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

   1 #ifndef NO_BCACHEFS_FS
   2
   3 #include "bcachefs.h"
   4 #include "acl.h"
   5 #include "btree_update.h"
   6 #include "buckets.h"
   7 #include "chardev.h"
   8 #include "dirent.h"
   9 #include "extents.h"
  10 #include "fs.h"
  11 #include "fs-io.h"
  12 #include "fs-ioctl.h"
  13 #include "fsck.h"
  14 #include "inode.h"
  15 #include "io.h"
  16 #include "journal.h"
  17 #include "keylist.h"
  18 #include "quota.h"
  19 #include "super.h"
  20 #include "xattr.h"
  21
  22 #include <linux/aio.h>
  23 #include <linux/backing-dev.h>
  24 #include <linux/exportfs.h>
  25 #include <linux/module.h>
  26 #include <linux/posix_acl.h>
  27 #include <linux/random.h>
  28 #include <linux/statfs.h>
  29 #include <linux/xattr.h>
  30
  31 static struct kmem_cache *bch2_inode_cache;
  32
  33 static void bch2_vfs_inode_init(struct bch_fs *,
  34                                 struct bch_inode_info *,
  35                                 struct bch_inode_unpacked *);
  36
  37 /*
  38  * I_SIZE_DIRTY requires special handling:
  39  *
  40  * To the recovery code, the flag means that there is stale data past i_size
  41  * that needs to be deleted; it's used for implementing atomic appends and
  42  * truncates.
  43  *
  44  * On append, we set I_SIZE_DIRTY before doing the write, then after the write
  45  * we clear I_SIZE_DIRTY atomically with updating i_size to the new larger size
  46  * that exposes the data we just wrote.
  47  *
  48  * On truncate, it's the reverse: We set I_SIZE_DIRTY atomically with setting
  49  * i_size to the new smaller size, then we delete the data that we just made
  50  * invisible, and then we clear I_SIZE_DIRTY.
  51  *
  52  * Because there can be multiple appends in flight at a time, we need a refcount
  53  * (i_size_dirty_count) instead of manipulating the flag directly. Nonzero
  54  * refcount means I_SIZE_DIRTY is set, zero means it's cleared.
  55  *
  56  * Because write_inode() can be called at any time, i_size_dirty_count means
  57  * something different to the runtime code - it means to write_inode() "don't
  58  * update i_size yet".
  59  *
  60  * We don't clear I_SIZE_DIRTY directly, we let write_inode() clear it when
  61  * i_size_dirty_count is zero - but the reverse is not true, I_SIZE_DIRTY must
  62  * be set explicitly.
  63  */
  64
  65 int __must_check __bch2_write_inode(struct bch_fs *c,
  66                                     struct bch_inode_info *inode,
  67                                     inode_set_fn set,
  68                                     void *p)
  69 {
  70         struct btree_iter iter;
  71         struct bch_inode_unpacked inode_u;
  72         struct bkey_inode_buf inode_p;
  73         u64 inum = inode->v.i_ino;
  74         unsigned i_nlink = READ_ONCE(inode->v.i_nlink);
  75         int ret;
  76
  77         /*
  78          * We can't write an inode with i_nlink == 0 because it's stored biased;
  79          * however, we don't need to because if i_nlink is 0 the inode is
  80          * getting deleted when it's evicted.
  81          */
  82         if (!i_nlink)
  83                 return 0;
  84
  85         lockdep_assert_held(&inode->ei_update_lock);
  86
  87         bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(inum, 0),
  88                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
  89
  90         do {
  91                 struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
  92
  93                 if ((ret = btree_iter_err(k)))
  94                         goto out;
  95
  96                 if (WARN_ONCE(k.k->type != BCH_INODE_FS,
  97                               "inode %llu not found when updating", inum)) {
  98                         bch2_btree_iter_unlock(&iter);
  99                         return -ENOENT;
 100                 }
 101
 102                 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u);
 103                 if (WARN_ONCE(ret,
 104                               "error %i unpacking inode %llu", ret, inum)) {
 105                         ret = -ENOENT;
 106                         break;
 107                 }
 108
 109                 if (set) {
 110                         ret = set(inode, &inode_u, p);
 111                         if (ret)
 112                                 goto out;
 113                 }
 114
 115                 BUG_ON(i_nlink < nlink_bias(inode->v.i_mode));
 116
 117                 inode_u.bi_mode = inode->v.i_mode;
 118                 inode_u.bi_uid  = i_uid_read(&inode->v);
 119                 inode_u.bi_gid  = i_gid_read(&inode->v);
 120                 inode_u.bi_project = inode->ei_qid.q[QTYP_PRJ];
 121                 inode_u.bi_nlink= i_nlink - nlink_bias(inode->v.i_mode);
 122                 inode_u.bi_dev  = inode->v.i_rdev;
 123                 inode_u.bi_atime= timespec_to_bch2_time(c, inode->v.i_atime);
 124                 inode_u.bi_mtime= timespec_to_bch2_time(c, inode->v.i_mtime);
 125                 inode_u.bi_ctime= timespec_to_bch2_time(c, inode->v.i_ctime);
 126
 127                 bch2_inode_pack(&inode_p, &inode_u);
 128
 129                 ret = bch2_btree_insert_at(c, NULL, NULL,
 130                                 &inode->ei_journal_seq,
 131                                 BTREE_INSERT_ATOMIC|
 132                                 BTREE_INSERT_NOFAIL,
 133                                 BTREE_INSERT_ENTRY(&iter, &inode_p.inode.k_i));
 134         } while (ret == -EINTR);
 135
 136         if (!ret) {
 137                 inode->ei_inode = inode_u;
 138                 inode->ei_qid   = bch_qid(&inode_u);
 139         }
 140 out:
 141         bch2_btree_iter_unlock(&iter);
 142
 143         return ret < 0 ? ret : 0;
 144 }
 145
 146 int __must_check bch2_write_inode(struct bch_fs *c,
 147                                   struct bch_inode_info *inode)
 148 {
 149         return __bch2_write_inode(c, inode, NULL, NULL);
 150 }
 151
 152 static int bch2_inc_nlink(struct bch_fs *c, struct bch_inode_info *inode)
 153 {
 154         int ret;
 155
 156         mutex_lock(&inode->ei_update_lock);
 157         inc_nlink(&inode->v);
 158         ret = bch2_write_inode(c, inode);
 159         mutex_unlock(&inode->ei_update_lock);
 160
 161         return ret;
 162 }
 163
 164 static int bch2_dec_nlink(struct bch_fs *c, struct bch_inode_info *inode)
 165 {
 166         int ret = 0;
 167
 168         mutex_lock(&inode->ei_update_lock);
 169         drop_nlink(&inode->v);
 170         ret = bch2_write_inode(c, inode);
 171         mutex_unlock(&inode->ei_update_lock);
 172
 173         return ret;
 174 }
 175
 176 static struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 177 {
 178         struct bch_inode_unpacked inode_u;
 179         struct bch_inode_info *inode;
 180         int ret;
 181
 182         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 183         if (unlikely(!inode))
 184                 return ERR_PTR(-ENOMEM);
 185         if (!(inode->v.i_state & I_NEW))
 186                 return &inode->v;
 187
 188         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 189         if (ret) {
 190                 iget_failed(&inode->v);
 191                 return ERR_PTR(ret);
 192         }
 193
 194         bch2_vfs_inode_init(c, inode, &inode_u);
 195
 196         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 197
 198         unlock_new_inode(&inode->v);
 199
 200         return &inode->v;
 201 }
 202
 203 static struct bch_inode_info *bch2_vfs_inode_create(struct bch_fs *c,
 204                                                     struct bch_inode_info *dir,
 205                                                     umode_t mode, dev_t rdev)
 206 {
 207         struct posix_acl *default_acl = NULL, *acl = NULL;
 208         struct bch_inode_info *inode;
 209         struct bch_inode_unpacked inode_u;
 210         int ret;
 211
 212         inode = to_bch_ei(new_inode(c->vfs_sb));
 213         if (unlikely(!inode))
 214                 return ERR_PTR(-ENOMEM);
 215
 216         inode_init_owner(&inode->v, &dir->v, mode);
 217
 218 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 219         ret = posix_acl_create(&dir->v, &inode->v.i_mode, &default_acl, &acl);
 220         if (ret)
 221                 goto err_make_bad;
 222 #endif
 223
 224         bch2_inode_init(c, &inode_u,
 225                         i_uid_read(&inode->v),
 226                         i_gid_read(&inode->v),
 227                         inode->v.i_mode, rdev,
 228                         &dir->ei_inode);
 229
 230         inode_u.bi_project = dir->ei_qid.q[QTYP_PRJ];
 231
 232         ret = bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1, BCH_QUOTA_PREALLOC);
 233         if (ret)
 234                 goto err_make_bad;
 235
 236         ret = bch2_inode_create(c, &inode_u,
 237                                 BLOCKDEV_INODE_MAX, 0,
 238                                 &c->unused_inode_hint);
 239         if (unlikely(ret))
 240                 goto err_acct_quota;
 241
 242         bch2_vfs_inode_init(c, inode, &inode_u);
 243         atomic_long_inc(&c->nr_inodes);
 244
 245         if (default_acl) {
 246                 ret = bch2_set_acl(&inode->v, default_acl, ACL_TYPE_DEFAULT);
 247                 if (unlikely(ret))
 248                         goto err;
 249         }
 250
 251         if (acl) {
 252                 ret = bch2_set_acl(&inode->v, acl, ACL_TYPE_ACCESS);
 253                 if (unlikely(ret))
 254                         goto err;
 255         }
 256
 257         insert_inode_hash(&inode->v);
 258 out:
 259         posix_acl_release(default_acl);
 260         posix_acl_release(acl);
 261         return inode;
 262 err_acct_quota:
 263         bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1, BCH_QUOTA_WARN);
 264 err_make_bad:
 265         /*
 266          * indicate to bch_evict_inode that the inode was never actually
 267          * created:
 268          */
 269         make_bad_inode(&inode->v);
 270 err:
 271         clear_nlink(&inode->v);
 272         iput(&inode->v);
 273         inode = ERR_PTR(ret);
 274         goto out;
 275 }
 276
 277 static int bch2_vfs_dirent_create(struct bch_fs *c,
 278                                   struct bch_inode_info *dir,
 279                                   u8 type, const struct qstr *name,
 280                                   u64 dst)
 281 {
 282         int ret;
 283
 284         ret = bch2_dirent_create(c, dir->v.i_ino, &dir->ei_str_hash,
 285                                 type, name, dst,
 286                                 &dir->ei_journal_seq,
 287                                 BCH_HASH_SET_MUST_CREATE);
 288         if (unlikely(ret))
 289                 return ret;
 290
 291         dir->v.i_mtime = dir->v.i_ctime = current_time(&dir->v);
 292         mark_inode_dirty_sync(&dir->v);
 293         return 0;
 294 }
 295
 296 static int __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 297                          umode_t mode, dev_t rdev)
 298 {
 299         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 300         struct bch_inode_info *inode;
 301         int ret;
 302
 303         inode = bch2_vfs_inode_create(c, dir, mode, rdev);
 304         if (unlikely(IS_ERR(inode)))
 305                 return PTR_ERR(inode);
 306
 307         ret = bch2_vfs_dirent_create(c, dir, mode_to_type(mode),
 308                                      &dentry->d_name, inode->v.i_ino);
 309         if (unlikely(ret)) {
 310                 clear_nlink(&inode->v);
 311                 iput(&inode->v);
 312                 return ret;
 313         }
 314
 315         if (dir->ei_journal_seq > inode->ei_journal_seq)
 316                 inode->ei_journal_seq = dir->ei_journal_seq;
 317
 318         d_instantiate(dentry, &inode->v);
 319         return 0;
 320 }
 321
 322 /* methods */
 323
 324 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 325                                   unsigned int flags)
 326 {
 327         struct bch_fs *c = vdir->i_sb->s_fs_info;
 328         struct bch_inode_info *dir = to_bch_ei(vdir);
 329         struct inode *vinode = NULL;
 330         u64 inum;
 331
 332         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 333                                   &dir->ei_str_hash,
 334                                   &dentry->d_name);
 335
 336         if (inum)
 337                 vinode = bch2_vfs_inode_get(c, inum);
 338
 339         return d_splice_alias(vinode, dentry);
 340 }
 341
 342 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 343                        umode_t mode, bool excl)
 344 {
 345         return __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0);
 346 }
 347
 348 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 349                      struct dentry *dentry)
 350 {
 351         struct bch_fs *c = vdir->i_sb->s_fs_info;
 352         struct bch_inode_info *dir = to_bch_ei(vdir);
 353         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 354         int ret;
 355
 356         lockdep_assert_held(&inode->v.i_rwsem);
 357
 358         inode->v.i_ctime = current_time(&dir->v);
 359
 360         ret = bch2_inc_nlink(c, inode);
 361         if (ret)
 362                 return ret;
 363
 364         ihold(&inode->v);
 365
 366         ret = bch2_vfs_dirent_create(c, dir, mode_to_type(inode->v.i_mode),
 367                                      &dentry->d_name, inode->v.i_ino);
 368         if (unlikely(ret)) {
 369                 bch2_dec_nlink(c, inode);
 370                 iput(&inode->v);
 371                 return ret;
 372         }
 373
 374         d_instantiate(dentry, &inode->v);
 375         return 0;
 376 }
 377
 378 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 379 {
 380         struct bch_fs *c = vdir->i_sb->s_fs_info;
 381         struct bch_inode_info *dir = to_bch_ei(vdir);
 382         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 383         int ret;
 384
 385         lockdep_assert_held(&inode->v.i_rwsem);
 386
 387         ret = bch2_dirent_delete(c, dir->v.i_ino, &dir->ei_str_hash,
 388                                  &dentry->d_name, &dir->ei_journal_seq);
 389         if (ret)
 390                 return ret;
 391
 392         if (dir->ei_journal_seq > inode->ei_journal_seq)
 393                 inode->ei_journal_seq = dir->ei_journal_seq;
 394
 395         inode->v.i_ctime = dir->v.i_ctime;
 396
 397         if (S_ISDIR(inode->v.i_mode)) {
 398                 bch2_dec_nlink(c, dir);
 399                 drop_nlink(&inode->v);
 400         }
 401
 402         bch2_dec_nlink(c, inode);
 403
 404         return 0;
 405 }
 406
 407 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 408                         const char *symname)
 409 {
 410         struct bch_fs *c = vdir->i_sb->s_fs_info;
 411         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 412         int ret;
 413
 414         inode = bch2_vfs_inode_create(c, dir, S_IFLNK|S_IRWXUGO, 0);
 415         if (unlikely(IS_ERR(inode)))
 416                 return PTR_ERR(inode);
 417
 418         inode_lock(&inode->v);
 419         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 420         inode_unlock(&inode->v);
 421
 422         if (unlikely(ret))
 423                 goto err;
 424
 425         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 426         if (unlikely(ret))
 427                 goto err;
 428
 429         /* XXX: racy */
 430         if (dir->ei_journal_seq < inode->ei_journal_seq)
 431                 dir->ei_journal_seq = inode->ei_journal_seq;
 432
 433         ret = bch2_vfs_dirent_create(c, dir, DT_LNK, &dentry->d_name,
 434                                      inode->v.i_ino);
 435         if (unlikely(ret))
 436                 goto err;
 437
 438         d_instantiate(dentry, &inode->v);
 439         return 0;
 440 err:
 441         clear_nlink(&inode->v);
 442         iput(&inode->v);
 443         return ret;
 444 }
 445
 446 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 447 {
 448         struct bch_fs *c = vdir->i_sb->s_fs_info;
 449         struct bch_inode_info *dir = to_bch_ei(vdir);
 450         int ret;
 451
 452         lockdep_assert_held(&dir->v.i_rwsem);
 453
 454         ret = __bch2_create(dir, dentry, mode|S_IFDIR, 0);
 455         if (unlikely(ret))
 456                 return ret;
 457
 458         bch2_inc_nlink(c, dir);
 459
 460         return 0;
 461 }
 462
 463 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 464 {
 465         struct bch_fs *c = vdir->i_sb->s_fs_info;
 466
 467         if (bch2_empty_dir(c, dentry->d_inode->i_ino))
 468                 return -ENOTEMPTY;
 469
 470         return bch2_unlink(vdir, dentry);
 471 }
 472
 473 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 474                       umode_t mode, dev_t rdev)
 475 {
 476         return __bch2_create(to_bch_ei(vdir), dentry, mode, rdev);
 477 }
 478
 479 static int bch2_rename(struct bch_fs *c,
 480                        struct bch_inode_info *old_dir,
 481                        struct dentry *old_dentry,
 482                        struct bch_inode_info *new_dir,
 483                        struct dentry *new_dentry)
 484 {
 485         struct bch_inode_info *old_inode = to_bch_ei(old_dentry->d_inode);
 486         struct bch_inode_info *new_inode = to_bch_ei(new_dentry->d_inode);
 487         struct timespec now = current_time(&old_dir->v);
 488         int ret;
 489
 490         lockdep_assert_held(&old_dir->v.i_rwsem);
 491         lockdep_assert_held(&new_dir->v.i_rwsem);
 492
 493         if (new_inode)
 494                 filemap_write_and_wait_range(old_inode->v.i_mapping,
 495                                              0, LLONG_MAX);
 496
 497         if (new_inode && S_ISDIR(old_inode->v.i_mode)) {
 498                 lockdep_assert_held(&new_inode->v.i_rwsem);
 499
 500                 if (!S_ISDIR(new_inode->v.i_mode))
 501                         return -ENOTDIR;
 502
 503                 if (bch2_empty_dir(c, new_inode->v.i_ino))
 504                         return -ENOTEMPTY;
 505
 506                 ret = bch2_dirent_rename(c,
 507                                 old_dir, &old_dentry->d_name,
 508                                 new_dir, &new_dentry->d_name,
 509                                 &old_inode->ei_journal_seq, BCH_RENAME_OVERWRITE);
 510                 if (unlikely(ret))
 511                         return ret;
 512
 513                 clear_nlink(&new_inode->v);
 514                 bch2_dec_nlink(c, old_dir);
 515         } else if (new_inode) {
 516                 lockdep_assert_held(&new_inode->v.i_rwsem);
 517
 518                 ret = bch2_dirent_rename(c,
 519                                 old_dir, &old_dentry->d_name,
 520                                 new_dir, &new_dentry->d_name,
 521                                 &old_inode->ei_journal_seq, BCH_RENAME_OVERWRITE);
 522                 if (unlikely(ret))
 523                         return ret;
 524
 525                 new_inode->v.i_ctime = now;
 526                 bch2_dec_nlink(c, new_inode);
 527         } else if (S_ISDIR(old_inode->v.i_mode)) {
 528                 ret = bch2_dirent_rename(c,
 529                                 old_dir, &old_dentry->d_name,
 530                                 new_dir, &new_dentry->d_name,
 531                                 &old_inode->ei_journal_seq, BCH_RENAME);
 532                 if (unlikely(ret))
 533                         return ret;
 534
 535                 bch2_inc_nlink(c, new_dir);
 536                 bch2_dec_nlink(c, old_dir);
 537         } else {
 538                 ret = bch2_dirent_rename(c,
 539                                 old_dir, &old_dentry->d_name,
 540                                 new_dir, &new_dentry->d_name,
 541                                 &old_inode->ei_journal_seq, BCH_RENAME);
 542                 if (unlikely(ret))
 543                         return ret;
 544         }
 545
 546         old_dir->v.i_ctime = old_dir->v.i_mtime = now;
 547         new_dir->v.i_ctime = new_dir->v.i_mtime = now;
 548         mark_inode_dirty_sync(&old_dir->v);
 549         mark_inode_dirty_sync(&new_dir->v);
 550
 551         old_inode->v.i_ctime = now;
 552         mark_inode_dirty_sync(&old_inode->v);
 553
 554         return 0;
 555 }
 556
 557 static int bch2_rename_exchange(struct bch_fs *c,
 558                                 struct bch_inode_info *old_dir,
 559                                 struct dentry *old_dentry,
 560                                 struct bch_inode_info *new_dir,
 561                                 struct dentry *new_dentry)
 562 {
 563         struct bch_inode_info *old_inode = to_bch_ei(old_dentry->d_inode);
 564         struct bch_inode_info *new_inode = to_bch_ei(new_dentry->d_inode);
 565         struct timespec now = current_time(&old_dir->v);
 566         int ret;
 567
 568         ret = bch2_dirent_rename(c,
 569                                  old_dir, &old_dentry->d_name,
 570                                  new_dir, &new_dentry->d_name,
 571                                  &old_inode->ei_journal_seq, BCH_RENAME_EXCHANGE);
 572         if (unlikely(ret))
 573                 return ret;
 574
 575         if (S_ISDIR(old_inode->v.i_mode) !=
 576             S_ISDIR(new_inode->v.i_mode)) {
 577                 if (S_ISDIR(old_inode->v.i_mode)) {
 578                         bch2_inc_nlink(c, new_dir);
 579                         bch2_dec_nlink(c, old_dir);
 580                 } else {
 581                         bch2_dec_nlink(c, new_dir);
 582                         bch2_inc_nlink(c, old_dir);
 583                 }
 584         }
 585
 586         old_dir->v.i_ctime = old_dir->v.i_mtime = now;
 587         new_dir->v.i_ctime = new_dir->v.i_mtime = now;
 588         mark_inode_dirty_sync(&old_dir->v);
 589         mark_inode_dirty_sync(&new_dir->v);
 590
 591         old_inode->v.i_ctime = now;
 592         new_inode->v.i_ctime = now;
 593         mark_inode_dirty_sync(&old_inode->v);
 594         mark_inode_dirty_sync(&new_inode->v);
 595
 596         return 0;
 597 }
 598
 599 static int bch2_rename2(struct inode *old_vdir, struct dentry *old_dentry,
 600                         struct inode *new_vdir, struct dentry *new_dentry,
 601                         unsigned flags)
 602 {
 603         struct bch_fs *c = old_vdir->i_sb->s_fs_info;
 604         struct bch_inode_info *old_dir = to_bch_ei(old_vdir);
 605         struct bch_inode_info *new_dir = to_bch_ei(new_vdir);
 606
 607         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 608                 return -EINVAL;
 609
 610         if (flags & RENAME_EXCHANGE)
 611                 return bch2_rename_exchange(c, old_dir, old_dentry,
 612                                             new_dir, new_dentry);
 613
 614         return bch2_rename(c, old_dir, old_dentry, new_dir, new_dentry);
 615 }
 616
 617 static int bch2_setattr_nonsize(struct bch_inode_info *inode, struct iattr *iattr)
 618 {
 619         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 620         struct bch_qid qid = inode->ei_qid;
 621         unsigned qtypes = 0;
 622         int ret;
 623
 624         mutex_lock(&inode->ei_update_lock);
 625
 626         if (c->opts.usrquota &&
 627             (iattr->ia_valid & ATTR_UID) &&
 628             !uid_eq(iattr->ia_uid, inode->v.i_uid)) {
 629                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, iattr->ia_uid),
 630                 qtypes |= 1 << QTYP_USR;
 631         }
 632
 633         if (c->opts.grpquota &&
 634             (iattr->ia_valid & ATTR_GID) &&
 635             !gid_eq(iattr->ia_gid, inode->v.i_gid)) {
 636                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, iattr->ia_gid);
 637                 qtypes |= 1 << QTYP_GRP;
 638         }
 639
 640         if (qtypes) {
 641                 ret = bch2_quota_transfer(c, qtypes, qid, inode->ei_qid,
 642                                           inode->v.i_blocks +
 643                                           inode->ei_quota_reserved);
 644                 if (ret)
 645                         goto out_unlock;
 646         }
 647
 648         setattr_copy(&inode->v, iattr);
 649
 650         ret = bch2_write_inode(c, inode);
 651 out_unlock:
 652         mutex_unlock(&inode->ei_update_lock);
 653
 654         if (!ret &&
 655             iattr->ia_valid & ATTR_MODE)
 656                 ret = posix_acl_chmod(&inode->v, inode->v.i_mode);
 657
 658         return ret;
 659 }
 660
 661 static int bch2_getattr(const struct path *path, struct kstat *stat,
 662                         u32 request_mask, unsigned query_flags)
 663 {
 664         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 665         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 666
 667         stat->dev       = inode->v.i_sb->s_dev;
 668         stat->ino       = inode->v.i_ino;
 669         stat->mode      = inode->v.i_mode;
 670         stat->nlink     = inode->v.i_nlink;
 671         stat->uid       = inode->v.i_uid;
 672         stat->gid       = inode->v.i_gid;
 673         stat->rdev      = inode->v.i_rdev;
 674         stat->size      = i_size_read(&inode->v);
 675         stat->atime     = inode->v.i_atime;
 676         stat->mtime     = inode->v.i_mtime;
 677         stat->ctime     = inode->v.i_ctime;
 678         stat->blksize   = block_bytes(c);
 679         stat->blocks    = inode->v.i_blocks;
 680
 681         if (request_mask & STATX_BTIME) {
 682                 stat->result_mask |= STATX_BTIME;
 683                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 684         }
 685
 686         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 687                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 688         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 689                 stat->attributes |= STATX_ATTR_APPEND;
 690         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 691                 stat->attributes |= STATX_ATTR_NODUMP;
 692
 693         return 0;
 694 }
 695
 696 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 697 {
 698         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 699         int ret;
 700
 701         lockdep_assert_held(&inode->v.i_rwsem);
 702
 703         ret = setattr_prepare(dentry, iattr);
 704         if (ret)
 705                 return ret;
 706
 707         return iattr->ia_valid & ATTR_SIZE
 708                 ? bch2_truncate(inode, iattr)
 709                 : bch2_setattr_nonsize(inode, iattr);
 710 }
 711
 712 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
 713 {
 714         struct bch_fs *c = vdir->i_sb->s_fs_info;
 715         struct bch_inode_info *dir = to_bch_ei(vdir);
 716         struct bch_inode_info *inode;
 717
 718         /* XXX: i_nlink should be 0? */
 719         inode = bch2_vfs_inode_create(c, dir, mode, 0);
 720         if (unlikely(IS_ERR(inode)))
 721                 return PTR_ERR(inode);
 722
 723         d_tmpfile(dentry, &inode->v);
 724         return 0;
 725 }
 726
 727 static int bch2_fill_extent(struct fiemap_extent_info *info,
 728                             const struct bkey_i *k, unsigned flags)
 729 {
 730         if (bkey_extent_is_data(&k->k)) {
 731                 struct bkey_s_c_extent e = bkey_i_to_s_c_extent(k);
 732                 const struct bch_extent_ptr *ptr;
 733                 struct bch_extent_crc_unpacked crc;
 734                 int ret;
 735
 736                 extent_for_each_ptr_crc(e, ptr, crc) {
 737                         int flags2 = 0;
 738                         u64 offset = ptr->offset;
 739
 740                         if (crc.compression_type)
 741                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 742                         else
 743                                 offset += crc.offset;
 744
 745                         if ((offset & (PAGE_SECTORS - 1)) ||
 746                             (e.k->size & (PAGE_SECTORS - 1)))
 747                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 748
 749                         ret = fiemap_fill_next_extent(info,
 750                                                       bkey_start_offset(e.k) << 9,
 751                                                       offset << 9,
 752                                                       e.k->size << 9, flags|flags2);
 753                         if (ret)
 754                                 return ret;
 755                 }
 756
 757                 return 0;
 758         } else if (k->k.type == BCH_RESERVATION) {
 759                 return fiemap_fill_next_extent(info,
 760                                                bkey_start_offset(&k->k) << 9,
 761                                                0, k->k.size << 9,
 762                                                flags|
 763                                                FIEMAP_EXTENT_DELALLOC|
 764                                                FIEMAP_EXTENT_UNWRITTEN);
 765         } else {
 766                 BUG();
 767         }
 768 }
 769
 770 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 771                        u64 start, u64 len)
 772 {
 773         struct bch_fs *c = vinode->i_sb->s_fs_info;
 774         struct bch_inode_info *ei = to_bch_ei(vinode);
 775         struct btree_iter iter;
 776         struct bkey_s_c k;
 777         BKEY_PADDED(k) tmp;
 778         bool have_extent = false;
 779         int ret = 0;
 780
 781         if (start + len < start)
 782                 return -EINVAL;
 783
 784         for_each_btree_key(&iter, c, BTREE_ID_EXTENTS,
 785                            POS(ei->v.i_ino, start >> 9), 0, k)
 786                 if (bkey_extent_is_data(k.k) ||
 787                     k.k->type == BCH_RESERVATION) {
 788                         if (bkey_cmp(bkey_start_pos(k.k),
 789                                      POS(ei->v.i_ino, (start + len) >> 9)) >= 0)
 790                                 break;
 791
 792                         if (have_extent) {
 793                                 ret = bch2_fill_extent(info, &tmp.k, 0);
 794                                 if (ret)
 795                                         goto out;
 796                         }
 797
 798                         bkey_reassemble(&tmp.k, k);
 799                         have_extent = true;
 800                 }
 801
 802         if (have_extent)
 803                 ret = bch2_fill_extent(info, &tmp.k, FIEMAP_EXTENT_LAST);
 804 out:
 805         bch2_btree_iter_unlock(&iter);
 806         return ret < 0 ? ret : 0;
 807 }
 808
 809 static const struct vm_operations_struct bch_vm_ops = {
 810         .fault          = filemap_fault,
 811         .map_pages      = filemap_map_pages,
 812         .page_mkwrite   = bch2_page_mkwrite,
 813 };
 814
 815 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 816 {
 817         file_accessed(file);
 818
 819         vma->vm_ops = &bch_vm_ops;
 820         return 0;
 821 }
 822
 823 /* Directories: */
 824
 825 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 826 {
 827         return generic_file_llseek_size(file, offset, whence,
 828                                         S64_MAX, S64_MAX);
 829 }
 830
 831 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 832 {
 833         struct bch_fs *c = file_inode(file)->i_sb->s_fs_info;
 834
 835         return bch2_readdir(c, file, ctx);
 836 }
 837
 838 static const struct file_operations bch_file_operations = {
 839         .llseek         = bch2_llseek,
 840         .read_iter      = generic_file_read_iter,
 841         .write_iter     = bch2_write_iter,
 842         .mmap           = bch2_mmap,
 843         .open           = generic_file_open,
 844         .fsync          = bch2_fsync,
 845         .splice_read    = generic_file_splice_read,
 846         .splice_write   = iter_file_splice_write,
 847         .fallocate      = bch2_fallocate_dispatch,
 848         .unlocked_ioctl = bch2_fs_file_ioctl,
 849 #ifdef CONFIG_COMPAT
 850         .compat_ioctl   = bch2_compat_fs_ioctl,
 851 #endif
 852 };
 853
 854 static const struct inode_operations bch_file_inode_operations = {
 855         .getattr        = bch2_getattr,
 856         .setattr        = bch2_setattr,
 857         .fiemap         = bch2_fiemap,
 858         .listxattr      = bch2_xattr_list,
 859 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 860         .get_acl        = bch2_get_acl,
 861         .set_acl        = bch2_set_acl,
 862 #endif
 863 };
 864
 865 static const struct inode_operations bch_dir_inode_operations = {
 866         .lookup         = bch2_lookup,
 867         .create         = bch2_create,
 868         .link           = bch2_link,
 869         .unlink         = bch2_unlink,
 870         .symlink        = bch2_symlink,
 871         .mkdir          = bch2_mkdir,
 872         .rmdir          = bch2_rmdir,
 873         .mknod          = bch2_mknod,
 874         .rename         = bch2_rename2,
 875         .getattr        = bch2_getattr,
 876         .setattr        = bch2_setattr,
 877         .tmpfile        = bch2_tmpfile,
 878         .listxattr      = bch2_xattr_list,
 879 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 880         .get_acl        = bch2_get_acl,
 881         .set_acl        = bch2_set_acl,
 882 #endif
 883 };
 884
 885 static const struct file_operations bch_dir_file_operations = {
 886         .llseek         = bch2_dir_llseek,
 887         .read           = generic_read_dir,
 888         .iterate        = bch2_vfs_readdir,
 889         .fsync          = bch2_fsync,
 890         .unlocked_ioctl = bch2_fs_file_ioctl,
 891 #ifdef CONFIG_COMPAT
 892         .compat_ioctl   = bch2_compat_fs_ioctl,
 893 #endif
 894 };
 895
 896 static const struct inode_operations bch_symlink_inode_operations = {
 897         .get_link       = page_get_link,
 898         .getattr        = bch2_getattr,
 899         .setattr        = bch2_setattr,
 900         .listxattr      = bch2_xattr_list,
 901 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 902         .get_acl        = bch2_get_acl,
 903         .set_acl        = bch2_set_acl,
 904 #endif
 905 };
 906
 907 static const struct inode_operations bch_special_inode_operations = {
 908         .getattr        = bch2_getattr,
 909         .setattr        = bch2_setattr,
 910         .listxattr      = bch2_xattr_list,
 911 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 912         .get_acl        = bch2_get_acl,
 913         .set_acl        = bch2_set_acl,
 914 #endif
 915 };
 916
 917 static const struct address_space_operations bch_address_space_operations = {
 918         .writepage      = bch2_writepage,
 919         .readpage       = bch2_readpage,
 920         .writepages     = bch2_writepages,
 921         .readpages      = bch2_readpages,
 922         .set_page_dirty = bch2_set_page_dirty,
 923         .write_begin    = bch2_write_begin,
 924         .write_end      = bch2_write_end,
 925         .invalidatepage = bch2_invalidatepage,
 926         .releasepage    = bch2_releasepage,
 927         .direct_IO      = bch2_direct_IO,
 928 #ifdef CONFIG_MIGRATION
 929         .migratepage    = bch2_migrate_page,
 930 #endif
 931         .error_remove_page = generic_error_remove_page,
 932 };
 933
 934 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
 935                 u64 ino, u32 generation)
 936 {
 937         struct bch_fs *c = sb->s_fs_info;
 938         struct inode *vinode;
 939
 940         if (ino < BCACHEFS_ROOT_INO)
 941                 return ERR_PTR(-ESTALE);
 942
 943         vinode = bch2_vfs_inode_get(c, ino);
 944         if (IS_ERR(vinode))
 945                 return ERR_CAST(vinode);
 946         if (generation && vinode->i_generation != generation) {
 947                 /* we didn't find the right inode.. */
 948                 iput(vinode);
 949                 return ERR_PTR(-ESTALE);
 950         }
 951         return vinode;
 952 }
 953
 954 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
 955                 int fh_len, int fh_type)
 956 {
 957         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 958                                     bch2_nfs_get_inode);
 959 }
 960
 961 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
 962                 int fh_len, int fh_type)
 963 {
 964         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 965                                     bch2_nfs_get_inode);
 966 }
 967
 968 static const struct export_operations bch_export_ops = {
 969         .fh_to_dentry   = bch2_fh_to_dentry,
 970         .fh_to_parent   = bch2_fh_to_parent,
 971         //.get_parent   = bch2_get_parent,
 972 };
 973
 974 static void bch2_vfs_inode_init(struct bch_fs *c,
 975                                 struct bch_inode_info *inode,
 976                                 struct bch_inode_unpacked *bi)
 977 {
 978         inode->v.i_mode         = bi->bi_mode;
 979         i_uid_write(&inode->v, bi->bi_uid);
 980         i_gid_write(&inode->v, bi->bi_gid);
 981         inode->v.i_blocks       = bi->bi_sectors;
 982         inode->v.i_ino          = bi->bi_inum;
 983         set_nlink(&inode->v, bi->bi_nlink + nlink_bias(inode->v.i_mode));
 984         inode->v.i_rdev         = bi->bi_dev;
 985         inode->v.i_generation   = bi->bi_generation;
 986         inode->v.i_size         = bi->bi_size;
 987         inode->v.i_atime        = bch2_time_to_timespec(c, bi->bi_atime);
 988         inode->v.i_mtime        = bch2_time_to_timespec(c, bi->bi_mtime);
 989         inode->v.i_ctime        = bch2_time_to_timespec(c, bi->bi_ctime);
 990
 991         inode->ei_journal_seq   = 0;
 992         inode->ei_quota_reserved = 0;
 993         inode->ei_qid           = bch_qid(bi);
 994         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
 995         inode->ei_inode         = *bi;
 996
 997         bch2_inode_flags_to_vfs(inode);
 998
 999         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1000
1001         switch (inode->v.i_mode & S_IFMT) {
1002         case S_IFREG:
1003                 inode->v.i_op   = &bch_file_inode_operations;
1004                 inode->v.i_fop  = &bch_file_operations;
1005                 break;
1006         case S_IFDIR:
1007                 inode->v.i_op   = &bch_dir_inode_operations;
1008                 inode->v.i_fop  = &bch_dir_file_operations;
1009                 break;
1010         case S_IFLNK:
1011                 inode_nohighmem(&inode->v);
1012                 inode->v.i_op   = &bch_symlink_inode_operations;
1013                 break;
1014         default:
1015                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1016                 inode->v.i_op   = &bch_special_inode_operations;
1017                 break;
1018         }
1019 }
1020
1021 static struct inode *bch2_alloc_inode(struct super_block *sb)
1022 {
1023         struct bch_inode_info *inode;
1024
1025         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1026         if (!inode)
1027                 return NULL;
1028
1029         inode_init_once(&inode->v);
1030         mutex_init(&inode->ei_update_lock);
1031         inode->ei_journal_seq = 0;
1032
1033         return &inode->v;
1034 }
1035
1036 static void bch2_i_callback(struct rcu_head *head)
1037 {
1038         struct inode *vinode = container_of(head, struct inode, i_rcu);
1039         struct bch_inode_info *inode = to_bch_ei(vinode);
1040
1041         kmem_cache_free(bch2_inode_cache, inode);
1042 }
1043
1044 static void bch2_destroy_inode(struct inode *vinode)
1045 {
1046         call_rcu(&vinode->i_rcu, bch2_i_callback);
1047 }
1048
1049 static int bch2_vfs_write_inode(struct inode *vinode,
1050                                 struct writeback_control *wbc)
1051 {
1052         struct bch_fs *c = vinode->i_sb->s_fs_info;
1053         struct bch_inode_info *inode = to_bch_ei(vinode);
1054         int ret;
1055
1056         mutex_lock(&inode->ei_update_lock);
1057         ret = bch2_write_inode(c, inode);
1058         mutex_unlock(&inode->ei_update_lock);
1059
1060         if (c->opts.journal_flush_disabled)
1061                 return ret;
1062
1063         if (!ret && wbc->sync_mode == WB_SYNC_ALL)
1064                 ret = bch2_journal_flush_seq(&c->journal, inode->ei_journal_seq);
1065
1066         return ret;
1067 }
1068
1069 static void bch2_evict_inode(struct inode *vinode)
1070 {
1071         struct bch_fs *c = vinode->i_sb->s_fs_info;
1072         struct bch_inode_info *inode = to_bch_ei(vinode);
1073
1074         truncate_inode_pages_final(&inode->v.i_data);
1075
1076         clear_inode(&inode->v);
1077
1078         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1079
1080         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1081                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1082                                 BCH_QUOTA_WARN);
1083                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1084                                 BCH_QUOTA_WARN);
1085                 bch2_inode_rm(c, inode->v.i_ino);
1086                 atomic_long_dec(&c->nr_inodes);
1087         }
1088 }
1089
1090 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1091 {
1092         struct super_block *sb = dentry->d_sb;
1093         struct bch_fs *c = sb->s_fs_info;
1094         u64 fsid;
1095
1096         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1097         buf->f_bsize    = sb->s_blocksize;
1098         buf->f_blocks   = c->capacity >> PAGE_SECTOR_SHIFT;
1099         buf->f_bfree    = bch2_fs_sectors_free(c, bch2_fs_usage_read(c)) >>
1100                            PAGE_SECTOR_SHIFT;
1101         buf->f_bavail   = buf->f_bfree;
1102         buf->f_files    = atomic_long_read(&c->nr_inodes);
1103         buf->f_ffree    = U64_MAX;
1104
1105         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1106                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1107         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1108         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1109         buf->f_namelen  = NAME_MAX;
1110
1111         return 0;
1112 }
1113
1114 static int bch2_sync_fs(struct super_block *sb, int wait)
1115 {
1116         struct bch_fs *c = sb->s_fs_info;
1117
1118         if (!wait) {
1119                 bch2_journal_flush_async(&c->journal, NULL);
1120                 return 0;
1121         }
1122
1123         return bch2_journal_flush(&c->journal);
1124 }
1125
1126 static struct bch_fs *bch2_path_to_fs(const char *dev)
1127 {
1128         struct bch_fs *c;
1129         struct block_device *bdev = lookup_bdev(dev);
1130
1131         if (IS_ERR(bdev))
1132                 return ERR_CAST(bdev);
1133
1134         c = bch2_bdev_to_fs(bdev);
1135         bdput(bdev);
1136         return c ?: ERR_PTR(-ENOENT);
1137 }
1138
1139 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1140                                                unsigned nr_devs, struct bch_opts opts)
1141 {
1142         struct bch_fs *c, *c1, *c2;
1143         size_t i;
1144
1145         if (!nr_devs)
1146                 return ERR_PTR(-EINVAL);
1147
1148         c = bch2_fs_open(devs, nr_devs, opts);
1149
1150         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1151                 /*
1152                  * Already open?
1153                  * Look up each block device, make sure they all belong to a
1154                  * filesystem and they all belong to the _same_ filesystem
1155                  */
1156
1157                 c1 = bch2_path_to_fs(devs[0]);
1158                 if (!c1)
1159                         return c;
1160
1161                 for (i = 1; i < nr_devs; i++) {
1162                         c2 = bch2_path_to_fs(devs[i]);
1163                         if (!IS_ERR(c2))
1164                                 closure_put(&c2->cl);
1165
1166                         if (c1 != c2) {
1167                                 closure_put(&c1->cl);
1168                                 return c;
1169                         }
1170                 }
1171
1172                 c = c1;
1173         }
1174
1175         if (IS_ERR(c))
1176                 return c;
1177
1178         mutex_lock(&c->state_lock);
1179
1180         if (!bch2_fs_running(c)) {
1181                 mutex_unlock(&c->state_lock);
1182                 closure_put(&c->cl);
1183                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1184                 return ERR_PTR(-EINVAL);
1185         }
1186
1187         mutex_unlock(&c->state_lock);
1188
1189         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1190         return c;
1191 }
1192
1193 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1194                                              struct bch_opts opts)
1195 {
1196         char *dev_name = NULL, **devs = NULL, *s;
1197         struct bch_fs *c = ERR_PTR(-ENOMEM);
1198         size_t i, nr_devs = 0;
1199
1200         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1201         if (!dev_name)
1202                 goto err;
1203
1204         for (s = dev_name; s; s = strchr(s + 1, ':'))
1205                 nr_devs++;
1206
1207         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1208         if (!devs)
1209                 goto err;
1210
1211         for (i = 0, s = dev_name;
1212              s;
1213              (s = strchr(s, ':')) && (*s++ = '\0'))
1214                 devs[i++] = s;
1215
1216         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1217 err:
1218         kfree(devs);
1219         kfree(dev_name);
1220         return c;
1221 }
1222
1223 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1224 {
1225         struct bch_fs *c = sb->s_fs_info;
1226         struct bch_opts opts = bch2_opts_empty();
1227         int ret;
1228
1229         opt_set(opts, read_only, (*flags & MS_RDONLY) != 0);
1230
1231         ret = bch2_parse_mount_opts(&opts, data);
1232         if (ret)
1233                 return ret;
1234
1235         if (opts.read_only != c->opts.read_only) {
1236                 const char *err = NULL;
1237
1238                 mutex_lock(&c->state_lock);
1239
1240                 if (opts.read_only) {
1241                         bch2_fs_read_only(c);
1242
1243                         sb->s_flags |= MS_RDONLY;
1244                 } else {
1245                         err = bch2_fs_read_write(c);
1246                         if (err) {
1247                                 bch_err(c, "error going rw: %s", err);
1248                                 return -EINVAL;
1249                         }
1250
1251                         sb->s_flags &= ~MS_RDONLY;
1252                 }
1253
1254                 c->opts.read_only = opts.read_only;
1255
1256                 mutex_unlock(&c->state_lock);
1257         }
1258
1259         if (opts.errors >= 0)
1260                 c->opts.errors = opts.errors;
1261
1262         return ret;
1263 }
1264
1265 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1266 {
1267         struct bch_fs *c = root->d_sb->s_fs_info;
1268         enum bch_opt_id i;
1269
1270         for (i = 0; i < bch2_opts_nr; i++) {
1271                 const struct bch_option *opt = &bch2_opt_table[i];
1272                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1273
1274                 if (opt->mode < OPT_MOUNT)
1275                         continue;
1276
1277                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1278                         continue;
1279
1280                 switch (opt->type) {
1281                 case BCH_OPT_BOOL:
1282                         seq_printf(seq, ",%s%s", v ? "" : "no", opt->attr.name);
1283                         break;
1284                 case BCH_OPT_UINT:
1285                         seq_printf(seq, ",%s=%llu", opt->attr.name, v);
1286                         break;
1287                 case BCH_OPT_STR:
1288                         seq_printf(seq, ",%s=%s", opt->attr.name, opt->choices[v]);
1289                         break;
1290                 }
1291         }
1292
1293         return 0;
1294
1295 }
1296
1297 static const struct super_operations bch_super_operations = {
1298         .alloc_inode    = bch2_alloc_inode,
1299         .destroy_inode  = bch2_destroy_inode,
1300         .write_inode    = bch2_vfs_write_inode,
1301         .evict_inode    = bch2_evict_inode,
1302         .sync_fs        = bch2_sync_fs,
1303         .statfs         = bch2_statfs,
1304         .show_options   = bch2_show_options,
1305         .remount_fs     = bch2_remount,
1306 #if 0
1307         .put_super      = bch2_put_super,
1308         .freeze_fs      = bch2_freeze,
1309         .unfreeze_fs    = bch2_unfreeze,
1310 #endif
1311 };
1312
1313 static int bch2_test_super(struct super_block *s, void *data)
1314 {
1315         return s->s_fs_info == data;
1316 }
1317
1318 static int bch2_set_super(struct super_block *s, void *data)
1319 {
1320         s->s_fs_info = data;
1321         return 0;
1322 }
1323
1324 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1325                                  int flags, const char *dev_name, void *data)
1326 {
1327         struct bch_fs *c;
1328         struct bch_dev *ca;
1329         struct super_block *sb;
1330         struct inode *vinode;
1331         struct bch_opts opts = bch2_opts_empty();
1332         unsigned i;
1333         int ret;
1334
1335         opt_set(opts, read_only, (flags & MS_RDONLY) != 0);
1336
1337         ret = bch2_parse_mount_opts(&opts, data);
1338         if (ret)
1339                 return ERR_PTR(ret);
1340
1341         c = bch2_open_as_blockdevs(dev_name, opts);
1342         if (IS_ERR(c))
1343                 return ERR_CAST(c);
1344
1345         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|MS_NOSEC, c);
1346         if (IS_ERR(sb)) {
1347                 closure_put(&c->cl);
1348                 return ERR_CAST(sb);
1349         }
1350
1351         BUG_ON(sb->s_fs_info != c);
1352
1353         if (sb->s_root) {
1354                 closure_put(&c->cl);
1355
1356                 if ((flags ^ sb->s_flags) & MS_RDONLY) {
1357                         ret = -EBUSY;
1358                         goto err_put_super;
1359                 }
1360                 goto out;
1361         }
1362
1363         /* XXX: blocksize */
1364         sb->s_blocksize         = PAGE_SIZE;
1365         sb->s_blocksize_bits    = PAGE_SHIFT;
1366         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1367         sb->s_op                = &bch_super_operations;
1368         sb->s_export_op         = &bch_export_ops;
1369 #ifdef CONFIG_BCACHEFS_QUOTA
1370         sb->s_qcop              = &bch2_quotactl_operations;
1371         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1372 #endif
1373         sb->s_xattr             = bch2_xattr_handlers;
1374         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1375         sb->s_time_gran         = c->sb.time_precision;
1376         c->vfs_sb               = sb;
1377         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1378
1379         ret = super_setup_bdi(sb);
1380         if (ret)
1381                 goto err_put_super;
1382
1383         sb->s_bdi->congested_fn         = bch2_congested;
1384         sb->s_bdi->congested_data       = c;
1385         sb->s_bdi->ra_pages             = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1386
1387         for_each_online_member(ca, c, i) {
1388                 struct block_device *bdev = ca->disk_sb.bdev;
1389
1390                 /* XXX: create an anonymous device for multi device filesystems */
1391                 sb->s_bdev      = bdev;
1392                 sb->s_dev       = bdev->bd_dev;
1393                 percpu_ref_put(&ca->io_ref);
1394                 break;
1395         }
1396
1397 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1398         if (c->opts.acl)
1399                 sb->s_flags     |= MS_POSIXACL;
1400 #endif
1401
1402         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1403         if (IS_ERR(vinode)) {
1404                 ret = PTR_ERR(vinode);
1405                 goto err_put_super;
1406         }
1407
1408         sb->s_root = d_make_root(vinode);
1409         if (!sb->s_root) {
1410                 ret = -ENOMEM;
1411                 goto err_put_super;
1412         }
1413
1414         sb->s_flags |= MS_ACTIVE;
1415 out:
1416         return dget(sb->s_root);
1417
1418 err_put_super:
1419         deactivate_locked_super(sb);
1420         return ERR_PTR(ret);
1421 }
1422
1423 static void bch2_kill_sb(struct super_block *sb)
1424 {
1425         struct bch_fs *c = sb->s_fs_info;
1426
1427         generic_shutdown_super(sb);
1428
1429         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1430                 bch2_fs_stop(c);
1431         else
1432                 closure_put(&c->cl);
1433 }
1434
1435 static struct file_system_type bcache_fs_type = {
1436         .owner          = THIS_MODULE,
1437         .name           = "bcachefs",
1438         .mount          = bch2_mount,
1439         .kill_sb        = bch2_kill_sb,
1440         .fs_flags       = FS_REQUIRES_DEV,
1441 };
1442
1443 MODULE_ALIAS_FS("bcachefs");
1444
1445 void bch2_vfs_exit(void)
1446 {
1447         unregister_filesystem(&bcache_fs_type);
1448         if (bch2_inode_cache)
1449                 kmem_cache_destroy(bch2_inode_cache);
1450 }
1451
1452 int __init bch2_vfs_init(void)
1453 {
1454         int ret = -ENOMEM;
1455
1456         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1457         if (!bch2_inode_cache)
1458                 goto err;
1459
1460         ret = register_filesystem(&bcache_fs_type);
1461         if (ret)
1462                 goto err;
1463
1464         return 0;
1465 err:
1466         bch2_vfs_exit();
1467         return ret;
1468 }
1469
1470 #endif /* NO_BCACHEFS_FS */