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