fs/bcachefs/fs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "bkey_buf.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "chardev.h"
  10 #include "dirent.h"
  11 #include "errcode.h"
  12 #include "extents.h"
  13 #include "fs.h"
  14 #include "fs-common.h"
  15 #include "fs-io.h"
  16 #include "fs-ioctl.h"
  17 #include "fs-io-buffered.h"
  18 #include "fs-io-direct.h"
  19 #include "fs-io-pagecache.h"
  20 #include "fsck.h"
  21 #include "inode.h"
  22 #include "io_read.h"
  23 #include "journal.h"
  24 #include "keylist.h"
  25 #include "quota.h"
  26 #include "snapshot.h"
  27 #include "super.h"
  28 #include "xattr.h"
  29
  30 #include <linux/aio.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/fiemap.h>
  34 #include <linux/module.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/posix_acl.h>
  37 #include <linux/random.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/statfs.h>
  40 #include <linux/string.h>
  41 #include <linux/xattr.h>
  42
  43 static struct kmem_cache *bch2_inode_cache;
  44
  45 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
  46                                 struct bch_inode_info *,
  47                                 struct bch_inode_unpacked *,
  48                                 struct bch_subvolume *);
  49
  50 void bch2_inode_update_after_write(struct btree_trans *trans,
  51                                    struct bch_inode_info *inode,
  52                                    struct bch_inode_unpacked *bi,
  53                                    unsigned fields)
  54 {
  55         struct bch_fs *c = trans->c;
  56
  57         BUG_ON(bi->bi_inum != inode->v.i_ino);
  58
  59         bch2_assert_pos_locked(trans, BTREE_ID_inodes,
  60                                POS(0, bi->bi_inum),
  61                                c->opts.inodes_use_key_cache);
  62
  63         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
  64         i_uid_write(&inode->v, bi->bi_uid);
  65         i_gid_write(&inode->v, bi->bi_gid);
  66         inode->v.i_mode = bi->bi_mode;
  67
  68         if (fields & ATTR_ATIME)
  69                 inode_set_atime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_atime));
  70         if (fields & ATTR_MTIME)
  71                 inode_set_mtime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_mtime));
  72         if (fields & ATTR_CTIME)
  73                 inode_set_ctime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_ctime));
  74
  75         inode->ei_inode         = *bi;
  76
  77         bch2_inode_flags_to_vfs(inode);
  78 }
  79
  80 int __must_check bch2_write_inode(struct bch_fs *c,
  81                                   struct bch_inode_info *inode,
  82                                   inode_set_fn set,
  83                                   void *p, unsigned fields)
  84 {
  85         struct btree_trans *trans = bch2_trans_get(c);
  86         struct btree_iter iter = { NULL };
  87         struct bch_inode_unpacked inode_u;
  88         int ret;
  89 retry:
  90         bch2_trans_begin(trans);
  91
  92         ret   = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
  93                                 BTREE_ITER_INTENT) ?:
  94                 (set ? set(trans, inode, &inode_u, p) : 0) ?:
  95                 bch2_inode_write(trans, &iter, &inode_u) ?:
  96                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
  97
  98         /*
  99          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 100          * this is important for inode updates via bchfs_write_index_update
 101          */
 102         if (!ret)
 103                 bch2_inode_update_after_write(trans, inode, &inode_u, fields);
 104
 105         bch2_trans_iter_exit(trans, &iter);
 106
 107         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 108                 goto retry;
 109
 110         bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
 111                              "%s: inode %u:%llu not found when updating",
 112                              bch2_err_str(ret),
 113                              inode_inum(inode).subvol,
 114                              inode_inum(inode).inum);
 115
 116         bch2_trans_put(trans);
 117         return ret < 0 ? ret : 0;
 118 }
 119
 120 int bch2_fs_quota_transfer(struct bch_fs *c,
 121                            struct bch_inode_info *inode,
 122                            struct bch_qid new_qid,
 123                            unsigned qtypes,
 124                            enum quota_acct_mode mode)
 125 {
 126         unsigned i;
 127         int ret;
 128
 129         qtypes &= enabled_qtypes(c);
 130
 131         for (i = 0; i < QTYP_NR; i++)
 132                 if (new_qid.q[i] == inode->ei_qid.q[i])
 133                         qtypes &= ~(1U << i);
 134
 135         if (!qtypes)
 136                 return 0;
 137
 138         mutex_lock(&inode->ei_quota_lock);
 139
 140         ret = bch2_quota_transfer(c, qtypes, new_qid,
 141                                   inode->ei_qid,
 142                                   inode->v.i_blocks +
 143                                   inode->ei_quota_reserved,
 144                                   mode);
 145         if (!ret)
 146                 for (i = 0; i < QTYP_NR; i++)
 147                         if (qtypes & (1 << i))
 148                                 inode->ei_qid.q[i] = new_qid.q[i];
 149
 150         mutex_unlock(&inode->ei_quota_lock);
 151
 152         return ret;
 153 }
 154
 155 static int bch2_iget5_test(struct inode *vinode, void *p)
 156 {
 157         struct bch_inode_info *inode = to_bch_ei(vinode);
 158         subvol_inum *inum = p;
 159
 160         return inode->ei_subvol == inum->subvol &&
 161                 inode->ei_inode.bi_inum == inum->inum;
 162 }
 163
 164 static int bch2_iget5_set(struct inode *vinode, void *p)
 165 {
 166         struct bch_inode_info *inode = to_bch_ei(vinode);
 167         subvol_inum *inum = p;
 168
 169         inode->v.i_ino          = inum->inum;
 170         inode->ei_subvol        = inum->subvol;
 171         inode->ei_inode.bi_inum = inum->inum;
 172         return 0;
 173 }
 174
 175 static unsigned bch2_inode_hash(subvol_inum inum)
 176 {
 177         return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
 178 }
 179
 180 static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_inode_info *inode)
 181 {
 182         subvol_inum inum = inode_inum(inode);
 183         struct bch_inode_info *old = to_bch_ei(inode_insert5(&inode->v,
 184                                       bch2_inode_hash(inum),
 185                                       bch2_iget5_test,
 186                                       bch2_iget5_set,
 187                                       &inum));
 188         BUG_ON(!old);
 189
 190         if (unlikely(old != inode)) {
 191                 __destroy_inode(&inode->v);
 192                 kmem_cache_free(bch2_inode_cache, inode);
 193                 inode = old;
 194         } else {
 195                 mutex_lock(&c->vfs_inodes_lock);
 196                 list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
 197                 mutex_unlock(&c->vfs_inodes_lock);
 198                 /*
 199                  * we really don't want insert_inode_locked2() to be setting
 200                  * I_NEW...
 201                  */
 202                 unlock_new_inode(&inode->v);
 203         }
 204
 205         return inode;
 206 }
 207
 208 #define memalloc_flags_do(_flags, _do)                                          \
 209 ({                                                                              \
 210         unsigned _saved_flags = memalloc_flags_save(_flags);                    \
 211         typeof(_do) _ret = _do;                                                 \
 212         memalloc_noreclaim_restore(_saved_flags);                               \
 213         _ret;                                                                   \
 214 })
 215
 216 /*
 217  * Allocate a new inode, dropping/retaking btree locks if necessary:
 218  */
 219 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
 220 {
 221         struct bch_fs *c = trans->c;
 222
 223         struct bch_inode_info *inode =
 224                 memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
 225                                   to_bch_ei(new_inode(c->vfs_sb)));
 226
 227         if (unlikely(!inode)) {
 228                 int ret = drop_locks_do(trans, (inode = to_bch_ei(new_inode(c->vfs_sb))) ? 0 : -ENOMEM);
 229                 if (ret && inode) {
 230                         __destroy_inode(&inode->v);
 231                         kmem_cache_free(bch2_inode_cache, inode);
 232                 }
 233                 if (ret)
 234                         return ERR_PTR(ret);
 235         }
 236
 237         return inode;
 238 }
 239
 240 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 241 {
 242         struct bch_inode_info *inode =
 243                 to_bch_ei(ilookup5_nowait(c->vfs_sb,
 244                                           bch2_inode_hash(inum),
 245                                           bch2_iget5_test,
 246                                           &inum));
 247         if (inode)
 248                 return &inode->v;
 249
 250         struct btree_trans *trans = bch2_trans_get(c);
 251
 252         struct bch_inode_unpacked inode_u;
 253         struct bch_subvolume subvol;
 254         int ret = lockrestart_do(trans,
 255                 bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
 256                 bch2_inode_find_by_inum_trans(trans, inum, &inode_u)) ?:
 257                 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
 258         if (!ret) {
 259                 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 260                 inode = bch2_inode_insert(c, inode);
 261         }
 262         bch2_trans_put(trans);
 263
 264         return ret ? ERR_PTR(ret) : &inode->v;
 265 }
 266
 267 struct bch_inode_info *
 268 __bch2_create(struct mnt_idmap *idmap,
 269               struct bch_inode_info *dir, struct dentry *dentry,
 270               umode_t mode, dev_t rdev, subvol_inum snapshot_src,
 271               unsigned flags)
 272 {
 273         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 274         struct btree_trans *trans;
 275         struct bch_inode_unpacked dir_u;
 276         struct bch_inode_info *inode;
 277         struct bch_inode_unpacked inode_u;
 278         struct posix_acl *default_acl = NULL, *acl = NULL;
 279         subvol_inum inum;
 280         struct bch_subvolume subvol;
 281         u64 journal_seq = 0;
 282         int ret;
 283
 284         /*
 285          * preallocate acls + vfs inode before btree transaction, so that
 286          * nothing can fail after the transaction succeeds:
 287          */
 288 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 289         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 290         if (ret)
 291                 return ERR_PTR(ret);
 292 #endif
 293         inode = to_bch_ei(new_inode(c->vfs_sb));
 294         if (unlikely(!inode)) {
 295                 inode = ERR_PTR(-ENOMEM);
 296                 goto err;
 297         }
 298
 299         bch2_inode_init_early(c, &inode_u);
 300
 301         if (!(flags & BCH_CREATE_TMPFILE))
 302                 mutex_lock(&dir->ei_update_lock);
 303
 304         trans = bch2_trans_get(c);
 305 retry:
 306         bch2_trans_begin(trans);
 307
 308         ret   = bch2_subvol_is_ro_trans(trans, dir->ei_subvol) ?:
 309                 bch2_create_trans(trans,
 310                                   inode_inum(dir), &dir_u, &inode_u,
 311                                   !(flags & BCH_CREATE_TMPFILE)
 312                                   ? &dentry->d_name : NULL,
 313                                   from_kuid(i_user_ns(&dir->v), current_fsuid()),
 314                                   from_kgid(i_user_ns(&dir->v), current_fsgid()),
 315                                   mode, rdev,
 316                                   default_acl, acl, snapshot_src, flags) ?:
 317                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 318                                 KEY_TYPE_QUOTA_PREALLOC);
 319         if (unlikely(ret))
 320                 goto err_before_quota;
 321
 322         inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
 323         inum.inum = inode_u.bi_inum;
 324
 325         ret   = bch2_subvolume_get(trans, inum.subvol, true,
 326                                    BTREE_ITER_WITH_UPDATES, &subvol) ?:
 327                 bch2_trans_commit(trans, NULL, &journal_seq, 0);
 328         if (unlikely(ret)) {
 329                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 330                                 KEY_TYPE_QUOTA_WARN);
 331 err_before_quota:
 332                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 333                         goto retry;
 334                 goto err_trans;
 335         }
 336
 337         if (!(flags & BCH_CREATE_TMPFILE)) {
 338                 bch2_inode_update_after_write(trans, dir, &dir_u,
 339                                               ATTR_MTIME|ATTR_CTIME);
 340                 mutex_unlock(&dir->ei_update_lock);
 341         }
 342
 343         bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 344
 345         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 346         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 347
 348         /*
 349          * we must insert the new inode into the inode cache before calling
 350          * bch2_trans_exit() and dropping locks, else we could race with another
 351          * thread pulling the inode in and modifying it:
 352          */
 353         inode = bch2_inode_insert(c, inode);
 354         bch2_trans_put(trans);
 355 err:
 356         posix_acl_release(default_acl);
 357         posix_acl_release(acl);
 358         return inode;
 359 err_trans:
 360         if (!(flags & BCH_CREATE_TMPFILE))
 361                 mutex_unlock(&dir->ei_update_lock);
 362
 363         bch2_trans_put(trans);
 364         make_bad_inode(&inode->v);
 365         iput(&inode->v);
 366         inode = ERR_PTR(ret);
 367         goto err;
 368 }
 369
 370 /* methods */
 371
 372 static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
 373                         subvol_inum dir, struct bch_hash_info *dir_hash_info,
 374                         const struct qstr *name)
 375 {
 376         struct bch_fs *c = trans->c;
 377         struct btree_iter dirent_iter = {};
 378         subvol_inum inum = {};
 379
 380         int ret = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
 381                                    dir_hash_info, dir, name, 0);
 382         if (ret)
 383                 return ERR_PTR(ret);
 384
 385         struct bkey_s_c k = bch2_btree_iter_peek_slot(&dirent_iter);
 386         ret = bkey_err(k);
 387         if (ret)
 388                 goto err;
 389
 390         ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
 391         if (ret > 0)
 392                 ret = -ENOENT;
 393         if (ret)
 394                 goto err;
 395
 396         struct bch_inode_info *inode =
 397                 to_bch_ei(ilookup5_nowait(c->vfs_sb,
 398                                           bch2_inode_hash(inum),
 399                                           bch2_iget5_test,
 400                                           &inum));
 401         if (inode)
 402                 goto out;
 403
 404         struct bch_subvolume subvol;
 405         struct bch_inode_unpacked inode_u;
 406         ret =   bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
 407                 bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
 408                 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
 409         if (bch2_err_matches(ret, ENOENT)) {
 410                 struct printbuf buf = PRINTBUF;
 411
 412                 bch2_bkey_val_to_text(&buf, c, k);
 413                 bch_err(c, "%s points to missing inode", buf.buf);
 414                 printbuf_exit(&buf);
 415         }
 416         if (ret)
 417                 goto err;
 418
 419         bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
 420         inode = bch2_inode_insert(c, inode);
 421 out:
 422         bch2_trans_iter_exit(trans, &dirent_iter);
 423         return inode;
 424 err:
 425         inode = ERR_PTR(ret);
 426         goto out;
 427 }
 428
 429 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 430                                   unsigned int flags)
 431 {
 432         struct bch_fs *c = vdir->i_sb->s_fs_info;
 433         struct bch_inode_info *dir = to_bch_ei(vdir);
 434         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 435
 436         struct bch_inode_info *inode;
 437         bch2_trans_do(c, NULL, NULL, 0,
 438                 PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
 439                                                           &hash, &dentry->d_name)));
 440         if (IS_ERR(inode))
 441                 inode = NULL;
 442
 443         return d_splice_alias(&inode->v, dentry);
 444 }
 445
 446 static int bch2_mknod(struct mnt_idmap *idmap,
 447                       struct inode *vdir, struct dentry *dentry,
 448                       umode_t mode, dev_t rdev)
 449 {
 450         struct bch_inode_info *inode =
 451                 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
 452                               (subvol_inum) { 0 }, 0);
 453
 454         if (IS_ERR(inode))
 455                 return bch2_err_class(PTR_ERR(inode));
 456
 457         d_instantiate(dentry, &inode->v);
 458         return 0;
 459 }
 460
 461 static int bch2_create(struct mnt_idmap *idmap,
 462                        struct inode *vdir, struct dentry *dentry,
 463                        umode_t mode, bool excl)
 464 {
 465         return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
 466 }
 467
 468 static int __bch2_link(struct bch_fs *c,
 469                        struct bch_inode_info *inode,
 470                        struct bch_inode_info *dir,
 471                        struct dentry *dentry)
 472 {
 473         struct btree_trans *trans = bch2_trans_get(c);
 474         struct bch_inode_unpacked dir_u, inode_u;
 475         int ret;
 476
 477         mutex_lock(&inode->ei_update_lock);
 478
 479         ret = commit_do(trans, NULL, NULL, 0,
 480                         bch2_link_trans(trans,
 481                                         inode_inum(dir),   &dir_u,
 482                                         inode_inum(inode), &inode_u,
 483                                         &dentry->d_name));
 484
 485         if (likely(!ret)) {
 486                 bch2_inode_update_after_write(trans, dir, &dir_u,
 487                                               ATTR_MTIME|ATTR_CTIME);
 488                 bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
 489         }
 490
 491         bch2_trans_put(trans);
 492         mutex_unlock(&inode->ei_update_lock);
 493         return ret;
 494 }
 495
 496 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 497                      struct dentry *dentry)
 498 {
 499         struct bch_fs *c = vdir->i_sb->s_fs_info;
 500         struct bch_inode_info *dir = to_bch_ei(vdir);
 501         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 502         int ret;
 503
 504         lockdep_assert_held(&inode->v.i_rwsem);
 505
 506         ret   = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
 507                 bch2_subvol_is_ro(c, inode->ei_subvol) ?:
 508                 __bch2_link(c, inode, dir, dentry);
 509         if (unlikely(ret))
 510                 return bch2_err_class(ret);
 511
 512         ihold(&inode->v);
 513         d_instantiate(dentry, &inode->v);
 514         return 0;
 515 }
 516
 517 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 518                   bool deleting_snapshot)
 519 {
 520         struct bch_fs *c = vdir->i_sb->s_fs_info;
 521         struct bch_inode_info *dir = to_bch_ei(vdir);
 522         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 523         struct bch_inode_unpacked dir_u, inode_u;
 524         struct btree_trans *trans = bch2_trans_get(c);
 525         int ret;
 526
 527         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 528
 529         ret = commit_do(trans, NULL, NULL,
 530                         BCH_TRANS_COMMIT_no_enospc,
 531                 bch2_unlink_trans(trans,
 532                                   inode_inum(dir), &dir_u,
 533                                   &inode_u, &dentry->d_name,
 534                                   deleting_snapshot));
 535         if (unlikely(ret))
 536                 goto err;
 537
 538         bch2_inode_update_after_write(trans, dir, &dir_u,
 539                                       ATTR_MTIME|ATTR_CTIME);
 540         bch2_inode_update_after_write(trans, inode, &inode_u,
 541                                       ATTR_MTIME);
 542
 543         if (inode_u.bi_subvol) {
 544                 /*
 545                  * Subvolume deletion is asynchronous, but we still want to tell
 546                  * the VFS that it's been deleted here:
 547                  */
 548                 set_nlink(&inode->v, 0);
 549         }
 550 err:
 551         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 552         bch2_trans_put(trans);
 553
 554         return ret;
 555 }
 556
 557 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 558 {
 559         struct bch_inode_info *dir= to_bch_ei(vdir);
 560         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 561
 562         int ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
 563                 __bch2_unlink(vdir, dentry, false);
 564         return bch2_err_class(ret);
 565 }
 566
 567 static int bch2_symlink(struct mnt_idmap *idmap,
 568                         struct inode *vdir, struct dentry *dentry,
 569                         const char *symname)
 570 {
 571         struct bch_fs *c = vdir->i_sb->s_fs_info;
 572         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 573         int ret;
 574
 575         inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
 576                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 577         if (IS_ERR(inode))
 578                 return bch2_err_class(PTR_ERR(inode));
 579
 580         inode_lock(&inode->v);
 581         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 582         inode_unlock(&inode->v);
 583
 584         if (unlikely(ret))
 585                 goto err;
 586
 587         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 588         if (unlikely(ret))
 589                 goto err;
 590
 591         ret = __bch2_link(c, inode, dir, dentry);
 592         if (unlikely(ret))
 593                 goto err;
 594
 595         d_instantiate(dentry, &inode->v);
 596         return 0;
 597 err:
 598         iput(&inode->v);
 599         return bch2_err_class(ret);
 600 }
 601
 602 static int bch2_mkdir(struct mnt_idmap *idmap,
 603                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 604 {
 605         return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
 606 }
 607
 608 static int bch2_rename2(struct mnt_idmap *idmap,
 609                         struct inode *src_vdir, struct dentry *src_dentry,
 610                         struct inode *dst_vdir, struct dentry *dst_dentry,
 611                         unsigned flags)
 612 {
 613         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 614         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 615         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 616         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 617         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 618         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 619         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 620         struct btree_trans *trans;
 621         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 622                 ? BCH_RENAME_EXCHANGE
 623                 : dst_dentry->d_inode
 624                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 625         int ret;
 626
 627         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 628                 return -EINVAL;
 629
 630         if (mode == BCH_RENAME_OVERWRITE) {
 631                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 632                                                    0, LLONG_MAX);
 633                 if (ret)
 634                         return ret;
 635         }
 636
 637         trans = bch2_trans_get(c);
 638
 639         bch2_lock_inodes(INODE_UPDATE_LOCK,
 640                          src_dir,
 641                          dst_dir,
 642                          src_inode,
 643                          dst_inode);
 644
 645         ret   = bch2_subvol_is_ro_trans(trans, src_dir->ei_subvol) ?:
 646                 bch2_subvol_is_ro_trans(trans, dst_dir->ei_subvol);
 647         if (ret)
 648                 goto err;
 649
 650         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 651                 ret = bch2_fs_quota_transfer(c, src_inode,
 652                                              dst_dir->ei_qid,
 653                                              1 << QTYP_PRJ,
 654                                              KEY_TYPE_QUOTA_PREALLOC);
 655                 if (ret)
 656                         goto err;
 657         }
 658
 659         if (mode == BCH_RENAME_EXCHANGE &&
 660             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 661                 ret = bch2_fs_quota_transfer(c, dst_inode,
 662                                              src_dir->ei_qid,
 663                                              1 << QTYP_PRJ,
 664                                              KEY_TYPE_QUOTA_PREALLOC);
 665                 if (ret)
 666                         goto err;
 667         }
 668
 669         ret = commit_do(trans, NULL, NULL, 0,
 670                         bch2_rename_trans(trans,
 671                                           inode_inum(src_dir), &src_dir_u,
 672                                           inode_inum(dst_dir), &dst_dir_u,
 673                                           &src_inode_u,
 674                                           &dst_inode_u,
 675                                           &src_dentry->d_name,
 676                                           &dst_dentry->d_name,
 677                                           mode));
 678         if (unlikely(ret))
 679                 goto err;
 680
 681         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 682         BUG_ON(dst_inode &&
 683                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 684
 685         bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
 686                                       ATTR_MTIME|ATTR_CTIME);
 687
 688         if (src_dir != dst_dir)
 689                 bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
 690                                               ATTR_MTIME|ATTR_CTIME);
 691
 692         bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
 693                                       ATTR_CTIME);
 694
 695         if (dst_inode)
 696                 bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
 697                                               ATTR_CTIME);
 698 err:
 699         bch2_trans_put(trans);
 700
 701         bch2_fs_quota_transfer(c, src_inode,
 702                                bch_qid(&src_inode->ei_inode),
 703                                1 << QTYP_PRJ,
 704                                KEY_TYPE_QUOTA_NOCHECK);
 705         if (dst_inode)
 706                 bch2_fs_quota_transfer(c, dst_inode,
 707                                        bch_qid(&dst_inode->ei_inode),
 708                                        1 << QTYP_PRJ,
 709                                        KEY_TYPE_QUOTA_NOCHECK);
 710
 711         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 712                            src_dir,
 713                            dst_dir,
 714                            src_inode,
 715                            dst_inode);
 716
 717         return bch2_err_class(ret);
 718 }
 719
 720 static void bch2_setattr_copy(struct mnt_idmap *idmap,
 721                               struct bch_inode_info *inode,
 722                               struct bch_inode_unpacked *bi,
 723                               struct iattr *attr)
 724 {
 725         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 726         unsigned int ia_valid = attr->ia_valid;
 727
 728         if (ia_valid & ATTR_UID)
 729                 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 730         if (ia_valid & ATTR_GID)
 731                 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 732
 733         if (ia_valid & ATTR_SIZE)
 734                 bi->bi_size = attr->ia_size;
 735
 736         if (ia_valid & ATTR_ATIME)
 737                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 738         if (ia_valid & ATTR_MTIME)
 739                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 740         if (ia_valid & ATTR_CTIME)
 741                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 742
 743         if (ia_valid & ATTR_MODE) {
 744                 umode_t mode = attr->ia_mode;
 745                 kgid_t gid = ia_valid & ATTR_GID
 746                         ? attr->ia_gid
 747                         : inode->v.i_gid;
 748
 749                 if (!in_group_p(gid) &&
 750                     !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
 751                         mode &= ~S_ISGID;
 752                 bi->bi_mode = mode;
 753         }
 754 }
 755
 756 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
 757                          struct bch_inode_info *inode,
 758                          struct iattr *attr)
 759 {
 760         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 761         struct bch_qid qid;
 762         struct btree_trans *trans;
 763         struct btree_iter inode_iter = { NULL };
 764         struct bch_inode_unpacked inode_u;
 765         struct posix_acl *acl = NULL;
 766         int ret;
 767
 768         mutex_lock(&inode->ei_update_lock);
 769
 770         qid = inode->ei_qid;
 771
 772         if (attr->ia_valid & ATTR_UID)
 773                 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 774
 775         if (attr->ia_valid & ATTR_GID)
 776                 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 777
 778         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 779                                      KEY_TYPE_QUOTA_PREALLOC);
 780         if (ret)
 781                 goto err;
 782
 783         trans = bch2_trans_get(c);
 784 retry:
 785         bch2_trans_begin(trans);
 786         kfree(acl);
 787         acl = NULL;
 788
 789         ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
 790                               BTREE_ITER_INTENT);
 791         if (ret)
 792                 goto btree_err;
 793
 794         bch2_setattr_copy(idmap, inode, &inode_u, attr);
 795
 796         if (attr->ia_valid & ATTR_MODE) {
 797                 ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
 798                                      inode_u.bi_mode, &acl);
 799                 if (ret)
 800                         goto btree_err;
 801         }
 802
 803         ret =   bch2_inode_write(trans, &inode_iter, &inode_u) ?:
 804                 bch2_trans_commit(trans, NULL, NULL,
 805                                   BCH_TRANS_COMMIT_no_enospc);
 806 btree_err:
 807         bch2_trans_iter_exit(trans, &inode_iter);
 808
 809         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 810                 goto retry;
 811         if (unlikely(ret))
 812                 goto err_trans;
 813
 814         bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
 815
 816         if (acl)
 817                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 818 err_trans:
 819         bch2_trans_put(trans);
 820 err:
 821         mutex_unlock(&inode->ei_update_lock);
 822
 823         return bch2_err_class(ret);
 824 }
 825
 826 static int bch2_getattr(struct mnt_idmap *idmap,
 827                         const struct path *path, struct kstat *stat,
 828                         u32 request_mask, unsigned query_flags)
 829 {
 830         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 831         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 832
 833         stat->dev       = inode->v.i_sb->s_dev;
 834         stat->ino       = inode->v.i_ino;
 835         stat->mode      = inode->v.i_mode;
 836         stat->nlink     = inode->v.i_nlink;
 837         stat->uid       = inode->v.i_uid;
 838         stat->gid       = inode->v.i_gid;
 839         stat->rdev      = inode->v.i_rdev;
 840         stat->size      = i_size_read(&inode->v);
 841         stat->atime     = inode_get_atime(&inode->v);
 842         stat->mtime     = inode_get_mtime(&inode->v);
 843         stat->ctime     = inode_get_ctime(&inode->v);
 844         stat->blksize   = block_bytes(c);
 845         stat->blocks    = inode->v.i_blocks;
 846
 847         if (request_mask & STATX_BTIME) {
 848                 stat->result_mask |= STATX_BTIME;
 849                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 850         }
 851
 852         if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
 853                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 854         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 855
 856         if (inode->ei_inode.bi_flags & BCH_INODE_append)
 857                 stat->attributes |= STATX_ATTR_APPEND;
 858         stat->attributes_mask    |= STATX_ATTR_APPEND;
 859
 860         if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
 861                 stat->attributes |= STATX_ATTR_NODUMP;
 862         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 863
 864         return 0;
 865 }
 866
 867 static int bch2_setattr(struct mnt_idmap *idmap,
 868                         struct dentry *dentry, struct iattr *iattr)
 869 {
 870         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 871         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 872         int ret;
 873
 874         lockdep_assert_held(&inode->v.i_rwsem);
 875
 876         ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
 877                 setattr_prepare(idmap, dentry, iattr);
 878         if (ret)
 879                 return ret;
 880
 881         return iattr->ia_valid & ATTR_SIZE
 882                 ? bchfs_truncate(idmap, inode, iattr)
 883                 : bch2_setattr_nonsize(idmap, inode, iattr);
 884 }
 885
 886 static int bch2_tmpfile(struct mnt_idmap *idmap,
 887                         struct inode *vdir, struct file *file, umode_t mode)
 888 {
 889         struct bch_inode_info *inode =
 890                 __bch2_create(idmap, to_bch_ei(vdir),
 891                               file->f_path.dentry, mode, 0,
 892                               (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
 893
 894         if (IS_ERR(inode))
 895                 return bch2_err_class(PTR_ERR(inode));
 896
 897         d_mark_tmpfile(file, &inode->v);
 898         d_instantiate(file->f_path.dentry, &inode->v);
 899         return finish_open_simple(file, 0);
 900 }
 901
 902 static int bch2_fill_extent(struct bch_fs *c,
 903                             struct fiemap_extent_info *info,
 904                             struct bkey_s_c k, unsigned flags)
 905 {
 906         if (bkey_extent_is_direct_data(k.k)) {
 907                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 908                 const union bch_extent_entry *entry;
 909                 struct extent_ptr_decoded p;
 910                 int ret;
 911
 912                 if (k.k->type == KEY_TYPE_reflink_v)
 913                         flags |= FIEMAP_EXTENT_SHARED;
 914
 915                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 916                         int flags2 = 0;
 917                         u64 offset = p.ptr.offset;
 918
 919                         if (p.ptr.unwritten)
 920                                 flags2 |= FIEMAP_EXTENT_UNWRITTEN;
 921
 922                         if (p.crc.compression_type)
 923                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 924                         else
 925                                 offset += p.crc.offset;
 926
 927                         if ((offset & (block_sectors(c) - 1)) ||
 928                             (k.k->size & (block_sectors(c) - 1)))
 929                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 930
 931                         ret = fiemap_fill_next_extent(info,
 932                                                 bkey_start_offset(k.k) << 9,
 933                                                 offset << 9,
 934                                                 k.k->size << 9, flags|flags2);
 935                         if (ret)
 936                                 return ret;
 937                 }
 938
 939                 return 0;
 940         } else if (bkey_extent_is_inline_data(k.k)) {
 941                 return fiemap_fill_next_extent(info,
 942                                                bkey_start_offset(k.k) << 9,
 943                                                0, k.k->size << 9,
 944                                                flags|
 945                                                FIEMAP_EXTENT_DATA_INLINE);
 946         } else if (k.k->type == KEY_TYPE_reservation) {
 947                 return fiemap_fill_next_extent(info,
 948                                                bkey_start_offset(k.k) << 9,
 949                                                0, k.k->size << 9,
 950                                                flags|
 951                                                FIEMAP_EXTENT_DELALLOC|
 952                                                FIEMAP_EXTENT_UNWRITTEN);
 953         } else {
 954                 BUG();
 955         }
 956 }
 957
 958 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 959                        u64 start, u64 len)
 960 {
 961         struct bch_fs *c = vinode->i_sb->s_fs_info;
 962         struct bch_inode_info *ei = to_bch_ei(vinode);
 963         struct btree_trans *trans;
 964         struct btree_iter iter;
 965         struct bkey_s_c k;
 966         struct bkey_buf cur, prev;
 967         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 968         unsigned offset_into_extent, sectors;
 969         bool have_extent = false;
 970         u32 snapshot;
 971         int ret = 0;
 972
 973         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 974         if (ret)
 975                 return ret;
 976
 977         if (start + len < start)
 978                 return -EINVAL;
 979
 980         start >>= 9;
 981
 982         bch2_bkey_buf_init(&cur);
 983         bch2_bkey_buf_init(&prev);
 984         trans = bch2_trans_get(c);
 985 retry:
 986         bch2_trans_begin(trans);
 987
 988         ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
 989         if (ret)
 990                 goto err;
 991
 992         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 993                              SPOS(ei->v.i_ino, start, snapshot), 0);
 994
 995         while (!(ret = btree_trans_too_many_iters(trans)) &&
 996                (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
 997                !(ret = bkey_err(k))) {
 998                 enum btree_id data_btree = BTREE_ID_extents;
 999
1000                 if (!bkey_extent_is_data(k.k) &&
1001                     k.k->type != KEY_TYPE_reservation) {
1002                         bch2_btree_iter_advance(&iter);
1003                         continue;
1004                 }
1005
1006                 offset_into_extent      = iter.pos.offset -
1007                         bkey_start_offset(k.k);
1008                 sectors                 = k.k->size - offset_into_extent;
1009
1010                 bch2_bkey_buf_reassemble(&cur, c, k);
1011
1012                 ret = bch2_read_indirect_extent(trans, &data_btree,
1013                                         &offset_into_extent, &cur);
1014                 if (ret)
1015                         break;
1016
1017                 k = bkey_i_to_s_c(cur.k);
1018                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
1019
1020                 sectors = min(sectors, k.k->size - offset_into_extent);
1021
1022                 bch2_cut_front(POS(k.k->p.inode,
1023                                    bkey_start_offset(k.k) +
1024                                    offset_into_extent),
1025                                cur.k);
1026                 bch2_key_resize(&cur.k->k, sectors);
1027                 cur.k->k.p = iter.pos;
1028                 cur.k->k.p.offset += cur.k->k.size;
1029
1030                 if (have_extent) {
1031                         bch2_trans_unlock(trans);
1032                         ret = bch2_fill_extent(c, info,
1033                                         bkey_i_to_s_c(prev.k), 0);
1034                         if (ret)
1035                                 break;
1036                 }
1037
1038                 bkey_copy(prev.k, cur.k);
1039                 have_extent = true;
1040
1041                 bch2_btree_iter_set_pos(&iter,
1042                         POS(iter.pos.inode, iter.pos.offset + sectors));
1043         }
1044         start = iter.pos.offset;
1045         bch2_trans_iter_exit(trans, &iter);
1046 err:
1047         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1048                 goto retry;
1049
1050         if (!ret && have_extent) {
1051                 bch2_trans_unlock(trans);
1052                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
1053                                        FIEMAP_EXTENT_LAST);
1054         }
1055
1056         bch2_trans_put(trans);
1057         bch2_bkey_buf_exit(&cur, c);
1058         bch2_bkey_buf_exit(&prev, c);
1059         return ret < 0 ? ret : 0;
1060 }
1061
1062 static const struct vm_operations_struct bch_vm_ops = {
1063         .fault          = bch2_page_fault,
1064         .map_pages      = filemap_map_pages,
1065         .page_mkwrite   = bch2_page_mkwrite,
1066 };
1067
1068 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1069 {
1070         file_accessed(file);
1071
1072         vma->vm_ops = &bch_vm_ops;
1073         return 0;
1074 }
1075
1076 /* Directories: */
1077
1078 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1079 {
1080         return generic_file_llseek_size(file, offset, whence,
1081                                         S64_MAX, S64_MAX);
1082 }
1083
1084 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1085 {
1086         struct bch_inode_info *inode = file_bch_inode(file);
1087         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1088
1089         if (!dir_emit_dots(file, ctx))
1090                 return 0;
1091
1092         int ret = bch2_readdir(c, inode_inum(inode), ctx);
1093
1094         bch_err_fn(c, ret);
1095         return bch2_err_class(ret);
1096 }
1097
1098 static int bch2_open(struct inode *vinode, struct file *file)
1099 {
1100         if (file->f_flags & (O_WRONLY|O_RDWR)) {
1101                 struct bch_inode_info *inode = to_bch_ei(vinode);
1102                 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1103
1104                 int ret = bch2_subvol_is_ro(c, inode->ei_subvol);
1105                 if (ret)
1106                         return ret;
1107         }
1108
1109         return generic_file_open(vinode, file);
1110 }
1111
1112 static const struct file_operations bch_file_operations = {
1113         .open           = bch2_open,
1114         .llseek         = bch2_llseek,
1115         .read_iter      = bch2_read_iter,
1116         .write_iter     = bch2_write_iter,
1117         .mmap           = bch2_mmap,
1118         .fsync          = bch2_fsync,
1119         .splice_read    = filemap_splice_read,
1120         .splice_write   = iter_file_splice_write,
1121         .fallocate      = bch2_fallocate_dispatch,
1122         .unlocked_ioctl = bch2_fs_file_ioctl,
1123 #ifdef CONFIG_COMPAT
1124         .compat_ioctl   = bch2_compat_fs_ioctl,
1125 #endif
1126         .remap_file_range = bch2_remap_file_range,
1127 };
1128
1129 static const struct inode_operations bch_file_inode_operations = {
1130         .getattr        = bch2_getattr,
1131         .setattr        = bch2_setattr,
1132         .fiemap         = bch2_fiemap,
1133         .listxattr      = bch2_xattr_list,
1134 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1135         .get_acl        = bch2_get_acl,
1136         .set_acl        = bch2_set_acl,
1137 #endif
1138 };
1139
1140 static const struct inode_operations bch_dir_inode_operations = {
1141         .lookup         = bch2_lookup,
1142         .create         = bch2_create,
1143         .link           = bch2_link,
1144         .unlink         = bch2_unlink,
1145         .symlink        = bch2_symlink,
1146         .mkdir          = bch2_mkdir,
1147         .rmdir          = bch2_unlink,
1148         .mknod          = bch2_mknod,
1149         .rename         = bch2_rename2,
1150         .getattr        = bch2_getattr,
1151         .setattr        = bch2_setattr,
1152         .tmpfile        = bch2_tmpfile,
1153         .listxattr      = bch2_xattr_list,
1154 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1155         .get_acl        = bch2_get_acl,
1156         .set_acl        = bch2_set_acl,
1157 #endif
1158 };
1159
1160 static const struct file_operations bch_dir_file_operations = {
1161         .llseek         = bch2_dir_llseek,
1162         .read           = generic_read_dir,
1163         .iterate_shared = bch2_vfs_readdir,
1164         .fsync          = bch2_fsync,
1165         .unlocked_ioctl = bch2_fs_file_ioctl,
1166 #ifdef CONFIG_COMPAT
1167         .compat_ioctl   = bch2_compat_fs_ioctl,
1168 #endif
1169 };
1170
1171 static const struct inode_operations bch_symlink_inode_operations = {
1172         .get_link       = page_get_link,
1173         .getattr        = bch2_getattr,
1174         .setattr        = bch2_setattr,
1175         .listxattr      = bch2_xattr_list,
1176 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1177         .get_acl        = bch2_get_acl,
1178         .set_acl        = bch2_set_acl,
1179 #endif
1180 };
1181
1182 static const struct inode_operations bch_special_inode_operations = {
1183         .getattr        = bch2_getattr,
1184         .setattr        = bch2_setattr,
1185         .listxattr      = bch2_xattr_list,
1186 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1187         .get_acl        = bch2_get_acl,
1188         .set_acl        = bch2_set_acl,
1189 #endif
1190 };
1191
1192 static const struct address_space_operations bch_address_space_operations = {
1193         .read_folio     = bch2_read_folio,
1194         .writepages     = bch2_writepages,
1195         .readahead      = bch2_readahead,
1196         .dirty_folio    = filemap_dirty_folio,
1197         .write_begin    = bch2_write_begin,
1198         .write_end      = bch2_write_end,
1199         .invalidate_folio = bch2_invalidate_folio,
1200         .release_folio  = bch2_release_folio,
1201         .direct_IO      = noop_direct_IO,
1202 #ifdef CONFIG_MIGRATION
1203         .migrate_folio  = filemap_migrate_folio,
1204 #endif
1205         .error_remove_folio = generic_error_remove_folio,
1206 };
1207
1208 struct bcachefs_fid {
1209         u64             inum;
1210         u32             subvol;
1211         u32             gen;
1212 } __packed;
1213
1214 struct bcachefs_fid_with_parent {
1215         struct bcachefs_fid     fid;
1216         struct bcachefs_fid     dir;
1217 } __packed;
1218
1219 static int bcachefs_fid_valid(int fh_len, int fh_type)
1220 {
1221         switch (fh_type) {
1222         case FILEID_BCACHEFS_WITHOUT_PARENT:
1223                 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1224         case FILEID_BCACHEFS_WITH_PARENT:
1225                 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1226         default:
1227                 return false;
1228         }
1229 }
1230
1231 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1232 {
1233         return (struct bcachefs_fid) {
1234                 .inum   = inode->ei_inode.bi_inum,
1235                 .subvol = inode->ei_subvol,
1236                 .gen    = inode->ei_inode.bi_generation,
1237         };
1238 }
1239
1240 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1241                           struct inode *vdir)
1242 {
1243         struct bch_inode_info *inode    = to_bch_ei(vinode);
1244         struct bch_inode_info *dir      = to_bch_ei(vdir);
1245         int min_len;
1246
1247         if (!S_ISDIR(inode->v.i_mode) && dir) {
1248                 struct bcachefs_fid_with_parent *fid = (void *) fh;
1249
1250                 min_len = sizeof(*fid) / sizeof(u32);
1251                 if (*len < min_len) {
1252                         *len = min_len;
1253                         return FILEID_INVALID;
1254                 }
1255
1256                 fid->fid = bch2_inode_to_fid(inode);
1257                 fid->dir = bch2_inode_to_fid(dir);
1258
1259                 *len = min_len;
1260                 return FILEID_BCACHEFS_WITH_PARENT;
1261         } else {
1262                 struct bcachefs_fid *fid = (void *) fh;
1263
1264                 min_len = sizeof(*fid) / sizeof(u32);
1265                 if (*len < min_len) {
1266                         *len = min_len;
1267                         return FILEID_INVALID;
1268                 }
1269                 *fid = bch2_inode_to_fid(inode);
1270
1271                 *len = min_len;
1272                 return FILEID_BCACHEFS_WITHOUT_PARENT;
1273         }
1274 }
1275
1276 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1277                                         struct bcachefs_fid fid)
1278 {
1279         struct bch_fs *c = sb->s_fs_info;
1280         struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1281                                     .subvol = fid.subvol,
1282                                     .inum = fid.inum,
1283         });
1284         if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1285                 iput(vinode);
1286                 vinode = ERR_PTR(-ESTALE);
1287         }
1288         return vinode;
1289 }
1290
1291 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1292                 int fh_len, int fh_type)
1293 {
1294         struct bcachefs_fid *fid = (void *) _fid;
1295
1296         if (!bcachefs_fid_valid(fh_len, fh_type))
1297                 return NULL;
1298
1299         return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1300 }
1301
1302 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1303                 int fh_len, int fh_type)
1304 {
1305         struct bcachefs_fid_with_parent *fid = (void *) _fid;
1306
1307         if (!bcachefs_fid_valid(fh_len, fh_type) ||
1308             fh_type != FILEID_BCACHEFS_WITH_PARENT)
1309                 return NULL;
1310
1311         return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1312 }
1313
1314 static struct dentry *bch2_get_parent(struct dentry *child)
1315 {
1316         struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1317         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1318         subvol_inum parent_inum = {
1319                 .subvol = inode->ei_inode.bi_parent_subvol ?:
1320                         inode->ei_subvol,
1321                 .inum = inode->ei_inode.bi_dir,
1322         };
1323
1324         return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1325 }
1326
1327 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1328 {
1329         struct bch_inode_info *inode    = to_bch_ei(child->d_inode);
1330         struct bch_inode_info *dir      = to_bch_ei(parent->d_inode);
1331         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1332         struct btree_trans *trans;
1333         struct btree_iter iter1;
1334         struct btree_iter iter2;
1335         struct bkey_s_c k;
1336         struct bkey_s_c_dirent d;
1337         struct bch_inode_unpacked inode_u;
1338         subvol_inum target;
1339         u32 snapshot;
1340         struct qstr dirent_name;
1341         unsigned name_len = 0;
1342         int ret;
1343
1344         if (!S_ISDIR(dir->v.i_mode))
1345                 return -EINVAL;
1346
1347         trans = bch2_trans_get(c);
1348
1349         bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1350                              POS(dir->ei_inode.bi_inum, 0), 0);
1351         bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1352                              POS(dir->ei_inode.bi_inum, 0), 0);
1353 retry:
1354         bch2_trans_begin(trans);
1355
1356         ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1357         if (ret)
1358                 goto err;
1359
1360         bch2_btree_iter_set_snapshot(&iter1, snapshot);
1361         bch2_btree_iter_set_snapshot(&iter2, snapshot);
1362
1363         ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1364         if (ret)
1365                 goto err;
1366
1367         if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1368                 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1369
1370                 k = bch2_btree_iter_peek_slot(&iter1);
1371                 ret = bkey_err(k);
1372                 if (ret)
1373                         goto err;
1374
1375                 if (k.k->type != KEY_TYPE_dirent) {
1376                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1377                         goto err;
1378                 }
1379
1380                 d = bkey_s_c_to_dirent(k);
1381                 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1382                 if (ret > 0)
1383                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1384                 if (ret)
1385                         goto err;
1386
1387                 if (target.subvol       == inode->ei_subvol &&
1388                     target.inum         == inode->ei_inode.bi_inum)
1389                         goto found;
1390         } else {
1391                 /*
1392                  * File with multiple hardlinks and our backref is to the wrong
1393                  * directory - linear search:
1394                  */
1395                 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1396                         if (k.k->p.inode > dir->ei_inode.bi_inum)
1397                                 break;
1398
1399                         if (k.k->type != KEY_TYPE_dirent)
1400                                 continue;
1401
1402                         d = bkey_s_c_to_dirent(k);
1403                         ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1404                         if (ret < 0)
1405                                 break;
1406                         if (ret)
1407                                 continue;
1408
1409                         if (target.subvol       == inode->ei_subvol &&
1410                             target.inum         == inode->ei_inode.bi_inum)
1411                                 goto found;
1412                 }
1413         }
1414
1415         ret = -ENOENT;
1416         goto err;
1417 found:
1418         dirent_name = bch2_dirent_get_name(d);
1419
1420         name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1421         memcpy(name, dirent_name.name, name_len);
1422         name[name_len] = '\0';
1423 err:
1424         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1425                 goto retry;
1426
1427         bch2_trans_iter_exit(trans, &iter1);
1428         bch2_trans_iter_exit(trans, &iter2);
1429         bch2_trans_put(trans);
1430
1431         return ret;
1432 }
1433
1434 static const struct export_operations bch_export_ops = {
1435         .encode_fh      = bch2_encode_fh,
1436         .fh_to_dentry   = bch2_fh_to_dentry,
1437         .fh_to_parent   = bch2_fh_to_parent,
1438         .get_parent     = bch2_get_parent,
1439         .get_name       = bch2_get_name,
1440 };
1441
1442 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1443                                 struct bch_inode_info *inode,
1444                                 struct bch_inode_unpacked *bi,
1445                                 struct bch_subvolume *subvol)
1446 {
1447         bch2_iget5_set(&inode->v, &inum);
1448         bch2_inode_update_after_write(trans, inode, bi, ~0);
1449
1450         if (BCH_SUBVOLUME_SNAP(subvol))
1451                 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1452         else
1453                 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1454
1455         inode->v.i_blocks       = bi->bi_sectors;
1456         inode->v.i_ino          = bi->bi_inum;
1457         inode->v.i_rdev         = bi->bi_dev;
1458         inode->v.i_generation   = bi->bi_generation;
1459         inode->v.i_size         = bi->bi_size;
1460
1461         inode->ei_flags         = 0;
1462         inode->ei_quota_reserved = 0;
1463         inode->ei_qid           = bch_qid(bi);
1464         inode->ei_subvol        = inum.subvol;
1465
1466         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1467
1468         switch (inode->v.i_mode & S_IFMT) {
1469         case S_IFREG:
1470                 inode->v.i_op   = &bch_file_inode_operations;
1471                 inode->v.i_fop  = &bch_file_operations;
1472                 break;
1473         case S_IFDIR:
1474                 inode->v.i_op   = &bch_dir_inode_operations;
1475                 inode->v.i_fop  = &bch_dir_file_operations;
1476                 break;
1477         case S_IFLNK:
1478                 inode_nohighmem(&inode->v);
1479                 inode->v.i_op   = &bch_symlink_inode_operations;
1480                 break;
1481         default:
1482                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1483                 inode->v.i_op   = &bch_special_inode_operations;
1484                 break;
1485         }
1486
1487         mapping_set_large_folios(inode->v.i_mapping);
1488 }
1489
1490 static struct inode *bch2_alloc_inode(struct super_block *sb)
1491 {
1492         struct bch_inode_info *inode;
1493
1494         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1495         if (!inode)
1496                 return NULL;
1497
1498         inode_init_once(&inode->v);
1499         mutex_init(&inode->ei_update_lock);
1500         two_state_lock_init(&inode->ei_pagecache_lock);
1501         INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
1502         mutex_init(&inode->ei_quota_lock);
1503
1504         return &inode->v;
1505 }
1506
1507 static void bch2_i_callback(struct rcu_head *head)
1508 {
1509         struct inode *vinode = container_of(head, struct inode, i_rcu);
1510         struct bch_inode_info *inode = to_bch_ei(vinode);
1511
1512         kmem_cache_free(bch2_inode_cache, inode);
1513 }
1514
1515 static void bch2_destroy_inode(struct inode *vinode)
1516 {
1517         call_rcu(&vinode->i_rcu, bch2_i_callback);
1518 }
1519
1520 static int inode_update_times_fn(struct btree_trans *trans,
1521                                  struct bch_inode_info *inode,
1522                                  struct bch_inode_unpacked *bi,
1523                                  void *p)
1524 {
1525         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1526
1527         bi->bi_atime    = timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1528         bi->bi_mtime    = timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1529         bi->bi_ctime    = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1530
1531         return 0;
1532 }
1533
1534 static int bch2_vfs_write_inode(struct inode *vinode,
1535                                 struct writeback_control *wbc)
1536 {
1537         struct bch_fs *c = vinode->i_sb->s_fs_info;
1538         struct bch_inode_info *inode = to_bch_ei(vinode);
1539         int ret;
1540
1541         mutex_lock(&inode->ei_update_lock);
1542         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1543                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1544         mutex_unlock(&inode->ei_update_lock);
1545
1546         return bch2_err_class(ret);
1547 }
1548
1549 static void bch2_evict_inode(struct inode *vinode)
1550 {
1551         struct bch_fs *c = vinode->i_sb->s_fs_info;
1552         struct bch_inode_info *inode = to_bch_ei(vinode);
1553
1554         truncate_inode_pages_final(&inode->v.i_data);
1555
1556         clear_inode(&inode->v);
1557
1558         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1559
1560         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1561                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1562                                 KEY_TYPE_QUOTA_WARN);
1563                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1564                                 KEY_TYPE_QUOTA_WARN);
1565                 bch2_inode_rm(c, inode_inum(inode));
1566         }
1567
1568         mutex_lock(&c->vfs_inodes_lock);
1569         list_del_init(&inode->ei_vfs_inode_list);
1570         mutex_unlock(&c->vfs_inodes_lock);
1571 }
1572
1573 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1574 {
1575         struct bch_inode_info *inode;
1576         DARRAY(struct bch_inode_info *) grabbed;
1577         bool clean_pass = false, this_pass_clean;
1578
1579         /*
1580          * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1581          * be pruned with d_mark_dontcache().
1582          *
1583          * Once we've had a clean pass where we didn't find any inodes without
1584          * I_DONTCACHE, we wait for them to be freed:
1585          */
1586
1587         darray_init(&grabbed);
1588         darray_make_room(&grabbed, 1024);
1589 again:
1590         cond_resched();
1591         this_pass_clean = true;
1592
1593         mutex_lock(&c->vfs_inodes_lock);
1594         list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1595                 if (!snapshot_list_has_id(s, inode->ei_subvol))
1596                         continue;
1597
1598                 if (!(inode->v.i_state & I_DONTCACHE) &&
1599                     !(inode->v.i_state & I_FREEING) &&
1600                     igrab(&inode->v)) {
1601                         this_pass_clean = false;
1602
1603                         if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1604                                 iput(&inode->v);
1605                                 break;
1606                         }
1607                 } else if (clean_pass && this_pass_clean) {
1608                         wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1609                         DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1610
1611                         prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1612                         mutex_unlock(&c->vfs_inodes_lock);
1613
1614                         schedule();
1615                         finish_wait(wq, &wait.wq_entry);
1616                         goto again;
1617                 }
1618         }
1619         mutex_unlock(&c->vfs_inodes_lock);
1620
1621         darray_for_each(grabbed, i) {
1622                 inode = *i;
1623                 d_mark_dontcache(&inode->v);
1624                 d_prune_aliases(&inode->v);
1625                 iput(&inode->v);
1626         }
1627         grabbed.nr = 0;
1628
1629         if (!clean_pass || !this_pass_clean) {
1630                 clean_pass = this_pass_clean;
1631                 goto again;
1632         }
1633
1634         darray_exit(&grabbed);
1635 }
1636
1637 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1638 {
1639         struct super_block *sb = dentry->d_sb;
1640         struct bch_fs *c = sb->s_fs_info;
1641         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1642         unsigned shift = sb->s_blocksize_bits - 9;
1643         /*
1644          * this assumes inodes take up 64 bytes, which is a decent average
1645          * number:
1646          */
1647         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1648
1649         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1650         buf->f_bsize    = sb->s_blocksize;
1651         buf->f_blocks   = usage.capacity >> shift;
1652         buf->f_bfree    = usage.free >> shift;
1653         buf->f_bavail   = avail_factor(usage.free) >> shift;
1654
1655         buf->f_files    = usage.nr_inodes + avail_inodes;
1656         buf->f_ffree    = avail_inodes;
1657
1658         buf->f_fsid     = uuid_to_fsid(c->sb.user_uuid.b);
1659         buf->f_namelen  = BCH_NAME_MAX;
1660
1661         return 0;
1662 }
1663
1664 static int bch2_sync_fs(struct super_block *sb, int wait)
1665 {
1666         struct bch_fs *c = sb->s_fs_info;
1667         int ret;
1668
1669         if (c->opts.journal_flush_disabled)
1670                 return 0;
1671
1672         if (!wait) {
1673                 bch2_journal_flush_async(&c->journal, NULL);
1674                 return 0;
1675         }
1676
1677         ret = bch2_journal_flush(&c->journal);
1678         return bch2_err_class(ret);
1679 }
1680
1681 static struct bch_fs *bch2_path_to_fs(const char *path)
1682 {
1683         struct bch_fs *c;
1684         dev_t dev;
1685         int ret;
1686
1687         ret = lookup_bdev(path, &dev);
1688         if (ret)
1689                 return ERR_PTR(ret);
1690
1691         c = bch2_dev_to_fs(dev);
1692         if (c)
1693                 closure_put(&c->cl);
1694         return c ?: ERR_PTR(-ENOENT);
1695 }
1696
1697 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1698 {
1699         struct bch_fs *c = sb->s_fs_info;
1700         struct bch_opts opts = bch2_opts_empty();
1701         int ret;
1702
1703         ret = bch2_parse_mount_opts(c, &opts, data);
1704         if (ret)
1705                 goto err;
1706
1707         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1708
1709         if (opts.read_only != c->opts.read_only) {
1710                 down_write(&c->state_lock);
1711
1712                 if (opts.read_only) {
1713                         bch2_fs_read_only(c);
1714
1715                         sb->s_flags |= SB_RDONLY;
1716                 } else {
1717                         ret = bch2_fs_read_write(c);
1718                         if (ret) {
1719                                 bch_err(c, "error going rw: %i", ret);
1720                                 up_write(&c->state_lock);
1721                                 ret = -EINVAL;
1722                                 goto err;
1723                         }
1724
1725                         sb->s_flags &= ~SB_RDONLY;
1726                 }
1727
1728                 c->opts.read_only = opts.read_only;
1729
1730                 up_write(&c->state_lock);
1731         }
1732
1733         if (opt_defined(opts, errors))
1734                 c->opts.errors = opts.errors;
1735 err:
1736         return bch2_err_class(ret);
1737 }
1738
1739 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1740 {
1741         struct bch_fs *c = root->d_sb->s_fs_info;
1742         bool first = true;
1743
1744         for_each_online_member(c, ca) {
1745                 if (!first)
1746                         seq_putc(seq, ':');
1747                 first = false;
1748                 seq_puts(seq, ca->disk_sb.sb_name);
1749         }
1750
1751         return 0;
1752 }
1753
1754 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1755 {
1756         struct bch_fs *c = root->d_sb->s_fs_info;
1757         enum bch_opt_id i;
1758         struct printbuf buf = PRINTBUF;
1759         int ret = 0;
1760
1761         for (i = 0; i < bch2_opts_nr; i++) {
1762                 const struct bch_option *opt = &bch2_opt_table[i];
1763                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1764
1765                 if (!(opt->flags & OPT_MOUNT))
1766                         continue;
1767
1768                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1769                         continue;
1770
1771                 printbuf_reset(&buf);
1772                 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1773                                  OPT_SHOW_MOUNT_STYLE);
1774                 seq_putc(seq, ',');
1775                 seq_puts(seq, buf.buf);
1776         }
1777
1778         if (buf.allocation_failure)
1779                 ret = -ENOMEM;
1780         printbuf_exit(&buf);
1781         return ret;
1782 }
1783
1784 static void bch2_put_super(struct super_block *sb)
1785 {
1786         struct bch_fs *c = sb->s_fs_info;
1787
1788         __bch2_fs_stop(c);
1789 }
1790
1791 /*
1792  * bcachefs doesn't currently integrate intwrite freeze protection but the
1793  * internal write references serve the same purpose. Therefore reuse the
1794  * read-only transition code to perform the quiesce. The caveat is that we don't
1795  * currently have the ability to block tasks that want a write reference while
1796  * the superblock is frozen. This is fine for now, but we should either add
1797  * blocking support or find a way to integrate sb_start_intwrite() and friends.
1798  */
1799 static int bch2_freeze(struct super_block *sb)
1800 {
1801         struct bch_fs *c = sb->s_fs_info;
1802
1803         down_write(&c->state_lock);
1804         bch2_fs_read_only(c);
1805         up_write(&c->state_lock);
1806         return 0;
1807 }
1808
1809 static int bch2_unfreeze(struct super_block *sb)
1810 {
1811         struct bch_fs *c = sb->s_fs_info;
1812         int ret;
1813
1814         if (test_bit(BCH_FS_emergency_ro, &c->flags))
1815                 return 0;
1816
1817         down_write(&c->state_lock);
1818         ret = bch2_fs_read_write(c);
1819         up_write(&c->state_lock);
1820         return ret;
1821 }
1822
1823 static const struct super_operations bch_super_operations = {
1824         .alloc_inode    = bch2_alloc_inode,
1825         .destroy_inode  = bch2_destroy_inode,
1826         .write_inode    = bch2_vfs_write_inode,
1827         .evict_inode    = bch2_evict_inode,
1828         .sync_fs        = bch2_sync_fs,
1829         .statfs         = bch2_statfs,
1830         .show_devname   = bch2_show_devname,
1831         .show_options   = bch2_show_options,
1832         .remount_fs     = bch2_remount,
1833         .put_super      = bch2_put_super,
1834         .freeze_fs      = bch2_freeze,
1835         .unfreeze_fs    = bch2_unfreeze,
1836 };
1837
1838 static int bch2_set_super(struct super_block *s, void *data)
1839 {
1840         s->s_fs_info = data;
1841         return 0;
1842 }
1843
1844 static int bch2_noset_super(struct super_block *s, void *data)
1845 {
1846         return -EBUSY;
1847 }
1848
1849 typedef DARRAY(struct bch_fs *) darray_fs;
1850
1851 static int bch2_test_super(struct super_block *s, void *data)
1852 {
1853         struct bch_fs *c = s->s_fs_info;
1854         darray_fs *d = data;
1855
1856         if (!c)
1857                 return false;
1858
1859         darray_for_each(*d, i)
1860                 if (c != *i)
1861                         return false;
1862         return true;
1863 }
1864
1865 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1866                                  int flags, const char *dev_name, void *data)
1867 {
1868         struct bch_fs *c;
1869         struct super_block *sb;
1870         struct inode *vinode;
1871         struct bch_opts opts = bch2_opts_empty();
1872         int ret;
1873
1874         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1875
1876         ret = bch2_parse_mount_opts(NULL, &opts, data);
1877         if (ret) {
1878                 ret = bch2_err_class(ret);
1879                 return ERR_PTR(ret);
1880         }
1881
1882         if (!dev_name || strlen(dev_name) == 0)
1883                 return ERR_PTR(-EINVAL);
1884
1885         darray_str devs;
1886         ret = bch2_split_devs(dev_name, &devs);
1887         if (ret)
1888                 return ERR_PTR(ret);
1889
1890         darray_fs devs_to_fs = {};
1891         darray_for_each(devs, i) {
1892                 ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
1893                 if (ret) {
1894                         sb = ERR_PTR(ret);
1895                         goto got_sb;
1896                 }
1897         }
1898
1899         sb = sget(fs_type, bch2_test_super, bch2_noset_super, flags|SB_NOSEC, &devs_to_fs);
1900         if (!IS_ERR(sb))
1901                 goto got_sb;
1902
1903         c = bch2_fs_open(devs.data, devs.nr, opts);
1904         if (IS_ERR(c)) {
1905                 sb = ERR_CAST(c);
1906                 goto got_sb;
1907         }
1908
1909         /* Some options can't be parsed until after the fs is started: */
1910         ret = bch2_parse_mount_opts(c, &opts, data);
1911         if (ret) {
1912                 bch2_fs_stop(c);
1913                 sb = ERR_PTR(ret);
1914                 goto got_sb;
1915         }
1916
1917         bch2_opts_apply(&c->opts, opts);
1918
1919         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1920         if (IS_ERR(sb))
1921                 bch2_fs_stop(c);
1922 got_sb:
1923         darray_exit(&devs_to_fs);
1924         bch2_darray_str_exit(&devs);
1925
1926         if (IS_ERR(sb)) {
1927                 ret = PTR_ERR(sb);
1928                 ret = bch2_err_class(ret);
1929                 return ERR_PTR(ret);
1930         }
1931
1932         c = sb->s_fs_info;
1933
1934         if (sb->s_root) {
1935                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1936                         ret = -EBUSY;
1937                         goto err_put_super;
1938                 }
1939                 goto out;
1940         }
1941
1942         sb->s_blocksize         = block_bytes(c);
1943         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1944         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1945         sb->s_op                = &bch_super_operations;
1946         sb->s_export_op         = &bch_export_ops;
1947 #ifdef CONFIG_BCACHEFS_QUOTA
1948         sb->s_qcop              = &bch2_quotactl_operations;
1949         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1950 #endif
1951         sb->s_xattr             = bch2_xattr_handlers;
1952         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1953         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1954         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1955         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1956         sb->s_uuid              = c->sb.user_uuid;
1957         c->vfs_sb               = sb;
1958         strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1959
1960         ret = super_setup_bdi(sb);
1961         if (ret)
1962                 goto err_put_super;
1963
1964         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1965
1966         for_each_online_member(c, ca) {
1967                 struct block_device *bdev = ca->disk_sb.bdev;
1968
1969                 /* XXX: create an anonymous device for multi device filesystems */
1970                 sb->s_bdev      = bdev;
1971                 sb->s_dev       = bdev->bd_dev;
1972                 percpu_ref_put(&ca->io_ref);
1973                 break;
1974         }
1975
1976         c->dev = sb->s_dev;
1977
1978 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1979         if (c->opts.acl)
1980                 sb->s_flags     |= SB_POSIXACL;
1981 #endif
1982
1983         sb->s_shrink->seeks = 0;
1984
1985         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
1986         ret = PTR_ERR_OR_ZERO(vinode);
1987         bch_err_msg(c, ret, "mounting: error getting root inode");
1988         if (ret)
1989                 goto err_put_super;
1990
1991         sb->s_root = d_make_root(vinode);
1992         if (!sb->s_root) {
1993                 bch_err(c, "error mounting: error allocating root dentry");
1994                 ret = -ENOMEM;
1995                 goto err_put_super;
1996         }
1997
1998         sb->s_flags |= SB_ACTIVE;
1999 out:
2000         return dget(sb->s_root);
2001
2002 err_put_super:
2003         __bch2_fs_stop(c);
2004         deactivate_locked_super(sb);
2005         return ERR_PTR(bch2_err_class(ret));
2006 }
2007
2008 static void bch2_kill_sb(struct super_block *sb)
2009 {
2010         struct bch_fs *c = sb->s_fs_info;
2011
2012         generic_shutdown_super(sb);
2013         bch2_fs_free(c);
2014 }
2015
2016 static struct file_system_type bcache_fs_type = {
2017         .owner          = THIS_MODULE,
2018         .name           = "bcachefs",
2019         .mount          = bch2_mount,
2020         .kill_sb        = bch2_kill_sb,
2021         .fs_flags       = FS_REQUIRES_DEV,
2022 };
2023
2024 MODULE_ALIAS_FS("bcachefs");
2025
2026 void bch2_vfs_exit(void)
2027 {
2028         unregister_filesystem(&bcache_fs_type);
2029         kmem_cache_destroy(bch2_inode_cache);
2030 }
2031
2032 int __init bch2_vfs_init(void)
2033 {
2034         int ret = -ENOMEM;
2035
2036         bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
2037         if (!bch2_inode_cache)
2038                 goto err;
2039
2040         ret = register_filesystem(&bcache_fs_type);
2041         if (ret)
2042                 goto err;
2043
2044         return 0;
2045 err:
2046         bch2_vfs_exit();
2047         return ret;
2048 }
2049
2050 #endif /* NO_BCACHEFS_FS */