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 static 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_SIZE)
 664                 bi->bi_size = attr->ia_size;
 665
 666         if (ia_valid & ATTR_ATIME)
 667                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 668         if (ia_valid & ATTR_MTIME)
 669                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 670         if (ia_valid & ATTR_CTIME)
 671                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 672
 673         if (ia_valid & ATTR_MODE) {
 674                 umode_t mode = attr->ia_mode;
 675                 kgid_t gid = ia_valid & ATTR_GID
 676                         ? attr->ia_gid
 677                         : inode->v.i_gid;
 678
 679                 if (!in_group_p(gid) &&
 680                     !capable_wrt_inode_uidgid(mnt_userns, &inode->v, CAP_FSETID))
 681                         mode &= ~S_ISGID;
 682                 bi->bi_mode = mode;
 683         }
 684 }
 685
 686 int bch2_setattr_nonsize(struct user_namespace *mnt_userns,
 687                          struct bch_inode_info *inode,
 688                          struct iattr *attr)
 689 {
 690         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 691         struct bch_qid qid;
 692         struct btree_trans trans;
 693         struct btree_iter *inode_iter;
 694         struct bch_inode_unpacked inode_u;
 695         struct posix_acl *acl = NULL;
 696         int ret;
 697
 698         mutex_lock(&inode->ei_update_lock);
 699
 700         qid = inode->ei_qid;
 701
 702         if (attr->ia_valid & ATTR_UID)
 703                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 704
 705         if (attr->ia_valid & ATTR_GID)
 706                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 707
 708         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 709                                      KEY_TYPE_QUOTA_PREALLOC);
 710         if (ret)
 711                 goto err;
 712
 713         bch2_trans_init(&trans, c, 0, 0);
 714 retry:
 715         bch2_trans_begin(&trans);
 716         kfree(acl);
 717         acl = NULL;
 718
 719         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 720                                      BTREE_ITER_INTENT);
 721         ret = PTR_ERR_OR_ZERO(inode_iter);
 722         if (ret)
 723                 goto btree_err;
 724
 725         bch2_setattr_copy(mnt_userns, inode, &inode_u, attr);
 726
 727         if (attr->ia_valid & ATTR_MODE) {
 728                 ret = bch2_acl_chmod(&trans, &inode_u, inode_u.bi_mode, &acl);
 729                 if (ret)
 730                         goto btree_err;
 731         }
 732
 733         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 734                 bch2_trans_commit(&trans, NULL,
 735                                   &inode->ei_journal_seq,
 736                                   BTREE_INSERT_NOUNLOCK|
 737                                   BTREE_INSERT_NOFAIL);
 738 btree_err:
 739         bch2_trans_iter_put(&trans, inode_iter);
 740
 741         if (ret == -EINTR)
 742                 goto retry;
 743         if (unlikely(ret))
 744                 goto err_trans;
 745
 746         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 747
 748         if (acl)
 749                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 750 err_trans:
 751         bch2_trans_exit(&trans);
 752 err:
 753         mutex_unlock(&inode->ei_update_lock);
 754
 755         return ret;
 756 }
 757
 758 static int bch2_getattr(struct user_namespace *mnt_userns,
 759                         const struct path *path, struct kstat *stat,
 760                         u32 request_mask, unsigned query_flags)
 761 {
 762         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 763         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 764
 765         stat->dev       = inode->v.i_sb->s_dev;
 766         stat->ino       = inode->v.i_ino;
 767         stat->mode      = inode->v.i_mode;
 768         stat->nlink     = inode->v.i_nlink;
 769         stat->uid       = inode->v.i_uid;
 770         stat->gid       = inode->v.i_gid;
 771         stat->rdev      = inode->v.i_rdev;
 772         stat->size      = i_size_read(&inode->v);
 773         stat->atime     = inode->v.i_atime;
 774         stat->mtime     = inode->v.i_mtime;
 775         stat->ctime     = inode->v.i_ctime;
 776         stat->blksize   = block_bytes(c);
 777         stat->blocks    = inode->v.i_blocks;
 778
 779         if (request_mask & STATX_BTIME) {
 780                 stat->result_mask |= STATX_BTIME;
 781                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 782         }
 783
 784         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 785                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 786         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 787
 788         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 789                 stat->attributes |= STATX_ATTR_APPEND;
 790         stat->attributes_mask    |= STATX_ATTR_APPEND;
 791
 792         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 793                 stat->attributes |= STATX_ATTR_NODUMP;
 794         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 795
 796         return 0;
 797 }
 798
 799 static int bch2_setattr(struct user_namespace *mnt_userns,
 800                         struct dentry *dentry, struct iattr *iattr)
 801 {
 802         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 803         int ret;
 804
 805         lockdep_assert_held(&inode->v.i_rwsem);
 806
 807         ret = setattr_prepare(mnt_userns, dentry, iattr);
 808         if (ret)
 809                 return ret;
 810
 811         return iattr->ia_valid & ATTR_SIZE
 812                 ? bch2_truncate(mnt_userns, inode, iattr)
 813                 : bch2_setattr_nonsize(mnt_userns, inode, iattr);
 814 }
 815
 816 static int bch2_tmpfile(struct user_namespace *mnt_userns,
 817                         struct inode *vdir, struct dentry *dentry, umode_t mode)
 818 {
 819         struct bch_inode_info *inode =
 820                 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, 0, true);
 821
 822         if (IS_ERR(inode))
 823                 return PTR_ERR(inode);
 824
 825         d_mark_tmpfile(dentry, &inode->v);
 826         d_instantiate(dentry, &inode->v);
 827         return 0;
 828 }
 829
 830 static int bch2_fill_extent(struct bch_fs *c,
 831                             struct fiemap_extent_info *info,
 832                             struct bkey_s_c k, unsigned flags)
 833 {
 834         if (bkey_extent_is_direct_data(k.k)) {
 835                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 836                 const union bch_extent_entry *entry;
 837                 struct extent_ptr_decoded p;
 838                 int ret;
 839
 840                 if (k.k->type == KEY_TYPE_reflink_v)
 841                         flags |= FIEMAP_EXTENT_SHARED;
 842
 843                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 844                         int flags2 = 0;
 845                         u64 offset = p.ptr.offset;
 846
 847                         if (p.crc.compression_type)
 848                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 849                         else
 850                                 offset += p.crc.offset;
 851
 852                         if ((offset & (c->opts.block_size - 1)) ||
 853                             (k.k->size & (c->opts.block_size - 1)))
 854                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 855
 856                         ret = fiemap_fill_next_extent(info,
 857                                                 bkey_start_offset(k.k) << 9,
 858                                                 offset << 9,
 859                                                 k.k->size << 9, flags|flags2);
 860                         if (ret)
 861                                 return ret;
 862                 }
 863
 864                 return 0;
 865         } else if (bkey_extent_is_inline_data(k.k)) {
 866                 return fiemap_fill_next_extent(info,
 867                                                bkey_start_offset(k.k) << 9,
 868                                                0, k.k->size << 9,
 869                                                flags|
 870                                                FIEMAP_EXTENT_DATA_INLINE);
 871         } else if (k.k->type == KEY_TYPE_reservation) {
 872                 return fiemap_fill_next_extent(info,
 873                                                bkey_start_offset(k.k) << 9,
 874                                                0, k.k->size << 9,
 875                                                flags|
 876                                                FIEMAP_EXTENT_DELALLOC|
 877                                                FIEMAP_EXTENT_UNWRITTEN);
 878         } else {
 879                 BUG();
 880         }
 881 }
 882
 883 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 884                        u64 start, u64 len)
 885 {
 886         struct bch_fs *c = vinode->i_sb->s_fs_info;
 887         struct bch_inode_info *ei = to_bch_ei(vinode);
 888         struct btree_trans trans;
 889         struct btree_iter *iter;
 890         struct bkey_s_c k;
 891         struct bkey_buf cur, prev;
 892         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 893         unsigned offset_into_extent, sectors;
 894         bool have_extent = false;
 895         int ret = 0;
 896
 897         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 898         if (ret)
 899                 return ret;
 900
 901         if (start + len < start)
 902                 return -EINVAL;
 903
 904         bch2_bkey_buf_init(&cur);
 905         bch2_bkey_buf_init(&prev);
 906         bch2_trans_init(&trans, c, 0, 0);
 907
 908         iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
 909                                    POS(ei->v.i_ino, start >> 9), 0);
 910 retry:
 911         while ((k = bch2_btree_iter_peek(iter)).k &&
 912                !(ret = bkey_err(k)) &&
 913                bkey_cmp(iter->pos, end) < 0) {
 914                 enum btree_id data_btree = BTREE_ID_extents;
 915
 916                 if (!bkey_extent_is_data(k.k) &&
 917                     k.k->type != KEY_TYPE_reservation) {
 918                         bch2_btree_iter_advance(iter);
 919                         continue;
 920                 }
 921
 922                 offset_into_extent      = iter->pos.offset -
 923                         bkey_start_offset(k.k);
 924                 sectors                 = k.k->size - offset_into_extent;
 925
 926                 bch2_bkey_buf_reassemble(&cur, c, k);
 927
 928                 ret = bch2_read_indirect_extent(&trans, &data_btree,
 929                                         &offset_into_extent, &cur);
 930                 if (ret)
 931                         break;
 932
 933                 k = bkey_i_to_s_c(cur.k);
 934                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 935
 936                 sectors = min(sectors, k.k->size - offset_into_extent);
 937
 938                 bch2_cut_front(POS(k.k->p.inode,
 939                                    bkey_start_offset(k.k) +
 940                                    offset_into_extent),
 941                                cur.k);
 942                 bch2_key_resize(&cur.k->k, sectors);
 943                 cur.k->k.p = iter->pos;
 944                 cur.k->k.p.offset += cur.k->k.size;
 945
 946                 if (have_extent) {
 947                         ret = bch2_fill_extent(c, info,
 948                                         bkey_i_to_s_c(prev.k), 0);
 949                         if (ret)
 950                                 break;
 951                 }
 952
 953                 bkey_copy(prev.k, cur.k);
 954                 have_extent = true;
 955
 956                 bch2_btree_iter_set_pos(iter,
 957                         POS(iter->pos.inode, iter->pos.offset + sectors));
 958         }
 959
 960         if (ret == -EINTR)
 961                 goto retry;
 962
 963         if (!ret && have_extent)
 964                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 965                                        FIEMAP_EXTENT_LAST);
 966
 967         bch2_trans_iter_put(&trans, iter);
 968         ret = bch2_trans_exit(&trans) ?: ret;
 969         bch2_bkey_buf_exit(&cur, c);
 970         bch2_bkey_buf_exit(&prev, c);
 971         return ret < 0 ? ret : 0;
 972 }
 973
 974 static const struct vm_operations_struct bch_vm_ops = {
 975         .fault          = bch2_page_fault,
 976         .map_pages      = filemap_map_pages,
 977         .page_mkwrite   = bch2_page_mkwrite,
 978 };
 979
 980 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 981 {
 982         file_accessed(file);
 983
 984         vma->vm_ops = &bch_vm_ops;
 985         return 0;
 986 }
 987
 988 /* Directories: */
 989
 990 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 991 {
 992         return generic_file_llseek_size(file, offset, whence,
 993                                         S64_MAX, S64_MAX);
 994 }
 995
 996 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 997 {
 998         struct bch_inode_info *inode = file_bch_inode(file);
 999         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1000
1001         if (!dir_emit_dots(file, ctx))
1002                 return 0;
1003
1004         return bch2_readdir(c, inode->v.i_ino, ctx);
1005 }
1006
1007 static const struct file_operations bch_file_operations = {
1008         .llseek         = bch2_llseek,
1009         .read_iter      = bch2_read_iter,
1010         .write_iter     = bch2_write_iter,
1011         .mmap           = bch2_mmap,
1012         .open           = generic_file_open,
1013         .fsync          = bch2_fsync,
1014         .splice_read    = generic_file_splice_read,
1015         .splice_write   = iter_file_splice_write,
1016         .fallocate      = bch2_fallocate_dispatch,
1017         .unlocked_ioctl = bch2_fs_file_ioctl,
1018 #ifdef CONFIG_COMPAT
1019         .compat_ioctl   = bch2_compat_fs_ioctl,
1020 #endif
1021         .remap_file_range = bch2_remap_file_range,
1022 };
1023
1024 static const struct inode_operations bch_file_inode_operations = {
1025         .getattr        = bch2_getattr,
1026         .setattr        = bch2_setattr,
1027         .fiemap         = bch2_fiemap,
1028         .listxattr      = bch2_xattr_list,
1029 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1030         .get_acl        = bch2_get_acl,
1031         .set_acl        = bch2_set_acl,
1032 #endif
1033 };
1034
1035 static const struct inode_operations bch_dir_inode_operations = {
1036         .lookup         = bch2_lookup,
1037         .create         = bch2_create,
1038         .link           = bch2_link,
1039         .unlink         = bch2_unlink,
1040         .symlink        = bch2_symlink,
1041         .mkdir          = bch2_mkdir,
1042         .rmdir          = bch2_unlink,
1043         .mknod          = bch2_mknod,
1044         .rename         = bch2_rename2,
1045         .getattr        = bch2_getattr,
1046         .setattr        = bch2_setattr,
1047         .tmpfile        = bch2_tmpfile,
1048         .listxattr      = bch2_xattr_list,
1049 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1050         .get_acl        = bch2_get_acl,
1051         .set_acl        = bch2_set_acl,
1052 #endif
1053 };
1054
1055 static const struct file_operations bch_dir_file_operations = {
1056         .llseek         = bch2_dir_llseek,
1057         .read           = generic_read_dir,
1058         .iterate_shared = bch2_vfs_readdir,
1059         .fsync          = bch2_fsync,
1060         .unlocked_ioctl = bch2_fs_file_ioctl,
1061 #ifdef CONFIG_COMPAT
1062         .compat_ioctl   = bch2_compat_fs_ioctl,
1063 #endif
1064 };
1065
1066 static const struct inode_operations bch_symlink_inode_operations = {
1067         .get_link       = page_get_link,
1068         .getattr        = bch2_getattr,
1069         .setattr        = bch2_setattr,
1070         .listxattr      = bch2_xattr_list,
1071 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1072         .get_acl        = bch2_get_acl,
1073         .set_acl        = bch2_set_acl,
1074 #endif
1075 };
1076
1077 static const struct inode_operations bch_special_inode_operations = {
1078         .getattr        = bch2_getattr,
1079         .setattr        = bch2_setattr,
1080         .listxattr      = bch2_xattr_list,
1081 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1082         .get_acl        = bch2_get_acl,
1083         .set_acl        = bch2_set_acl,
1084 #endif
1085 };
1086
1087 static const struct address_space_operations bch_address_space_operations = {
1088         .writepage      = bch2_writepage,
1089         .readpage       = bch2_readpage,
1090         .writepages     = bch2_writepages,
1091         .readahead      = bch2_readahead,
1092         .set_page_dirty = __set_page_dirty_nobuffers,
1093         .write_begin    = bch2_write_begin,
1094         .write_end      = bch2_write_end,
1095         .invalidatepage = bch2_invalidatepage,
1096         .releasepage    = bch2_releasepage,
1097         .direct_IO      = noop_direct_IO,
1098 #ifdef CONFIG_MIGRATION
1099         .migratepage    = bch2_migrate_page,
1100 #endif
1101         .error_remove_page = generic_error_remove_page,
1102 };
1103
1104 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1105                 u64 ino, u32 generation)
1106 {
1107         struct bch_fs *c = sb->s_fs_info;
1108         struct inode *vinode;
1109
1110         if (ino < BCACHEFS_ROOT_INO)
1111                 return ERR_PTR(-ESTALE);
1112
1113         vinode = bch2_vfs_inode_get(c, ino);
1114         if (IS_ERR(vinode))
1115                 return ERR_CAST(vinode);
1116         if (generation && vinode->i_generation != generation) {
1117                 /* we didn't find the right inode.. */
1118                 iput(vinode);
1119                 return ERR_PTR(-ESTALE);
1120         }
1121         return vinode;
1122 }
1123
1124 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1125                 int fh_len, int fh_type)
1126 {
1127         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1128                                     bch2_nfs_get_inode);
1129 }
1130
1131 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1132                 int fh_len, int fh_type)
1133 {
1134         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1135                                     bch2_nfs_get_inode);
1136 }
1137
1138 static const struct export_operations bch_export_ops = {
1139         .fh_to_dentry   = bch2_fh_to_dentry,
1140         .fh_to_parent   = bch2_fh_to_parent,
1141         //.get_parent   = bch2_get_parent,
1142 };
1143
1144 static void bch2_vfs_inode_init(struct bch_fs *c,
1145                                 struct bch_inode_info *inode,
1146                                 struct bch_inode_unpacked *bi)
1147 {
1148         bch2_inode_update_after_write(c, inode, bi, ~0);
1149
1150         inode->v.i_blocks       = bi->bi_sectors;
1151         inode->v.i_ino          = bi->bi_inum;
1152         inode->v.i_rdev         = bi->bi_dev;
1153         inode->v.i_generation   = bi->bi_generation;
1154         inode->v.i_size         = bi->bi_size;
1155
1156         inode->ei_flags         = 0;
1157         inode->ei_journal_seq   = 0;
1158         inode->ei_quota_reserved = 0;
1159         inode->ei_qid           = bch_qid(bi);
1160
1161         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1162
1163         switch (inode->v.i_mode & S_IFMT) {
1164         case S_IFREG:
1165                 inode->v.i_op   = &bch_file_inode_operations;
1166                 inode->v.i_fop  = &bch_file_operations;
1167                 break;
1168         case S_IFDIR:
1169                 inode->v.i_op   = &bch_dir_inode_operations;
1170                 inode->v.i_fop  = &bch_dir_file_operations;
1171                 break;
1172         case S_IFLNK:
1173                 inode_nohighmem(&inode->v);
1174                 inode->v.i_op   = &bch_symlink_inode_operations;
1175                 break;
1176         default:
1177                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1178                 inode->v.i_op   = &bch_special_inode_operations;
1179                 break;
1180         }
1181 }
1182
1183 static struct inode *bch2_alloc_inode(struct super_block *sb)
1184 {
1185         struct bch_inode_info *inode;
1186
1187         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1188         if (!inode)
1189                 return NULL;
1190
1191         inode_init_once(&inode->v);
1192         mutex_init(&inode->ei_update_lock);
1193         pagecache_lock_init(&inode->ei_pagecache_lock);
1194         mutex_init(&inode->ei_quota_lock);
1195         inode->ei_journal_seq = 0;
1196
1197         return &inode->v;
1198 }
1199
1200 static void bch2_i_callback(struct rcu_head *head)
1201 {
1202         struct inode *vinode = container_of(head, struct inode, i_rcu);
1203         struct bch_inode_info *inode = to_bch_ei(vinode);
1204
1205         kmem_cache_free(bch2_inode_cache, inode);
1206 }
1207
1208 static void bch2_destroy_inode(struct inode *vinode)
1209 {
1210         call_rcu(&vinode->i_rcu, bch2_i_callback);
1211 }
1212
1213 static int inode_update_times_fn(struct bch_inode_info *inode,
1214                                  struct bch_inode_unpacked *bi,
1215                                  void *p)
1216 {
1217         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1218
1219         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1220         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1221         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1222
1223         return 0;
1224 }
1225
1226 static int bch2_vfs_write_inode(struct inode *vinode,
1227                                 struct writeback_control *wbc)
1228 {
1229         struct bch_fs *c = vinode->i_sb->s_fs_info;
1230         struct bch_inode_info *inode = to_bch_ei(vinode);
1231         int ret;
1232
1233         mutex_lock(&inode->ei_update_lock);
1234         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1235                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1236         mutex_unlock(&inode->ei_update_lock);
1237
1238         return ret;
1239 }
1240
1241 static void bch2_evict_inode(struct inode *vinode)
1242 {
1243         struct bch_fs *c = vinode->i_sb->s_fs_info;
1244         struct bch_inode_info *inode = to_bch_ei(vinode);
1245
1246         truncate_inode_pages_final(&inode->v.i_data);
1247
1248         clear_inode(&inode->v);
1249
1250         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1251
1252         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1253                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1254                                 KEY_TYPE_QUOTA_WARN);
1255                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1256                                 KEY_TYPE_QUOTA_WARN);
1257                 bch2_inode_rm(c, inode->v.i_ino, true);
1258         }
1259 }
1260
1261 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1262 {
1263         struct super_block *sb = dentry->d_sb;
1264         struct bch_fs *c = sb->s_fs_info;
1265         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1266         unsigned shift = sb->s_blocksize_bits - 9;
1267         /*
1268          * this assumes inodes take up 64 bytes, which is a decent average
1269          * number:
1270          */
1271         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1272         u64 fsid;
1273
1274         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1275         buf->f_bsize    = sb->s_blocksize;
1276         buf->f_blocks   = usage.capacity >> shift;
1277         buf->f_bfree    = usage.free >> shift;
1278         buf->f_bavail   = avail_factor(usage.free) >> shift;
1279
1280         buf->f_files    = usage.nr_inodes + avail_inodes;
1281         buf->f_ffree    = avail_inodes;
1282
1283         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1284                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1285         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1286         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1287         buf->f_namelen  = BCH_NAME_MAX;
1288
1289         return 0;
1290 }
1291
1292 static int bch2_sync_fs(struct super_block *sb, int wait)
1293 {
1294         struct bch_fs *c = sb->s_fs_info;
1295
1296         if (c->opts.journal_flush_disabled)
1297                 return 0;
1298
1299         if (!wait) {
1300                 bch2_journal_flush_async(&c->journal, NULL);
1301                 return 0;
1302         }
1303
1304         return bch2_journal_flush(&c->journal);
1305 }
1306
1307 static struct bch_fs *bch2_path_to_fs(const char *path)
1308 {
1309         struct bch_fs *c;
1310         dev_t dev;
1311         int ret;
1312
1313         ret = lookup_bdev(path, &dev);
1314         if (ret)
1315                 return ERR_PTR(ret);
1316
1317         c = bch2_dev_to_fs(dev);
1318         if (c)
1319                 closure_put(&c->cl);
1320         return c ?: ERR_PTR(-ENOENT);
1321 }
1322
1323 static char **split_devs(const char *_dev_name, unsigned *nr)
1324 {
1325         char *dev_name = NULL, **devs = NULL, *s;
1326         size_t i, nr_devs = 0;
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         if (!dev_name || strlen(dev_name) == 0)
1504                 return ERR_PTR(-EINVAL);
1505
1506         devs = split_devs(dev_name, &nr_devs);
1507         if (!devs)
1508                 return ERR_PTR(-ENOMEM);
1509
1510         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1511         if (!devs_to_fs) {
1512                 sb = ERR_PTR(-ENOMEM);
1513                 goto got_sb;
1514         }
1515
1516         for (i = 0; i < nr_devs; i++)
1517                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1518
1519         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1520                   flags|SB_NOSEC, devs_to_fs);
1521         if (!IS_ERR(sb))
1522                 goto got_sb;
1523
1524         c = bch2_fs_open(devs, nr_devs, opts);
1525         if (IS_ERR(c)) {
1526                 sb = ERR_CAST(c);
1527                 goto got_sb;
1528         }
1529
1530         /* Some options can't be parsed until after the fs is started: */
1531         ret = bch2_parse_mount_opts(c, &opts, data);
1532         if (ret) {
1533                 bch2_fs_stop(c);
1534                 sb = ERR_PTR(ret);
1535                 goto got_sb;
1536         }
1537
1538         bch2_opts_apply(&c->opts, opts);
1539
1540         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1541         if (IS_ERR(sb))
1542                 bch2_fs_stop(c);
1543 got_sb:
1544         kfree(devs_to_fs);
1545         kfree(devs[0]);
1546         kfree(devs);
1547
1548         if (IS_ERR(sb))
1549                 return ERR_CAST(sb);
1550
1551         c = sb->s_fs_info;
1552
1553         if (sb->s_root) {
1554                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1555                         ret = -EBUSY;
1556                         goto err_put_super;
1557                 }
1558                 goto out;
1559         }
1560
1561         sb->s_blocksize         = block_bytes(c);
1562         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1563         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1564         sb->s_op                = &bch_super_operations;
1565         sb->s_export_op         = &bch_export_ops;
1566 #ifdef CONFIG_BCACHEFS_QUOTA
1567         sb->s_qcop              = &bch2_quotactl_operations;
1568         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1569 #endif
1570         sb->s_xattr             = bch2_xattr_handlers;
1571         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1572         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1573         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1574         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1575         c->vfs_sb               = sb;
1576         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1577
1578         ret = super_setup_bdi(sb);
1579         if (ret)
1580                 goto err_put_super;
1581
1582         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1583
1584         for_each_online_member(ca, c, i) {
1585                 struct block_device *bdev = ca->disk_sb.bdev;
1586
1587                 /* XXX: create an anonymous device for multi device filesystems */
1588                 sb->s_bdev      = bdev;
1589                 sb->s_dev       = bdev->bd_dev;
1590                 percpu_ref_put(&ca->io_ref);
1591                 break;
1592         }
1593
1594         c->dev = sb->s_dev;
1595
1596 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1597         if (c->opts.acl)
1598                 sb->s_flags     |= SB_POSIXACL;
1599 #endif
1600
1601         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1602         if (IS_ERR(vinode)) {
1603                 bch_err(c, "error mounting: error getting root inode %i",
1604                         (int) PTR_ERR(vinode));
1605                 ret = PTR_ERR(vinode);
1606                 goto err_put_super;
1607         }
1608
1609         sb->s_root = d_make_root(vinode);
1610         if (!sb->s_root) {
1611                 bch_err(c, "error mounting: error allocating root dentry");
1612                 ret = -ENOMEM;
1613                 goto err_put_super;
1614         }
1615
1616         sb->s_flags |= SB_ACTIVE;
1617 out:
1618         return dget(sb->s_root);
1619
1620 err_put_super:
1621         deactivate_locked_super(sb);
1622         return ERR_PTR(ret);
1623 }
1624
1625 static void bch2_kill_sb(struct super_block *sb)
1626 {
1627         struct bch_fs *c = sb->s_fs_info;
1628
1629         generic_shutdown_super(sb);
1630         bch2_fs_free(c);
1631 }
1632
1633 static struct file_system_type bcache_fs_type = {
1634         .owner          = THIS_MODULE,
1635         .name           = "bcachefs",
1636         .mount          = bch2_mount,
1637         .kill_sb        = bch2_kill_sb,
1638         .fs_flags       = FS_REQUIRES_DEV,
1639 };
1640
1641 MODULE_ALIAS_FS("bcachefs");
1642
1643 void bch2_vfs_exit(void)
1644 {
1645         unregister_filesystem(&bcache_fs_type);
1646         if (bch2_inode_cache)
1647                 kmem_cache_destroy(bch2_inode_cache);
1648 }
1649
1650 int __init bch2_vfs_init(void)
1651 {
1652         int ret = -ENOMEM;
1653
1654         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1655         if (!bch2_inode_cache)
1656                 goto err;
1657
1658         ret = register_filesystem(&bcache_fs_type);
1659         if (ret)
1660                 goto err;
1661
1662         return 0;
1663 err:
1664         bch2_vfs_exit();
1665         return ret;
1666 }
1667
1668 #endif /* NO_BCACHEFS_FS */