_inode_info *dir,
 437                        struct dentry *dentry)
 438 {
 439         struct btree_trans trans;
 440         struct bch_inode_unpacked dir_u, inode_u;
 441         int ret;
 442
 443         mutex_lock(&inode->ei_update_lock);
 444         bch2_trans_init(&trans, c, 4, 1024);
 445
 446         ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 447                         bch2_link_trans(&trans,
 448                                         inode_inum(dir),   &dir_u,
 449                                         inode_inum(inode), &inode_u,
 450                                         &dentry->d_name));
 451
 452         if (likely(!ret)) {
 453                 bch2_inode_update_after_write(&trans, dir, &dir_u,
 454                                               ATTR_MTIME|ATTR_CTIME);
 455                 bch2_inode_update_after_write(&trans, inode, &inode_u, ATTR_CTIME);
 456         }
 457
 458         bch2_trans_exit(&trans);
 459         mutex_unlock(&inode->ei_update_lock);
 460         return ret;
 461 }
 462
 463 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 464                      struct dentry *dentry)
 465 {
 466         struct bch_fs *c = vdir->i_sb->s_fs_info;
 467         struct bch_inode_info *dir = to_bch_ei(vdir);
 468         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 469         int ret;
 470
 471         lockdep_assert_held(&inode->v.i_rwsem);
 472
 473         ret = __bch2_link(c, inode, dir, dentry);
 474         if (unlikely(ret))
 475                 return ret;
 476
 477         ihold(&inode->v);
 478         d_instantiate(dentry, &inode->v);
 479         return 0;
 480 }
 481
 482 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 483                   bool deleting_snapshot)
 484 {
 485         struct bch_fs *c = vdir->i_sb->s_fs_info;
 486         struct bch_inode_info *dir = to_bch_ei(vdir);
 487         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 488         struct bch_inode_unpacked dir_u, inode_u;
 489         struct btree_trans trans;
 490         int ret;
 491
 492         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 493         bch2_trans_init(&trans, c, 4, 1024);
 494
 495         ret = __bch2_trans_do(&trans, NULL, NULL,
 496                               BTREE_INSERT_NOFAIL,
 497                         bch2_unlink_trans(&trans,
 498                                           inode_inum(dir), &dir_u,
 499                                           &inode_u, &dentry->d_name,
 500                                           deleting_snapshot));
 501
 502         if (likely(!ret)) {
 503                 bch2_inode_update_after_write(&trans, dir, &dir_u,
 504                                               ATTR_MTIME|ATTR_CTIME);
 505                 bch2_inode_update_after_write(&trans, inode, &inode_u,
 506                                               ATTR_MTIME);
 507         }
 508
 509         bch2_trans_exit(&trans);
 510         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 511
 512         return ret;
 513 }
 514
 515 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 516 {
 517         return __bch2_unlink(vdir, dentry, false);
 518 }
 519
 520 static int bch2_symlink(struct user_namespace *mnt_userns,
 521                         struct inode *vdir, struct dentry *dentry,
 522                         const char *symname)
 523 {
 524         struct bch_fs *c = vdir->i_sb->s_fs_info;
 525         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 526         int ret;
 527
 528         inode = __bch2_create(mnt_userns, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
 529                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 530         if (unlikely(IS_ERR(inode)))
 531                 return PTR_ERR(inode);
 532
 533         inode_lock(&inode->v);
 534         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 535         inode_unlock(&inode->v);
 536
 537         if (unlikely(ret))
 538                 goto err;
 539
 540         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 541         if (unlikely(ret))
 542                 goto err;
 543
 544         ret = __bch2_link(c, inode, dir, dentry);
 545         if (unlikely(ret))
 546                 goto err;
 547
 548         d_instantiate(dentry, &inode->v);
 549         return 0;
 550 err:
 551         iput(&inode->v);
 552         return ret;
 553 }
 554
 555 static int bch2_mkdir(struct user_namespace *mnt_userns,
 556                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 557 {
 558         return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFDIR, 0);
 559 }
 560
 561 static int bch2_rename2(struct user_namespace *mnt_userns,
 562                         struct inode *src_vdir, struct dentry *src_dentry,
 563                         struct inode *dst_vdir, struct dentry *dst_dentry,
 564                         unsigned flags)
 565 {
 566         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 567         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 568         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 569         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 570         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 571         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 572         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 573         struct btree_trans trans;
 574         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 575                 ? BCH_RENAME_EXCHANGE
 576                 : dst_dentry->d_inode
 577                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 578         int ret;
 579
 580         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 581                 return -EINVAL;
 582
 583         if (mode == BCH_RENAME_OVERWRITE) {
 584                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 585                                                    0, LLONG_MAX);
 586                 if (ret)
 587                         return ret;
 588         }
 589
 590         bch2_trans_init(&trans, c, 8, 2048);
 591
 592         bch2_lock_inodes(INODE_UPDATE_LOCK,
 593                          src_dir,
 594                          dst_dir,
 595                          src_inode,
 596                          dst_inode);
 597
 598         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 599                 ret = bch2_fs_quota_transfer(c, src_inode,
 600                                              dst_dir->ei_qid,
 601                                              1 << QTYP_PRJ,
 602                                              KEY_TYPE_QUOTA_PREALLOC);
 603                 if (ret)
 604                         goto err;
 605         }
 606
 607         if (mode == BCH_RENAME_EXCHANGE &&
 608             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 609                 ret = bch2_fs_quota_transfer(c, dst_inode,
 610                                              src_dir->ei_qid,
 611                                              1 << QTYP_PRJ,
 612                                              KEY_TYPE_QUOTA_PREALLOC);
 613                 if (ret)
 614                         goto err;
 615         }
 616
 617         ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 618                         bch2_rename_trans(&trans,
 619                                           inode_inum(src_dir), &src_dir_u,
 620                                           inode_inum(dst_dir), &dst_dir_u,
 621                                           &src_inode_u,
 622                                           &dst_inode_u,
 623                                           &src_dentry->d_name,
 624                                           &dst_dentry->d_name,
 625                                           mode));
 626         if (unlikely(ret))
 627                 goto err;
 628
 629         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 630         BUG_ON(dst_inode &&
 631                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 632
 633         bch2_inode_update_after_write(&trans, src_dir, &src_dir_u,
 634                                       ATTR_MTIME|ATTR_CTIME);
 635
 636         if (src_dir != dst_dir)
 637                 bch2_inode_update_after_write(&trans, dst_dir, &dst_dir_u,
 638                                               ATTR_MTIME|ATTR_CTIME);
 639
 640         bch2_inode_update_after_write(&trans, src_inode, &src_inode_u,
 641                                       ATTR_CTIME);
 642
 643         if (dst_inode)
 644                 bch2_inode_update_after_write(&trans, dst_inode, &dst_inode_u,
 645                                               ATTR_CTIME);
 646 err:
 647         bch2_trans_exit(&trans);
 648
 649         bch2_fs_quota_transfer(c, src_inode,
 650                                bch_qid(&src_inode->ei_inode),
 651                                1 << QTYP_PRJ,
 652                                KEY_TYPE_QUOTA_NOCHECK);
 653         if (dst_inode)
 654                 bch2_fs_quota_transfer(c, dst_inode,
 655                                        bch_qid(&dst_inode->ei_inode),
 656                                        1 << QTYP_PRJ,
 657                                        KEY_TYPE_QUOTA_NOCHECK);
 658
 659         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 660                            src_dir,
 661                            dst_dir,
 662                            src_inode,
 663                            dst_inode);
 664
 665         return ret;
 666 }
 667
 668 static void bch2_setattr_copy(struct user_namespace *mnt_userns,
 669                               struct bch_inode_info *inode,
 670                               struct bch_inode_unpacked *bi,
 671                               struct iattr *attr)
 672 {
 673         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 674         unsigned int ia_valid = attr->ia_valid;
 675
 676         if (ia_valid & ATTR_UID)
 677                 bi->bi_uid = from_kuid(mnt_userns, attr->ia_uid);
 678         if (ia_valid & ATTR_GID)
 679                 bi->bi_gid = from_kgid(mnt_userns, attr->ia_gid);
 680
 681         if (ia_valid & ATTR_SIZE)
 682                 bi->bi_size = attr->ia_size;
 683
 684         if (ia_valid & ATTR_ATIME)
 685                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 686         if (ia_valid & ATTR_MTIME)
 687                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 688         if (ia_valid & ATTR_CTIME)
 689                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 690
 691         if (ia_valid & ATTR_MODE) {
 692                 umode_t mode = attr->ia_mode;
 693                 kgid_t gid = ia_valid & ATTR_GID
 694                         ? attr->ia_gid
 695                         : inode->v.i_gid;
 696
 697                 if (!in_group_p(gid) &&
 698                     !capable_wrt_inode_uidgid(mnt_userns, &inode->v, CAP_FSETID))
 699                         mode &= ~S_ISGID;
 700                 bi->bi_mode = mode;
 701         }
 702 }
 703
 704 int bch2_setattr_nonsize(struct user_namespace *mnt_userns,
 705                          struct bch_inode_info *inode,
 706                          struct iattr *attr)
 707 {
 708         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 709         struct bch_qid qid;
 710         struct btree_trans trans;
 711         struct btree_iter inode_iter = { NULL };
 712         struct bch_inode_unpacked inode_u;
 713         struct posix_acl *acl = NULL;
 714         int ret;
 715
 716         mutex_lock(&inode->ei_update_lock);
 717
 718         qid = inode->ei_qid;
 719
 720         if (attr->ia_valid & ATTR_UID)
 721                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 722
 723         if (attr->ia_valid & ATTR_GID)
 724                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 725
 726         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 727                                      KEY_TYPE_QUOTA_PREALLOC);
 728         if (ret)
 729                 goto err;
 730
 731         bch2_trans_init(&trans, c, 0, 0);
 732 retry:
 733         bch2_trans_begin(&trans);
 734         kfree(acl);
 735         acl = NULL;
 736
 737         ret = bch2_inode_peek(&trans, &inode_iter, &inode_u, inode_inum(inode),
 738                               BTREE_ITER_INTENT);
 739         if (ret)
 740                 goto btree_err;
 741
 742         bch2_setattr_copy(mnt_userns, inode, &inode_u, attr);
 743
 744         if (attr->ia_valid & ATTR_MODE) {
 745                 ret = bch2_acl_chmod(&trans, inode_inum(inode), &inode_u,
 746                                      inode_u.bi_mode, &acl);
 747                 if (ret)
 748                         goto btree_err;
 749         }
 750
 751         ret =   bch2_inode_write(&trans, &inode_iter, &inode_u) ?:
 752                 bch2_trans_commit(&trans, NULL, NULL,
 753                                   BTREE_INSERT_NOFAIL);
 754 btree_err:
 755         bch2_trans_iter_exit(&trans, &inode_iter);
 756
 757         if (ret == -EINTR)
 758                 goto retry;
 759         if (unlikely(ret))
 760                 goto err_trans;
 761
 762         bch2_inode_update_after_write(&trans, inode, &inode_u, attr->ia_valid);
 763
 764         if (acl)
 765                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 766 err_trans:
 767         bch2_trans_exit(&trans);
 768 err:
 769         mutex_unlock(&inode->ei_update_lock);
 770
 771         return ret;
 772 }
 773
 774 static int bch2_getattr(struct user_namespace *mnt_userns,
 775                         const struct path *path, struct kstat *stat,
 776                         u32 request_mask, unsigned query_flags)
 777 {
 778         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 779         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 780
 781         stat->dev       = inode->v.i_sb->s_dev;
 782         stat->ino       = inode->v.i_ino;
 783         stat->mode      = inode->v.i_mode;
 784         stat->nlink     = inode->v.i_nlink;
 785         stat->uid       = inode->v.i_uid;
 786         stat->gid       = inode->v.i_gid;
 787         stat->rdev      = inode->v.i_rdev;
 788         stat->size      = i_size_read(&inode->v);
 789         stat->atime     = inode->v.i_atime;
 790         stat->mtime     = inode->v.i_mtime;
 791         stat->ctime     = inode->v.i_ctime;
 792         stat->blksize   = block_bytes(c);
 793         stat->blocks    = inode->v.i_blocks;
 794
 795         if (request_mask & STATX_BTIME) {
 796                 stat->result_mask |= STATX_BTIME;
 797                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 798         }
 799
 800         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 801                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 802         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 803
 804         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 805                 stat->attributes |= STATX_ATTR_APPEND;
 806         stat->attributes_mask    |= STATX_ATTR_APPEND;
 807
 808         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 809                 stat->attributes |= STATX_ATTR_NODUMP;
 810         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 811
 812         return 0;
 813 }
 814
 815 static int bch2_setattr(struct user_namespace *mnt_userns,
 816                         struct dentry *dentry, struct iattr *iattr)
 817 {
 818         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 819         int ret;
 820
 821         lockdep_assert_held(&inode->v.i_rwsem);
 822
 823         ret = setattr_prepare(mnt_userns, dentry, iattr);
 824         if (ret)
 825                 return ret;
 826
 827         return iattr->ia_valid & ATTR_SIZE
 828                 ? bch2_truncate(mnt_userns, inode, iattr)
 829                 : bch2_setattr_nonsize(mnt_userns, inode, iattr);
 830 }
 831
 832 static int bch2_tmpfile(struct user_namespace *mnt_userns,
 833                         struct inode *vdir, struct dentry *dentry, umode_t mode)
 834 {
 835         struct bch_inode_info *inode =
 836                 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, 0,
 837                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 838
 839         if (IS_ERR(inode))
 840                 return PTR_ERR(inode);
 841
 842         d_mark_tmpfile(dentry, &inode->v);
 843         d_instantiate(dentry, &inode->v);
 844         return 0;
 845 }
 846
 847 static int bch2_fill_extent(struct bch_fs *c,
 848                             struct fiemap_extent_info *info,
 849                             struct bkey_s_c k, unsigned flags)
 850 {
 851         if (bkey_extent_is_direct_data(k.k)) {
 852                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 853                 const union bch_extent_entry *entry;
 854                 struct extent_ptr_decoded p;
 855                 int ret;
 856
 857                 if (k.k->type == KEY_TYPE_reflink_v)
 858                         flags |= FIEMAP_EXTENT_SHARED;
 859
 860                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 861                         int flags2 = 0;
 862                         u64 offset = p.ptr.offset;
 863
 864                         if (p.crc.compression_type)
 865                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 866                         else
 867                                 offset += p.crc.offset;
 868
 869                         if ((offset & (block_sectors(c) - 1)) ||
 870                             (k.k->size & (block_sectors(c) - 1)))
 871                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 872
 873                         ret = fiemap_fill_next_extent(info,
 874                                                 bkey_start_offset(k.k) << 9,
 875                                                 offset << 9,
 876                                                 k.k->size << 9, flags|flags2);
 877                         if (ret)
 878                                 return ret;
 879                 }
 880
 881                 return 0;
 882         } else if (bkey_extent_is_inline_data(k.k)) {
 883                 return fiemap_fill_next_extent(info,
 884                                                bkey_start_offset(k.k) << 9,
 885                                                0, k.k->size << 9,
 886                                                flags|
 887                                                FIEMAP_EXTENT_DATA_INLINE);
 888         } else if (k.k->type == KEY_TYPE_reservation) {
 889                 return fiemap_fill_next_extent(info,
 890                                                bkey_start_offset(k.k) << 9,
 891                                                0, k.k->size << 9,
 892                                                flags|
 893                                                FIEMAP_EXTENT_DELALLOC|
 894                                                FIEMAP_EXTENT_UNWRITTEN);
 895         } else {
 896                 BUG();
 897         }
 898 }
 899
 900 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 901                        u64 start, u64 len)
 902 {
 903         struct bch_fs *c = vinode->i_sb->s_fs_info;
 904         struct bch_inode_info *ei = to_bch_ei(vinode);
 905         struct btree_trans trans;
 906         struct btree_iter iter;
 907         struct bkey_s_c k;
 908         struct bkey_buf cur, prev;
 909         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 910         unsigned offset_into_extent, sectors;
 911         bool have_extent = false;
 912         u32 snapshot;
 913         int ret = 0;
 914
 915         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 916         if (ret)
 917                 return ret;
 918
 919         if (start + len < start)
 920                 return -EINVAL;
 921
 922         start >>= 9;
 923
 924         bch2_bkey_buf_init(&cur);
 925         bch2_bkey_buf_init(&prev);
 926         bch2_trans_init(&trans, c, 0, 0);
 927 retry:
 928         bch2_trans_begin(&trans);
 929
 930         ret = bch2_subvolume_get_snapshot(&trans, ei->ei_subvol, &snapshot);
 931         if (ret)
 932                 goto err;
 933
 934         bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
 935                              SPOS(ei->v.i_ino, start, snapshot), 0);
 936
 937         while (!(ret = btree_trans_too_many_iters(&trans)) &&
 938                (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
 939                !(ret = bkey_err(k))) {
 940                 enum btree_id data_btree = BTREE_ID_extents;
 941
 942                 if (!bkey_extent_is_data(k.k) &&
 943                     k.k->type != KEY_TYPE_reservation) {
 944                         bch2_btree_iter_advance(&iter);
 945                         continue;
 946                 }
 947
 948                 offset_into_extent      = iter.pos.offset -
 949                         bkey_start_offset(k.k);
 950                 sectors                 = k.k->size - offset_into_extent;
 951
 952                 bch2_bkey_buf_reassemble(&cur, c, k);
 953
 954                 ret = bch2_read_indirect_extent(&trans, &data_btree,
 955                                         &offset_into_extent, &cur);
 956                 if (ret)
 957                         break;
 958
 959                 k = bkey_i_to_s_c(cur.k);
 960                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 961
 962                 sectors = min(sectors, k.k->size - offset_into_extent);
 963
 964                 bch2_cut_front(POS(k.k->p.inode,
 965                                    bkey_start_offset(k.k) +
 966                                    offset_into_extent),
 967                                cur.k);
 968                 bch2_key_resize(&cur.k->k, sectors);
 969                 cur.k->k.p = iter.pos;
 970                 cur.k->k.p.offset += cur.k->k.size;
 971
 972                 if (have_extent) {
 973                         ret = bch2_fill_extent(c, info,
 974                                         bkey_i_to_s_c(prev.k), 0);
 975                         if (ret)
 976                                 break;
 977                 }
 978
 979                 bkey_copy(prev.k, cur.k);
 980                 have_extent = true;
 981
 982                 bch2_btree_iter_set_pos(&iter,
 983                         POS(iter.pos.inode, iter.pos.offset + sectors));
 984         }
 985         start = iter.pos.offset;
 986         bch2_trans_iter_exit(&trans, &iter);
 987 err:
 988         if (ret == -EINTR)
 989                 goto retry;
 990
 991         if (!ret && have_extent)
 992                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 993                                        FIEMAP_EXTENT_LAST);
 994
 995         bch2_trans_exit(&trans);
 996         bch2_bkey_buf_exit(&cur, c);
 997         bch2_bkey_buf_exit(&prev, c);
 998         return ret < 0 ? ret : 0;
 999 }
1000
1001 static const struct vm_operations_struct bch_vm_ops = {
1002         .fault          = bch2_page_fault,
1003         .map_pages      = filemap_map_pages,
1004         .page_mkwrite   = bch2_page_mkwrite,
1005 };
1006
1007 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1008 {
1009         file_accessed(file);
1010
1011         vma->vm_ops = &bch_vm_ops;
1012         return 0;
1013 }
1014
1015 /* Directories: */
1016
1017 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1018 {
1019         return generic_file_llseek_size(file, offset, whence,
1020                                         S64_MAX, S64_MAX);
1021 }
1022
1023 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1024 {
1025         struct bch_inode_info *inode = file_bch_inode(file);
1026         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1027
1028         if (!dir_emit_dots(file, ctx))
1029                 return 0;
1030
1031         return bch2_readdir(c, inode_inum(inode), ctx);
1032 }
1033
1034 static const struct file_operations bch_file_operations = {
1035         .llseek         = bch2_llseek,
1036         .read_iter      = bch2_read_iter,
1037         .write_iter     = bch2_write_iter,
1038         .mmap           = bch2_mmap,
1039         .open           = generic_file_open,
1040         .fsync          = bch2_fsync,
1041         .splice_read    = generic_file_splice_read,
1042         .splice_write   = iter_file_splice_write,
1043         .fallocate      = bch2_fallocate_dispatch,
1044         .unlocked_ioctl = bch2_fs_file_ioctl,
1045 #ifdef CONFIG_COMPAT
1046         .compat_ioctl   = bch2_compat_fs_ioctl,
1047 #endif
1048         .remap_file_range = bch2_remap_file_range,
1049 };
1050
1051 static const struct inode_operations bch_file_inode_operations = {
1052         .getattr        = bch2_getattr,
1053         .setattr        = bch2_setattr,
1054         .fiemap         = bch2_fiemap,
1055         .listxattr      = bch2_xattr_list,
1056 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1057         .get_acl        = bch2_get_acl,
1058         .set_acl        = bch2_set_acl,
1059 #endif
1060 };
1061
1062 static const struct inode_operations bch_dir_inode_operations = {
1063         .lookup         = bch2_lookup,
1064         .create         = bch2_create,
1065         .link           = bch2_link,
1066         .unlink         = bch2_unlink,
1067         .symlink        = bch2_symlink,
1068         .mkdir          = bch2_mkdir,
1069         .rmdir          = bch2_unlink,
1070         .mknod          = bch2_mknod,
1071         .rename         = bch2_rename2,
1072         .getattr        = bch2_getattr,
1073         .setattr        = bch2_setattr,
1074         .tmpfile        = bch2_tmpfile,
1075         .listxattr      = bch2_xattr_list,
1076 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1077         .get_acl        = bch2_get_acl,
1078         .set_acl        = bch2_set_acl,
1079 #endif
1080 };
1081
1082 static const struct file_operations bch_dir_file_operations = {
1083         .llseek         = bch2_dir_llseek,
1084         .read           = generic_read_dir,
1085         .iterate_shared = bch2_vfs_readdir,
1086         .fsync          = bch2_fsync,
1087         .unlocked_ioctl = bch2_fs_file_ioctl,
1088 #ifdef CONFIG_COMPAT
1089         .compat_ioctl   = bch2_compat_fs_ioctl,
1090 #endif
1091 };
1092
1093 static const struct inode_operations bch_symlink_inode_operations = {
1094         .get_link       = page_get_link,
1095         .getattr        = bch2_getattr,
1096         .setattr        = bch2_setattr,
1097         .listxattr      = bch2_xattr_list,
1098 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1099         .get_acl        = bch2_get_acl,
1100         .set_acl        = bch2_set_acl,
1101 #endif
1102 };
1103
1104 static const struct inode_operations bch_special_inode_operations = {
1105         .getattr        = bch2_getattr,
1106         .setattr        = bch2_setattr,
1107         .listxattr      = bch2_xattr_list,
1108 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1109         .get_acl        = bch2_get_acl,
1110         .set_acl        = bch2_set_acl,
1111 #endif
1112 };
1113
1114 static const struct address_space_operations bch_address_space_operations = {
1115         .readpage       = bch2_readpage,
1116         .writepages     = bch2_writepages,
1117         .readahead      = bch2_readahead,
1118         .set_page_dirty = __set_page_dirty_nobuffers,
1119         .write_begin    = bch2_write_begin,
1120         .write_end      = bch2_write_end,
1121         .invalidatepage = bch2_invalidatepage,
1122         .releasepage    = bch2_releasepage,
1123         .direct_IO      = noop_direct_IO,
1124 #ifdef CONFIG_MIGRATION
1125         .migratepage    = bch2_migrate_page,
1126 #endif
1127         .error_remove_page = generic_error_remove_page,
1128 };
1129
1130 struct bcachefs_fid {
1131         u64             inum;
1132         u32             subvol;
1133         u32             gen;
1134 } __packed;
1135
1136 struct bcachefs_fid_with_parent {
1137         struct bcachefs_fid     fid;
1138         struct bcachefs_fid     dir;
1139 } __packed;
1140
1141 static int bcachefs_fid_valid(int fh_len, int fh_type)
1142 {
1143         switch (fh_type) {
1144         case FILEID_BCACHEFS_WITHOUT_PARENT:
1145                 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1146         case FILEID_BCACHEFS_WITH_PARENT:
1147                 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1148         default:
1149                 return false;
1150         }
1151 }
1152
1153 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1154 {
1155         return (struct bcachefs_fid) {
1156                 .inum   = inode->ei_inode.bi_inum,
1157                 .subvol = inode->ei_subvol,
1158                 .gen    = inode->ei_inode.bi_generation,
1159         };
1160 }
1161
1162 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1163                           struct inode *vdir)
1164 {
1165         struct bch_inode_info *inode    = to_bch_ei(vinode);
1166         struct bch_inode_info *dir      = to_bch_ei(vdir);
1167
1168         if (*len < sizeof(struct bcachefs_fid_with_parent) / sizeof(u32))
1169                 return FILEID_INVALID;
1170
1171         if (!S_ISDIR(inode->v.i_mode) && dir) {
1172                 struct bcachefs_fid_with_parent *fid = (void *) fh;
1173
1174                 fid->fid = bch2_inode_to_fid(inode);
1175                 fid->dir = bch2_inode_to_fid(dir);
1176
1177                 *len = sizeof(*fid) / sizeof(u32);
1178                 return FILEID_BCACHEFS_WITH_PARENT;
1179         } else {
1180                 struct bcachefs_fid *fid = (void *) fh;
1181
1182                 *fid = bch2_inode_to_fid(inode);
1183
1184                 *len = sizeof(*fid) / sizeof(u32);
1185                 return FILEID_BCACHEFS_WITHOUT_PARENT;
1186         }
1187 }
1188
1189 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1190                                         struct bcachefs_fid fid)
1191 {
1192         struct bch_fs *c = sb->s_fs_info;
1193         struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1194                                     .subvol = fid.subvol,
1195                                     .inum = fid.inum,
1196         });
1197         if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1198                 iput(vinode);
1199                 vinode = ERR_PTR(-ESTALE);
1200         }
1201         return vinode;
1202 }
1203
1204 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1205                 int fh_len, int fh_type)
1206 {
1207         struct bcachefs_fid *fid = (void *) _fid;
1208
1209         if (!bcachefs_fid_valid(fh_len, fh_type))
1210                 return NULL;
1211
1212         return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1213 }
1214
1215 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1216                 int fh_len, int fh_type)
1217 {
1218         struct bcachefs_fid_with_parent *fid = (void *) _fid;
1219
1220         if (!bcachefs_fid_valid(fh_len, fh_type) ||
1221             fh_type != FILEID_BCACHEFS_WITH_PARENT)
1222                 return NULL;
1223
1224         return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1225 }
1226
1227 static struct dentry *bch2_get_parent(struct dentry *child)
1228 {
1229         struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1230         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1231         subvol_inum parent_inum = {
1232                 .subvol = inode->ei_inode.bi_parent_subvol ?:
1233                         inode->ei_subvol,
1234                 .inum = inode->ei_inode.bi_dir,
1235         };
1236
1237         if (!parent_inum.inum)
1238                 return NULL;
1239
1240         return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1241 }
1242
1243 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1244 {
1245         struct bch_inode_info *inode    = to_bch_ei(child->d_inode);
1246         struct bch_inode_info *dir      = to_bch_ei(parent->d_inode);
1247         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1248         struct btree_trans trans;
1249         struct btree_iter iter1;
1250         struct btree_iter iter2;
1251         struct bkey_s_c k;
1252         struct bkey_s_c_dirent d;
1253         struct bch_inode_unpacked inode_u;
1254         subvol_inum target;
1255         u32 snapshot;
1256         unsigned name_len;
1257         int ret;
1258
1259         if (!S_ISDIR(dir->v.i_mode))
1260                 return -EINVAL;
1261
1262         bch2_trans_init(&trans, c, 0, 0);
1263
1264         bch2_trans_iter_init(&trans, &iter1, BTREE_ID_dirents,
1265                              POS(dir->ei_inode.bi_inum, 0), 0);
1266         bch2_trans_iter_init(&trans, &iter2, BTREE_ID_dirents,
1267                              POS(dir->ei_inode.bi_inum, 0), 0);
1268 retry:
1269         bch2_trans_begin(&trans);
1270
1271         ret = bch2_subvolume_get_snapshot(&trans, dir->ei_subvol, &snapshot);
1272         if (ret)
1273                 goto err;
1274
1275         bch2_btree_iter_set_snapshot(&iter1, snapshot);
1276         bch2_btree_iter_set_snapshot(&iter2, snapshot);
1277
1278         ret = bch2_inode_find_by_inum_trans(&trans, inode_inum(inode), &inode_u);
1279         if (ret)
1280                 goto err;
1281
1282         if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1283                 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1284
1285                 k = bch2_btree_iter_peek_slot(&iter1);
1286                 ret = bkey_err(k);
1287                 if (ret)
1288                         goto err;
1289
1290                 if (k.k->type != KEY_TYPE_dirent) {
1291                         ret = -ENOENT;
1292                         goto err;
1293                 }
1294
1295                 d = bkey_s_c_to_dirent(k);
1296                 ret = bch2_dirent_read_target(&trans, inode_inum(dir), d, &target);
1297                 if (ret > 0)
1298                         ret = -ENOENT;
1299                 if (ret)
1300                         goto err;
1301
1302                 if (target.subvol       == inode->ei_subvol &&
1303                     target.inum         == inode->ei_inode.bi_inum)
1304                         goto found;
1305         } else {
1306                 /*
1307                  * File with multiple hardlinks and our backref is to the wrong
1308                  * directory - linear search:
1309                  */
1310                 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1311                         if (k.k->p.inode > dir->ei_inode.bi_inum)
1312                                 break;
1313
1314                         if (k.k->type != KEY_TYPE_dirent)
1315                                 continue;
1316
1317                         d = bkey_s_c_to_dirent(k);
1318                         ret = bch2_dirent_read_target(&trans, inode_inum(dir), d, &target);
1319                         if (ret < 0)
1320                                 break;
1321                         if (ret)
1322                                 continue;
1323
1324                         if (target.subvol       == inode->ei_subvol &&
1325                             target.inum         == inode->ei_inode.bi_inum)
1326                                 goto found;
1327                 }
1328         }
1329
1330         ret = -ENOENT;
1331         goto err;
1332 found:
1333         name_len = min_t(unsigned, bch2_dirent_name_bytes(d), NAME_MAX);
1334
1335         memcpy(name, d.v->d_name, name_len);
1336         name[name_len] = '\0';
1337 err:
1338         if (ret == -EINTR)
1339                 goto retry;
1340
1341         bch2_trans_iter_exit(&trans, &iter1);
1342         bch2_trans_iter_exit(&trans, &iter2);
1343         bch2_trans_exit(&trans);
1344
1345         return ret;
1346 }
1347
1348 static const struct export_operations bch_export_ops = {
1349         .encode_fh      = bch2_encode_fh,
1350         .fh_to_dentry   = bch2_fh_to_dentry,
1351         .fh_to_parent   = bch2_fh_to_parent,
1352         .get_parent     = bch2_get_parent,
1353         .get_name       = bch2_get_name,
1354 };
1355
1356 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1357                                 struct bch_inode_info *inode,
1358                                 struct bch_inode_unpacked *bi,
1359                                 struct bch_subvolume *subvol)
1360 {
1361         bch2_inode_update_after_write(trans, inode, bi, ~0);
1362
1363         if (BCH_SUBVOLUME_SNAP(subvol))
1364                 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1365         else
1366                 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1367
1368         inode->v.i_blocks       = bi->bi_sectors;
1369         inode->v.i_ino          = bi->bi_inum;
1370         inode->v.i_rdev         = bi->bi_dev;
1371         inode->v.i_generation   = bi->bi_generation;
1372         inode->v.i_size         = bi->bi_size;
1373
1374         inode->ei_flags         = 0;
1375         inode->ei_quota_reserved = 0;
1376         inode->ei_qid           = bch_qid(bi);
1377         inode->ei_subvol        = inum.subvol;
1378
1379         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1380
1381         switch (inode->v.i_mode & S_IFMT) {
1382         case S_IFREG:
1383                 inode->v.i_op   = &bch_file_inode_operations;
1384                 inode->v.i_fop  = &bch_file_operations;
1385                 break;
1386         case S_IFDIR:
1387                 inode->v.i_op   = &bch_dir_inode_operations;
1388                 inode->v.i_fop  = &bch_dir_file_operations;
1389                 break;
1390         case S_IFLNK:
1391                 inode_nohighmem(&inode->v);
1392                 inode->v.i_op   = &bch_symlink_inode_operations;
1393                 break;
1394         default:
1395                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1396                 inode->v.i_op   = &bch_special_inode_operations;
1397                 break;
1398         }
1399 }
1400
1401 static struct inode *bch2_alloc_inode(struct super_block *sb)
1402 {
1403         struct bch_inode_info *inode;
1404
1405         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1406         if (!inode)
1407                 return NULL;
1408
1409         inode_init_once(&inode->v);
1410         mutex_init(&inode->ei_update_lock);
1411         pagecache_lock_init(&inode->ei_pagecache_lock);
1412         mutex_init(&inode->ei_quota_lock);
1413
1414         return &inode->v;
1415 }
1416
1417 static void bch2_i_callback(struct rcu_head *head)
1418 {
1419         struct inode *vinode = container_of(head, struct inode, i_rcu);
1420         struct bch_inode_info *inode = to_bch_ei(vinode);
1421
1422         kmem_cache_free(bch2_inode_cache, inode);
1423 }
1424
1425 static void bch2_destroy_inode(struct inode *vinode)
1426 {
1427         call_rcu(&vinode->i_rcu, bch2_i_callback);
1428 }
1429
1430 static int inode_update_times_fn(struct bch_inode_info *inode,
1431                                  struct bch_inode_unpacked *bi,
1432                                  void *p)
1433 {
1434         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1435
1436         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1437         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1438         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1439
1440         return 0;
1441 }
1442
1443 static int bch2_vfs_write_inode(struct inode *vinode,
1444                                 struct writeback_control *wbc)
1445 {
1446         struct bch_fs *c = vinode->i_sb->s_fs_info;
1447         struct bch_inode_info *inode = to_bch_ei(vinode);
1448         int ret;
1449
1450         mutex_lock(&inode->ei_update_lock);
1451         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1452                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1453         mutex_unlock(&inode->ei_update_lock);
1454
1455         return ret;
1456 }
1457
1458 static void bch2_evict_inode(struct inode *vinode)
1459 {
1460         struct bch_fs *c = vinode->i_sb->s_fs_info;
1461         struct bch_inode_info *inode = to_bch_ei(vinode);
1462
1463         truncate_inode_pages_final(&inode->v.i_data);
1464
1465         clear_inode(&inode->v);
1466
1467         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1468
1469         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1470                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1471                                 KEY_TYPE_QUOTA_WARN);
1472                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1473                                 KEY_TYPE_QUOTA_WARN);
1474                 bch2_inode_rm(c, inode_inum(inode));
1475         }
1476 }
1477
1478 void bch2_evict_subvolume_inodes(struct bch_fs *c,
1479                                  snapshot_id_list *s)
1480 {
1481         struct super_block *sb = c->vfs_sb;
1482         struct inode *inode;
1483
1484         spin_lock(&sb->s_inode_list_lock);
1485         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1486                 if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
1487                     (inode->i_state & I_FREEING))
1488                         continue;
1489
1490                 d_mark_dontcache(inode);
1491                 d_prune_aliases(inode);
1492         }
1493         spin_unlock(&sb->s_inode_list_lock);
1494 again:
1495         cond_resched();
1496         spin_lock(&sb->s_inode_list_lock);
1497         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1498                 if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
1499                     (inode->i_state & I_FREEING))
1500                         continue;
1501
1502                 if (!(inode->i_state & I_DONTCACHE)) {
1503                         d_mark_dontcache(inode);
1504                         d_prune_aliases(inode);
1505                 }
1506
1507                 spin_lock(&inode->i_lock);
1508                 if (snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) &&
1509                     !(inode->i_state & I_FREEING)) {
1510                         wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_NEW);
1511                         DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1512                         prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1513                         spin_unlock(&inode->i_lock);
1514                         spin_unlock(&sb->s_inode_list_lock);
1515                         schedule();
1516                         finish_wait(wq, &wait.wq_entry);
1517                         goto again;
1518                 }
1519
1520                 spin_unlock(&inode->i_lock);
1521         }
1522         spin_unlock(&sb->s_inode_list_lock);
1523 }
1524
1525 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1526 {
1527         struct super_block *sb = dentry->d_sb;
1528         struct bch_fs *c = sb->s_fs_info;
1529         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1530         unsigned shift = sb->s_blocksize_bits - 9;
1531         /*
1532          * this assumes inodes take up 64 bytes, which is a decent average
1533          * number:
1534          */
1535         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1536         u64 fsid;
1537
1538         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1539         buf->f_bsize    = sb->s_blocksize;
1540         buf->f_blocks   = usage.capacity >> shift;
1541         buf->f_bfree    = usage.free >> shift;
1542         buf->f_bavail   = avail_factor(usage.free) >> shift;
1543
1544         buf->f_files    = usage.nr_inodes + avail_inodes;
1545         buf->f_ffree    = avail_inodes;
1546
1547         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1548                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1549         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1550         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1551         buf->f_namelen  = BCH_NAME_MAX;
1552
1553         return 0;
1554 }
1555
1556 static int bch2_sync_fs(struct super_block *sb, int wait)
1557 {
1558         struct bch_fs *c = sb->s_fs_info;
1559
1560         if (c->opts.journal_flush_disabled)
1561                 return 0;
1562
1563         if (!wait) {
1564                 bch2_journal_flush_async(&c->journal, NULL);
1565                 return 0;
1566         }
1567
1568         return bch2_journal_flush(&c->journal);
1569 }
1570
1571 static struct bch_fs *bch2_path_to_fs(const char *path)
1572 {
1573         struct bch_fs *c;
1574         dev_t dev;
1575         int ret;
1576
1577         ret = lookup_bdev(path, &dev);
1578         if (ret)
1579                 return ERR_PTR(ret);
1580
1581         c = bch2_dev_to_fs(dev);
1582         if (c)
1583                 closure_put(&c->cl);
1584         return c ?: ERR_PTR(-ENOENT);
1585 }
1586
1587 static char **split_devs(const char *_dev_name, unsigned *nr)
1588 {
1589         char *dev_name = NULL, **devs = NULL, *s;
1590         size_t i, nr_devs = 0;
1591
1592         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1593         if (!dev_name)
1594                 return NULL;
1595
1596         for (s = dev_name; s; s = strchr(s + 1, ':'))
1597                 nr_devs++;
1598
1599         devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1600         if (!devs) {
1601                 kfree(dev_name);
1602                 return NULL;
1603         }
1604
1605         for (i = 0, s = dev_name;
1606              s;
1607              (s = strchr(s, ':')) && (*s++ = '\0'))
1608                 devs[i++] = s;
1609
1610         *nr = nr_devs;
1611         return devs;
1612 }
1613
1614 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1615 {
1616         struct bch_fs *c = sb->s_fs_info;
1617         struct bch_opts opts = bch2_opts_empty();
1618         int ret;
1619
1620         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1621
1622         ret = bch2_parse_mount_opts(c, &opts, data);
1623         if (ret)
1624                 return ret;
1625
1626         if (opts.read_only != c->opts.read_only) {
1627                 down_write(&c->state_lock);
1628
1629                 if (opts.read_only) {
1630                         bch2_fs_read_only(c);
1631
1632                         sb->s_flags |= SB_RDONLY;
1633                 } else {
1634                         ret = bch2_fs_read_write(c);
1635                         if (ret) {
1636                                 bch_err(c, "error going rw: %i", ret);
1637                                 up_write(&c->state_lock);
1638                                 return -EINVAL;
1639                         }
1640
1641                         sb->s_flags &= ~SB_RDONLY;
1642                 }
1643
1644                 c->opts.read_only = opts.read_only;
1645
1646                 up_write(&c->state_lock);
1647         }
1648
1649         if (opts.errors >= 0)
1650                 c->opts.errors = opts.errors;
1651
1652         return ret;
1653 }
1654
1655 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1656 {
1657         struct bch_fs *c = root->d_sb->s_fs_info;
1658         struct bch_dev *ca;
1659         unsigned i;
1660         bool first = true;
1661
1662         for_each_online_member(ca, c, i) {
1663                 if (!first)
1664                         seq_putc(seq, ':');
1665                 first = false;
1666                 seq_puts(seq, "/dev/");
1667                 seq_puts(seq, ca->name);
1668         }
1669
1670         return 0;
1671 }
1672
1673 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1674 {
1675         struct bch_fs *c = root->d_sb->s_fs_info;
1676         enum bch_opt_id i;
1677         struct printbuf buf = PRINTBUF;
1678         int ret = 0;
1679
1680         for (i = 0; i < bch2_opts_nr; i++) {
1681                 const struct bch_option *opt = &bch2_opt_table[i];
1682                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1683
1684                 if (!(opt->flags & OPT_MOUNT))
1685                         continue;
1686
1687                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1688                         continue;
1689
1690                 printbuf_reset(&buf);
1691                 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1692                                  OPT_SHOW_MOUNT_STYLE);
1693                 seq_putc(seq, ',');
1694                 seq_puts(seq, buf.buf);
1695         }
1696
1697         if (buf.allocation_failure)
1698                 ret = -ENOMEM;
1699         printbuf_exit(&buf);
1700         return ret;
1701 }
1702
1703 static void bch2_put_super(struct super_block *sb)
1704 {
1705         struct bch_fs *c = sb->s_fs_info;
1706
1707         __bch2_fs_stop(c);
1708 }
1709
1710 static const struct super_operations bch_super_operations = {
1711         .alloc_inode    = bch2_alloc_inode,
1712         .destroy_inode  = bch2_destroy_inode,
1713         .write_inode    = bch2_vfs_write_inode,
1714         .evict_inode    = bch2_evict_inode,
1715         .sync_fs        = bch2_sync_fs,
1716         .statfs         = bch2_statfs,
1717         .show_devname   = bch2_show_devname,
1718         .show_options   = bch2_show_options,
1719         .remount_fs     = bch2_remount,
1720         .put_super      = bch2_put_super,
1721 #if 0
1722         .freeze_fs      = bch2_freeze,
1723         .unfreeze_fs    = bch2_unfreeze,
1724 #endif
1725 };
1726
1727 static int bch2_set_super(struct super_block *s, void *data)
1728 {
1729         s->s_fs_info = data;
1730         return 0;
1731 }
1732
1733 static int bch2_noset_super(struct super_block *s, void *data)
1734 {
1735         return -EBUSY;
1736 }
1737
1738 static int bch2_test_super(struct super_block *s, void *data)
1739 {
1740         struct bch_fs *c = s->s_fs_info;
1741         struct bch_fs **devs = data;
1742         unsigned i;
1743
1744         if (!c)
1745                 return false;
1746
1747         for (i = 0; devs[i]; i++)
1748                 if (c != devs[i])
1749                         return false;
1750         return true;
1751 }
1752
1753 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1754                                  int flags, const char *dev_name, void *data)
1755 {
1756         struct bch_fs *c;
1757         struct bch_dev *ca;
1758         struct super_block *sb;
1759         struct inode *vinode;
1760         struct bch_opts opts = bch2_opts_empty();
1761         char **devs;
1762         struct bch_fs **devs_to_fs = NULL;
1763         unsigned i, nr_devs;
1764         int ret;
1765
1766         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1767
1768         ret = bch2_parse_mount_opts(NULL, &opts, data);
1769         if (ret)
1770                 return ERR_PTR(ret);
1771
1772         if (!dev_name || strlen(dev_name) == 0)
1773                 return ERR_PTR(-EINVAL);
1774
1775         devs = split_devs(dev_name, &nr_devs);
1776         if (!devs)
1777                 return ERR_PTR(-ENOMEM);
1778
1779         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1780         if (!devs_to_fs) {
1781                 sb = ERR_PTR(-ENOMEM);
1782                 goto got_sb;
1783         }
1784
1785         for (i = 0; i < nr_devs; i++)
1786                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1787
1788         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1789                   flags|SB_NOSEC, devs_to_fs);
1790         if (!IS_ERR(sb))
1791                 goto got_sb;
1792
1793         c = bch2_fs_open(devs, nr_devs, opts);
1794         if (IS_ERR(c)) {
1795                 sb = ERR_CAST(c);
1796                 goto got_sb;
1797         }
1798
1799         /* Some options can't be parsed until after the fs is started: */
1800         ret = bch2_parse_mount_opts(c, &opts, data);
1801         if (ret) {
1802                 bch2_fs_stop(c);
1803                 sb = ERR_PTR(ret);
1804                 goto got_sb;
1805         }
1806
1807         bch2_opts_apply(&c->opts, opts);
1808
1809         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1810         if (IS_ERR(sb))
1811                 bch2_fs_stop(c);
1812 got_sb:
1813         kfree(devs_to_fs);
1814         kfree(devs[0]);
1815         kfree(devs);
1816
1817         if (IS_ERR(sb))
1818                 return ERR_CAST(sb);
1819
1820         c = sb->s_fs_info;
1821
1822         if (sb->s_root) {
1823                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1824                         ret = -EBUSY;
1825                         goto err_put_super;
1826                 }
1827                 goto out;
1828         }
1829
1830         sb->s_blocksize         = block_bytes(c);
1831         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1832         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1833         sb->s_op                = &bch_super_operations;
1834         sb->s_export_op         = &bch_export_ops;
1835 #ifdef CONFIG_BCACHEFS_QUOTA
1836         sb->s_qcop              = &bch2_quotactl_operations;
1837         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1838 #endif
1839         sb->s_xattr             = bch2_xattr_handlers;
1840         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1841         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1842         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1843         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1844         c->vfs_sb               = sb;
1845         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1846
1847         ret = super_setup_bdi(sb);
1848         if (ret)
1849                 goto err_put_super;
1850
1851         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1852
1853         for_each_online_member(ca, c, i) {
1854                 struct block_device *bdev = ca->disk_sb.bdev;
1855
1856                 /* XXX: create an anonymous device for multi device filesystems */
1857                 sb->s_bdev      = bdev;
1858                 sb->s_dev       = bdev->bd_dev;
1859                 percpu_ref_put(&ca->io_ref);
1860                 break;
1861         }
1862
1863         c->dev = sb->s_dev;
1864
1865 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1866         if (c->opts.acl)
1867                 sb->s_flags     |= SB_POSIXACL;
1868 #endif
1869
1870         sb->s_shrink.seeks = 0;
1871
1872         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
1873         if (IS_ERR(vinode)) {
1874                 bch_err(c, "error mounting: error getting root inode %i",
1875                         (int) PTR_ERR(vinode));
1876                 ret = PTR_ERR(vinode);
1877                 goto err_put_super;
1878         }
1879
1880         sb->s_root = d_make_root(vinode);
1881         if (!sb->s_root) {
1882                 bch_err(c, "error mounting: error allocating root dentry");
1883                 ret = -ENOMEM;
1884                 goto err_put_super;
1885         }
1886
1887         sb->s_flags |= SB_ACTIVE;
1888 out:
1889         return dget(sb->s_root);
1890
1891 err_put_super:
1892         deactivate_locked_super(sb);
1893         return ERR_PTR(ret);
1894 }
1895
1896 static void bch2_kill_sb(struct super_block *sb)
1897 {
1898         struct bch_fs *c = sb->s_fs_info;
1899
1900         generic_shutdown_super(sb);
1901         bch2_fs_free(c);
1902 }
1903
1904 static struct file_system_type bcache_fs_type = {
1905         .owner          = THIS_MODULE,
1906         .name           = "bcachefs",
1907         .mount          = bch2_mount,
1908         .kill_sb        = bch2_kill_sb,
1909         .fs_flags       = FS_REQUIRES_DEV,
1910 };
1911
1912 MODULE_ALIAS_FS("bcachefs");
1913
1914 void bch2_vfs_exit(void)
1915 {
1916         unregister_filesystem(&bcache_fs_type);
1917         if (bch2_inode_cache)
1918                 kmem_cache_destroy(bch2_inode_cache);
1919 }
1920
1921 int __init bch2_vfs_init(void)
1922 {
1923         int ret = -ENOMEM;
1924
1925         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1926         if (!bch2_inode_cache)
1927                 goto err;
1928
1929         ret = register_filesystem(&bcache_fs_type);
1930         if (ret)
1931                 goto err;
1932
1933         return 0;
1934 err:
1935         bch2_vfs_exit();
1936         return ret;
1937 }
1938
1939 #endif /* NO_BCACHEFS_FS */