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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "chardev.h"
   9 #include "dirent.h"
  10 #include "extents.h"
  11 #include "fs.h"
  12 #include "fs-io.h"
  13 #include "fs-ioctl.h"
  14 #include "fsck.h"
  15 #include "inode.h"
  16 #include "io.h"
  17 #include "journal.h"
  18 #include "keylist.h"
  19 #include "quota.h"
  20 #include "super.h"
  21 #include "xattr.h"
  22
  23 #include <linux/aio.h>
  24 #include <linux/backing-dev.h>
  25 #include <linux/exportfs.h>
  26 #include <linux/module.h>
  27 #include <linux/posix_acl.h>
  28 #include <linux/random.h>
  29 #include <linux/statfs.h>
  30 #include <linux/xattr.h>
  31
  32 static struct kmem_cache *bch2_inode_cache;
  33
  34 static void bch2_vfs_inode_init(struct bch_fs *,
  35                                 struct bch_inode_info *,
  36                                 struct bch_inode_unpacked *);
  37
  38 static void journal_seq_copy(struct bch_inode_info *dst,
  39                              u64 journal_seq)
  40 {
  41         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  42
  43         do {
  44                 old = v;
  45
  46                 if (old >= journal_seq)
  47                         break;
  48         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  49 }
  50
  51 /*
  52  * I_SIZE_DIRTY requires special handling:
  53  *
  54  * To the recovery code, the flag means that there is stale data past i_size
  55  * that needs to be deleted; it's used for implementing atomic appends and
  56  * truncates.
  57  *
  58  * On append, we set I_SIZE_DIRTY before doing the write, then after the write
  59  * we clear I_SIZE_DIRTY atomically with updating i_size to the new larger size
  60  * that exposes the data we just wrote.
  61  *
  62  * On truncate, it's the reverse: We set I_SIZE_DIRTY atomically with setting
  63  * i_size to the new smaller size, then we delete the data that we just made
  64  * invisible, and then we clear I_SIZE_DIRTY.
  65  *
  66  * Because there can be multiple appends in flight at a time, we need a refcount
  67  * (i_size_dirty_count) instead of manipulating the flag directly. Nonzero
  68  * refcount means I_SIZE_DIRTY is set, zero means it's cleared.
  69  *
  70  * Because write_inode() can be called at any time, i_size_dirty_count means
  71  * something different to the runtime code - it means to write_inode() "don't
  72  * update i_size yet".
  73  *
  74  * We don't clear I_SIZE_DIRTY directly, we let write_inode() clear it when
  75  * i_size_dirty_count is zero - but the reverse is not true, I_SIZE_DIRTY must
  76  * be set explicitly.
  77  */
  78
  79 void bch2_inode_update_after_write(struct bch_fs *c,
  80                                    struct bch_inode_info *inode,
  81                                    struct bch_inode_unpacked *bi,
  82                                    unsigned fields)
  83 {
  84         set_nlink(&inode->v, bi->bi_flags & BCH_INODE_UNLINKED
  85                   ? 0
  86                   : bi->bi_nlink + nlink_bias(inode->v.i_mode));
  87         i_uid_write(&inode->v, bi->bi_uid);
  88         i_gid_write(&inode->v, bi->bi_gid);
  89         inode->v.i_mode = bi->bi_mode;
  90
  91         if (fields & ATTR_ATIME)
  92                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
  93         if (fields & ATTR_MTIME)
  94                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
  95         if (fields & ATTR_CTIME)
  96                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
  97
  98         inode->ei_inode         = *bi;
  99
 100         bch2_inode_flags_to_vfs(inode);
 101 }
 102
 103 int __must_check bch2_write_inode_trans(struct btree_trans *trans,
 104                                 struct bch_inode_info *inode,
 105                                 struct bch_inode_unpacked *inode_u,
 106                                 inode_set_fn set,
 107                                 void *p)
 108 {
 109         struct bch_fs *c = trans->c;
 110         struct btree_iter *iter = NULL;
 111         struct bkey_inode_buf *inode_p;
 112         int ret;
 113
 114         lockdep_assert_held(&inode->ei_update_lock);
 115
 116         if (c->opts.new_inode_updates) {
 117         /* XXX: Don't do this with btree locks held */
 118         if (!inode->ei_inode_update)
 119                 inode->ei_inode_update =
 120                         bch2_deferred_update_alloc(c, BTREE_ID_INODES, 64);
 121         } else {
 122                 iter = bch2_trans_get_iter(trans, BTREE_ID_INODES,
 123                                            POS(inode->v.i_ino, 0),
 124                                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 125                 if (IS_ERR(iter))
 126                         return PTR_ERR(iter);
 127
 128                 /* The btree node lock is our lock on the inode: */
 129                 ret = bch2_btree_iter_traverse(iter);
 130                 if (ret)
 131                         return ret;
 132         }
 133
 134         *inode_u = inode->ei_inode;
 135
 136         if (set) {
 137                 ret = set(inode, inode_u, p);
 138                 if (ret)
 139                         return ret;
 140         }
 141
 142         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 143         if (IS_ERR(inode_p))
 144                 return PTR_ERR(inode_p);
 145
 146         bch2_inode_pack(inode_p, inode_u);
 147
 148         if (!inode->ei_inode_update)
 149                 bch2_trans_update(trans,
 150                         BTREE_INSERT_ENTRY(iter, &inode_p->inode.k_i));
 151         else
 152                 bch2_trans_update(trans,
 153                         BTREE_INSERT_DEFERRED(inode->ei_inode_update,
 154                                               &inode_p->inode.k_i));
 155
 156         return 0;
 157 }
 158
 159 int __must_check bch2_write_inode(struct bch_fs *c,
 160                                   struct bch_inode_info *inode,
 161                                   inode_set_fn set,
 162                                   void *p, unsigned fields)
 163 {
 164         struct btree_trans trans;
 165         struct bch_inode_unpacked inode_u;
 166         int ret;
 167
 168         bch2_trans_init(&trans, c, 0, 0);
 169 retry:
 170         bch2_trans_begin(&trans);
 171
 172         ret = bch2_write_inode_trans(&trans, inode, &inode_u, set, p) ?:
 173                 bch2_trans_commit(&trans, NULL,
 174                                   &inode->ei_journal_seq,
 175                                   BTREE_INSERT_ATOMIC|
 176                                   BTREE_INSERT_NOUNLOCK|
 177                                   BTREE_INSERT_NOFAIL);
 178         if (ret == -EINTR)
 179                 goto retry;
 180
 181         /*
 182          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 183          * this is important for inode updates via bchfs_write_index_update
 184          */
 185         if (!ret)
 186                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 187
 188         bch2_trans_exit(&trans);
 189         return ret < 0 ? ret : 0;
 190 }
 191
 192 int bch2_fs_quota_transfer(struct bch_fs *c,
 193                            struct bch_inode_info *inode,
 194                            struct bch_qid new_qid,
 195                            unsigned qtypes,
 196                            enum quota_acct_mode mode)
 197 {
 198         unsigned i;
 199         int ret;
 200
 201         qtypes &= enabled_qtypes(c);
 202
 203         for (i = 0; i < QTYP_NR; i++)
 204                 if (new_qid.q[i] == inode->ei_qid.q[i])
 205                         qtypes &= ~(1U << i);
 206
 207         if (!qtypes)
 208                 return 0;
 209
 210         mutex_lock(&inode->ei_quota_lock);
 211
 212         ret = bch2_quota_transfer(c, qtypes, new_qid,
 213                                   inode->ei_qid,
 214                                   inode->v.i_blocks +
 215                                   inode->ei_quota_reserved,
 216                                   mode);
 217         if (!ret)
 218                 for (i = 0; i < QTYP_NR; i++)
 219                         if (qtypes & (1 << i))
 220                                 inode->ei_qid.q[i] = new_qid.q[i];
 221
 222         mutex_unlock(&inode->ei_quota_lock);
 223
 224         return ret;
 225 }
 226
 227 int bch2_reinherit_attrs_fn(struct bch_inode_info *inode,
 228                             struct bch_inode_unpacked *bi,
 229                             void *p)
 230 {
 231         struct bch_inode_info *dir = p;
 232         u64 src, dst;
 233         unsigned id;
 234         int ret = 1;
 235
 236         for (id = 0; id < Inode_opt_nr; id++) {
 237                 if (bi->bi_fields_set & (1 << id))
 238                         continue;
 239
 240                 src = bch2_inode_opt_get(&dir->ei_inode, id);
 241                 dst = bch2_inode_opt_get(bi, id);
 242
 243                 if (src == dst)
 244                         continue;
 245
 246                 bch2_inode_opt_set(bi, id, src);
 247                 ret = 0;
 248         }
 249
 250         return ret;
 251 }
 252
 253 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 254 {
 255         struct bch_inode_unpacked inode_u;
 256         struct bch_inode_info *inode;
 257         int ret;
 258
 259         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 260         if (unlikely(!inode))
 261                 return ERR_PTR(-ENOMEM);
 262         if (!(inode->v.i_state & I_NEW))
 263                 return &inode->v;
 264
 265         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 266         if (ret) {
 267                 iget_failed(&inode->v);
 268                 return ERR_PTR(ret);
 269         }
 270
 271         bch2_vfs_inode_init(c, inode, &inode_u);
 272
 273         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 274
 275         unlock_new_inode(&inode->v);
 276
 277         return &inode->v;
 278 }
 279
 280 static void bch2_inode_init_owner(struct bch_inode_unpacked *inode_u,
 281                                   const struct inode *dir, umode_t mode)
 282 {
 283         kuid_t uid = current_fsuid();
 284         kgid_t gid;
 285
 286         if (dir && dir->i_mode & S_ISGID) {
 287                 gid = dir->i_gid;
 288                 if (S_ISDIR(mode))
 289                         mode |= S_ISGID;
 290         } else
 291                 gid = current_fsgid();
 292
 293         inode_u->bi_uid         = from_kuid(dir->i_sb->s_user_ns, uid);
 294         inode_u->bi_gid         = from_kgid(dir->i_sb->s_user_ns, gid);
 295         inode_u->bi_mode        = mode;
 296 }
 297
 298 static int inode_update_for_create_fn(struct bch_inode_info *inode,
 299                                       struct bch_inode_unpacked *bi,
 300                                       void *p)
 301 {
 302         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 303         struct bch_inode_unpacked *new_inode = p;
 304
 305         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 306
 307         if (S_ISDIR(new_inode->bi_mode))
 308                 bi->bi_nlink++;
 309
 310         return 0;
 311 }
 312
 313 static struct bch_inode_info *
 314 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 315               umode_t mode, dev_t rdev, bool tmpfile)
 316 {
 317         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 318         struct btree_trans trans;
 319         struct bch_inode_unpacked dir_u;
 320         struct bch_inode_info *inode, *old;
 321         struct bch_inode_unpacked inode_u;
 322         struct bch_hash_info hash_info;
 323         struct posix_acl *default_acl = NULL, *acl = NULL;
 324         u64 journal_seq = 0;
 325         int ret;
 326
 327         bch2_inode_init(c, &inode_u, 0, 0, 0, rdev, &dir->ei_inode);
 328         bch2_inode_init_owner(&inode_u, &dir->v, mode);
 329
 330         hash_info = bch2_hash_info_init(c, &inode_u);
 331
 332         if (tmpfile)
 333                 inode_u.bi_flags |= BCH_INODE_UNLINKED;
 334
 335         ret = bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 336                               KEY_TYPE_QUOTA_PREALLOC);
 337         if (ret)
 338                 return ERR_PTR(ret);
 339
 340 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 341         ret = posix_acl_create(&dir->v, &inode_u.bi_mode, &default_acl, &acl);
 342         if (ret)
 343                 goto err;
 344 #endif
 345
 346         /*
 347          * preallocate vfs inode before btree transaction, so that nothing can
 348          * fail after the transaction succeeds:
 349          */
 350         inode = to_bch_ei(new_inode(c->vfs_sb));
 351         if (unlikely(!inode)) {
 352                 ret = -ENOMEM;
 353                 goto err;
 354         }
 355
 356         if (!tmpfile)
 357                 mutex_lock(&dir->ei_update_lock);
 358
 359         bch2_trans_init(&trans, c, 8, 1024);
 360 retry:
 361         bch2_trans_begin(&trans);
 362
 363         ret   = __bch2_inode_create(&trans, &inode_u,
 364                                     BLOCKDEV_INODE_MAX, 0,
 365                                     &c->unused_inode_hint) ?:
 366                 (default_acl
 367                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 368                                       default_acl, ACL_TYPE_DEFAULT)
 369                  : 0) ?:
 370                 (acl
 371                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 372                                       acl, ACL_TYPE_ACCESS)
 373                  : 0) ?:
 374                 (!tmpfile
 375                  ? __bch2_dirent_create(&trans, dir->v.i_ino,
 376                                         &dir->ei_str_hash,
 377                                         mode_to_type(mode),
 378                                         &dentry->d_name,
 379                                         inode_u.bi_inum,
 380                                         BCH_HASH_SET_MUST_CREATE)
 381                 : 0) ?:
 382                 (!tmpfile
 383                  ? bch2_write_inode_trans(&trans, dir, &dir_u,
 384                                           inode_update_for_create_fn,
 385                                           &inode_u)
 386                  : 0) ?:
 387                 bch2_trans_commit(&trans, NULL,
 388                                   &journal_seq,
 389                                   BTREE_INSERT_ATOMIC|
 390                                   BTREE_INSERT_NOUNLOCK);
 391         if (ret == -EINTR)
 392                 goto retry;
 393         if (unlikely(ret))
 394                 goto err_trans;
 395
 396         if (!tmpfile) {
 397                 bch2_inode_update_after_write(c, dir, &dir_u,
 398                                               ATTR_MTIME|ATTR_CTIME);
 399                 journal_seq_copy(dir, journal_seq);
 400                 mutex_unlock(&dir->ei_update_lock);
 401         }
 402
 403         bch2_vfs_inode_init(c, inode, &inode_u);
 404         journal_seq_copy(inode, journal_seq);
 405
 406         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 407         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 408
 409         /*
 410          * we must insert the new inode into the inode cache before calling
 411          * bch2_trans_exit() and dropping locks, else we could race with another
 412          * thread pulling the inode in and modifying it:
 413          */
 414
 415         old = to_bch_ei(insert_inode_locked2(&inode->v));
 416         if (unlikely(old)) {
 417                 /*
 418                  * We raced, another process pulled the new inode into cache
 419                  * before us:
 420                  */
 421                 old->ei_journal_seq = inode->ei_journal_seq;
 422                 make_bad_inode(&inode->v);
 423                 iput(&inode->v);
 424
 425                 inode = old;
 426         } else {
 427                 /*
 428                  * we really don't want insert_inode_locked2() to be setting
 429                  * I_NEW...
 430                  */
 431                 unlock_new_inode(&inode->v);
 432         }
 433
 434         bch2_trans_exit(&trans);
 435 out:
 436         posix_acl_release(default_acl);
 437         posix_acl_release(acl);
 438         return inode;
 439 err_trans:
 440         if (!tmpfile)
 441                 mutex_unlock(&dir->ei_update_lock);
 442
 443         bch2_trans_exit(&trans);
 444         make_bad_inode(&inode->v);
 445         iput(&inode->v);
 446 err:
 447         bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1, KEY_TYPE_QUOTA_WARN);
 448         inode = ERR_PTR(ret);
 449         goto out;
 450 }
 451
 452 /* methods */
 453
 454 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 455                                   unsigned int flags)
 456 {
 457         struct bch_fs *c = vdir->i_sb->s_fs_info;
 458         struct bch_inode_info *dir = to_bch_ei(vdir);
 459         struct inode *vinode = NULL;
 460         u64 inum;
 461
 462         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 463                                   &dir->ei_str_hash,
 464                                   &dentry->d_name);
 465
 466         if (inum)
 467                 vinode = bch2_vfs_inode_get(c, inum);
 468
 469         return d_splice_alias(vinode, dentry);
 470 }
 471
 472 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 473                        umode_t mode, bool excl)
 474 {
 475         struct bch_inode_info *inode =
 476                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0, false);
 477
 478         if (IS_ERR(inode))
 479                 return PTR_ERR(inode);
 480
 481         d_instantiate(dentry, &inode->v);
 482         return 0;
 483 }
 484
 485 static int inode_update_for_link_fn(struct bch_inode_info *inode,
 486                                     struct bch_inode_unpacked *bi,
 487                                     void *p)
 488 {
 489         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 490
 491         bi->bi_ctime = bch2_current_time(c);
 492
 493         if (bi->bi_flags & BCH_INODE_UNLINKED)
 494                 bi->bi_flags &= ~BCH_INODE_UNLINKED;
 495         else
 496                 bi->bi_nlink++;
 497
 498         return 0;
 499 }
 500
 501 static int __bch2_link(struct bch_fs *c,
 502                        struct bch_inode_info *inode,
 503                        struct bch_inode_info *dir,
 504                        struct dentry *dentry)
 505 {
 506         struct btree_trans trans;
 507         struct bch_inode_unpacked inode_u;
 508         int ret;
 509
 510         mutex_lock(&inode->ei_update_lock);
 511         bch2_trans_init(&trans, c, 4, 1024);
 512 retry:
 513         bch2_trans_begin(&trans);
 514
 515         ret   = __bch2_dirent_create(&trans, dir->v.i_ino,
 516                                      &dir->ei_str_hash,
 517                                      mode_to_type(inode->v.i_mode),
 518                                      &dentry->d_name,
 519                                      inode->v.i_ino,
 520                                      BCH_HASH_SET_MUST_CREATE) ?:
 521                 bch2_write_inode_trans(&trans, inode, &inode_u,
 522                                        inode_update_for_link_fn,
 523                                        NULL) ?:
 524                 bch2_trans_commit(&trans, NULL,
 525                                   &inode->ei_journal_seq,
 526                                   BTREE_INSERT_ATOMIC|
 527                                   BTREE_INSERT_NOUNLOCK);
 528
 529         if (ret == -EINTR)
 530                 goto retry;
 531
 532         if (likely(!ret))
 533                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 534
 535         bch2_trans_exit(&trans);
 536         mutex_unlock(&inode->ei_update_lock);
 537         return ret;
 538 }
 539
 540 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 541                      struct dentry *dentry)
 542 {
 543         struct bch_fs *c = vdir->i_sb->s_fs_info;
 544         struct bch_inode_info *dir = to_bch_ei(vdir);
 545         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 546         int ret;
 547
 548         lockdep_assert_held(&inode->v.i_rwsem);
 549
 550         ret = __bch2_link(c, inode, dir, dentry);
 551         if (unlikely(ret))
 552                 return ret;
 553
 554         ihold(&inode->v);
 555         d_instantiate(dentry, &inode->v);
 556         return 0;
 557 }
 558
 559 static int inode_update_dir_for_unlink_fn(struct bch_inode_info *inode,
 560                                           struct bch_inode_unpacked *bi,
 561                                           void *p)
 562 {
 563         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 564         struct bch_inode_info *unlink_inode = p;
 565
 566         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 567
 568         bi->bi_nlink -= S_ISDIR(unlink_inode->v.i_mode);
 569
 570         return 0;
 571 }
 572
 573 static int inode_update_for_unlink_fn(struct bch_inode_info *inode,
 574                                       struct bch_inode_unpacked *bi,
 575                                       void *p)
 576 {
 577         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 578
 579         bi->bi_ctime = bch2_current_time(c);
 580         if (bi->bi_nlink)
 581                 bi->bi_nlink--;
 582         else
 583                 bi->bi_flags |= BCH_INODE_UNLINKED;
 584
 585         return 0;
 586 }
 587
 588 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 589 {
 590         struct bch_fs *c = vdir->i_sb->s_fs_info;
 591         struct bch_inode_info *dir = to_bch_ei(vdir);
 592         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 593         struct bch_inode_unpacked dir_u, inode_u;
 594         struct btree_trans trans;
 595         int ret;
 596
 597         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 598         bch2_trans_init(&trans, c, 4, 1024);
 599 retry:
 600         bch2_trans_begin(&trans);
 601
 602         ret   = __bch2_dirent_delete(&trans, dir->v.i_ino,
 603                                      &dir->ei_str_hash,
 604                                      &dentry->d_name) ?:
 605                 bch2_write_inode_trans(&trans, dir, &dir_u,
 606                                        inode_update_dir_for_unlink_fn,
 607                                        inode) ?:
 608                 bch2_write_inode_trans(&trans, inode, &inode_u,
 609                                        inode_update_for_unlink_fn,
 610                                        NULL) ?:
 611                 bch2_trans_commit(&trans, NULL,
 612                                   &dir->ei_journal_seq,
 613                                   BTREE_INSERT_ATOMIC|
 614                                   BTREE_INSERT_NOUNLOCK|
 615                                   BTREE_INSERT_NOFAIL);
 616         if (ret == -EINTR)
 617                 goto retry;
 618         if (ret)
 619                 goto err;
 620
 621         if (dir->ei_journal_seq > inode->ei_journal_seq)
 622                 inode->ei_journal_seq = dir->ei_journal_seq;
 623
 624         bch2_inode_update_after_write(c, dir, &dir_u,
 625                                       ATTR_MTIME|ATTR_CTIME);
 626         bch2_inode_update_after_write(c, inode, &inode_u,
 627                                       ATTR_MTIME);
 628 err:
 629         bch2_trans_exit(&trans);
 630         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 631
 632         return ret;
 633 }
 634
 635 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 636                         const char *symname)
 637 {
 638         struct bch_fs *c = vdir->i_sb->s_fs_info;
 639         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 640         int ret;
 641
 642         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 643         if (unlikely(IS_ERR(inode)))
 644                 return PTR_ERR(inode);
 645
 646         inode_lock(&inode->v);
 647         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 648         inode_unlock(&inode->v);
 649
 650         if (unlikely(ret))
 651                 goto err;
 652
 653         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 654         if (unlikely(ret))
 655                 goto err;
 656
 657         journal_seq_copy(dir, inode->ei_journal_seq);
 658
 659         ret = __bch2_link(c, inode, dir, dentry);
 660         if (unlikely(ret))
 661                 goto err;
 662
 663         d_instantiate(dentry, &inode->v);
 664         return 0;
 665 err:
 666         iput(&inode->v);
 667         return ret;
 668 }
 669
 670 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 671 {
 672         struct bch_inode_info *inode =
 673                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFDIR, 0, false);
 674
 675         if (IS_ERR(inode))
 676                 return PTR_ERR(inode);
 677
 678         d_instantiate(dentry, &inode->v);
 679         return 0;
 680 }
 681
 682 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 683 {
 684         struct bch_fs *c = vdir->i_sb->s_fs_info;
 685
 686         if (bch2_empty_dir(c, dentry->d_inode->i_ino))
 687                 return -ENOTEMPTY;
 688
 689         return bch2_unlink(vdir, dentry);
 690 }
 691
 692 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 693                       umode_t mode, dev_t rdev)
 694 {
 695         struct bch_inode_info *inode =
 696                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 697
 698         if (IS_ERR(inode))
 699                 return PTR_ERR(inode);
 700
 701         d_instantiate(dentry, &inode->v);
 702         return 0;
 703 }
 704
 705 struct rename_info {
 706         u64                     now;
 707         struct bch_inode_info   *src_dir;
 708         struct bch_inode_info   *dst_dir;
 709         struct bch_inode_info   *src_inode;
 710         struct bch_inode_info   *dst_inode;
 711         enum bch_rename_mode    mode;
 712 };
 713
 714 static int inode_update_for_rename_fn(struct bch_inode_info *inode,
 715                                       struct bch_inode_unpacked *bi,
 716                                       void *p)
 717 {
 718         struct rename_info *info = p;
 719         int ret;
 720
 721         if (inode == info->src_dir) {
 722                 bi->bi_nlink -= S_ISDIR(info->src_inode->v.i_mode);
 723                 bi->bi_nlink += info->dst_inode &&
 724                         S_ISDIR(info->dst_inode->v.i_mode) &&
 725                         info->mode == BCH_RENAME_EXCHANGE;
 726         }
 727
 728         if (inode == info->dst_dir) {
 729                 bi->bi_nlink += S_ISDIR(info->src_inode->v.i_mode);
 730                 bi->bi_nlink -= info->dst_inode &&
 731                         S_ISDIR(info->dst_inode->v.i_mode);
 732         }
 733
 734         if (inode == info->src_inode) {
 735                 ret = bch2_reinherit_attrs_fn(inode, bi, info->dst_dir);
 736
 737                 BUG_ON(!ret && S_ISDIR(info->src_inode->v.i_mode));
 738         }
 739
 740         if (inode == info->dst_inode &&
 741             info->mode == BCH_RENAME_EXCHANGE) {
 742                 ret = bch2_reinherit_attrs_fn(inode, bi, info->src_dir);
 743
 744                 BUG_ON(!ret && S_ISDIR(info->dst_inode->v.i_mode));
 745         }
 746
 747         if (inode == info->dst_inode &&
 748             info->mode == BCH_RENAME_OVERWRITE) {
 749                 BUG_ON(bi->bi_nlink &&
 750                        S_ISDIR(info->dst_inode->v.i_mode));
 751
 752                 if (bi->bi_nlink)
 753                         bi->bi_nlink--;
 754                 else
 755                         bi->bi_flags |= BCH_INODE_UNLINKED;
 756         }
 757
 758         if (inode == info->src_dir ||
 759             inode == info->dst_dir)
 760                 bi->bi_mtime = info->now;
 761         bi->bi_ctime = info->now;
 762
 763         return 0;
 764 }
 765
 766 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 767                         struct inode *dst_vdir, struct dentry *dst_dentry,
 768                         unsigned flags)
 769 {
 770         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 771         struct rename_info i = {
 772                 .src_dir        = to_bch_ei(src_vdir),
 773                 .dst_dir        = to_bch_ei(dst_vdir),
 774                 .src_inode      = to_bch_ei(src_dentry->d_inode),
 775                 .dst_inode      = to_bch_ei(dst_dentry->d_inode),
 776                 .mode           = flags & RENAME_EXCHANGE
 777                                 ? BCH_RENAME_EXCHANGE
 778                         : dst_dentry->d_inode
 779                                 ? BCH_RENAME_OVERWRITE : BCH_RENAME,
 780         };
 781         struct btree_trans trans;
 782         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 783         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 784         u64 journal_seq = 0;
 785         int ret;
 786
 787         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 788                 return -EINVAL;
 789
 790         if (i.mode == BCH_RENAME_OVERWRITE) {
 791                 if (S_ISDIR(i.src_inode->v.i_mode) !=
 792                     S_ISDIR(i.dst_inode->v.i_mode))
 793                         return -ENOTDIR;
 794
 795                 if (S_ISDIR(i.src_inode->v.i_mode) &&
 796                     bch2_empty_dir(c, i.dst_inode->v.i_ino))
 797                         return -ENOTEMPTY;
 798
 799                 ret = filemap_write_and_wait_range(i.src_inode->v.i_mapping,
 800                                                    0, LLONG_MAX);
 801                 if (ret)
 802                         return ret;
 803         }
 804
 805         bch2_trans_init(&trans, c, 8, 2048);
 806
 807         bch2_lock_inodes(INODE_UPDATE_LOCK,
 808                          i.src_dir,
 809                          i.dst_dir,
 810                          i.src_inode,
 811                          i.dst_inode);
 812
 813         if (S_ISDIR(i.src_inode->v.i_mode) &&
 814             inode_attrs_changing(i.dst_dir, i.src_inode)) {
 815                 ret = -EXDEV;
 816                 goto err;
 817         }
 818
 819         if (i.mode == BCH_RENAME_EXCHANGE &&
 820             S_ISDIR(i.dst_inode->v.i_mode) &&
 821             inode_attrs_changing(i.src_dir, i.dst_inode)) {
 822                 ret = -EXDEV;
 823                 goto err;
 824         }
 825
 826         if (inode_attr_changing(i.dst_dir, i.src_inode, Inode_opt_project)) {
 827                 ret = bch2_fs_quota_transfer(c, i.src_inode,
 828                                              i.dst_dir->ei_qid,
 829                                              1 << QTYP_PRJ,
 830                                              KEY_TYPE_QUOTA_PREALLOC);
 831                 if (ret)
 832                         goto err;
 833         }
 834
 835         if (i.mode == BCH_RENAME_EXCHANGE &&
 836             inode_attr_changing(i.src_dir, i.dst_inode, Inode_opt_project)) {
 837                 ret = bch2_fs_quota_transfer(c, i.dst_inode,
 838                                              i.src_dir->ei_qid,
 839                                              1 << QTYP_PRJ,
 840                                              KEY_TYPE_QUOTA_PREALLOC);
 841                 if (ret)
 842                         goto err;
 843         }
 844
 845 retry:
 846         bch2_trans_begin(&trans);
 847         i.now = bch2_current_time(c);
 848
 849         ret   = bch2_dirent_rename(&trans,
 850                                    i.src_dir, &src_dentry->d_name,
 851                                    i.dst_dir, &dst_dentry->d_name,
 852                                    i.mode) ?:
 853                 bch2_write_inode_trans(&trans, i.src_dir, &src_dir_u,
 854                                        inode_update_for_rename_fn, &i) ?:
 855                 (i.src_dir != i.dst_dir
 856                  ? bch2_write_inode_trans(&trans, i.dst_dir, &dst_dir_u,
 857                                        inode_update_for_rename_fn, &i)
 858                  : 0 ) ?:
 859                 bch2_write_inode_trans(&trans, i.src_inode, &src_inode_u,
 860                                        inode_update_for_rename_fn, &i) ?:
 861                 (i.dst_inode
 862                  ? bch2_write_inode_trans(&trans, i.dst_inode, &dst_inode_u,
 863                                        inode_update_for_rename_fn, &i)
 864                  : 0 ) ?:
 865                 bch2_trans_commit(&trans, NULL,
 866                                   &journal_seq,
 867                                   BTREE_INSERT_ATOMIC|
 868                                   BTREE_INSERT_NOUNLOCK);
 869         if (ret == -EINTR)
 870                 goto retry;
 871         if (unlikely(ret))
 872                 goto err;
 873
 874         bch2_inode_update_after_write(c, i.src_dir, &src_dir_u,
 875                                       ATTR_MTIME|ATTR_CTIME);
 876         journal_seq_copy(i.src_dir, journal_seq);
 877
 878         if (i.src_dir != i.dst_dir) {
 879                 bch2_inode_update_after_write(c, i.dst_dir, &dst_dir_u,
 880                                               ATTR_MTIME|ATTR_CTIME);
 881                 journal_seq_copy(i.dst_dir, journal_seq);
 882         }
 883
 884         journal_seq_copy(i.src_inode, journal_seq);
 885         if (i.dst_inode)
 886                 journal_seq_copy(i.dst_inode, journal_seq);
 887
 888         bch2_inode_update_after_write(c, i.src_inode, &src_inode_u,
 889                                       ATTR_CTIME);
 890         if (i.dst_inode)
 891                 bch2_inode_update_after_write(c, i.dst_inode, &dst_inode_u,
 892                                               ATTR_CTIME);
 893 err:
 894         bch2_trans_exit(&trans);
 895
 896         bch2_fs_quota_transfer(c, i.src_inode,
 897                                bch_qid(&i.src_inode->ei_inode),
 898                                1 << QTYP_PRJ,
 899                                KEY_TYPE_QUOTA_NOCHECK);
 900         if (i.dst_inode)
 901                 bch2_fs_quota_transfer(c, i.dst_inode,
 902                                        bch_qid(&i.dst_inode->ei_inode),
 903                                        1 << QTYP_PRJ,
 904                                        KEY_TYPE_QUOTA_NOCHECK);
 905
 906         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 907                            i.src_dir,
 908                            i.dst_dir,
 909                            i.src_inode,
 910                            i.dst_inode);
 911
 912         return ret;
 913 }
 914
 915 static int inode_update_for_setattr_fn(struct bch_inode_info *inode,
 916                                        struct bch_inode_unpacked *bi,
 917                                        void *p)
 918 {
 919         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 920         struct iattr *attr = p;
 921         unsigned int ia_valid = attr->ia_valid;
 922
 923         if (ia_valid & ATTR_UID)
 924                 bi->bi_uid = from_kuid(inode->v.i_sb->s_user_ns, attr->ia_uid);
 925         if (ia_valid & ATTR_GID)
 926                 bi->bi_gid = from_kgid(inode->v.i_sb->s_user_ns, attr->ia_gid);
 927
 928         if (ia_valid & ATTR_ATIME)
 929                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 930         if (ia_valid & ATTR_MTIME)
 931                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 932         if (ia_valid & ATTR_CTIME)
 933                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 934
 935         if (ia_valid & ATTR_MODE) {
 936                 umode_t mode = attr->ia_mode;
 937                 kgid_t gid = ia_valid & ATTR_GID
 938                         ? attr->ia_gid
 939                         : inode->v.i_gid;
 940
 941                 if (!in_group_p(gid) &&
 942                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 943                         mode &= ~S_ISGID;
 944                 bi->bi_mode = mode;
 945         }
 946
 947         return 0;
 948 }
 949
 950 static int bch2_setattr_nonsize(struct bch_inode_info *inode, struct iattr *iattr)
 951 {
 952         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 953         struct bch_qid qid;
 954         struct btree_trans trans;
 955         struct bch_inode_unpacked inode_u;
 956         struct posix_acl *acl = NULL;
 957         int ret;
 958
 959         mutex_lock(&inode->ei_update_lock);
 960
 961         qid = inode->ei_qid;
 962
 963         if (iattr->ia_valid & ATTR_UID)
 964                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, iattr->ia_uid);
 965
 966         if (iattr->ia_valid & ATTR_GID)
 967                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, iattr->ia_gid);
 968
 969         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 970                                      KEY_TYPE_QUOTA_PREALLOC);
 971         if (ret)
 972                 goto err;
 973
 974         bch2_trans_init(&trans, c, 0, 0);
 975 retry:
 976         bch2_trans_begin(&trans);
 977         kfree(acl);
 978         acl = NULL;
 979
 980         ret = bch2_write_inode_trans(&trans, inode, &inode_u,
 981                                 inode_update_for_setattr_fn, iattr) ?:
 982                 (iattr->ia_valid & ATTR_MODE
 983                  ? bch2_acl_chmod(&trans, inode, iattr->ia_mode, &acl)
 984                  : 0) ?:
 985                 bch2_trans_commit(&trans, NULL,
 986                                   &inode->ei_journal_seq,
 987                                   BTREE_INSERT_ATOMIC|
 988                                   BTREE_INSERT_NOUNLOCK|
 989                                   BTREE_INSERT_NOFAIL);
 990         if (ret == -EINTR)
 991                 goto retry;
 992         if (unlikely(ret))
 993                 goto err_trans;
 994
 995         bch2_inode_update_after_write(c, inode, &inode_u, iattr->ia_valid);
 996
 997         if (acl)
 998                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 999 err_trans:
1000         bch2_trans_exit(&trans);
1001 err:
1002         mutex_unlock(&inode->ei_update_lock);
1003
1004         return ret;
1005 }
1006
1007 static int bch2_getattr(const struct path *path, struct kstat *stat,
1008                         u32 request_mask, unsigned query_flags)
1009 {
1010         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
1011         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1012
1013         stat->dev       = inode->v.i_sb->s_dev;
1014         stat->ino       = inode->v.i_ino;
1015         stat->mode      = inode->v.i_mode;
1016         stat->nlink     = inode->v.i_nlink;
1017         stat->uid       = inode->v.i_uid;
1018         stat->gid       = inode->v.i_gid;
1019         stat->rdev      = inode->v.i_rdev;
1020         stat->size      = i_size_read(&inode->v);
1021         stat->atime     = inode->v.i_atime;
1022         stat->mtime     = inode->v.i_mtime;
1023         stat->ctime     = inode->v.i_ctime;
1024         stat->blksize   = block_bytes(c);
1025         stat->blocks    = inode->v.i_blocks;
1026
1027         if (request_mask & STATX_BTIME) {
1028                 stat->result_mask |= STATX_BTIME;
1029                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
1030         }
1031
1032         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
1033                 stat->attributes |= STATX_ATTR_IMMUTABLE;
1034         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
1035                 stat->attributes |= STATX_ATTR_APPEND;
1036         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
1037                 stat->attributes |= STATX_ATTR_NODUMP;
1038
1039         return 0;
1040 }
1041
1042 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
1043 {
1044         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
1045         int ret;
1046
1047         lockdep_assert_held(&inode->v.i_rwsem);
1048
1049         ret = setattr_prepare(dentry, iattr);
1050         if (ret)
1051                 return ret;
1052
1053         return iattr->ia_valid & ATTR_SIZE
1054                 ? bch2_truncate(inode, iattr)
1055                 : bch2_setattr_nonsize(inode, iattr);
1056 }
1057
1058 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
1059 {
1060         struct bch_inode_info *inode =
1061                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
1062
1063         if (IS_ERR(inode))
1064                 return PTR_ERR(inode);
1065
1066         d_mark_tmpfile(dentry, &inode->v);
1067         d_instantiate(dentry, &inode->v);
1068         return 0;
1069 }
1070
1071 static int bch2_fill_extent(struct bch_fs *c,
1072                             struct fiemap_extent_info *info,
1073                             struct bkey_s_c k, unsigned flags)
1074 {
1075         if (bkey_extent_is_data(k.k)) {
1076                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1077                 const union bch_extent_entry *entry;
1078                 struct extent_ptr_decoded p;
1079                 int ret;
1080
1081                 if (k.k->type == KEY_TYPE_reflink_v)
1082                         flags |= FIEMAP_EXTENT_SHARED;
1083
1084                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1085                         int flags2 = 0;
1086                         u64 offset = p.ptr.offset;
1087
1088                         if (p.crc.compression_type)
1089                                 flags2 |= FIEMAP_EXTENT_ENCODED;
1090                         else
1091                                 offset += p.crc.offset;
1092
1093                         if ((offset & (c->opts.block_size - 1)) ||
1094                             (k.k->size & (c->opts.block_size - 1)))
1095                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
1096
1097                         ret = fiemap_fill_next_extent(info,
1098                                                 bkey_start_offset(k.k) << 9,
1099                                                 offset << 9,
1100                                                 k.k->size << 9, flags|flags2);
1101                         if (ret)
1102                                 return ret;
1103                 }
1104
1105                 return 0;
1106         } else if (k.k->type == KEY_TYPE_reservation) {
1107                 return fiemap_fill_next_extent(info,
1108                                                bkey_start_offset(k.k) << 9,
1109                                                0, k.k->size << 9,
1110                                                flags|
1111                                                FIEMAP_EXTENT_DELALLOC|
1112                                                FIEMAP_EXTENT_UNWRITTEN);
1113         } else {
1114                 BUG();
1115         }
1116 }
1117
1118 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
1119                        u64 start, u64 len)
1120 {
1121         struct bch_fs *c = vinode->i_sb->s_fs_info;
1122         struct bch_inode_info *ei = to_bch_ei(vinode);
1123         struct btree_trans trans;
1124         struct btree_iter *iter;
1125         struct bkey_s_c k;
1126         BKEY_PADDED(k) cur, prev;
1127         unsigned offset_into_extent, sectors;
1128         bool have_extent = false;
1129         int ret = 0;
1130
1131         if (start + len < start)
1132                 return -EINVAL;
1133
1134         bch2_trans_init(&trans, c, 0, 0);
1135
1136         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
1137                                    POS(ei->v.i_ino, start >> 9),
1138                                    BTREE_ITER_SLOTS);
1139
1140         while (bkey_cmp(iter->pos, POS(ei->v.i_ino, (start + len) >> 9)) < 0) {
1141                 k = bch2_btree_iter_peek_slot(iter);
1142                 ret = bkey_err(k);
1143                 if (ret)
1144                         goto err;
1145
1146                 bkey_reassemble(&cur.k, k);
1147                 k = bkey_i_to_s_c(&cur.k);
1148
1149                 offset_into_extent      = iter->pos.offset -
1150                         bkey_start_offset(k.k);
1151                 sectors                 = k.k->size - offset_into_extent;
1152
1153                 ret = bch2_read_indirect_extent(&trans, iter,
1154                                         &offset_into_extent, &cur.k);
1155                 if (ret)
1156                         break;
1157
1158                 sectors = min(sectors, k.k->size - offset_into_extent);
1159
1160                 bch2_cut_front(POS(k.k->p.inode,
1161                                    bkey_start_offset(k.k) + offset_into_extent),
1162                                &cur.k);
1163                 bch2_key_resize(&cur.k.k, sectors);
1164                 cur.k.k.p.offset = iter->pos.offset + cur.k.k.size;
1165
1166                 if (bkey_extent_is_data(k.k) ||
1167                     k.k->type == KEY_TYPE_reservation) {
1168                         if (have_extent) {
1169                                 ret = bch2_fill_extent(c, info,
1170                                                 bkey_i_to_s_c(&prev.k), 0);
1171                                 if (ret)
1172                                         break;
1173                         }
1174
1175                         bkey_copy(&prev.k, &cur.k);
1176                         have_extent = true;
1177                 }
1178
1179                 bch2_btree_iter_set_pos(iter,
1180                                 POS(iter->pos.inode,
1181                                     iter->pos.offset + sectors));
1182         }
1183
1184         if (!ret && have_extent)
1185                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(&prev.k),
1186                                        FIEMAP_EXTENT_LAST);
1187 err:
1188         ret = bch2_trans_exit(&trans) ?: ret;
1189         return ret < 0 ? ret : 0;
1190 }
1191
1192 static const struct vm_operations_struct bch_vm_ops = {
1193         .fault          = filemap_fault,
1194         .map_pages      = filemap_map_pages,
1195         .page_mkwrite   = bch2_page_mkwrite,
1196 };
1197
1198 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1199 {
1200         file_accessed(file);
1201
1202         vma->vm_ops = &bch_vm_ops;
1203         return 0;
1204 }
1205
1206 /* Directories: */
1207
1208 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1209 {
1210         return generic_file_llseek_size(file, offset, whence,
1211                                         S64_MAX, S64_MAX);
1212 }
1213
1214 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1215 {
1216         struct bch_fs *c = file_inode(file)->i_sb->s_fs_info;
1217
1218         return bch2_readdir(c, file, ctx);
1219 }
1220
1221 static const struct file_operations bch_file_operations = {
1222         .llseek         = bch2_llseek,
1223         .read_iter      = generic_file_read_iter,
1224         .write_iter     = bch2_write_iter,
1225         .mmap           = bch2_mmap,
1226         .open           = generic_file_open,
1227         .fsync          = bch2_fsync,
1228         .splice_read    = generic_file_splice_read,
1229         .splice_write   = iter_file_splice_write,
1230         .fallocate      = bch2_fallocate_dispatch,
1231         .unlocked_ioctl = bch2_fs_file_ioctl,
1232 #ifdef CONFIG_COMPAT
1233         .compat_ioctl   = bch2_compat_fs_ioctl,
1234 #endif
1235         .remap_file_range = bch2_remap_file_range,
1236 };
1237
1238 static const struct inode_operations bch_file_inode_operations = {
1239         .getattr        = bch2_getattr,
1240         .setattr        = bch2_setattr,
1241         .fiemap         = bch2_fiemap,
1242         .listxattr      = bch2_xattr_list,
1243 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1244         .get_acl        = bch2_get_acl,
1245         .set_acl        = bch2_set_acl,
1246 #endif
1247 };
1248
1249 static const struct inode_operations bch_dir_inode_operations = {
1250         .lookup         = bch2_lookup,
1251         .create         = bch2_create,
1252         .link           = bch2_link,
1253         .unlink         = bch2_unlink,
1254         .symlink        = bch2_symlink,
1255         .mkdir          = bch2_mkdir,
1256         .rmdir          = bch2_rmdir,
1257         .mknod          = bch2_mknod,
1258         .rename         = bch2_rename2,
1259         .getattr        = bch2_getattr,
1260         .setattr        = bch2_setattr,
1261         .tmpfile        = bch2_tmpfile,
1262         .listxattr      = bch2_xattr_list,
1263 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1264         .get_acl        = bch2_get_acl,
1265         .set_acl        = bch2_set_acl,
1266 #endif
1267 };
1268
1269 static const struct file_operations bch_dir_file_operations = {
1270         .llseek         = bch2_dir_llseek,
1271         .read           = generic_read_dir,
1272         .iterate        = bch2_vfs_readdir,
1273         .fsync          = bch2_fsync,
1274         .unlocked_ioctl = bch2_fs_file_ioctl,
1275 #ifdef CONFIG_COMPAT
1276         .compat_ioctl   = bch2_compat_fs_ioctl,
1277 #endif
1278 };
1279
1280 static const struct inode_operations bch_symlink_inode_operations = {
1281         .get_link       = page_get_link,
1282         .getattr        = bch2_getattr,
1283         .setattr        = bch2_setattr,
1284         .listxattr      = bch2_xattr_list,
1285 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1286         .get_acl        = bch2_get_acl,
1287         .set_acl        = bch2_set_acl,
1288 #endif
1289 };
1290
1291 static const struct inode_operations bch_special_inode_operations = {
1292         .getattr        = bch2_getattr,
1293         .setattr        = bch2_setattr,
1294         .listxattr      = bch2_xattr_list,
1295 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1296         .get_acl        = bch2_get_acl,
1297         .set_acl        = bch2_set_acl,
1298 #endif
1299 };
1300
1301 static const struct address_space_operations bch_address_space_operations = {
1302         .writepage      = bch2_writepage,
1303         .readpage       = bch2_readpage,
1304         .writepages     = bch2_writepages,
1305         .readpages      = bch2_readpages,
1306         .set_page_dirty = __set_page_dirty_nobuffers,
1307         .write_begin    = bch2_write_begin,
1308         .write_end      = bch2_write_end,
1309         .invalidatepage = bch2_invalidatepage,
1310         .releasepage    = bch2_releasepage,
1311         .direct_IO      = bch2_direct_IO,
1312 #ifdef CONFIG_MIGRATION
1313         .migratepage    = bch2_migrate_page,
1314 #endif
1315         .error_remove_page = generic_error_remove_page,
1316 };
1317
1318 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1319                 u64 ino, u32 generation)
1320 {
1321         struct bch_fs *c = sb->s_fs_info;
1322         struct inode *vinode;
1323
1324         if (ino < BCACHEFS_ROOT_INO)
1325                 return ERR_PTR(-ESTALE);
1326
1327         vinode = bch2_vfs_inode_get(c, ino);
1328         if (IS_ERR(vinode))
1329                 return ERR_CAST(vinode);
1330         if (generation && vinode->i_generation != generation) {
1331                 /* we didn't find the right inode.. */
1332                 iput(vinode);
1333                 return ERR_PTR(-ESTALE);
1334         }
1335         return vinode;
1336 }
1337
1338 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1339                 int fh_len, int fh_type)
1340 {
1341         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1342                                     bch2_nfs_get_inode);
1343 }
1344
1345 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1346                 int fh_len, int fh_type)
1347 {
1348         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1349                                     bch2_nfs_get_inode);
1350 }
1351
1352 static const struct export_operations bch_export_ops = {
1353         .fh_to_dentry   = bch2_fh_to_dentry,
1354         .fh_to_parent   = bch2_fh_to_parent,
1355         //.get_parent   = bch2_get_parent,
1356 };
1357
1358 static void bch2_vfs_inode_init(struct bch_fs *c,
1359                                 struct bch_inode_info *inode,
1360                                 struct bch_inode_unpacked *bi)
1361 {
1362         bch2_inode_update_after_write(c, inode, bi, ~0);
1363
1364         inode->v.i_blocks       = bi->bi_sectors;
1365         inode->v.i_ino          = bi->bi_inum;
1366         inode->v.i_rdev         = bi->bi_dev;
1367         inode->v.i_generation   = bi->bi_generation;
1368         inode->v.i_size         = bi->bi_size;
1369
1370         inode->ei_journal_seq   = 0;
1371         inode->ei_quota_reserved = 0;
1372         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1373         inode->ei_qid           = bch_qid(bi);
1374
1375         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1376
1377         switch (inode->v.i_mode & S_IFMT) {
1378         case S_IFREG:
1379                 inode->v.i_op   = &bch_file_inode_operations;
1380                 inode->v.i_fop  = &bch_file_operations;
1381                 break;
1382         case S_IFDIR:
1383                 inode->v.i_op   = &bch_dir_inode_operations;
1384                 inode->v.i_fop  = &bch_dir_file_operations;
1385                 break;
1386         case S_IFLNK:
1387                 inode_nohighmem(&inode->v);
1388                 inode->v.i_op   = &bch_symlink_inode_operations;
1389                 break;
1390         default:
1391                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1392                 inode->v.i_op   = &bch_special_inode_operations;
1393                 break;
1394         }
1395 }
1396
1397 static struct inode *bch2_alloc_inode(struct super_block *sb)
1398 {
1399         struct bch_inode_info *inode;
1400
1401         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1402         if (!inode)
1403                 return NULL;
1404
1405         inode_init_once(&inode->v);
1406         mutex_init(&inode->ei_update_lock);
1407         mutex_init(&inode->ei_quota_lock);
1408         inode->ei_inode_update = NULL;
1409         inode->ei_journal_seq = 0;
1410
1411         return &inode->v;
1412 }
1413
1414 static void bch2_i_callback(struct rcu_head *head)
1415 {
1416         struct inode *vinode = container_of(head, struct inode, i_rcu);
1417         struct bch_inode_info *inode = to_bch_ei(vinode);
1418
1419         kmem_cache_free(bch2_inode_cache, inode);
1420 }
1421
1422 static void bch2_destroy_inode(struct inode *vinode)
1423 {
1424         call_rcu(&vinode->i_rcu, bch2_i_callback);
1425 }
1426
1427 static int inode_update_times_fn(struct bch_inode_info *inode,
1428                                  struct bch_inode_unpacked *bi,
1429                                  void *p)
1430 {
1431         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1432
1433         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1434         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1435         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1436
1437         return 0;
1438 }
1439
1440 static int bch2_vfs_write_inode(struct inode *vinode,
1441                                 struct writeback_control *wbc)
1442 {
1443         struct bch_fs *c = vinode->i_sb->s_fs_info;
1444         struct bch_inode_info *inode = to_bch_ei(vinode);
1445         int ret;
1446
1447         mutex_lock(&inode->ei_update_lock);
1448         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1449                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1450         mutex_unlock(&inode->ei_update_lock);
1451
1452         if (c->opts.journal_flush_disabled)
1453                 return ret;
1454
1455         if (!ret && wbc->sync_mode == WB_SYNC_ALL)
1456                 ret = bch2_journal_flush_seq(&c->journal, inode->ei_journal_seq);
1457
1458         return ret;
1459 }
1460
1461 static void bch2_evict_inode(struct inode *vinode)
1462 {
1463         struct bch_fs *c = vinode->i_sb->s_fs_info;
1464         struct bch_inode_info *inode = to_bch_ei(vinode);
1465
1466         truncate_inode_pages_final(&inode->v.i_data);
1467
1468         clear_inode(&inode->v);
1469
1470         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1471
1472         if (inode->ei_inode_update)
1473                 bch2_deferred_update_free(c, inode->ei_inode_update);
1474         inode->ei_inode_update = NULL;
1475
1476         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1477                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1478                                 KEY_TYPE_QUOTA_WARN);
1479                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1480                                 KEY_TYPE_QUOTA_WARN);
1481                 bch2_inode_rm(c, inode->v.i_ino);
1482         }
1483 }
1484
1485 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1486 {
1487         struct super_block *sb = dentry->d_sb;
1488         struct bch_fs *c = sb->s_fs_info;
1489         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1490         unsigned shift = sb->s_blocksize_bits - 9;
1491         u64 fsid;
1492
1493         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1494         buf->f_bsize    = sb->s_blocksize;
1495         buf->f_blocks   = usage.capacity >> shift;
1496         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1497         buf->f_bavail   = buf->f_bfree;
1498         buf->f_files    = usage.nr_inodes;
1499         buf->f_ffree    = U64_MAX;
1500
1501         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1502                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1503         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1504         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1505         buf->f_namelen  = BCH_NAME_MAX;
1506
1507         return 0;
1508 }
1509
1510 static int bch2_sync_fs(struct super_block *sb, int wait)
1511 {
1512         struct bch_fs *c = sb->s_fs_info;
1513
1514         if (!wait) {
1515                 bch2_journal_flush_async(&c->journal, NULL);
1516                 return 0;
1517         }
1518
1519         return bch2_journal_flush(&c->journal);
1520 }
1521
1522 static struct bch_fs *bch2_path_to_fs(const char *dev)
1523 {
1524         struct bch_fs *c;
1525         struct block_device *bdev = lookup_bdev(dev);
1526
1527         if (IS_ERR(bdev))
1528                 return ERR_CAST(bdev);
1529
1530         c = bch2_bdev_to_fs(bdev);
1531         bdput(bdev);
1532         return c ?: ERR_PTR(-ENOENT);
1533 }
1534
1535 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1536                                                unsigned nr_devs, struct bch_opts opts)
1537 {
1538         struct bch_fs *c, *c1, *c2;
1539         size_t i;
1540
1541         if (!nr_devs)
1542                 return ERR_PTR(-EINVAL);
1543
1544         c = bch2_fs_open(devs, nr_devs, opts);
1545
1546         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1547                 /*
1548                  * Already open?
1549                  * Look up each block device, make sure they all belong to a
1550                  * filesystem and they all belong to the _same_ filesystem
1551                  */
1552
1553                 c1 = bch2_path_to_fs(devs[0]);
1554                 if (IS_ERR(c1))
1555                         return c;
1556
1557                 for (i = 1; i < nr_devs; i++) {
1558                         c2 = bch2_path_to_fs(devs[i]);
1559                         if (!IS_ERR(c2))
1560                                 closure_put(&c2->cl);
1561
1562                         if (c1 != c2) {
1563                                 closure_put(&c1->cl);
1564                                 return c;
1565                         }
1566                 }
1567
1568                 c = c1;
1569         }
1570
1571         if (IS_ERR(c))
1572                 return c;
1573
1574         mutex_lock(&c->state_lock);
1575
1576         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1577                 mutex_unlock(&c->state_lock);
1578                 closure_put(&c->cl);
1579                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1580                 return ERR_PTR(-EINVAL);
1581         }
1582
1583         mutex_unlock(&c->state_lock);
1584
1585         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1586         return c;
1587 }
1588
1589 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1590                                              struct bch_opts opts)
1591 {
1592         char *dev_name = NULL, **devs = NULL, *s;
1593         struct bch_fs *c = ERR_PTR(-ENOMEM);
1594         size_t i, nr_devs = 0;
1595
1596         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1597         if (!dev_name)
1598                 goto err;
1599
1600         for (s = dev_name; s; s = strchr(s + 1, ':'))
1601                 nr_devs++;
1602
1603         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1604         if (!devs)
1605                 goto err;
1606
1607         for (i = 0, s = dev_name;
1608              s;
1609              (s = strchr(s, ':')) && (*s++ = '\0'))
1610                 devs[i++] = s;
1611
1612         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1613 err:
1614         kfree(devs);
1615         kfree(dev_name);
1616         return c;
1617 }
1618
1619 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1620 {
1621         struct bch_fs *c = sb->s_fs_info;
1622         struct bch_opts opts = bch2_opts_empty();
1623         int ret;
1624
1625         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1626
1627         ret = bch2_parse_mount_opts(&opts, data);
1628         if (ret)
1629                 return ret;
1630
1631         if (opts.read_only != c->opts.read_only) {
1632                 mutex_lock(&c->state_lock);
1633
1634                 if (opts.read_only) {
1635                         bch2_fs_read_only(c);
1636
1637                         sb->s_flags |= SB_RDONLY;
1638                 } else {
1639                         ret = bch2_fs_read_write(c);
1640                         if (ret) {
1641                                 bch_err(c, "error going rw: %i", ret);
1642                                 mutex_unlock(&c->state_lock);
1643                                 return -EINVAL;
1644                         }
1645
1646                         sb->s_flags &= ~SB_RDONLY;
1647                 }
1648
1649                 c->opts.read_only = opts.read_only;
1650
1651                 mutex_unlock(&c->state_lock);
1652         }
1653
1654         if (opts.errors >= 0)
1655                 c->opts.errors = opts.errors;
1656
1657         return ret;
1658 }
1659
1660 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1661 {
1662         struct bch_fs *c = root->d_sb->s_fs_info;
1663         enum bch_opt_id i;
1664         char buf[512];
1665
1666         for (i = 0; i < bch2_opts_nr; i++) {
1667                 const struct bch_option *opt = &bch2_opt_table[i];
1668                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1669
1670                 if (!(opt->mode & OPT_MOUNT))
1671                         continue;
1672
1673                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1674                         continue;
1675
1676                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1677                                  OPT_SHOW_MOUNT_STYLE);
1678                 seq_putc(seq, ',');
1679                 seq_puts(seq, buf);
1680         }
1681
1682         return 0;
1683
1684 }
1685
1686 static const struct super_operations bch_super_operations = {
1687         .alloc_inode    = bch2_alloc_inode,
1688         .destroy_inode  = bch2_destroy_inode,
1689         .write_inode    = bch2_vfs_write_inode,
1690         .evict_inode    = bch2_evict_inode,
1691         .sync_fs        = bch2_sync_fs,
1692         .statfs         = bch2_statfs,
1693         .show_options   = bch2_show_options,
1694         .remount_fs     = bch2_remount,
1695 #if 0
1696         .put_super      = bch2_put_super,
1697         .freeze_fs      = bch2_freeze,
1698         .unfreeze_fs    = bch2_unfreeze,
1699 #endif
1700 };
1701
1702 static int bch2_test_super(struct super_block *s, void *data)
1703 {
1704         return s->s_fs_info == data;
1705 }
1706
1707 static int bch2_set_super(struct super_block *s, void *data)
1708 {
1709         s->s_fs_info = data;
1710         return 0;
1711 }
1712
1713 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1714                                  int flags, const char *dev_name, void *data)
1715 {
1716         struct bch_fs *c;
1717         struct bch_dev *ca;
1718         struct super_block *sb;
1719         struct inode *vinode;
1720         struct bch_opts opts = bch2_opts_empty();
1721         unsigned i;
1722         int ret;
1723
1724         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1725
1726         ret = bch2_parse_mount_opts(&opts, data);
1727         if (ret)
1728                 return ERR_PTR(ret);
1729
1730         c = bch2_open_as_blockdevs(dev_name, opts);
1731         if (IS_ERR(c))
1732                 return ERR_CAST(c);
1733
1734         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1735         if (IS_ERR(sb)) {
1736                 closure_put(&c->cl);
1737                 return ERR_CAST(sb);
1738         }
1739
1740         BUG_ON(sb->s_fs_info != c);
1741
1742         if (sb->s_root) {
1743                 closure_put(&c->cl);
1744
1745                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1746                         ret = -EBUSY;
1747                         goto err_put_super;
1748                 }
1749                 goto out;
1750         }
1751
1752         sb->s_blocksize         = block_bytes(c);
1753         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1754         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1755         sb->s_op                = &bch_super_operations;
1756         sb->s_export_op         = &bch_export_ops;
1757 #ifdef CONFIG_BCACHEFS_QUOTA
1758         sb->s_qcop              = &bch2_quotactl_operations;
1759         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1760 #endif
1761         sb->s_xattr             = bch2_xattr_handlers;
1762         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1763         sb->s_time_gran         = c->sb.time_precision;
1764         c->vfs_sb               = sb;
1765         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1766
1767         ret = super_setup_bdi(sb);
1768         if (ret)
1769                 goto err_put_super;
1770
1771         sb->s_bdi->congested_fn         = bch2_congested;
1772         sb->s_bdi->congested_data       = c;
1773         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1774
1775         for_each_online_member(ca, c, i) {
1776                 struct block_device *bdev = ca->disk_sb.bdev;
1777
1778                 /* XXX: create an anonymous device for multi device filesystems */
1779                 sb->s_bdev      = bdev;
1780                 sb->s_dev       = bdev->bd_dev;
1781                 percpu_ref_put(&ca->io_ref);
1782                 break;
1783         }
1784
1785 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1786         if (c->opts.acl)
1787                 sb->s_flags     |= SB_POSIXACL;
1788 #endif
1789
1790         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1791         if (IS_ERR(vinode)) {
1792                 bch_err(c, "error mounting: error getting root inode %i",
1793                         (int) PTR_ERR(vinode));
1794                 ret = PTR_ERR(vinode);
1795                 goto err_put_super;
1796         }
1797
1798         sb->s_root = d_make_root(vinode);
1799         if (!sb->s_root) {
1800                 bch_err(c, "error mounting: error allocating root dentry");
1801                 ret = -ENOMEM;
1802                 goto err_put_super;
1803         }
1804
1805         sb->s_flags |= SB_ACTIVE;
1806 out:
1807         return dget(sb->s_root);
1808
1809 err_put_super:
1810         deactivate_locked_super(sb);
1811         return ERR_PTR(ret);
1812 }
1813
1814 static void bch2_kill_sb(struct super_block *sb)
1815 {
1816         struct bch_fs *c = sb->s_fs_info;
1817
1818         generic_shutdown_super(sb);
1819
1820         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1821                 bch2_fs_stop(c);
1822         else
1823                 closure_put(&c->cl);
1824 }
1825
1826 static struct file_system_type bcache_fs_type = {
1827         .owner          = THIS_MODULE,
1828         .name           = "bcachefs",
1829         .mount          = bch2_mount,
1830         .kill_sb        = bch2_kill_sb,
1831         .fs_flags       = FS_REQUIRES_DEV,
1832 };
1833
1834 MODULE_ALIAS_FS("bcachefs");
1835
1836 void bch2_vfs_exit(void)
1837 {
1838         unregister_filesystem(&bcache_fs_type);
1839         if (bch2_inode_cache)
1840                 kmem_cache_destroy(bch2_inode_cache);
1841 }
1842
1843 int __init bch2_vfs_init(void)
1844 {
1845         int ret = -ENOMEM;
1846
1847         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1848         if (!bch2_inode_cache)
1849                 goto err;
1850
1851         ret = register_filesystem(&bcache_fs_type);
1852         if (ret)
1853                 goto err;
1854
1855         return 0;
1856 err:
1857         bch2_vfs_exit();
1858         return ret;
1859 }
1860
1861 #endif /* NO_BCACHEFS_FS */