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

   1 #ifndef NO_BCACHEFS_FS
   2
   3 #include "bcachefs.h"
   4 #include "acl.h"
   5 #include "btree_update.h"
   6 #include "buckets.h"
   7 #include "chardev.h"
   8 #include "dirent.h"
   9 #include "extents.h"
  10 #include "fs.h"
  11 #include "fs-io.h"
  12 #include "fs-ioctl.h"
  13 #include "fsck.h"
  14 #include "inode.h"
  15 #include "io.h"
  16 #include "journal.h"
  17 #include "keylist.h"
  18 #include "quota.h"
  19 #include "super.h"
  20 #include "xattr.h"
  21
  22 #include <linux/aio.h>
  23 #include <linux/backing-dev.h>
  24 #include <linux/exportfs.h>
  25 #include <linux/module.h>
  26 #include <linux/posix_acl.h>
  27 #include <linux/random.h>
  28 #include <linux/statfs.h>
  29 #include <linux/xattr.h>
  30
  31 static struct kmem_cache *bch2_inode_cache;
  32
  33 static void bch2_vfs_inode_init(struct bch_fs *,
  34                                 struct bch_inode_info *,
  35                                 struct bch_inode_unpacked *);
  36
  37 static void journal_seq_copy(struct bch_inode_info *dst,
  38                              u64 journal_seq)
  39 {
  40         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  41
  42         do {
  43                 old = v;
  44
  45                 if (old >= journal_seq)
  46                         break;
  47         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  48 }
  49
  50 /*
  51  * I_SIZE_DIRTY requires special handling:
  52  *
  53  * To the recovery code, the flag means that there is stale data past i_size
  54  * that needs to be deleted; it's used for implementing atomic appends and
  55  * truncates.
  56  *
  57  * On append, we set I_SIZE_DIRTY before doing the write, then after the write
  58  * we clear I_SIZE_DIRTY atomically with updating i_size to the new larger size
  59  * that exposes the data we just wrote.
  60  *
  61  * On truncate, it's the reverse: We set I_SIZE_DIRTY atomically with setting
  62  * i_size to the new smaller size, then we delete the data that we just made
  63  * invisible, and then we clear I_SIZE_DIRTY.
  64  *
  65  * Because there can be multiple appends in flight at a time, we need a refcount
  66  * (i_size_dirty_count) instead of manipulating the flag directly. Nonzero
  67  * refcount means I_SIZE_DIRTY is set, zero means it's cleared.
  68  *
  69  * Because write_inode() can be called at any time, i_size_dirty_count means
  70  * something different to the runtime code - it means to write_inode() "don't
  71  * update i_size yet".
  72  *
  73  * We don't clear I_SIZE_DIRTY directly, we let write_inode() clear it when
  74  * i_size_dirty_count is zero - but the reverse is not true, I_SIZE_DIRTY must
  75  * be set explicitly.
  76  */
  77
  78 void bch2_inode_update_after_write(struct bch_fs *c,
  79                                    struct bch_inode_info *inode,
  80                                    struct bch_inode_unpacked *bi,
  81                                    unsigned fields)
  82 {
  83         set_nlink(&inode->v, bi->bi_flags & BCH_INODE_UNLINKED
  84                   ? 0
  85                   : bi->bi_nlink + nlink_bias(inode->v.i_mode));
  86         i_uid_write(&inode->v, bi->bi_uid);
  87         i_gid_write(&inode->v, bi->bi_gid);
  88         inode->v.i_mode = bi->bi_mode;
  89
  90         if (fields & ATTR_ATIME)
  91                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
  92         if (fields & ATTR_MTIME)
  93                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
  94         if (fields & ATTR_CTIME)
  95                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
  96
  97         inode->ei_inode         = *bi;
  98
  99         bch2_inode_flags_to_vfs(inode);
 100 }
 101
 102 int __must_check bch2_write_inode_trans(struct btree_trans *trans,
 103                                 struct bch_inode_info *inode,
 104                                 struct bch_inode_unpacked *inode_u,
 105                                 inode_set_fn set,
 106                                 void *p)
 107 {
 108         struct bch_fs *c = trans->c;
 109         struct btree_iter *iter;
 110         struct bkey_inode_buf *inode_p;
 111         int ret;
 112
 113         lockdep_assert_held(&inode->ei_update_lock);
 114
 115         if (c->opts.new_inode_updates) {
 116         /* XXX: Don't do this with btree locks held */
 117         if (!inode->ei_inode_update)
 118                 inode->ei_inode_update =
 119                         bch2_deferred_update_alloc(c, BTREE_ID_INODES, 64);
 120         } else {
 121                 iter = bch2_trans_get_iter(trans, BTREE_ID_INODES,
 122                                            POS(inode->v.i_ino, 0),
 123                                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 124                 if (IS_ERR(iter))
 125                         return PTR_ERR(iter);
 126
 127                 /* The btree node lock is our lock on the inode: */
 128                 ret = bch2_btree_iter_traverse(iter);
 129                 if (ret)
 130                         return ret;
 131         }
 132
 133         *inode_u = inode->ei_inode;
 134
 135         if (set) {
 136                 ret = set(inode, inode_u, p);
 137                 if (ret)
 138                         return ret;
 139         }
 140
 141         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 142         if (IS_ERR(inode_p))
 143                 return PTR_ERR(inode_p);
 144
 145         bch2_inode_pack(inode_p, inode_u);
 146
 147         if (!inode->ei_inode_update)
 148                 bch2_trans_update(trans,
 149                         BTREE_INSERT_ENTRY(iter, &inode_p->inode.k_i));
 150         else
 151                 bch2_trans_update(trans,
 152                         BTREE_INSERT_DEFERRED(inode->ei_inode_update,
 153                                               &inode_p->inode.k_i));
 154
 155         return 0;
 156 }
 157
 158 int __must_check bch2_write_inode(struct bch_fs *c,
 159                                   struct bch_inode_info *inode,
 160                                   inode_set_fn set,
 161                                   void *p, unsigned fields)
 162 {
 163         struct btree_trans trans;
 164         struct bch_inode_unpacked inode_u;
 165         int ret;
 166
 167         bch2_trans_init(&trans, c);
 168 retry:
 169         bch2_trans_begin(&trans);
 170
 171         ret = bch2_write_inode_trans(&trans, inode, &inode_u, set, p) ?:
 172                 bch2_trans_commit(&trans, NULL,
 173                                   &inode->ei_journal_seq,
 174                                   BTREE_INSERT_ATOMIC|
 175                                   BTREE_INSERT_NOUNLOCK|
 176                                   BTREE_INSERT_NOFAIL);
 177         if (ret == -EINTR)
 178                 goto retry;
 179
 180         /*
 181          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 182          * this is important for inode updates via bchfs_write_index_update
 183          */
 184         if (!ret)
 185                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 186
 187         bch2_trans_exit(&trans);
 188         return ret < 0 ? ret : 0;
 189 }
 190
 191 int bch2_fs_quota_transfer(struct bch_fs *c,
 192                            struct bch_inode_info *inode,
 193                            struct bch_qid new_qid,
 194                            unsigned qtypes,
 195                            enum quota_acct_mode mode)
 196 {
 197         unsigned i;
 198         int ret;
 199
 200         qtypes &= enabled_qtypes(c);
 201
 202         for (i = 0; i < QTYP_NR; i++)
 203                 if (new_qid.q[i] == inode->ei_qid.q[i])
 204                         qtypes &= ~(1U << i);
 205
 206         if (!qtypes)
 207                 return 0;
 208
 209         mutex_lock(&inode->ei_quota_lock);
 210
 211         ret = bch2_quota_transfer(c, qtypes, new_qid,
 212                                   inode->ei_qid,
 213                                   inode->v.i_blocks +
 214                                   inode->ei_quota_reserved,
 215                                   mode);
 216         if (!ret)
 217                 for (i = 0; i < QTYP_NR; i++)
 218                         if (qtypes & (1 << i))
 219                                 inode->ei_qid.q[i] = new_qid.q[i];
 220
 221         mutex_unlock(&inode->ei_quota_lock);
 222
 223         return ret;
 224 }
 225
 226 int bch2_reinherit_attrs_fn(struct bch_inode_info *inode,
 227                             struct bch_inode_unpacked *bi,
 228                             void *p)
 229 {
 230         struct bch_inode_info *dir = p;
 231         u64 src, dst;
 232         unsigned id;
 233         int ret = 1;
 234
 235         for (id = 0; id < Inode_opt_nr; id++) {
 236                 if (bi->bi_fields_set & (1 << id))
 237                         continue;
 238
 239                 src = bch2_inode_opt_get(&dir->ei_inode, id);
 240                 dst = bch2_inode_opt_get(bi, id);
 241
 242                 if (src == dst)
 243                         continue;
 244
 245                 bch2_inode_opt_set(bi, id, src);
 246                 ret = 0;
 247         }
 248
 249         return ret;
 250 }
 251
 252 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 253 {
 254         struct bch_inode_unpacked inode_u;
 255         struct bch_inode_info *inode;
 256         int ret;
 257
 258         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 259         if (unlikely(!inode))
 260                 return ERR_PTR(-ENOMEM);
 261         if (!(inode->v.i_state & I_NEW))
 262                 return &inode->v;
 263
 264         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 265         if (ret) {
 266                 iget_failed(&inode->v);
 267                 return ERR_PTR(ret);
 268         }
 269
 270         bch2_vfs_inode_init(c, inode, &inode_u);
 271
 272         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 273
 274         unlock_new_inode(&inode->v);
 275
 276         return &inode->v;
 277 }
 278
 279 static void bch2_inode_init_owner(struct bch_inode_unpacked *inode_u,
 280                                   const struct inode *dir, umode_t mode)
 281 {
 282         kuid_t uid = current_fsuid();
 283         kgid_t gid;
 284
 285         if (dir && dir->i_mode & S_ISGID) {
 286                 gid = dir->i_gid;
 287                 if (S_ISDIR(mode))
 288                         mode |= S_ISGID;
 289         } else
 290                 gid = current_fsgid();
 291
 292         inode_u->bi_uid         = from_kuid(dir->i_sb->s_user_ns, uid);
 293         inode_u->bi_gid         = from_kgid(dir->i_sb->s_user_ns, gid);
 294         inode_u->bi_mode        = mode;
 295 }
 296
 297 static int inode_update_for_create_fn(struct bch_inode_info *inode,
 298                                       struct bch_inode_unpacked *bi,
 299                                       void *p)
 300 {
 301         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 302         struct bch_inode_unpacked *new_inode = p;
 303
 304         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 305
 306         if (S_ISDIR(new_inode->bi_mode))
 307                 bi->bi_nlink++;
 308
 309         return 0;
 310 }
 311
 312 static struct bch_inode_info *
 313 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 314               umode_t mode, dev_t rdev, bool tmpfile)
 315 {
 316         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 317         struct btree_trans trans;
 318         struct bch_inode_unpacked dir_u;
 319         struct bch_inode_info *inode, *old;
 320         struct bch_inode_unpacked inode_u;
 321         struct bch_hash_info hash_info;
 322         struct posix_acl *default_acl = NULL, *acl = NULL;
 323         u64 journal_seq = 0;
 324         int ret;
 325
 326         bch2_inode_init(c, &inode_u, 0, 0, 0, rdev, &dir->ei_inode);
 327         bch2_inode_init_owner(&inode_u, &dir->v, mode);
 328
 329         hash_info = bch2_hash_info_init(c, &inode_u);
 330
 331         if (tmpfile)
 332                 inode_u.bi_flags |= BCH_INODE_UNLINKED;
 333
 334         ret = bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 335                               KEY_TYPE_QUOTA_PREALLOC);
 336         if (ret)
 337                 return ERR_PTR(ret);
 338
 339 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 340         ret = posix_acl_create(&dir->v, &inode_u.bi_mode, &default_acl, &acl);
 341         if (ret)
 342                 goto err;
 343 #endif
 344
 345         /*
 346          * preallocate vfs inode before btree transaction, so that nothing can
 347          * fail after the transaction succeeds:
 348          */
 349         inode = to_bch_ei(new_inode(c->vfs_sb));
 350         if (unlikely(!inode)) {
 351                 ret = -ENOMEM;
 352                 goto err;
 353         }
 354
 355         if (!tmpfile)
 356                 mutex_lock(&dir->ei_update_lock);
 357
 358         bch2_trans_init(&trans, c);
 359 retry:
 360         bch2_trans_begin(&trans);
 361
 362         ret   = __bch2_inode_create(&trans, &inode_u,
 363                                     BLOCKDEV_INODE_MAX, 0,
 364                                     &c->unused_inode_hint) ?:
 365                 (default_acl
 366                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 367                                       default_acl, ACL_TYPE_DEFAULT)
 368                  : 0) ?:
 369                 (acl
 370                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 371                                       acl, ACL_TYPE_ACCESS)
 372                  : 0) ?:
 373                 (!tmpfile
 374                  ? __bch2_dirent_create(&trans, dir->v.i_ino,
 375                                         &dir->ei_str_hash,
 376                                         mode_to_type(mode),
 377                                         &dentry->d_name,
 378                                         inode_u.bi_inum,
 379                                         BCH_HASH_SET_MUST_CREATE)
 380                 : 0) ?:
 381                 (!tmpfile
 382                  ? bch2_write_inode_trans(&trans, dir, &dir_u,
 383                                           inode_update_for_create_fn,
 384                                           &inode_u)
 385                  : 0) ?:
 386                 bch2_trans_commit(&trans, NULL,
 387                                   &journal_seq,
 388                                   BTREE_INSERT_ATOMIC|
 389                                   BTREE_INSERT_NOUNLOCK);
 390         if (ret == -EINTR)
 391                 goto retry;
 392         if (unlikely(ret))
 393                 goto err_trans;
 394
 395         if (!tmpfile) {
 396                 bch2_inode_update_after_write(c, dir, &dir_u,
 397                                               ATTR_MTIME|ATTR_CTIME);
 398                 journal_seq_copy(dir, inode->ei_journal_seq);
 399                 mutex_unlock(&dir->ei_update_lock);
 400         }
 401
 402         bch2_vfs_inode_init(c, inode, &inode_u);
 403         journal_seq_copy(inode, journal_seq);
 404
 405         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 406         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 407
 408         /*
 409          * we must insert the new inode into the inode cache before calling
 410          * bch2_trans_exit() and dropping locks, else we could race with another
 411          * thread pulling the inode in and modifying it:
 412          */
 413
 414         old = to_bch_ei(insert_inode_locked2(&inode->v));
 415         if (unlikely(old)) {
 416                 /*
 417                  * We raced, another process pulled the new inode into cache
 418                  * before us:
 419                  */
 420                 old->ei_journal_seq = inode->ei_journal_seq;
 421                 make_bad_inode(&inode->v);
 422                 iput(&inode->v);
 423
 424                 inode = old;
 425         } else {
 426                 /*
 427                  * we really don't want insert_inode_locked2() to be setting
 428                  * I_NEW...
 429                  */
 430                 unlock_new_inode(&inode->v);
 431         }
 432
 433         bch2_trans_exit(&trans);
 434 out:
 435         posix_acl_release(default_acl);
 436         posix_acl_release(acl);
 437         return inode;
 438 err_trans:
 439         if (!tmpfile)
 440                 mutex_unlock(&dir->ei_update_lock);
 441
 442         bch2_trans_exit(&trans);
 443         make_bad_inode(&inode->v);
 444         iput(&inode->v);
 445 err:
 446         bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1, KEY_TYPE_QUOTA_WARN);
 447         inode = ERR_PTR(ret);
 448         goto out;
 449 }
 450
 451 /* methods */
 452
 453 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 454                                   unsigned int flags)
 455 {
 456         struct bch_fs *c = vdir->i_sb->s_fs_info;
 457         struct bch_inode_info *dir = to_bch_ei(vdir);
 458         struct inode *vinode = NULL;
 459         u64 inum;
 460
 461         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 462                                   &dir->ei_str_hash,
 463                                   &dentry->d_name);
 464
 465         if (inum)
 466                 vinode = bch2_vfs_inode_get(c, inum);
 467
 468         return d_splice_alias(vinode, dentry);
 469 }
 470
 471 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 472                        umode_t mode, bool excl)
 473 {
 474         struct bch_inode_info *inode =
 475                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0, false);
 476
 477         if (IS_ERR(inode))
 478                 return PTR_ERR(inode);
 479
 480         d_instantiate(dentry, &inode->v);
 481         return 0;
 482 }
 483
 484 static int inode_update_for_link_fn(struct bch_inode_info *inode,
 485                                     struct bch_inode_unpacked *bi,
 486                                     void *p)
 487 {
 488         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 489
 490         bi->bi_ctime = bch2_current_time(c);
 491
 492         if (bi->bi_flags & BCH_INODE_UNLINKED)
 493                 bi->bi_flags &= ~BCH_INODE_UNLINKED;
 494         else
 495                 bi->bi_nlink++;
 496
 497         return 0;
 498 }
 499
 500 static int __bch2_link(struct bch_fs *c,
 501                        struct bch_inode_info *inode,
 502                        struct bch_inode_info *dir,
 503                        struct dentry *dentry)
 504 {
 505         struct btree_trans trans;
 506         struct bch_inode_unpacked inode_u;
 507         int ret;
 508
 509         mutex_lock(&inode->ei_update_lock);
 510         bch2_trans_init(&trans, c);
 511 retry:
 512         bch2_trans_begin(&trans);
 513
 514         ret   = __bch2_dirent_create(&trans, dir->v.i_ino,
 515                                      &dir->ei_str_hash,
 516                                      mode_to_type(inode->v.i_mode),
 517                                      &dentry->d_name,
 518                                      inode->v.i_ino,
 519                                      BCH_HASH_SET_MUST_CREATE) ?:
 520                 bch2_write_inode_trans(&trans, inode, &inode_u,
 521                                        inode_update_for_link_fn,
 522                                        NULL) ?:
 523                 bch2_trans_commit(&trans, NULL,
 524                                   &inode->ei_journal_seq,
 525                                   BTREE_INSERT_ATOMIC|
 526                                   BTREE_INSERT_NOUNLOCK);
 527
 528         if (ret == -EINTR)
 529                 goto retry;
 530
 531         if (likely(!ret))
 532                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 533
 534         bch2_trans_exit(&trans);
 535         mutex_unlock(&inode->ei_update_lock);
 536         return ret;
 537 }
 538
 539 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 540                      struct dentry *dentry)
 541 {
 542         struct bch_fs *c = vdir->i_sb->s_fs_info;
 543         struct bch_inode_info *dir = to_bch_ei(vdir);
 544         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 545         int ret;
 546
 547         lockdep_assert_held(&inode->v.i_rwsem);
 548
 549         ret = __bch2_link(c, inode, dir, dentry);
 550         if (unlikely(ret))
 551                 return ret;
 552
 553         ihold(&inode->v);
 554         d_instantiate(dentry, &inode->v);
 555         return 0;
 556 }
 557
 558 static int inode_update_dir_for_unlink_fn(struct bch_inode_info *inode,
 559                                           struct bch_inode_unpacked *bi,
 560                                           void *p)
 561 {
 562         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 563         struct bch_inode_info *unlink_inode = p;
 564
 565         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 566
 567         bi->bi_nlink -= S_ISDIR(unlink_inode->v.i_mode);
 568
 569         return 0;
 570 }
 571
 572 static int inode_update_for_unlink_fn(struct bch_inode_info *inode,
 573                                       struct bch_inode_unpacked *bi,
 574                                       void *p)
 575 {
 576         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 577
 578         bi->bi_ctime = bch2_current_time(c);
 579         if (bi->bi_nlink)
 580                 bi->bi_nlink--;
 581         else
 582                 bi->bi_flags |= BCH_INODE_UNLINKED;
 583
 584         return 0;
 585 }
 586
 587 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 588 {
 589         struct bch_fs *c = vdir->i_sb->s_fs_info;
 590         struct bch_inode_info *dir = to_bch_ei(vdir);
 591         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 592         struct bch_inode_unpacked dir_u, inode_u;
 593         struct btree_trans trans;
 594         int ret;
 595
 596         bch2_lock_inodes(dir, inode);
 597         bch2_trans_init(&trans, c);
 598 retry:
 599         bch2_trans_begin(&trans);
 600
 601         ret   = __bch2_dirent_delete(&trans, dir->v.i_ino,
 602                                      &dir->ei_str_hash,
 603                                      &dentry->d_name) ?:
 604                 bch2_write_inode_trans(&trans, dir, &dir_u,
 605                                        inode_update_dir_for_unlink_fn,
 606                                        inode) ?:
 607                 bch2_write_inode_trans(&trans, inode, &inode_u,
 608                                        inode_update_for_unlink_fn,
 609                                        NULL) ?:
 610                 bch2_trans_commit(&trans, NULL,
 611                                   &dir->ei_journal_seq,
 612                                   BTREE_INSERT_ATOMIC|
 613                                   BTREE_INSERT_NOUNLOCK|
 614                                   BTREE_INSERT_NOFAIL);
 615         if (ret == -EINTR)
 616                 goto retry;
 617         if (ret)
 618                 goto err;
 619
 620         if (dir->ei_journal_seq > inode->ei_journal_seq)
 621                 inode->ei_journal_seq = dir->ei_journal_seq;
 622
 623         bch2_inode_update_after_write(c, dir, &dir_u,
 624                                       ATTR_MTIME|ATTR_CTIME);
 625         bch2_inode_update_after_write(c, inode, &inode_u,
 626                                       ATTR_MTIME);
 627 err:
 628         bch2_trans_exit(&trans);
 629         bch2_unlock_inodes(dir, inode);
 630
 631         return ret;
 632 }
 633
 634 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 635                         const char *symname)
 636 {
 637         struct bch_fs *c = vdir->i_sb->s_fs_info;
 638         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 639         int ret;
 640
 641         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 642         if (unlikely(IS_ERR(inode)))
 643                 return PTR_ERR(inode);
 644
 645         inode_lock(&inode->v);
 646         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 647         inode_unlock(&inode->v);
 648
 649         if (unlikely(ret))
 650                 goto err;
 651
 652         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 653         if (unlikely(ret))
 654                 goto err;
 655
 656         journal_seq_copy(dir, inode->ei_journal_seq);
 657
 658         ret = __bch2_link(c, inode, dir, dentry);
 659         if (unlikely(ret))
 660                 goto err;
 661
 662         d_instantiate(dentry, &inode->v);
 663         return 0;
 664 err:
 665         iput(&inode->v);
 666         return ret;
 667 }
 668
 669 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 670 {
 671         struct bch_inode_info *inode =
 672                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFDIR, 0, false);
 673
 674         if (IS_ERR(inode))
 675                 return PTR_ERR(inode);
 676
 677         d_instantiate(dentry, &inode->v);
 678         return 0;
 679 }
 680
 681 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 682 {
 683         struct bch_fs *c = vdir->i_sb->s_fs_info;
 684
 685         if (bch2_empty_dir(c, dentry->d_inode->i_ino))
 686                 return -ENOTEMPTY;
 687
 688         return bch2_unlink(vdir, dentry);
 689 }
 690
 691 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 692                       umode_t mode, dev_t rdev)
 693 {
 694         struct bch_inode_info *inode =
 695                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 696
 697         if (IS_ERR(inode))
 698                 return PTR_ERR(inode);
 699
 700         d_instantiate(dentry, &inode->v);
 701         return 0;
 702 }
 703
 704 struct rename_info {
 705         u64                     now;
 706         struct bch_inode_info   *src_dir;
 707         struct bch_inode_info   *dst_dir;
 708         struct bch_inode_info   *src_inode;
 709         struct bch_inode_info   *dst_inode;
 710         enum bch_rename_mode    mode;
 711 };
 712
 713 static int inode_update_for_rename_fn(struct bch_inode_info *inode,
 714                                       struct bch_inode_unpacked *bi,
 715                                       void *p)
 716 {
 717         struct rename_info *info = p;
 718         int ret;
 719
 720         if (inode == info->src_dir) {
 721                 bi->bi_nlink -= S_ISDIR(info->src_inode->v.i_mode);
 722                 bi->bi_nlink += info->dst_inode &&
 723                         S_ISDIR(info->dst_inode->v.i_mode) &&
 724                         info->mode == BCH_RENAME_EXCHANGE;
 725         }
 726
 727         if (inode == info->dst_dir) {
 728                 bi->bi_nlink += S_ISDIR(info->src_inode->v.i_mode);
 729                 bi->bi_nlink -= info->dst_inode &&
 730                         S_ISDIR(info->dst_inode->v.i_mode);
 731         }
 732
 733         if (inode == info->src_inode) {
 734                 ret = bch2_reinherit_attrs_fn(inode, bi, info->dst_dir);
 735
 736                 BUG_ON(!ret && S_ISDIR(info->src_inode->v.i_mode));
 737         }
 738
 739         if (inode == info->dst_inode &&
 740             info->mode == BCH_RENAME_EXCHANGE) {
 741                 ret = bch2_reinherit_attrs_fn(inode, bi, info->src_dir);
 742
 743                 BUG_ON(!ret && S_ISDIR(info->dst_inode->v.i_mode));
 744         }
 745
 746         if (inode == info->dst_inode &&
 747             info->mode == BCH_RENAME_OVERWRITE) {
 748                 BUG_ON(bi->bi_nlink &&
 749                        S_ISDIR(info->dst_inode->v.i_mode));
 750
 751                 if (bi->bi_nlink)
 752                         bi->bi_nlink--;
 753                 else
 754                         bi->bi_flags |= BCH_INODE_UNLINKED;
 755         }
 756
 757         if (inode == info->src_dir ||
 758             inode == info->dst_dir)
 759                 bi->bi_mtime = info->now;
 760         bi->bi_ctime = info->now;
 761
 762         return 0;
 763 }
 764
 765 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 766                         struct inode *dst_vdir, struct dentry *dst_dentry,
 767                         unsigned flags)
 768 {
 769         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 770         struct rename_info i = {
 771                 .src_dir        = to_bch_ei(src_vdir),
 772                 .dst_dir        = to_bch_ei(dst_vdir),
 773                 .src_inode      = to_bch_ei(src_dentry->d_inode),
 774                 .dst_inode      = to_bch_ei(dst_dentry->d_inode),
 775                 .mode           = flags & RENAME_EXCHANGE
 776                                 ? BCH_RENAME_EXCHANGE
 777                         : dst_dentry->d_inode
 778                                 ? BCH_RENAME_OVERWRITE : BCH_RENAME,
 779         };
 780         struct btree_trans trans;
 781         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 782         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 783         u64 journal_seq = 0;
 784         int ret;
 785
 786         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 787                 return -EINVAL;
 788
 789         if (i.mode == BCH_RENAME_OVERWRITE) {
 790                 if (S_ISDIR(i.src_inode->v.i_mode) !=
 791                     S_ISDIR(i.dst_inode->v.i_mode))
 792                         return -ENOTDIR;
 793
 794                 if (S_ISDIR(i.src_inode->v.i_mode) &&
 795                     bch2_empty_dir(c, i.dst_inode->v.i_ino))
 796                         return -ENOTEMPTY;
 797
 798                 ret = filemap_write_and_wait_range(i.src_inode->v.i_mapping,
 799                                                    0, LLONG_MAX);
 800                 if (ret)
 801                         return ret;
 802         }
 803
 804         bch2_lock_inodes(i.src_dir,
 805                          i.dst_dir,
 806                          i.src_inode,
 807                          i.dst_inode);
 808
 809         bch2_trans_init(&trans, c);
 810
 811         if (S_ISDIR(i.src_inode->v.i_mode) &&
 812             inode_attrs_changing(i.dst_dir, i.src_inode)) {
 813                 ret = -EXDEV;
 814                 goto err;
 815         }
 816
 817         if (i.mode == BCH_RENAME_EXCHANGE &&
 818             S_ISDIR(i.dst_inode->v.i_mode) &&
 819             inode_attrs_changing(i.src_dir, i.dst_inode)) {
 820                 ret = -EXDEV;
 821                 goto err;
 822         }
 823
 824         if (inode_attr_changing(i.dst_dir, i.src_inode, Inode_opt_project)) {
 825                 ret = bch2_fs_quota_transfer(c, i.src_inode,
 826                                              i.dst_dir->ei_qid,
 827                                              1 << QTYP_PRJ,
 828                                              KEY_TYPE_QUOTA_PREALLOC);
 829                 if (ret)
 830                         goto err;
 831         }
 832
 833         if (i.mode == BCH_RENAME_EXCHANGE &&
 834             inode_attr_changing(i.src_dir, i.dst_inode, Inode_opt_project)) {
 835                 ret = bch2_fs_quota_transfer(c, i.dst_inode,
 836                                              i.src_dir->ei_qid,
 837                                              1 << QTYP_PRJ,
 838                                              KEY_TYPE_QUOTA_PREALLOC);
 839                 if (ret)
 840                         goto err;
 841         }
 842
 843 retry:
 844         bch2_trans_begin(&trans);
 845         i.now = bch2_current_time(c);
 846
 847         ret   = bch2_dirent_rename(&trans,
 848                                    i.src_dir, &src_dentry->d_name,
 849                                    i.dst_dir, &dst_dentry->d_name,
 850                                    i.mode) ?:
 851                 bch2_write_inode_trans(&trans, i.src_dir, &src_dir_u,
 852                                        inode_update_for_rename_fn, &i) ?:
 853                 (i.src_dir != i.dst_dir
 854                  ? bch2_write_inode_trans(&trans, i.dst_dir, &dst_dir_u,
 855                                        inode_update_for_rename_fn, &i)
 856                  : 0 ) ?:
 857                 bch2_write_inode_trans(&trans, i.src_inode, &src_inode_u,
 858                                        inode_update_for_rename_fn, &i) ?:
 859                 (i.dst_inode
 860                  ? bch2_write_inode_trans(&trans, i.dst_inode, &dst_inode_u,
 861                                        inode_update_for_rename_fn, &i)
 862                  : 0 ) ?:
 863                 bch2_trans_commit(&trans, NULL,
 864                                   &journal_seq,
 865                                   BTREE_INSERT_ATOMIC|
 866                                   BTREE_INSERT_NOUNLOCK);
 867         if (ret == -EINTR)
 868                 goto retry;
 869         if (unlikely(ret))
 870                 goto err;
 871
 872         bch2_inode_update_after_write(c, i.src_dir, &src_dir_u,
 873                                       ATTR_MTIME|ATTR_CTIME);
 874         journal_seq_copy(i.src_dir, journal_seq);
 875
 876         if (i.src_dir != i.dst_dir) {
 877                 bch2_inode_update_after_write(c, i.dst_dir, &dst_dir_u,
 878                                               ATTR_MTIME|ATTR_CTIME);
 879                 journal_seq_copy(i.dst_dir, journal_seq);
 880         }
 881
 882         journal_seq_copy(i.src_inode, journal_seq);
 883         if (i.dst_inode)
 884                 journal_seq_copy(i.dst_inode, journal_seq);
 885
 886         bch2_inode_update_after_write(c, i.src_inode, &src_inode_u,
 887                                       ATTR_CTIME);
 888         if (i.dst_inode)
 889                 bch2_inode_update_after_write(c, i.dst_inode, &dst_inode_u,
 890                                               ATTR_CTIME);
 891 err:
 892         bch2_trans_exit(&trans);
 893
 894         bch2_fs_quota_transfer(c, i.src_inode,
 895                                bch_qid(&i.src_inode->ei_inode),
 896                                1 << QTYP_PRJ,
 897                                KEY_TYPE_QUOTA_NOCHECK);
 898         if (i.dst_inode)
 899                 bch2_fs_quota_transfer(c, i.dst_inode,
 900                                        bch_qid(&i.dst_inode->ei_inode),
 901                                        1 << QTYP_PRJ,
 902                                        KEY_TYPE_QUOTA_NOCHECK);
 903
 904         bch2_unlock_inodes(i.src_dir,
 905                            i.dst_dir,
 906                            i.src_inode,
 907                            i.dst_inode);
 908
 909         return ret;
 910 }
 911
 912 static int inode_update_for_setattr_fn(struct bch_inode_info *inode,
 913                                        struct bch_inode_unpacked *bi,
 914                                        void *p)
 915 {
 916         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 917         struct iattr *attr = p;
 918         unsigned int ia_valid = attr->ia_valid;
 919
 920         if (ia_valid & ATTR_UID)
 921                 bi->bi_uid = from_kuid(inode->v.i_sb->s_user_ns, attr->ia_uid);
 922         if (ia_valid & ATTR_GID)
 923                 bi->bi_gid = from_kgid(inode->v.i_sb->s_user_ns, attr->ia_gid);
 924
 925         if (ia_valid & ATTR_ATIME)
 926                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 927         if (ia_valid & ATTR_MTIME)
 928                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 929         if (ia_valid & ATTR_CTIME)
 930                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 931
 932         if (ia_valid & ATTR_MODE) {
 933                 umode_t mode = attr->ia_mode;
 934                 kgid_t gid = ia_valid & ATTR_GID
 935                         ? attr->ia_gid
 936                         : inode->v.i_gid;
 937
 938                 if (!in_group_p(gid) &&
 939                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 940                         mode &= ~S_ISGID;
 941                 bi->bi_mode = mode;
 942         }
 943
 944         return 0;
 945 }
 946
 947 static int bch2_setattr_nonsize(struct bch_inode_info *inode, struct iattr *iattr)
 948 {
 949         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 950         struct bch_qid qid;
 951         struct btree_trans trans;
 952         struct bch_inode_unpacked inode_u;
 953         struct posix_acl *acl = NULL;
 954         int ret;
 955
 956         mutex_lock(&inode->ei_update_lock);
 957
 958         qid = inode->ei_qid;
 959
 960         if (iattr->ia_valid & ATTR_UID)
 961                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, iattr->ia_uid);
 962
 963         if (iattr->ia_valid & ATTR_GID)
 964                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, iattr->ia_gid);
 965
 966         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 967                                      KEY_TYPE_QUOTA_PREALLOC);
 968         if (ret)
 969                 goto err;
 970
 971         bch2_trans_init(&trans, c);
 972 retry:
 973         bch2_trans_begin(&trans);
 974         kfree(acl);
 975         acl = NULL;
 976
 977         ret = bch2_write_inode_trans(&trans, inode, &inode_u,
 978                                 inode_update_for_setattr_fn, iattr) ?:
 979                 (iattr->ia_valid & ATTR_MODE
 980                  ? bch2_acl_chmod(&trans, inode, iattr->ia_mode, &acl)
 981                  : 0) ?:
 982                 bch2_trans_commit(&trans, NULL,
 983                                   &inode->ei_journal_seq,
 984                                   BTREE_INSERT_ATOMIC|
 985                                   BTREE_INSERT_NOUNLOCK|
 986                                   BTREE_INSERT_NOFAIL);
 987         if (ret == -EINTR)
 988                 goto retry;
 989         if (unlikely(ret))
 990                 goto err_trans;
 991
 992         bch2_inode_update_after_write(c, inode, &inode_u, iattr->ia_valid);
 993
 994         if (acl)
 995                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 996 err_trans:
 997         bch2_trans_exit(&trans);
 998 err:
 999         mutex_unlock(&inode->ei_update_lock);
1000
1001         return ret;
1002 }
1003
1004 static int bch2_getattr(const struct path *path, struct kstat *stat,
1005                         u32 request_mask, unsigned query_flags)
1006 {
1007         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
1008         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1009
1010         stat->dev       = inode->v.i_sb->s_dev;
1011         stat->ino       = inode->v.i_ino;
1012         stat->mode      = inode->v.i_mode;
1013         stat->nlink     = inode->v.i_nlink;
1014         stat->uid       = inode->v.i_uid;
1015         stat->gid       = inode->v.i_gid;
1016         stat->rdev      = inode->v.i_rdev;
1017         stat->size      = i_size_read(&inode->v);
1018         stat->atime     = inode->v.i_atime;
1019         stat->mtime     = inode->v.i_mtime;
1020         stat->ctime     = inode->v.i_ctime;
1021         stat->blksize   = block_bytes(c);
1022         stat->blocks    = inode->v.i_blocks;
1023
1024         if (request_mask & STATX_BTIME) {
1025                 stat->result_mask |= STATX_BTIME;
1026                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
1027         }
1028
1029         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
1030                 stat->attributes |= STATX_ATTR_IMMUTABLE;
1031         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
1032                 stat->attributes |= STATX_ATTR_APPEND;
1033         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
1034                 stat->attributes |= STATX_ATTR_NODUMP;
1035
1036         return 0;
1037 }
1038
1039 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
1040 {
1041         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
1042         int ret;
1043
1044         lockdep_assert_held(&inode->v.i_rwsem);
1045
1046         ret = setattr_prepare(dentry, iattr);
1047         if (ret)
1048                 return ret;
1049
1050         return iattr->ia_valid & ATTR_SIZE
1051                 ? bch2_truncate(inode, iattr)
1052                 : bch2_setattr_nonsize(inode, iattr);
1053 }
1054
1055 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
1056 {
1057         struct bch_inode_info *inode =
1058                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
1059
1060         if (IS_ERR(inode))
1061                 return PTR_ERR(inode);
1062
1063         d_mark_tmpfile(dentry, &inode->v);
1064         d_instantiate(dentry, &inode->v);
1065         return 0;
1066 }
1067
1068 static int bch2_fill_extent(struct fiemap_extent_info *info,
1069                             const struct bkey_i *k, unsigned flags)
1070 {
1071         if (bkey_extent_is_data(&k->k)) {
1072                 struct bkey_s_c_extent e = bkey_i_to_s_c_extent(k);
1073                 const union bch_extent_entry *entry;
1074                 struct extent_ptr_decoded p;
1075                 int ret;
1076
1077                 extent_for_each_ptr_decode(e, p, entry) {
1078                         int flags2 = 0;
1079                         u64 offset = p.ptr.offset;
1080
1081                         if (p.crc.compression_type)
1082                                 flags2 |= FIEMAP_EXTENT_ENCODED;
1083                         else
1084                                 offset += p.crc.offset;
1085
1086                         if ((offset & (PAGE_SECTORS - 1)) ||
1087                             (e.k->size & (PAGE_SECTORS - 1)))
1088                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
1089
1090                         ret = fiemap_fill_next_extent(info,
1091                                                 bkey_start_offset(e.k) << 9,
1092                                                 offset << 9,
1093                                                 e.k->size << 9, flags|flags2);
1094                         if (ret)
1095                                 return ret;
1096                 }
1097
1098                 return 0;
1099         } else if (k->k.type == KEY_TYPE_reservation) {
1100                 return fiemap_fill_next_extent(info,
1101                                                bkey_start_offset(&k->k) << 9,
1102                                                0, k->k.size << 9,
1103                                                flags|
1104                                                FIEMAP_EXTENT_DELALLOC|
1105                                                FIEMAP_EXTENT_UNWRITTEN);
1106         } else {
1107                 BUG();
1108         }
1109 }
1110
1111 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
1112                        u64 start, u64 len)
1113 {
1114         struct bch_fs *c = vinode->i_sb->s_fs_info;
1115         struct bch_inode_info *ei = to_bch_ei(vinode);
1116         struct btree_iter iter;
1117         struct bkey_s_c k;
1118         BKEY_PADDED(k) tmp;
1119         bool have_extent = false;
1120         int ret = 0;
1121
1122         if (start + len < start)
1123                 return -EINVAL;
1124
1125         for_each_btree_key(&iter, c, BTREE_ID_EXTENTS,
1126                            POS(ei->v.i_ino, start >> 9), 0, k)
1127                 if (bkey_extent_is_data(k.k) ||
1128                     k.k->type == KEY_TYPE_reservation) {
1129                         if (bkey_cmp(bkey_start_pos(k.k),
1130                                      POS(ei->v.i_ino, (start + len) >> 9)) >= 0)
1131                                 break;
1132
1133                         if (have_extent) {
1134                                 ret = bch2_fill_extent(info, &tmp.k, 0);
1135                                 if (ret)
1136                                         goto out;
1137                         }
1138
1139                         bkey_reassemble(&tmp.k, k);
1140                         have_extent = true;
1141                 }
1142
1143         if (have_extent)
1144                 ret = bch2_fill_extent(info, &tmp.k, FIEMAP_EXTENT_LAST);
1145 out:
1146         bch2_btree_iter_unlock(&iter);
1147         return ret < 0 ? ret : 0;
1148 }
1149
1150 static const struct vm_operations_struct bch_vm_ops = {
1151         .fault          = filemap_fault,
1152         .map_pages      = filemap_map_pages,
1153         .page_mkwrite   = bch2_page_mkwrite,
1154 };
1155
1156 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1157 {
1158         file_accessed(file);
1159
1160         vma->vm_ops = &bch_vm_ops;
1161         return 0;
1162 }
1163
1164 /* Directories: */
1165
1166 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1167 {
1168         return generic_file_llseek_size(file, offset, whence,
1169                                         S64_MAX, S64_MAX);
1170 }
1171
1172 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1173 {
1174         struct bch_fs *c = file_inode(file)->i_sb->s_fs_info;
1175
1176         return bch2_readdir(c, file, ctx);
1177 }
1178
1179 static const struct file_operations bch_file_operations = {
1180         .llseek         = bch2_llseek,
1181         .read_iter      = generic_file_read_iter,
1182         .write_iter     = bch2_write_iter,
1183         .mmap           = bch2_mmap,
1184         .open           = generic_file_open,
1185         .fsync          = bch2_fsync,
1186         .splice_read    = generic_file_splice_read,
1187         .splice_write   = iter_file_splice_write,
1188         .fallocate      = bch2_fallocate_dispatch,
1189         .unlocked_ioctl = bch2_fs_file_ioctl,
1190 #ifdef CONFIG_COMPAT
1191         .compat_ioctl   = bch2_compat_fs_ioctl,
1192 #endif
1193 };
1194
1195 static const struct inode_operations bch_file_inode_operations = {
1196         .getattr        = bch2_getattr,
1197         .setattr        = bch2_setattr,
1198         .fiemap         = bch2_fiemap,
1199         .listxattr      = bch2_xattr_list,
1200 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1201         .get_acl        = bch2_get_acl,
1202         .set_acl        = bch2_set_acl,
1203 #endif
1204 };
1205
1206 static const struct inode_operations bch_dir_inode_operations = {
1207         .lookup         = bch2_lookup,
1208         .create         = bch2_create,
1209         .link           = bch2_link,
1210         .unlink         = bch2_unlink,
1211         .symlink        = bch2_symlink,
1212         .mkdir          = bch2_mkdir,
1213         .rmdir          = bch2_rmdir,
1214         .mknod          = bch2_mknod,
1215         .rename         = bch2_rename2,
1216         .getattr        = bch2_getattr,
1217         .setattr        = bch2_setattr,
1218         .tmpfile        = bch2_tmpfile,
1219         .listxattr      = bch2_xattr_list,
1220 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1221         .get_acl        = bch2_get_acl,
1222         .set_acl        = bch2_set_acl,
1223 #endif
1224 };
1225
1226 static const struct file_operations bch_dir_file_operations = {
1227         .llseek         = bch2_dir_llseek,
1228         .read           = generic_read_dir,
1229         .iterate        = bch2_vfs_readdir,
1230         .fsync          = bch2_fsync,
1231         .unlocked_ioctl = bch2_fs_file_ioctl,
1232 #ifdef CONFIG_COMPAT
1233         .compat_ioctl   = bch2_compat_fs_ioctl,
1234 #endif
1235 };
1236
1237 static const struct inode_operations bch_symlink_inode_operations = {
1238         .get_link       = page_get_link,
1239         .getattr        = bch2_getattr,
1240         .setattr        = bch2_setattr,
1241         .listxattr      = bch2_xattr_list,
1242 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1243         .get_acl        = bch2_get_acl,
1244         .set_acl        = bch2_set_acl,
1245 #endif
1246 };
1247
1248 static const struct inode_operations bch_special_inode_operations = {
1249         .getattr        = bch2_getattr,
1250         .setattr        = bch2_setattr,
1251         .listxattr      = bch2_xattr_list,
1252 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1253         .get_acl        = bch2_get_acl,
1254         .set_acl        = bch2_set_acl,
1255 #endif
1256 };
1257
1258 static const struct address_space_operations bch_address_space_operations = {
1259         .writepage      = bch2_writepage,
1260         .readpage       = bch2_readpage,
1261         .writepages     = bch2_writepages,
1262         .readpages      = bch2_readpages,
1263         .set_page_dirty = bch2_set_page_dirty,
1264         .write_begin    = bch2_write_begin,
1265         .write_end      = bch2_write_end,
1266         .invalidatepage = bch2_invalidatepage,
1267         .releasepage    = bch2_releasepage,
1268         .direct_IO      = bch2_direct_IO,
1269 #ifdef CONFIG_MIGRATION
1270         .migratepage    = bch2_migrate_page,
1271 #endif
1272         .error_remove_page = generic_error_remove_page,
1273 };
1274
1275 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1276                 u64 ino, u32 generation)
1277 {
1278         struct bch_fs *c = sb->s_fs_info;
1279         struct inode *vinode;
1280
1281         if (ino < BCACHEFS_ROOT_INO)
1282                 return ERR_PTR(-ESTALE);
1283
1284         vinode = bch2_vfs_inode_get(c, ino);
1285         if (IS_ERR(vinode))
1286                 return ERR_CAST(vinode);
1287         if (generation && vinode->i_generation != generation) {
1288                 /* we didn't find the right inode.. */
1289                 iput(vinode);
1290                 return ERR_PTR(-ESTALE);
1291         }
1292         return vinode;
1293 }
1294
1295 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1296                 int fh_len, int fh_type)
1297 {
1298         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1299                                     bch2_nfs_get_inode);
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         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1306                                     bch2_nfs_get_inode);
1307 }
1308
1309 static const struct export_operations bch_export_ops = {
1310         .fh_to_dentry   = bch2_fh_to_dentry,
1311         .fh_to_parent   = bch2_fh_to_parent,
1312         //.get_parent   = bch2_get_parent,
1313 };
1314
1315 static void bch2_vfs_inode_init(struct bch_fs *c,
1316                                 struct bch_inode_info *inode,
1317                                 struct bch_inode_unpacked *bi)
1318 {
1319         bch2_inode_update_after_write(c, inode, bi, ~0);
1320
1321         inode->v.i_blocks       = bi->bi_sectors;
1322         inode->v.i_ino          = bi->bi_inum;
1323         inode->v.i_rdev         = bi->bi_dev;
1324         inode->v.i_generation   = bi->bi_generation;
1325         inode->v.i_size         = bi->bi_size;
1326
1327         inode->ei_journal_seq   = 0;
1328         inode->ei_quota_reserved = 0;
1329         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1330         inode->ei_qid           = bch_qid(bi);
1331
1332         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1333
1334         switch (inode->v.i_mode & S_IFMT) {
1335         case S_IFREG:
1336                 inode->v.i_op   = &bch_file_inode_operations;
1337                 inode->v.i_fop  = &bch_file_operations;
1338                 break;
1339         case S_IFDIR:
1340                 inode->v.i_op   = &bch_dir_inode_operations;
1341                 inode->v.i_fop  = &bch_dir_file_operations;
1342                 break;
1343         case S_IFLNK:
1344                 inode_nohighmem(&inode->v);
1345                 inode->v.i_op   = &bch_symlink_inode_operations;
1346                 break;
1347         default:
1348                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1349                 inode->v.i_op   = &bch_special_inode_operations;
1350                 break;
1351         }
1352 }
1353
1354 static struct inode *bch2_alloc_inode(struct super_block *sb)
1355 {
1356         struct bch_inode_info *inode;
1357
1358         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1359         if (!inode)
1360                 return NULL;
1361
1362         inode_init_once(&inode->v);
1363         mutex_init(&inode->ei_update_lock);
1364         mutex_init(&inode->ei_quota_lock);
1365         inode->ei_inode_update = NULL;
1366         inode->ei_journal_seq = 0;
1367
1368         return &inode->v;
1369 }
1370
1371 static void bch2_i_callback(struct rcu_head *head)
1372 {
1373         struct inode *vinode = container_of(head, struct inode, i_rcu);
1374         struct bch_inode_info *inode = to_bch_ei(vinode);
1375
1376         kmem_cache_free(bch2_inode_cache, inode);
1377 }
1378
1379 static void bch2_destroy_inode(struct inode *vinode)
1380 {
1381         call_rcu(&vinode->i_rcu, bch2_i_callback);
1382 }
1383
1384 static int inode_update_times_fn(struct bch_inode_info *inode,
1385                                  struct bch_inode_unpacked *bi,
1386                                  void *p)
1387 {
1388         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1389
1390         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1391         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1392         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1393
1394         return 0;
1395 }
1396
1397 static int bch2_vfs_write_inode(struct inode *vinode,
1398                                 struct writeback_control *wbc)
1399 {
1400         struct bch_fs *c = vinode->i_sb->s_fs_info;
1401         struct bch_inode_info *inode = to_bch_ei(vinode);
1402         int ret;
1403
1404         mutex_lock(&inode->ei_update_lock);
1405         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1406                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1407         mutex_unlock(&inode->ei_update_lock);
1408
1409         if (c->opts.journal_flush_disabled)
1410                 return ret;
1411
1412         if (!ret && wbc->sync_mode == WB_SYNC_ALL)
1413                 ret = bch2_journal_flush_seq(&c->journal, inode->ei_journal_seq);
1414
1415         return ret;
1416 }
1417
1418 static void bch2_evict_inode(struct inode *vinode)
1419 {
1420         struct bch_fs *c = vinode->i_sb->s_fs_info;
1421         struct bch_inode_info *inode = to_bch_ei(vinode);
1422
1423         truncate_inode_pages_final(&inode->v.i_data);
1424
1425         clear_inode(&inode->v);
1426
1427         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1428
1429         if (inode->ei_inode_update)
1430                 bch2_deferred_update_free(c, inode->ei_inode_update);
1431         inode->ei_inode_update = NULL;
1432
1433         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1434                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1435                                 KEY_TYPE_QUOTA_WARN);
1436                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1437                                 KEY_TYPE_QUOTA_WARN);
1438                 bch2_inode_rm(c, inode->v.i_ino);
1439         }
1440 }
1441
1442 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1443 {
1444         struct super_block *sb = dentry->d_sb;
1445         struct bch_fs *c = sb->s_fs_info;
1446         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1447         unsigned shift = sb->s_blocksize_bits - 9;
1448         u64 fsid;
1449
1450         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1451         buf->f_bsize    = sb->s_blocksize;
1452         buf->f_blocks   = usage.capacity >> shift;
1453         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1454         buf->f_bavail   = buf->f_bfree;
1455         buf->f_files    = usage.nr_inodes;
1456         buf->f_ffree    = U64_MAX;
1457
1458         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1459                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1460         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1461         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1462         buf->f_namelen  = BCH_NAME_MAX;
1463
1464         return 0;
1465 }
1466
1467 static int bch2_sync_fs(struct super_block *sb, int wait)
1468 {
1469         struct bch_fs *c = sb->s_fs_info;
1470
1471         if (!wait) {
1472                 bch2_journal_flush_async(&c->journal, NULL);
1473                 return 0;
1474         }
1475
1476         return bch2_journal_flush(&c->journal);
1477 }
1478
1479 static struct bch_fs *bch2_path_to_fs(const char *dev)
1480 {
1481         struct bch_fs *c;
1482         struct block_device *bdev = lookup_bdev(dev);
1483
1484         if (IS_ERR(bdev))
1485                 return ERR_CAST(bdev);
1486
1487         c = bch2_bdev_to_fs(bdev);
1488         bdput(bdev);
1489         return c ?: ERR_PTR(-ENOENT);
1490 }
1491
1492 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1493                                                unsigned nr_devs, struct bch_opts opts)
1494 {
1495         struct bch_fs *c, *c1, *c2;
1496         size_t i;
1497
1498         if (!nr_devs)
1499                 return ERR_PTR(-EINVAL);
1500
1501         c = bch2_fs_open(devs, nr_devs, opts);
1502
1503         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1504                 /*
1505                  * Already open?
1506                  * Look up each block device, make sure they all belong to a
1507                  * filesystem and they all belong to the _same_ filesystem
1508                  */
1509
1510                 c1 = bch2_path_to_fs(devs[0]);
1511                 if (!c1)
1512                         return c;
1513
1514                 for (i = 1; i < nr_devs; i++) {
1515                         c2 = bch2_path_to_fs(devs[i]);
1516                         if (!IS_ERR(c2))
1517                                 closure_put(&c2->cl);
1518
1519                         if (c1 != c2) {
1520                                 closure_put(&c1->cl);
1521                                 return c;
1522                         }
1523                 }
1524
1525                 c = c1;
1526         }
1527
1528         if (IS_ERR(c))
1529                 return c;
1530
1531         mutex_lock(&c->state_lock);
1532
1533         if (!bch2_fs_running(c)) {
1534                 mutex_unlock(&c->state_lock);
1535                 closure_put(&c->cl);
1536                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1537                 return ERR_PTR(-EINVAL);
1538         }
1539
1540         mutex_unlock(&c->state_lock);
1541
1542         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1543         return c;
1544 }
1545
1546 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1547                                              struct bch_opts opts)
1548 {
1549         char *dev_name = NULL, **devs = NULL, *s;
1550         struct bch_fs *c = ERR_PTR(-ENOMEM);
1551         size_t i, nr_devs = 0;
1552
1553         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1554         if (!dev_name)
1555                 goto err;
1556
1557         for (s = dev_name; s; s = strchr(s + 1, ':'))
1558                 nr_devs++;
1559
1560         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1561         if (!devs)
1562                 goto err;
1563
1564         for (i = 0, s = dev_name;
1565              s;
1566              (s = strchr(s, ':')) && (*s++ = '\0'))
1567                 devs[i++] = s;
1568
1569         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1570 err:
1571         kfree(devs);
1572         kfree(dev_name);
1573         return c;
1574 }
1575
1576 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1577 {
1578         struct bch_fs *c = sb->s_fs_info;
1579         struct bch_opts opts = bch2_opts_empty();
1580         int ret;
1581
1582         opt_set(opts, read_only, (*flags & MS_RDONLY) != 0);
1583
1584         ret = bch2_parse_mount_opts(&opts, data);
1585         if (ret)
1586                 return ret;
1587
1588         if (opts.read_only != c->opts.read_only) {
1589                 const char *err = NULL;
1590
1591                 mutex_lock(&c->state_lock);
1592
1593                 if (opts.read_only) {
1594                         bch2_fs_read_only(c);
1595
1596                         sb->s_flags |= MS_RDONLY;
1597                 } else {
1598                         err = bch2_fs_read_write(c);
1599                         if (err) {
1600                                 bch_err(c, "error going rw: %s", err);
1601                                 return -EINVAL;
1602                         }
1603
1604                         sb->s_flags &= ~MS_RDONLY;
1605                 }
1606
1607                 c->opts.read_only = opts.read_only;
1608
1609                 mutex_unlock(&c->state_lock);
1610         }
1611
1612         if (opts.errors >= 0)
1613                 c->opts.errors = opts.errors;
1614
1615         return ret;
1616 }
1617
1618 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1619 {
1620         struct bch_fs *c = root->d_sb->s_fs_info;
1621         enum bch_opt_id i;
1622         char buf[512];
1623
1624         for (i = 0; i < bch2_opts_nr; i++) {
1625                 const struct bch_option *opt = &bch2_opt_table[i];
1626                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1627
1628                 if (!(opt->mode & OPT_MOUNT))
1629                         continue;
1630
1631                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1632                         continue;
1633
1634                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1635                                  OPT_SHOW_MOUNT_STYLE);
1636                 seq_putc(seq, ',');
1637                 seq_puts(seq, buf);
1638         }
1639
1640         return 0;
1641
1642 }
1643
1644 static const struct super_operations bch_super_operations = {
1645         .alloc_inode    = bch2_alloc_inode,
1646         .destroy_inode  = bch2_destroy_inode,
1647         .write_inode    = bch2_vfs_write_inode,
1648         .evict_inode    = bch2_evict_inode,
1649         .sync_fs        = bch2_sync_fs,
1650         .statfs         = bch2_statfs,
1651         .show_options   = bch2_show_options,
1652         .remount_fs     = bch2_remount,
1653 #if 0
1654         .put_super      = bch2_put_super,
1655         .freeze_fs      = bch2_freeze,
1656         .unfreeze_fs    = bch2_unfreeze,
1657 #endif
1658 };
1659
1660 static int bch2_test_super(struct super_block *s, void *data)
1661 {
1662         return s->s_fs_info == data;
1663 }
1664
1665 static int bch2_set_super(struct super_block *s, void *data)
1666 {
1667         s->s_fs_info = data;
1668         return 0;
1669 }
1670
1671 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1672                                  int flags, const char *dev_name, void *data)
1673 {
1674         struct bch_fs *c;
1675         struct bch_dev *ca;
1676         struct super_block *sb;
1677         struct inode *vinode;
1678         struct bch_opts opts = bch2_opts_empty();
1679         unsigned i;
1680         int ret;
1681
1682         opt_set(opts, read_only, (flags & MS_RDONLY) != 0);
1683
1684         ret = bch2_parse_mount_opts(&opts, data);
1685         if (ret)
1686                 return ERR_PTR(ret);
1687
1688         c = bch2_open_as_blockdevs(dev_name, opts);
1689         if (IS_ERR(c))
1690                 return ERR_CAST(c);
1691
1692         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|MS_NOSEC, c);
1693         if (IS_ERR(sb)) {
1694                 closure_put(&c->cl);
1695                 return ERR_CAST(sb);
1696         }
1697
1698         BUG_ON(sb->s_fs_info != c);
1699
1700         if (sb->s_root) {
1701                 closure_put(&c->cl);
1702
1703                 if ((flags ^ sb->s_flags) & MS_RDONLY) {
1704                         ret = -EBUSY;
1705                         goto err_put_super;
1706                 }
1707                 goto out;
1708         }
1709
1710         /* XXX: blocksize */
1711         sb->s_blocksize         = PAGE_SIZE;
1712         sb->s_blocksize_bits    = PAGE_SHIFT;
1713         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1714         sb->s_op                = &bch_super_operations;
1715         sb->s_export_op         = &bch_export_ops;
1716 #ifdef CONFIG_BCACHEFS_QUOTA
1717         sb->s_qcop              = &bch2_quotactl_operations;
1718         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1719 #endif
1720         sb->s_xattr             = bch2_xattr_handlers;
1721         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1722         sb->s_time_gran         = c->sb.time_precision;
1723         c->vfs_sb               = sb;
1724         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1725
1726         ret = super_setup_bdi(sb);
1727         if (ret)
1728                 goto err_put_super;
1729
1730         sb->s_bdi->congested_fn         = bch2_congested;
1731         sb->s_bdi->congested_data       = c;
1732         sb->s_bdi->ra_pages             = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1733
1734         for_each_online_member(ca, c, i) {
1735                 struct block_device *bdev = ca->disk_sb.bdev;
1736
1737                 /* XXX: create an anonymous device for multi device filesystems */
1738                 sb->s_bdev      = bdev;
1739                 sb->s_dev       = bdev->bd_dev;
1740                 percpu_ref_put(&ca->io_ref);
1741                 break;
1742         }
1743
1744 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1745         if (c->opts.acl)
1746                 sb->s_flags     |= MS_POSIXACL;
1747 #endif
1748
1749         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1750         if (IS_ERR(vinode)) {
1751                 ret = PTR_ERR(vinode);
1752                 goto err_put_super;
1753         }
1754
1755         sb->s_root = d_make_root(vinode);
1756         if (!sb->s_root) {
1757                 ret = -ENOMEM;
1758                 goto err_put_super;
1759         }
1760
1761         sb->s_flags |= MS_ACTIVE;
1762 out:
1763         return dget(sb->s_root);
1764
1765 err_put_super:
1766         deactivate_locked_super(sb);
1767         return ERR_PTR(ret);
1768 }
1769
1770 static void bch2_kill_sb(struct super_block *sb)
1771 {
1772         struct bch_fs *c = sb->s_fs_info;
1773
1774         generic_shutdown_super(sb);
1775
1776         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1777                 bch2_fs_stop(c);
1778         else
1779                 closure_put(&c->cl);
1780 }
1781
1782 static struct file_system_type bcache_fs_type = {
1783         .owner          = THIS_MODULE,
1784         .name           = "bcachefs",
1785         .mount          = bch2_mount,
1786         .kill_sb        = bch2_kill_sb,
1787         .fs_flags       = FS_REQUIRES_DEV,
1788 };
1789
1790 MODULE_ALIAS_FS("bcachefs");
1791
1792 void bch2_vfs_exit(void)
1793 {
1794         unregister_filesystem(&bcache_fs_type);
1795         if (bch2_inode_cache)
1796                 kmem_cache_destroy(bch2_inode_cache);
1797 }
1798
1799 int __init bch2_vfs_init(void)
1800 {
1801         int ret = -ENOMEM;
1802
1803         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1804         if (!bch2_inode_cache)
1805                 goto err;
1806
1807         ret = register_filesystem(&bcache_fs_type);
1808         if (ret)
1809                 goto err;
1810
1811         return 0;
1812 err:
1813         bch2_vfs_exit();
1814         return ret;
1815 }
1816
1817 #endif /* NO_BCACHEFS_FS */