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 "bkey_buf.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "chardev.h"
  10 #include "dirent.h"
  11 #include "extents.h"
  12 #include "fs.h"
  13 #include "fs-common.h"
  14 #include "fs-io.h"
  15 #include "fs-ioctl.h"
  16 #include "fsck.h"
  17 #include "inode.h"
  18 #include "io.h"
  19 #include "journal.h"
  20 #include "keylist.h"
  21 #include "quota.h"
  22 #include "super.h"
  23 #include "xattr.h"
  24
  25 #include <linux/aio.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/exportfs.h>
  28 #include <linux/fiemap.h>
  29 #include <linux/module.h>
  30 #include <linux/posix_acl.h>
  31 #include <linux/random.h>
  32 #include <linux/statfs.h>
  33 #include <linux/string.h>
  34 #include <linux/xattr.h>
  35
  36 static struct kmem_cache *bch2_inode_cache;
  37
  38 static void bch2_vfs_inode_init(struct bch_fs *,
  39                                 struct bch_inode_info *,
  40                                 struct bch_inode_unpacked *);
  41
  42 static void journal_seq_copy(struct bch_fs *c,
  43                              struct bch_inode_info *dst,
  44                              u64 journal_seq)
  45 {
  46         /*
  47          * atomic64_cmpxchg has a fallback for archs that don't support it,
  48          * cmpxchg does not:
  49          */
  50         atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
  51         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  52
  53         do {
  54                 old = v;
  55
  56                 if (old >= journal_seq)
  57                         break;
  58         } while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
  59
  60         bch2_journal_set_has_inum(&c->journal, dst->v.i_ino, journal_seq);
  61 }
  62
  63 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
  64 {
  65         BUG_ON(atomic_long_read(&lock->v) == 0);
  66
  67         if (atomic_long_sub_return_release(i, &lock->v) == 0)
  68                 wake_up_all(&lock->wait);
  69 }
  70
  71 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
  72 {
  73         long v = atomic_long_read(&lock->v), old;
  74
  75         do {
  76                 old = v;
  77
  78                 if (i > 0 ? v < 0 : v > 0)
  79                         return false;
  80         } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
  81                                         old, old + i)) != old);
  82         return true;
  83 }
  84
  85 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
  86 {
  87         wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
  88 }
  89
  90 void bch2_pagecache_add_put(struct pagecache_lock *lock)
  91 {
  92         __pagecache_lock_put(lock, 1);
  93 }
  94
  95 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
  96 {
  97         return __pagecache_lock_tryget(lock, 1);
  98 }
  99
 100 void bch2_pagecache_add_get(struct pagecache_lock *lock)
 101 {
 102         __pagecache_lock_get(lock, 1);
 103 }
 104
 105 void bch2_pagecache_block_put(struct pagecache_lock *lock)
 106 {
 107         __pagecache_lock_put(lock, -1);
 108 }
 109
 110 void bch2_pagecache_block_get(struct pagecache_lock *lock)
 111 {
 112         __pagecache_lock_get(lock, -1);
 113 }
 114
 115 void bch2_inode_update_after_write(struct bch_fs *c,
 116                                    struct bch_inode_info *inode,
 117                                    struct bch_inode_unpacked *bi,
 118                                    unsigned fields)
 119 {
 120         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
 121         i_uid_write(&inode->v, bi->bi_uid);
 122         i_gid_write(&inode->v, bi->bi_gid);
 123         inode->v.i_mode = bi->bi_mode;
 124
 125         if (fields & ATTR_ATIME)
 126                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
 127         if (fields & ATTR_MTIME)
 128                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
 129         if (fields & ATTR_CTIME)
 130                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
 131
 132         inode->ei_inode         = *bi;
 133
 134         bch2_inode_flags_to_vfs(inode);
 135 }
 136
 137 int __must_check bch2_write_inode(struct bch_fs *c,
 138                                   struct bch_inode_info *inode,
 139                                   inode_set_fn set,
 140                                   void *p, unsigned fields)
 141 {
 142         struct btree_trans trans;
 143         struct btree_iter *iter;
 144         struct bch_inode_unpacked inode_u;
 145         int ret;
 146
 147         bch2_trans_init(&trans, c, 0, 512);
 148 retry:
 149         bch2_trans_begin(&trans);
 150
 151         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 152                                BTREE_ITER_INTENT);
 153         ret   = PTR_ERR_OR_ZERO(iter) ?:
 154                 (set ? set(inode, &inode_u, p) : 0) ?:
 155                 bch2_inode_write(&trans, iter, &inode_u) ?:
 156                 bch2_trans_commit(&trans, NULL,
 157                                   &inode->ei_journal_seq,
 158                                   BTREE_INSERT_NOUNLOCK|
 159                                   BTREE_INSERT_NOFAIL);
 160
 161         /*
 162          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 163          * this is important for inode updates via bchfs_write_index_update
 164          */
 165         if (!ret)
 166                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 167
 168         bch2_trans_iter_put(&trans, iter);
 169
 170         if (ret == -EINTR)
 171                 goto retry;
 172
 173         bch2_trans_exit(&trans);
 174         return ret < 0 ? ret : 0;
 175 }
 176
 177 int bch2_fs_quota_transfer(struct bch_fs *c,
 178                            struct bch_inode_info *inode,
 179                            struct bch_qid new_qid,
 180                            unsigned qtypes,
 181                            enum quota_acct_mode mode)
 182 {
 183         unsigned i;
 184         int ret;
 185
 186         qtypes &= enabled_qtypes(c);
 187
 188         for (i = 0; i < QTYP_NR; i++)
 189                 if (new_qid.q[i] == inode->ei_qid.q[i])
 190                         qtypes &= ~(1U << i);
 191
 192         if (!qtypes)
 193                 return 0;
 194
 195         mutex_lock(&inode->ei_quota_lock);
 196
 197         ret = bch2_quota_transfer(c, qtypes, new_qid,
 198                                   inode->ei_qid,
 199                                   inode->v.i_blocks +
 200                                   inode->ei_quota_reserved,
 201                                   mode);
 202         if (!ret)
 203                 for (i = 0; i < QTYP_NR; i++)
 204                         if (qtypes & (1 << i))
 205                                 inode->ei_qid.q[i] = new_qid.q[i];
 206
 207         mutex_unlock(&inode->ei_quota_lock);
 208
 209         return ret;
 210 }
 211
 212 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 213 {
 214         struct bch_inode_unpacked inode_u;
 215         struct bch_inode_info *inode;
 216         int ret;
 217
 218         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 219         if (unlikely(!inode))
 220                 return ERR_PTR(-ENOMEM);
 221         if (!(inode->v.i_state & I_NEW))
 222                 return &inode->v;
 223
 224         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 225         if (ret) {
 226                 iget_failed(&inode->v);
 227                 return ERR_PTR(ret);
 228         }
 229
 230         bch2_vfs_inode_init(c, inode, &inode_u);
 231
 232         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 233
 234         unlock_new_inode(&inode->v);
 235
 236         return &inode->v;
 237 }
 238
 239 static int inum_test(struct inode *inode, void *p)
 240 {
 241         unsigned long *ino = p;
 242
 243         return *ino == inode->i_ino;
 244 }
 245
 246 static struct bch_inode_info *
 247 __bch2_create(struct user_namespace *mnt_userns,
 248               struct bch_inode_info *dir, struct dentry *dentry,
 249               umode_t mode, dev_t rdev, bool tmpfile)
 250 {
 251         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 252         struct btree_trans trans;
 253         struct bch_inode_unpacked dir_u;
 254         struct bch_inode_info *inode, *old;
 255         struct bch_inode_unpacked inode_u;
 256         struct posix_acl *default_acl = NULL, *acl = NULL;
 257         u64 journal_seq = 0;
 258         int ret;
 259
 260         /*
 261          * preallocate acls + vfs inode before btree transaction, so that
 262          * nothing can fail after the transaction succeeds:
 263          */
 264 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 265         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 266         if (ret)
 267                 return ERR_PTR(ret);
 268 #endif
 269         inode = to_bch_ei(new_inode(c->vfs_sb));
 270         if (unlikely(!inode)) {
 271                 inode = ERR_PTR(-ENOMEM);
 272                 goto err;
 273         }
 274
 275         bch2_inode_init_early(c, &inode_u);
 276
 277         if (!tmpfile)
 278                 mutex_lock(&dir->ei_update_lock);
 279
 280         bch2_trans_init(&trans, c, 8,
 281                         2048 + (!tmpfile ? dentry->d_name.len : 0));
 282 retry:
 283         bch2_trans_begin(&trans);
 284
 285         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 286                                   !tmpfile ? &dentry->d_name : NULL,
 287                                   from_kuid(mnt_userns, current_fsuid()),
 288                                   from_kgid(mnt_userns, current_fsgid()),
 289                                   mode, rdev,
 290                                   default_acl, acl) ?:
 291                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 292                                 KEY_TYPE_QUOTA_PREALLOC);
 293         if (unlikely(ret))
 294                 goto err_before_quota;
 295
 296         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 297                                   BTREE_INSERT_NOUNLOCK);
 298         if (unlikely(ret)) {
 299                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 300                                 KEY_TYPE_QUOTA_WARN);
 301 err_before_quota:
 302                 if (ret == -EINTR)
 303                         goto retry;
 304                 goto err_trans;
 305         }
 306
 307         if (!tmpfile) {
 308                 bch2_inode_update_after_write(c, dir, &dir_u,
 309                                               ATTR_MTIME|ATTR_CTIME);
 310                 journal_seq_copy(c, dir, journal_seq);
 311                 mutex_unlock(&dir->ei_update_lock);
 312         }
 313
 314         bch2_vfs_inode_init(c, inode, &inode_u);
 315         journal_seq_copy(c, inode, journal_seq);
 316
 317         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 318         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 319
 320         /*
 321          * we must insert the new inode into the inode cache before calling
 322          * bch2_trans_exit() and dropping locks, else we could race with another
 323          * thread pulling the inode in and modifying it:
 324          */
 325
 326         inode->v.i_state |= I_CREATING;
 327         old = to_bch_ei(inode_insert5(&inode->v, inode->v.i_ino,
 328                                       inum_test, NULL, &inode->v.i_ino));
 329         BUG_ON(!old);
 330
 331         if (unlikely(old != inode)) {
 332                 /*
 333                  * We raced, another process pulled the new inode into cache
 334                  * before us:
 335                  */
 336                 journal_seq_copy(c, old, journal_seq);
 337                 make_bad_inode(&inode->v);
 338                 iput(&inode->v);
 339
 340                 inode = old;
 341         } else {
 342                 /*
 343                  * we really don't want insert_inode_locked2() to be setting
 344                  * I_NEW...
 345                  */
 346                 unlock_new_inode(&inode->v);
 347         }
 348
 349         bch2_trans_exit(&trans);
 350 err:
 351         posix_acl_release(default_acl);
 352         posix_acl_release(acl);
 353         return inode;
 354 err_trans:
 355         if (!tmpfile)
 356                 mutex_unlock(&dir->ei_update_lock);
 357
 358         bch2_trans_exit(&trans);
 359         make_bad_inode(&inode->v);
 360         iput(&inode->v);
 361         inode = ERR_PTR(ret);
 362         goto err;
 363 }
 364
 365 /* methods */
 366
 367 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 368                                   unsigned int flags)
 369 {
 370         struct bch_fs *c = vdir->i_sb->s_fs_info;
 371         struct bch_inode_info *dir = to_bch_ei(vdir);
 372         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
 373         struct inode *vinode = NULL;
 374         u64 inum;
 375
 376         inum = bch2_dirent_lookup(c, dir->v.i_ino, &hash,
 377                                   &dentry->d_name);
 378
 379         if (inum)
 380                 vinode = bch2_vfs_inode_get(c, inum);
 381
 382         return d_splice_alias(vinode, dentry);
 383 }
 384
 385 static int bch2_mknod(struct user_namespace *mnt_userns,
 386                       struct inode *vdir, struct dentry *dentry,
 387                       umode_t mode, dev_t rdev)
 388 {
 389         struct bch_inode_info *inode =
 390                 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, rdev, false);
 391
 392         if (IS_ERR(inode))
 393                 return PTR_ERR(inode);
 394
 395         d_instantiate(dentry, &inode->v);
 396         return 0;
 397 }
 398
 399 static int bch2_create(struct user_namespace *mnt_userns,
 400                        struct inode *vdir, struct dentry *dentry,
 401                        umode_t mode, bool excl)
 402 {
 403         return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFREG, 0);
 404 }
 405
 406 static int __bch2_link(struct bch_fs *c,
 407                        struct bch_inode_info *inode,
 408                        struct bch_inode_info *dir,
 409                        struct dentry *dentry)
 410 {
 411         struct btree_trans trans;
 412         struct bch_inode_unpacked dir_u, inode_u;
 413         int ret;
 414
 415         mutex_lock(&inode->ei_update_lock);
 416         bch2_trans_init(&trans, c, 4, 1024);
 417
 418         ret = __bch2_trans_do(&trans, NULL, &inode->ei_journal_seq,
 419                               BTREE_INSERT_NOUNLOCK,
 420                         bch2_link_trans(&trans,
 421                                         dir->v.i_ino,
 422                                         inode->v.i_ino, &dir_u, &inode_u,
 423                                         &dentry->d_name));
 424
 425         if (likely(!ret)) {
 426                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 427
 428                 journal_seq_copy(c, inode, dir->ei_journal_seq);
 429                 bch2_inode_update_after_write(c, dir, &dir_u,
 430                                               ATTR_MTIME|ATTR_CTIME);
 431                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 432         }
 433
 434         bch2_trans_exit(&trans);
 435         mutex_unlock(&inode->ei_update_lock);
 436         return ret;
 437 }
 438
 439 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 440                      struct dentry *dentry)
 441 {
 442         struct bch_fs *c = vdir->i_sb->s_fs_info;
 443         struct bch_inode_info *dir = to_bch_ei(vdir);
 444         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 445         int ret;
 446
 447         lockdep_assert_held(&inode->v.i_rwsem);
 448
 449         ret = __bch2_link(c, inode, dir, dentry);
 450         if (unlikely(ret))
 451                 return ret;
 452
 453         ihold(&inode->v);
 454         d_instantiate(dentry, &inode->v);
 455         return 0;
 456 }
 457
 458 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 459 {
 460         struct bch_fs *c = vdir->i_sb->s_fs_info;
 461         struct bch_inode_info *dir = to_bch_ei(vdir);
 462         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 463         struct bch_inode_unpacked dir_u, inode_u;
 464         struct btree_trans trans;
 465         int ret;
 466
 467         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 468         bch2_trans_init(&trans, c, 4, 1024);
 469
 470         ret = __bch2_trans_do(&trans, NULL, &dir->ei_journal_seq,
 471                               BTREE_INSERT_NOUNLOCK|
 472                               BTREE_INSERT_NOFAIL,
 473                         bch2_unlink_trans(&trans,
 474                                           dir->v.i_ino, &dir_u,
 475                                           &inode_u, &dentry->d_name));
 476
 477         if (likely(!ret)) {
 478                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 479
 480                 journal_seq_copy(c, inode, dir->ei_journal_seq);
 481                 bch2_inode_update_after_write(c, dir, &dir_u,
 482                                               ATTR_MTIME|ATTR_CTIME);
 483                 bch2_inode_update_after_write(c, inode, &inode_u,
 484                                               ATTR_MTIME);
 485         }
 486
 487         bch2_trans_exit(&trans);
 488         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 489
 490         return ret;
 491 }
 492
 493 static int bch2_symlink(struct user_namespace *mnt_userns,
 494                         struct inode *vdir, struct dentry *dentry,
 495                         const char *symname)
 496 {
 497         struct bch_fs *c = vdir->i_sb->s_fs_info;
 498         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 499         int ret;
 500
 501         inode = __bch2_create(mnt_userns, dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 502         if (unlikely(IS_ERR(inode)))
 503                 return PTR_ERR(inode);
 504
 505         inode_lock(&inode->v);
 506         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 507         inode_unlock(&inode->v);
 508
 509         if (unlikely(ret))
 510                 goto err;
 511
 512         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 513         if (unlikely(ret))
 514                 goto err;
 515
 516         journal_seq_copy(c, dir, inode->ei_journal_seq);
 517
 518         ret = __bch2_link(c, inode, dir, dentry);
 519         if (unlikely(ret))
 520                 goto err;
 521
 522         d_instantiate(dentry, &inode->v);
 523         return 0;
 524 err:
 525         iput(&inode->v);
 526         return ret;
 527 }
 528
 529 static int bch2_mkdir(struct user_namespace *mnt_userns,
 530                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 531 {
 532         return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFDIR, 0);
 533 }
 534
 535 static int bch2_rename2(struct user_namespace *mnt_userns,
 536                         struct inode *src_vdir, struct dentry *src_dentry,
 537                         struct inode *dst_vdir, struct dentry *dst_dentry,
 538                         unsigned flags)
 539 {
 540         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 541         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 542         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 543         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 544         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 545         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 546         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 547         struct btree_trans trans;
 548         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 549                 ? BCH_RENAME_EXCHANGE
 550                 : dst_dentry->d_inode
 551                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 552         u64 journal_seq = 0;
 553         int ret;
 554
 555         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 556                 return -EINVAL;
 557
 558         if (mode == BCH_RENAME_OVERWRITE) {
 559                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 560                                                    0, LLONG_MAX);
 561                 if (ret)
 562                         return ret;
 563         }
 564
 565         bch2_trans_init(&trans, c, 8, 2048);
 566
 567         bch2_lock_inodes(INODE_UPDATE_LOCK,
 568                          src_dir,
 569                          dst_dir,
 570                          src_inode,
 571                          dst_inode);
 572
 573         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 574                 ret = bch2_fs_quota_transfer(c, src_inode,
 575                                              dst_dir->ei_qid,
 576                                              1 << QTYP_PRJ,
 577                                              KEY_TYPE_QUOTA_PREALLOC);
 578                 if (ret)
 579                         goto err;
 580         }
 581
 582         if (mode == BCH_RENAME_EXCHANGE &&
 583             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 584                 ret = bch2_fs_quota_transfer(c, dst_inode,
 585                                              src_dir->ei_qid,
 586                                              1 << QTYP_PRJ,
 587                                              KEY_TYPE_QUOTA_PREALLOC);
 588                 if (ret)
 589                         goto err;
 590         }
 591
 592         ret = __bch2_trans_do(&trans, NULL, &journal_seq,
 593                               BTREE_INSERT_NOUNLOCK,
 594                         bch2_rename_trans(&trans,
 595                                           src_dir->v.i_ino, &src_dir_u,
 596                                           dst_dir->v.i_ino, &dst_dir_u,
 597                                           &src_inode_u,
 598                                           &dst_inode_u,
 599                                           &src_dentry->d_name,
 600                                           &dst_dentry->d_name,
 601                                           mode));
 602         if (unlikely(ret))
 603                 goto err;
 604
 605         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 606         BUG_ON(dst_inode &&
 607                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 608
 609         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 610                                       ATTR_MTIME|ATTR_CTIME);
 611         journal_seq_copy(c, src_dir, journal_seq);
 612
 613         if (src_dir != dst_dir) {
 614                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 615                                               ATTR_MTIME|ATTR_CTIME);
 616                 journal_seq_copy(c, dst_dir, journal_seq);
 617         }
 618
 619         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 620                                       ATTR_CTIME);
 621         journal_seq_copy(c, src_inode, journal_seq);
 622
 623         if (dst_inode) {
 624                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 625                                               ATTR_CTIME);
 626                 journal_seq_copy(c, dst_inode, journal_seq);
 627         }
 628 err:
 629         bch2_trans_exit(&trans);
 630
 631         bch2_fs_quota_transfer(c, src_inode,
 632                                bch_qid(&src_inode->ei_inode),
 633                                1 << QTYP_PRJ,
 634                                KEY_TYPE_QUOTA_NOCHECK);
 635         if (dst_inode)
 636                 bch2_fs_quota_transfer(c, dst_inode,
 637                                        bch_qid(&dst_inode->ei_inode),
 638                                        1 << QTYP_PRJ,
 639                                        KEY_TYPE_QUOTA_NOCHECK);
 640
 641         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 642                            src_dir,
 643                            dst_dir,
 644                            src_inode,
 645                            dst_inode);
 646
 647         return ret;
 648 }
 649
 650 void bch2_setattr_copy(struct user_namespace *mnt_userns,
 651                        struct bch_inode_info *inode,
 652                        struct bch_inode_unpacked *bi,
 653                        struct iattr *attr)
 654 {
 655         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 656         unsigned int ia_valid = attr->ia_valid;
 657
 658         if (ia_valid & ATTR_UID)
 659                 bi->bi_uid = from_kuid(mnt_userns, attr->ia_uid);
 660         if (ia_valid & ATTR_GID)
 661                 bi->bi_gid = from_kgid(mnt_userns, attr->ia_gid);
 662
 663         if (ia_valid & ATTR_ATIME)
 664                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 665         if (ia_valid & ATTR_MTIME)
 666                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 667         if (ia_valid & ATTR_CTIME)
 668                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 669
 670         if (ia_valid & ATTR_MODE) {
 671                 umode_t mode = attr->ia_mode;
 672                 kgid_t gid = ia_valid & ATTR_GID
 673                         ? attr->ia_gid
 674                         : inode->v.i_gid;
 675
 676                 if (!in_group_p(gid) &&
 677                     !capable_wrt_inode_uidgid(mnt_userns, &inode->v, CAP_FSETID))
 678                         mode &= ~S_ISGID;
 679                 bi->bi_mode = mode;
 680         }
 681 }
 682
 683 static int bch2_setattr_nonsize(struct user_namespace *mnt_userns,
 684                                 struct bch_inode_info *inode,
 685                                 struct iattr *attr)
 686 {
 687         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 688         struct bch_qid qid;
 689         struct btree_trans trans;
 690         struct btree_iter *inode_iter;
 691         struct bch_inode_unpacked inode_u;
 692         struct posix_acl *acl = NULL;
 693         int ret;
 694
 695         mutex_lock(&inode->ei_update_lock);
 696
 697         qid = inode->ei_qid;
 698
 699         if (attr->ia_valid & ATTR_UID)
 700                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 701
 702         if (attr->ia_valid & ATTR_GID)
 703                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 704
 705         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 706                                      KEY_TYPE_QUOTA_PREALLOC);
 707         if (ret)
 708                 goto err;
 709
 710         bch2_trans_init(&trans, c, 0, 0);
 711 retry:
 712         bch2_trans_begin(&trans);
 713         kfree(acl);
 714         acl = NULL;
 715
 716         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 717                                      BTREE_ITER_INTENT);
 718         ret = PTR_ERR_OR_ZERO(inode_iter);
 719         if (ret)
 720                 goto btree_err;
 721
 722         bch2_setattr_copy(mnt_userns, inode, &inode_u, attr);
 723
 724         if (attr->ia_valid & ATTR_MODE) {
 725                 ret = bch2_acl_chmod(&trans, &inode_u, inode_u.bi_mode, &acl);
 726                 if (ret)
 727                         goto btree_err;
 728         }
 729
 730         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 731                 bch2_trans_commit(&trans, NULL,
 732                                   &inode->ei_journal_seq,
 733                                   BTREE_INSERT_NOUNLOCK|
 734                                   BTREE_INSERT_NOFAIL);
 735 btree_err:
 736         bch2_trans_iter_put(&trans, inode_iter);
 737
 738         if (ret == -EINTR)
 739                 goto retry;
 740         if (unlikely(ret))
 741                 goto err_trans;
 742
 743         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 744
 745         if (acl)
 746                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 747 err_trans:
 748         bch2_trans_exit(&trans);
 749 err:
 750         mutex_unlock(&inode->ei_update_lock);
 751
 752         return ret;
 753 }
 754
 755 static int bch2_getattr(struct user_namespace *mnt_userns,
 756                         const struct path *path, struct kstat *stat,
 757                         u32 request_mask, unsigned query_flags)
 758 {
 759         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 760         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 761
 762         stat->dev       = inode->v.i_sb->s_dev;
 763         stat->ino       = inode->v.i_ino;
 764         stat->mode      = inode->v.i_mode;
 765         stat->nlink     = inode->v.i_nlink;
 766         stat->uid       = inode->v.i_uid;
 767         stat->gid       = inode->v.i_gid;
 768         stat->rdev      = inode->v.i_rdev;
 769         stat->size      = i_size_read(&inode->v);
 770         stat->atime     = inode->v.i_atime;
 771         stat->mtime     = inode->v.i_mtime;
 772         stat->ctime     = inode->v.i_ctime;
 773         stat->blksize   = block_bytes(c);
 774         stat->blocks    = inode->v.i_blocks;
 775
 776         if (request_mask & STATX_BTIME) {
 777                 stat->result_mask |= STATX_BTIME;
 778                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 779         }
 780
 781         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 782                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 783         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 784
 785         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 786                 stat->attributes |= STATX_ATTR_APPEND;
 787         stat->attributes_mask    |= STATX_ATTR_APPEND;
 788
 789         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 790                 stat->attributes |= STATX_ATTR_NODUMP;
 791         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 792
 793         return 0;
 794 }
 795
 796 static int bch2_setattr(struct user_namespace *mnt_userns,
 797                         struct dentry *dentry, struct iattr *iattr)
 798 {
 799         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 800         int ret;
 801
 802         lockdep_assert_held(&inode->v.i_rwsem);
 803
 804         ret = setattr_prepare(mnt_userns, dentry, iattr);
 805         if (ret)
 806                 return ret;
 807
 808         return iattr->ia_valid & ATTR_SIZE
 809                 ? bch2_truncate(inode, iattr)
 810                 : bch2_setattr_nonsize(mnt_userns, inode, iattr);
 811 }
 812
 813 static int bch2_tmpfile(struct user_namespace *mnt_userns,
 814                         struct inode *vdir, struct dentry *dentry, umode_t mode)
 815 {
 816         struct bch_inode_info *inode =
 817                 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, 0, true);
 818
 819         if (IS_ERR(inode))
 820                 return PTR_ERR(inode);
 821
 822         d_mark_tmpfile(dentry, &inode->v);
 823         d_instantiate(dentry, &inode->v);
 824         return 0;
 825 }
 826
 827 static int bch2_fill_extent(struct bch_fs *c,
 828                             struct fiemap_extent_info *info,
 829                             struct bkey_s_c k, unsigned flags)
 830 {
 831         if (bkey_extent_is_direct_data(k.k)) {
 832                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 833                 const union bch_extent_entry *entry;
 834                 struct extent_ptr_decoded p;
 835                 int ret;
 836
 837                 if (k.k->type == KEY_TYPE_reflink_v)
 838                         flags |= FIEMAP_EXTENT_SHARED;
 839
 840                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 841                         int flags2 = 0;
 842                         u64 offset = p.ptr.offset;
 843
 844                         if (p.crc.compression_type)
 845                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 846                         else
 847                                 offset += p.crc.offset;
 848
 849                         if ((offset & (c->opts.block_size - 1)) ||
 850                             (k.k->size & (c->opts.block_size - 1)))
 851                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 852
 853                         ret = fiemap_fill_next_extent(info,
 854                                                 bkey_start_offset(k.k) << 9,
 855                                                 offset << 9,
 856                                                 k.k->size << 9, flags|flags2);
 857                         if (ret)
 858                                 return ret;
 859                 }
 860
 861                 return 0;
 862         } else if (bkey_extent_is_inline_data(k.k)) {
 863                 return fiemap_fill_next_extent(info,
 864                                                bkey_start_offset(k.k) << 9,
 865                                                0, k.k->size << 9,
 866                                                flags|
 867                                                FIEMAP_EXTENT_DATA_INLINE);
 868         } else if (k.k->type == KEY_TYPE_reservation) {
 869                 return fiemap_fill_next_extent(info,
 870                                                bkey_start_offset(k.k) << 9,
 871                                                0, k.k->size << 9,
 872                                                flags|
 873                                                FIEMAP_EXTENT_DELALLOC|
 874                                                FIEMAP_EXTENT_UNWRITTEN);
 875         } else {
 876                 BUG();
 877         }
 878 }
 879
 880 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 881                        u64 start, u64 len)
 882 {
 883         struct bch_fs *c = vinode->i_sb->s_fs_info;
 884         struct bch_inode_info *ei = to_bch_ei(vinode);
 885         struct btree_trans trans;
 886         struct btree_iter *iter;
 887         struct bkey_s_c k;
 888         struct bkey_buf cur, prev;
 889         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 890         unsigned offset_into_extent, sectors;
 891         bool have_extent = false;
 892         int ret = 0;
 893
 894         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 895         if (ret)
 896                 return ret;
 897
 898         if (start + len < start)
 899                 return -EINVAL;
 900
 901         bch2_bkey_buf_init(&cur);
 902         bch2_bkey_buf_init(&prev);
 903         bch2_trans_init(&trans, c, 0, 0);
 904
 905         iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
 906                                    POS(ei->v.i_ino, start >> 9), 0);
 907 retry:
 908         while ((k = bch2_btree_iter_peek(iter)).k &&
 909                !(ret = bkey_err(k)) &&
 910                bkey_cmp(iter->pos, end) < 0) {
 911                 enum btree_id data_btree = BTREE_ID_extents;
 912
 913                 if (!bkey_extent_is_data(k.k) &&
 914                     k.k->type != KEY_TYPE_reservation) {
 915                         bch2_btree_iter_advance(iter);
 916                         continue;
 917                 }
 918
 919                 offset_into_extent      = iter->pos.offset -
 920                         bkey_start_offset(k.k);
 921                 sectors                 = k.k->size - offset_into_extent;
 922
 923                 bch2_bkey_buf_reassemble(&cur, c, k);
 924
 925                 ret = bch2_read_indirect_extent(&trans, &data_btree,
 926                                         &offset_into_extent, &cur);
 927                 if (ret)
 928                         break;
 929
 930                 k = bkey_i_to_s_c(cur.k);
 931                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 932
 933                 sectors = min(sectors, k.k->size - offset_into_extent);
 934
 935                 bch2_cut_front(POS(k.k->p.inode,
 936                                    bkey_start_offset(k.k) +
 937                                    offset_into_extent),
 938                                cur.k);
 939                 bch2_key_resize(&cur.k->k, sectors);
 940                 cur.k->k.p = iter->pos;
 941                 cur.k->k.p.offset += cur.k->k.size;
 942
 943                 if (have_extent) {
 944                         ret = bch2_fill_extent(c, info,
 945                                         bkey_i_to_s_c(prev.k), 0);
 946                         if (ret)
 947                                 break;
 948                 }
 949
 950                 bkey_copy(prev.k, cur.k);
 951                 have_extent = true;
 952
 953                 bch2_btree_iter_set_pos(iter,
 954                         POS(iter->pos.inode, iter->pos.offset + sectors));
 955         }
 956
 957         if (ret == -EINTR)
 958                 goto retry;
 959
 960         if (!ret && have_extent)
 961                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 962                                        FIEMAP_EXTENT_LAST);
 963
 964         bch2_trans_iter_put(&trans, iter);
 965         ret = bch2_trans_exit(&trans) ?: ret;
 966         bch2_bkey_buf_exit(&cur, c);
 967         bch2_bkey_buf_exit(&prev, c);
 968         return ret < 0 ? ret : 0;
 969 }
 970
 971 static const struct vm_operations_struct bch_vm_ops = {
 972         .fault          = bch2_page_fault,
 973         .map_pages      = filemap_map_pages,
 974         .page_mkwrite   = bch2_page_mkwrite,
 975 };
 976
 977 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 978 {
 979         file_accessed(file);
 980
 981         vma->vm_ops = &bch_vm_ops;
 982         return 0;
 983 }
 984
 985 /* Directories: */
 986
 987 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 988 {
 989         return generic_file_llseek_size(file, offset, whence,
 990                                         S64_MAX, S64_MAX);
 991 }
 992
 993 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 994 {
 995         struct bch_inode_info *inode = file_bch_inode(file);
 996         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 997
 998         if (!dir_emit_dots(file, ctx))
 999                 return 0;
1000
1001         return bch2_readdir(c, inode->v.i_ino, ctx);
1002 }
1003
1004 static const struct file_operations bch_file_operations = {
1005         .llseek         = bch2_llseek,
1006         .read_iter      = bch2_read_iter,
1007         .write_iter     = bch2_write_iter,
1008         .mmap           = bch2_mmap,
1009         .open           = generic_file_open,
1010         .fsync          = bch2_fsync,
1011         .splice_read    = generic_file_splice_read,
1012         .splice_write   = iter_file_splice_write,
1013         .fallocate      = bch2_fallocate_dispatch,
1014         .unlocked_ioctl = bch2_fs_file_ioctl,
1015 #ifdef CONFIG_COMPAT
1016         .compat_ioctl   = bch2_compat_fs_ioctl,
1017 #endif
1018         .remap_file_range = bch2_remap_file_range,
1019 };
1020
1021 static const struct inode_operations bch_file_inode_operations = {
1022         .getattr        = bch2_getattr,
1023         .setattr        = bch2_setattr,
1024         .fiemap         = bch2_fiemap,
1025         .listxattr      = bch2_xattr_list,
1026 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1027         .get_acl        = bch2_get_acl,
1028         .set_acl        = bch2_set_acl,
1029 #endif
1030 };
1031
1032 static const struct inode_operations bch_dir_inode_operations = {
1033         .lookup         = bch2_lookup,
1034         .create         = bch2_create,
1035         .link           = bch2_link,
1036         .unlink         = bch2_unlink,
1037         .symlink        = bch2_symlink,
1038         .mkdir          = bch2_mkdir,
1039         .rmdir          = bch2_unlink,
1040         .mknod          = bch2_mknod,
1041         .rename         = bch2_rename2,
1042         .getattr        = bch2_getattr,
1043         .setattr        = bch2_setattr,
1044         .tmpfile        = bch2_tmpfile,
1045         .listxattr      = bch2_xattr_list,
1046 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1047         .get_acl        = bch2_get_acl,
1048         .set_acl        = bch2_set_acl,
1049 #endif
1050 };
1051
1052 static const struct file_operations bch_dir_file_operations = {
1053         .llseek         = bch2_dir_llseek,
1054         .read           = generic_read_dir,
1055         .iterate_shared = bch2_vfs_readdir,
1056         .fsync          = bch2_fsync,
1057         .unlocked_ioctl = bch2_fs_file_ioctl,
1058 #ifdef CONFIG_COMPAT
1059         .compat_ioctl   = bch2_compat_fs_ioctl,
1060 #endif
1061 };
1062
1063 static const struct inode_operations bch_symlink_inode_operations = {
1064         .get_link       = page_get_link,
1065         .getattr        = bch2_getattr,
1066         .setattr        = bch2_setattr,
1067         .listxattr      = bch2_xattr_list,
1068 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1069         .get_acl        = bch2_get_acl,
1070         .set_acl        = bch2_set_acl,
1071 #endif
1072 };
1073
1074 static const struct inode_operations bch_special_inode_operations = {
1075         .getattr        = bch2_getattr,
1076         .setattr        = bch2_setattr,
1077         .listxattr      = bch2_xattr_list,
1078 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1079         .get_acl        = bch2_get_acl,
1080         .set_acl        = bch2_set_acl,
1081 #endif
1082 };
1083
1084 static const struct address_space_operations bch_address_space_operations = {
1085         .writepage      = bch2_writepage,
1086         .readpage       = bch2_readpage,
1087         .writepages     = bch2_writepages,
1088         .readahead      = bch2_readahead,
1089         .set_page_dirty = __set_page_dirty_nobuffers,
1090         .write_begin    = bch2_write_begin,
1091         .write_end      = bch2_write_end,
1092         .invalidatepage = bch2_invalidatepage,
1093         .releasepage    = bch2_releasepage,
1094         .direct_IO      = noop_direct_IO,
1095 #ifdef CONFIG_MIGRATION
1096         .migratepage    = bch2_migrate_page,
1097 #endif
1098         .error_remove_page = generic_error_remove_page,
1099 };
1100
1101 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1102                 u64 ino, u32 generation)
1103 {
1104         struct bch_fs *c = sb->s_fs_info;
1105         struct inode *vinode;
1106
1107         if (ino < BCACHEFS_ROOT_INO)
1108                 return ERR_PTR(-ESTALE);
1109
1110         vinode = bch2_vfs_inode_get(c, ino);
1111         if (IS_ERR(vinode))
1112                 return ERR_CAST(vinode);
1113         if (generation && vinode->i_generation != generation) {
1114                 /* we didn't find the right inode.. */
1115                 iput(vinode);
1116                 return ERR_PTR(-ESTALE);
1117         }
1118         return vinode;
1119 }
1120
1121 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1122                 int fh_len, int fh_type)
1123 {
1124         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1125                                     bch2_nfs_get_inode);
1126 }
1127
1128 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1129                 int fh_len, int fh_type)
1130 {
1131         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1132                                     bch2_nfs_get_inode);
1133 }
1134
1135 static const struct export_operations bch_export_ops = {
1136         .fh_to_dentry   = bch2_fh_to_dentry,
1137         .fh_to_parent   = bch2_fh_to_parent,
1138         //.get_parent   = bch2_get_parent,
1139 };
1140
1141 static void bch2_vfs_inode_init(struct bch_fs *c,
1142                                 struct bch_inode_info *inode,
1143                                 struct bch_inode_unpacked *bi)
1144 {
1145         bch2_inode_update_after_write(c, inode, bi, ~0);
1146
1147         inode->v.i_blocks       = bi->bi_sectors;
1148         inode->v.i_ino          = bi->bi_inum;
1149         inode->v.i_rdev         = bi->bi_dev;
1150         inode->v.i_generation   = bi->bi_generation;
1151         inode->v.i_size         = bi->bi_size;
1152
1153         inode->ei_flags         = 0;
1154         inode->ei_journal_seq   = 0;
1155         inode->ei_quota_reserved = 0;
1156         inode->ei_qid           = bch_qid(bi);
1157
1158         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1159
1160         switch (inode->v.i_mode & S_IFMT) {
1161         case S_IFREG:
1162                 inode->v.i_op   = &bch_file_inode_operations;
1163                 inode->v.i_fop  = &bch_file_operations;
1164                 break;
1165         case S_IFDIR:
1166                 inode->v.i_op   = &bch_dir_inode_operations;
1167                 inode->v.i_fop  = &bch_dir_file_operations;
1168                 break;
1169         case S_IFLNK:
1170                 inode_nohighmem(&inode->v);
1171                 inode->v.i_op   = &bch_symlink_inode_operations;
1172                 break;
1173         default:
1174                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1175                 inode->v.i_op   = &bch_special_inode_operations;
1176                 break;
1177         }
1178 }
1179
1180 static struct inode *bch2_alloc_inode(struct super_block *sb)
1181 {
1182         struct bch_inode_info *inode;
1183
1184         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1185         if (!inode)
1186                 return NULL;
1187
1188         inode_init_once(&inode->v);
1189         mutex_init(&inode->ei_update_lock);
1190         pagecache_lock_init(&inode->ei_pagecache_lock);
1191         mutex_init(&inode->ei_quota_lock);
1192         inode->ei_journal_seq = 0;
1193
1194         return &inode->v;
1195 }
1196
1197 static void bch2_i_callback(struct rcu_head *head)
1198 {
1199         struct inode *vinode = container_of(head, struct inode, i_rcu);
1200         struct bch_inode_info *inode = to_bch_ei(vinode);
1201
1202         kmem_cache_free(bch2_inode_cache, inode);
1203 }
1204
1205 static void bch2_destroy_inode(struct inode *vinode)
1206 {
1207         call_rcu(&vinode->i_rcu, bch2_i_callback);
1208 }
1209
1210 static int inode_update_times_fn(struct bch_inode_info *inode,
1211                                  struct bch_inode_unpacked *bi,
1212                                  void *p)
1213 {
1214         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1215
1216         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1217         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1218         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1219
1220         return 0;
1221 }
1222
1223 static int bch2_vfs_write_inode(struct inode *vinode,
1224                                 struct writeback_control *wbc)
1225 {
1226         struct bch_fs *c = vinode->i_sb->s_fs_info;
1227         struct bch_inode_info *inode = to_bch_ei(vinode);
1228         int ret;
1229
1230         mutex_lock(&inode->ei_update_lock);
1231         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1232                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1233         mutex_unlock(&inode->ei_update_lock);
1234
1235         return ret;
1236 }
1237
1238 static void bch2_evict_inode(struct inode *vinode)
1239 {
1240         struct bch_fs *c = vinode->i_sb->s_fs_info;
1241         struct bch_inode_info *inode = to_bch_ei(vinode);
1242
1243         truncate_inode_pages_final(&inode->v.i_data);
1244
1245         clear_inode(&inode->v);
1246
1247         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1248
1249         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1250                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1251                                 KEY_TYPE_QUOTA_WARN);
1252                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1253                                 KEY_TYPE_QUOTA_WARN);
1254                 bch2_inode_rm(c, inode->v.i_ino, true);
1255         }
1256 }
1257
1258 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1259 {
1260         struct super_block *sb = dentry->d_sb;
1261         struct bch_fs *c = sb->s_fs_info;
1262         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1263         unsigned shift = sb->s_blocksize_bits - 9;
1264         /*
1265          * this assumes inodes take up 64 bytes, which is a decent average
1266          * number:
1267          */
1268         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1269         u64 fsid;
1270
1271         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1272         buf->f_bsize    = sb->s_blocksize;
1273         buf->f_blocks   = usage.capacity >> shift;
1274         buf->f_bfree    = usage.free >> shift;
1275         buf->f_bavail   = avail_factor(usage.free) >> shift;
1276
1277         buf->f_files    = usage.nr_inodes + avail_inodes;
1278         buf->f_ffree    = avail_inodes;
1279
1280         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1281                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1282         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1283         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1284         buf->f_namelen  = BCH_NAME_MAX;
1285
1286         return 0;
1287 }
1288
1289 static int bch2_sync_fs(struct super_block *sb, int wait)
1290 {
1291         struct bch_fs *c = sb->s_fs_info;
1292
1293         if (c->opts.journal_flush_disabled)
1294                 return 0;
1295
1296         if (!wait) {
1297                 bch2_journal_flush_async(&c->journal, NULL);
1298                 return 0;
1299         }
1300
1301         return bch2_journal_flush(&c->journal);
1302 }
1303
1304 static struct bch_fs *bch2_path_to_fs(const char *path)
1305 {
1306         struct bch_fs *c;
1307         dev_t dev;
1308         int ret;
1309
1310         ret = lookup_bdev(path, &dev);
1311         if (ret)
1312                 return ERR_PTR(ret);
1313
1314         c = bch2_dev_to_fs(dev);
1315         if (c)
1316                 closure_put(&c->cl);
1317         return c ?: ERR_PTR(-ENOENT);
1318 }
1319
1320 static char **split_devs(const char *_dev_name, unsigned *nr)
1321 {
1322         char *dev_name = NULL, **devs = NULL, *s;
1323         size_t i, nr_devs = 0;
1324
1325         if (strlen(_dev_name) == 0)
1326                 return NULL;
1327
1328         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1329         if (!dev_name)
1330                 return NULL;
1331
1332         for (s = dev_name; s; s = strchr(s + 1, ':'))
1333                 nr_devs++;
1334
1335         devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1336         if (!devs) {
1337                 kfree(dev_name);
1338                 return NULL;
1339         }
1340
1341         for (i = 0, s = dev_name;
1342              s;
1343              (s = strchr(s, ':')) && (*s++ = '\0'))
1344                 devs[i++] = s;
1345
1346         *nr = nr_devs;
1347         return devs;
1348 }
1349
1350 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1351 {
1352         struct bch_fs *c = sb->s_fs_info;
1353         struct bch_opts opts = bch2_opts_empty();
1354         int ret;
1355
1356         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1357
1358         ret = bch2_parse_mount_opts(c, &opts, data);
1359         if (ret)
1360                 return ret;
1361
1362         if (opts.read_only != c->opts.read_only) {
1363                 down_write(&c->state_lock);
1364
1365                 if (opts.read_only) {
1366                         bch2_fs_read_only(c);
1367
1368                         sb->s_flags |= SB_RDONLY;
1369                 } else {
1370                         ret = bch2_fs_read_write(c);
1371                         if (ret) {
1372                                 bch_err(c, "error going rw: %i", ret);
1373                                 up_write(&c->state_lock);
1374                                 return -EINVAL;
1375                         }
1376
1377                         sb->s_flags &= ~SB_RDONLY;
1378                 }
1379
1380                 c->opts.read_only = opts.read_only;
1381
1382                 up_write(&c->state_lock);
1383         }
1384
1385         if (opts.errors >= 0)
1386                 c->opts.errors = opts.errors;
1387
1388         return ret;
1389 }
1390
1391 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1392 {
1393         struct bch_fs *c = root->d_sb->s_fs_info;
1394         struct bch_dev *ca;
1395         unsigned i;
1396         bool first = true;
1397
1398         for_each_online_member(ca, c, i) {
1399                 if (!first)
1400                         seq_putc(seq, ':');
1401                 first = false;
1402                 seq_puts(seq, "/dev/");
1403                 seq_puts(seq, ca->name);
1404         }
1405
1406         return 0;
1407 }
1408
1409 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1410 {
1411         struct bch_fs *c = root->d_sb->s_fs_info;
1412         enum bch_opt_id i;
1413         char buf[512];
1414
1415         for (i = 0; i < bch2_opts_nr; i++) {
1416                 const struct bch_option *opt = &bch2_opt_table[i];
1417                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1418
1419                 if (!(opt->mode & OPT_MOUNT))
1420                         continue;
1421
1422                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1423                         continue;
1424
1425                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1426                                  OPT_SHOW_MOUNT_STYLE);
1427                 seq_putc(seq, ',');
1428                 seq_puts(seq, buf);
1429         }
1430
1431         return 0;
1432 }
1433
1434 static void bch2_put_super(struct super_block *sb)
1435 {
1436         struct bch_fs *c = sb->s_fs_info;
1437
1438         __bch2_fs_stop(c);
1439 }
1440
1441 static const struct super_operations bch_super_operations = {
1442         .alloc_inode    = bch2_alloc_inode,
1443         .destroy_inode  = bch2_destroy_inode,
1444         .write_inode    = bch2_vfs_write_inode,
1445         .evict_inode    = bch2_evict_inode,
1446         .sync_fs        = bch2_sync_fs,
1447         .statfs         = bch2_statfs,
1448         .show_devname   = bch2_show_devname,
1449         .show_options   = bch2_show_options,
1450         .remount_fs     = bch2_remount,
1451         .put_super      = bch2_put_super,
1452 #if 0
1453         .freeze_fs      = bch2_freeze,
1454         .unfreeze_fs    = bch2_unfreeze,
1455 #endif
1456 };
1457
1458 static int bch2_set_super(struct super_block *s, void *data)
1459 {
1460         s->s_fs_info = data;
1461         return 0;
1462 }
1463
1464 static int bch2_noset_super(struct super_block *s, void *data)
1465 {
1466         return -EBUSY;
1467 }
1468
1469 static int bch2_test_super(struct super_block *s, void *data)
1470 {
1471         struct bch_fs *c = s->s_fs_info;
1472         struct bch_fs **devs = data;
1473         unsigned i;
1474
1475         if (!c)
1476                 return false;
1477
1478         for (i = 0; devs[i]; i++)
1479                 if (c != devs[i])
1480                         return false;
1481         return true;
1482 }
1483
1484 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1485                                  int flags, const char *dev_name, void *data)
1486 {
1487         struct bch_fs *c;
1488         struct bch_dev *ca;
1489         struct super_block *sb;
1490         struct inode *vinode;
1491         struct bch_opts opts = bch2_opts_empty();
1492         char **devs;
1493         struct bch_fs **devs_to_fs = NULL;
1494         unsigned i, nr_devs;
1495         int ret;
1496
1497         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1498
1499         ret = bch2_parse_mount_opts(NULL, &opts, data);
1500         if (ret)
1501                 return ERR_PTR(ret);
1502
1503         devs = split_devs(dev_name, &nr_devs);
1504         if (!devs)
1505                 return ERR_PTR(-ENOMEM);
1506
1507         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1508         if (!devs_to_fs) {
1509                 sb = ERR_PTR(-ENOMEM);
1510                 goto got_sb;
1511         }
1512
1513         for (i = 0; i < nr_devs; i++)
1514                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1515
1516         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1517                   flags|SB_NOSEC, devs_to_fs);
1518         if (!IS_ERR(sb))
1519                 goto got_sb;
1520
1521         c = bch2_fs_open(devs, nr_devs, opts);
1522         if (IS_ERR(c)) {
1523                 sb = ERR_CAST(c);
1524                 goto got_sb;
1525         }
1526
1527         /* Some options can't be parsed until after the fs is started: */
1528         ret = bch2_parse_mount_opts(c, &opts, data);
1529         if (ret) {
1530                 bch2_fs_stop(c);
1531                 sb = ERR_PTR(ret);
1532                 goto got_sb;
1533         }
1534
1535         bch2_opts_apply(&c->opts, opts);
1536
1537         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1538         if (IS_ERR(sb))
1539                 bch2_fs_stop(c);
1540 got_sb:
1541         kfree(devs_to_fs);
1542         kfree(devs[0]);
1543         kfree(devs);
1544
1545         if (IS_ERR(sb))
1546                 return ERR_CAST(sb);
1547
1548         c = sb->s_fs_info;
1549
1550         if (sb->s_root) {
1551                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1552                         ret = -EBUSY;
1553                         goto err_put_super;
1554                 }
1555                 goto out;
1556         }
1557
1558         sb->s_blocksize         = block_bytes(c);
1559         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1560         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1561         sb->s_op                = &bch_super_operations;
1562         sb->s_export_op         = &bch_export_ops;
1563 #ifdef CONFIG_BCACHEFS_QUOTA
1564         sb->s_qcop              = &bch2_quotactl_operations;
1565         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1566 #endif
1567         sb->s_xattr             = bch2_xattr_handlers;
1568         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1569         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1570         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1571         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1572         c->vfs_sb               = sb;
1573         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1574
1575         ret = super_setup_bdi(sb);
1576         if (ret)
1577                 goto err_put_super;
1578
1579         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1580
1581         for_each_online_member(ca, c, i) {
1582                 struct block_device *bdev = ca->disk_sb.bdev;
1583
1584                 /* XXX: create an anonymous device for multi device filesystems */
1585                 sb->s_bdev      = bdev;
1586                 sb->s_dev       = bdev->bd_dev;
1587                 percpu_ref_put(&ca->io_ref);
1588                 break;
1589         }
1590
1591         c->dev = sb->s_dev;
1592
1593 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1594         if (c->opts.acl)
1595                 sb->s_flags     |= SB_POSIXACL;
1596 #endif
1597
1598         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1599         if (IS_ERR(vinode)) {
1600                 bch_err(c, "error mounting: error getting root inode %i",
1601                         (int) PTR_ERR(vinode));
1602                 ret = PTR_ERR(vinode);
1603                 goto err_put_super;
1604         }
1605
1606         sb->s_root = d_make_root(vinode);
1607         if (!sb->s_root) {
1608                 bch_err(c, "error mounting: error allocating root dentry");
1609                 ret = -ENOMEM;
1610                 goto err_put_super;
1611         }
1612
1613         sb->s_flags |= SB_ACTIVE;
1614 out:
1615         return dget(sb->s_root);
1616
1617 err_put_super:
1618         deactivate_locked_super(sb);
1619         return ERR_PTR(ret);
1620 }
1621
1622 static void bch2_kill_sb(struct super_block *sb)
1623 {
1624         struct bch_fs *c = sb->s_fs_info;
1625
1626         generic_shutdown_super(sb);
1627         bch2_fs_free(c);
1628 }
1629
1630 static struct file_system_type bcache_fs_type = {
1631         .owner          = THIS_MODULE,
1632         .name           = "bcachefs",
1633         .mount          = bch2_mount,
1634         .kill_sb        = bch2_kill_sb,
1635         .fs_flags       = FS_REQUIRES_DEV,
1636 };
1637
1638 MODULE_ALIAS_FS("bcachefs");
1639
1640 void bch2_vfs_exit(void)
1641 {
1642         unregister_filesystem(&bcache_fs_type);
1643         if (bch2_inode_cache)
1644                 kmem_cache_destroy(bch2_inode_cache);
1645 }
1646
1647 int __init bch2_vfs_init(void)
1648 {
1649         int ret = -ENOMEM;
1650
1651         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1652         if (!bch2_inode_cache)
1653                 goto err;
1654
1655         ret = register_filesystem(&bcache_fs_type);
1656         if (ret)
1657                 goto err;
1658
1659         return 0;
1660 err:
1661         bch2_vfs_exit();
1662         return ret;
1663 }
1664
1665 #endif /* NO_BCACHEFS_FS */