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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "chardev.h"
   9 #include "dirent.h"
  10 #include "extents.h"
  11 #include "fs.h"
  12 #include "fs-common.h"
  13 #include "fs-io.h"
  14 #include "fs-ioctl.h"
  15 #include "fsck.h"
  16 #include "inode.h"
  17 #include "io.h"
  18 #include "journal.h"
  19 #include "keylist.h"
  20 #include "quota.h"
  21 #include "super.h"
  22 #include "xattr.h"
  23
  24 #include <linux/aio.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/exportfs.h>
  27 #include <linux/module.h>
  28 #include <linux/posix_acl.h>
  29 #include <linux/random.h>
  30 #include <linux/statfs.h>
  31 #include <linux/xattr.h>
  32
  33 static struct kmem_cache *bch2_inode_cache;
  34
  35 static void bch2_vfs_inode_init(struct bch_fs *,
  36                                 struct bch_inode_info *,
  37                                 struct bch_inode_unpacked *);
  38
  39 static void journal_seq_copy(struct bch_inode_info *dst,
  40                              u64 journal_seq)
  41 {
  42         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  43
  44         do {
  45                 old = v;
  46
  47                 if (old >= journal_seq)
  48                         break;
  49         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  50 }
  51
  52 /*
  53  * I_SIZE_DIRTY requires special handling:
  54  *
  55  * To the recovery code, the flag means that there is stale data past i_size
  56  * that needs to be deleted; it's used for implementing atomic appends and
  57  * truncates.
  58  *
  59  * On append, we set I_SIZE_DIRTY before doing the write, then after the write
  60  * we clear I_SIZE_DIRTY atomically with updating i_size to the new larger size
  61  * that exposes the data we just wrote.
  62  *
  63  * On truncate, it's the reverse: We set I_SIZE_DIRTY atomically with setting
  64  * i_size to the new smaller size, then we delete the data that we just made
  65  * invisible, and then we clear I_SIZE_DIRTY.
  66  *
  67  * Because there can be multiple appends in flight at a time, we need a refcount
  68  * (i_size_dirty_count) instead of manipulating the flag directly. Nonzero
  69  * refcount means I_SIZE_DIRTY is set, zero means it's cleared.
  70  *
  71  * Because write_inode() can be called at any time, i_size_dirty_count means
  72  * something different to the runtime code - it means to write_inode() "don't
  73  * update i_size yet".
  74  *
  75  * We don't clear I_SIZE_DIRTY directly, we let write_inode() clear it when
  76  * i_size_dirty_count is zero - but the reverse is not true, I_SIZE_DIRTY must
  77  * be set explicitly.
  78  */
  79
  80 void bch2_inode_update_after_write(struct bch_fs *c,
  81                                    struct bch_inode_info *inode,
  82                                    struct bch_inode_unpacked *bi,
  83                                    unsigned fields)
  84 {
  85         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
  86         i_uid_write(&inode->v, bi->bi_uid);
  87         i_gid_write(&inode->v, bi->bi_gid);
  88         inode->v.i_mode = bi->bi_mode;
  89
  90         if (fields & ATTR_ATIME)
  91                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
  92         if (fields & ATTR_MTIME)
  93                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
  94         if (fields & ATTR_CTIME)
  95                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
  96
  97         inode->ei_inode         = *bi;
  98
  99         bch2_inode_flags_to_vfs(inode);
 100 }
 101
 102 int __must_check bch2_write_inode(struct bch_fs *c,
 103                                   struct bch_inode_info *inode,
 104                                   inode_set_fn set,
 105                                   void *p, unsigned fields)
 106 {
 107         struct btree_trans trans;
 108         struct btree_iter *iter;
 109         struct bch_inode_unpacked inode_u;
 110         int ret;
 111
 112         bch2_trans_init(&trans, c, 0, 0);
 113 retry:
 114         bch2_trans_begin(&trans);
 115
 116         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 117                                BTREE_ITER_INTENT);
 118         ret   = PTR_ERR_OR_ZERO(iter) ?:
 119                 (set ? set(inode, &inode_u, p) : 0) ?:
 120                 bch2_inode_write(&trans, iter, &inode_u) ?:
 121                 bch2_trans_commit(&trans, NULL,
 122                                   &inode->ei_journal_seq,
 123                                   BTREE_INSERT_ATOMIC|
 124                                   BTREE_INSERT_NOUNLOCK|
 125                                   BTREE_INSERT_NOFAIL);
 126         if (ret == -EINTR)
 127                 goto retry;
 128
 129         /*
 130          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 131          * this is important for inode updates via bchfs_write_index_update
 132          */
 133         if (!ret)
 134                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 135
 136         bch2_trans_exit(&trans);
 137         return ret < 0 ? ret : 0;
 138 }
 139
 140 int bch2_fs_quota_transfer(struct bch_fs *c,
 141                            struct bch_inode_info *inode,
 142                            struct bch_qid new_qid,
 143                            unsigned qtypes,
 144                            enum quota_acct_mode mode)
 145 {
 146         unsigned i;
 147         int ret;
 148
 149         qtypes &= enabled_qtypes(c);
 150
 151         for (i = 0; i < QTYP_NR; i++)
 152                 if (new_qid.q[i] == inode->ei_qid.q[i])
 153                         qtypes &= ~(1U << i);
 154
 155         if (!qtypes)
 156                 return 0;
 157
 158         mutex_lock(&inode->ei_quota_lock);
 159
 160         ret = bch2_quota_transfer(c, qtypes, new_qid,
 161                                   inode->ei_qid,
 162                                   inode->v.i_blocks +
 163                                   inode->ei_quota_reserved,
 164                                   mode);
 165         if (!ret)
 166                 for (i = 0; i < QTYP_NR; i++)
 167                         if (qtypes & (1 << i))
 168                                 inode->ei_qid.q[i] = new_qid.q[i];
 169
 170         mutex_unlock(&inode->ei_quota_lock);
 171
 172         return ret;
 173 }
 174
 175 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 176 {
 177         struct bch_inode_unpacked inode_u;
 178         struct bch_inode_info *inode;
 179         int ret;
 180
 181         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 182         if (unlikely(!inode))
 183                 return ERR_PTR(-ENOMEM);
 184         if (!(inode->v.i_state & I_NEW))
 185                 return &inode->v;
 186
 187         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 188         if (ret) {
 189                 iget_failed(&inode->v);
 190                 return ERR_PTR(ret);
 191         }
 192
 193         bch2_vfs_inode_init(c, inode, &inode_u);
 194
 195         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 196
 197         unlock_new_inode(&inode->v);
 198
 199         return &inode->v;
 200 }
 201
 202 static struct bch_inode_info *
 203 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 204               umode_t mode, dev_t rdev, bool tmpfile)
 205 {
 206         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 207         struct user_namespace *ns = dir->v.i_sb->s_user_ns;
 208         struct btree_trans trans;
 209         struct bch_inode_unpacked dir_u;
 210         struct bch_inode_info *inode, *old;
 211         struct bch_inode_unpacked inode_u;
 212         struct posix_acl *default_acl = NULL, *acl = NULL;
 213         u64 journal_seq = 0;
 214         int ret;
 215
 216         /*
 217          * preallocate acls + vfs inode before btree transaction, so that
 218          * nothing can fail after the transaction succeeds:
 219          */
 220 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 221         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 222         if (ret)
 223                 return ERR_PTR(ret);
 224 #endif
 225         inode = to_bch_ei(new_inode(c->vfs_sb));
 226         if (unlikely(!inode)) {
 227                 inode = ERR_PTR(-ENOMEM);
 228                 goto err;
 229         }
 230
 231         bch2_inode_init_early(c, &inode_u);
 232
 233         if (!tmpfile)
 234                 mutex_lock(&dir->ei_update_lock);
 235
 236         bch2_trans_init(&trans, c, 8, 1024);
 237 retry:
 238         bch2_trans_begin(&trans);
 239
 240         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 241                                   !tmpfile ? &dentry->d_name : NULL,
 242                                   from_kuid(ns, current_fsuid()),
 243                                   from_kgid(ns, current_fsgid()),
 244                                   mode, rdev,
 245                                   default_acl, acl) ?:
 246                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 247                                 KEY_TYPE_QUOTA_PREALLOC);
 248         if (unlikely(ret))
 249                 goto err_before_quota;
 250
 251         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 252                                   BTREE_INSERT_ATOMIC|
 253                                   BTREE_INSERT_NOUNLOCK);
 254         if (unlikely(ret)) {
 255                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 256                                 KEY_TYPE_QUOTA_WARN);
 257 err_before_quota:
 258                 if (ret == -EINTR)
 259                         goto retry;
 260                 goto err_trans;
 261         }
 262
 263         if (!tmpfile) {
 264                 bch2_inode_update_after_write(c, dir, &dir_u,
 265                                               ATTR_MTIME|ATTR_CTIME);
 266                 journal_seq_copy(dir, journal_seq);
 267                 mutex_unlock(&dir->ei_update_lock);
 268         }
 269
 270         bch2_vfs_inode_init(c, inode, &inode_u);
 271         journal_seq_copy(inode, journal_seq);
 272
 273         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 274         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 275
 276         /*
 277          * we must insert the new inode into the inode cache before calling
 278          * bch2_trans_exit() and dropping locks, else we could race with another
 279          * thread pulling the inode in and modifying it:
 280          */
 281
 282         old = to_bch_ei(insert_inode_locked2(&inode->v));
 283         if (unlikely(old)) {
 284                 /*
 285                  * We raced, another process pulled the new inode into cache
 286                  * before us:
 287                  */
 288                 journal_seq_copy(old, journal_seq);
 289                 make_bad_inode(&inode->v);
 290                 iput(&inode->v);
 291
 292                 inode = old;
 293         } else {
 294                 /*
 295                  * we really don't want insert_inode_locked2() to be setting
 296                  * I_NEW...
 297                  */
 298                 unlock_new_inode(&inode->v);
 299         }
 300
 301         bch2_trans_exit(&trans);
 302 err:
 303         posix_acl_release(default_acl);
 304         posix_acl_release(acl);
 305         return inode;
 306 err_trans:
 307         if (!tmpfile)
 308                 mutex_unlock(&dir->ei_update_lock);
 309
 310         bch2_trans_exit(&trans);
 311         make_bad_inode(&inode->v);
 312         iput(&inode->v);
 313         inode = ERR_PTR(ret);
 314         goto err;
 315 }
 316
 317 /* methods */
 318
 319 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 320                                   unsigned int flags)
 321 {
 322         struct bch_fs *c = vdir->i_sb->s_fs_info;
 323         struct bch_inode_info *dir = to_bch_ei(vdir);
 324         struct inode *vinode = NULL;
 325         u64 inum;
 326
 327         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 328                                   &dir->ei_str_hash,
 329                                   &dentry->d_name);
 330
 331         if (inum)
 332                 vinode = bch2_vfs_inode_get(c, inum);
 333
 334         return d_splice_alias(vinode, dentry);
 335 }
 336
 337 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 338                        umode_t mode, bool excl)
 339 {
 340         struct bch_inode_info *inode =
 341                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0, false);
 342
 343         if (IS_ERR(inode))
 344                 return PTR_ERR(inode);
 345
 346         d_instantiate(dentry, &inode->v);
 347         return 0;
 348 }
 349
 350 static int __bch2_link(struct bch_fs *c,
 351                        struct bch_inode_info *inode,
 352                        struct bch_inode_info *dir,
 353                        struct dentry *dentry)
 354 {
 355         struct btree_trans trans;
 356         struct bch_inode_unpacked inode_u;
 357         int ret;
 358
 359         mutex_lock(&inode->ei_update_lock);
 360         bch2_trans_init(&trans, c, 4, 1024);
 361
 362         do {
 363                 bch2_trans_begin(&trans);
 364                 ret   = bch2_link_trans(&trans,
 365                                         dir->v.i_ino,
 366                                         inode->v.i_ino, &inode_u,
 367                                         &dentry->d_name) ?:
 368                         bch2_trans_commit(&trans, NULL,
 369                                         &inode->ei_journal_seq,
 370                                         BTREE_INSERT_ATOMIC|
 371                                         BTREE_INSERT_NOUNLOCK);
 372         } while (ret == -EINTR);
 373
 374         if (likely(!ret))
 375                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 376
 377         bch2_trans_exit(&trans);
 378         mutex_unlock(&inode->ei_update_lock);
 379         return ret;
 380 }
 381
 382 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 383                      struct dentry *dentry)
 384 {
 385         struct bch_fs *c = vdir->i_sb->s_fs_info;
 386         struct bch_inode_info *dir = to_bch_ei(vdir);
 387         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 388         int ret;
 389
 390         lockdep_assert_held(&inode->v.i_rwsem);
 391
 392         ret = __bch2_link(c, inode, dir, dentry);
 393         if (unlikely(ret))
 394                 return ret;
 395
 396         ihold(&inode->v);
 397         d_instantiate(dentry, &inode->v);
 398         return 0;
 399 }
 400
 401 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 402 {
 403         struct bch_fs *c = vdir->i_sb->s_fs_info;
 404         struct bch_inode_info *dir = to_bch_ei(vdir);
 405         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 406         struct bch_inode_unpacked dir_u, inode_u;
 407         struct btree_trans trans;
 408         int ret;
 409
 410         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 411         bch2_trans_init(&trans, c, 4, 1024);
 412
 413         do {
 414                 bch2_trans_begin(&trans);
 415
 416                 ret   = bch2_unlink_trans(&trans,
 417                                           dir->v.i_ino, &dir_u,
 418                                           &inode_u, &dentry->d_name) ?:
 419                         bch2_trans_commit(&trans, NULL,
 420                                           &dir->ei_journal_seq,
 421                                           BTREE_INSERT_ATOMIC|
 422                                           BTREE_INSERT_NOUNLOCK|
 423                                           BTREE_INSERT_NOFAIL);
 424         } while (ret == -EINTR);
 425
 426         if (likely(!ret)) {
 427                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 428
 429                 journal_seq_copy(inode, dir->ei_journal_seq);
 430                 bch2_inode_update_after_write(c, dir, &dir_u,
 431                                               ATTR_MTIME|ATTR_CTIME);
 432                 bch2_inode_update_after_write(c, inode, &inode_u,
 433                                               ATTR_MTIME);
 434         }
 435
 436         bch2_trans_exit(&trans);
 437         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 438
 439         return ret;
 440 }
 441
 442 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 443                         const char *symname)
 444 {
 445         struct bch_fs *c = vdir->i_sb->s_fs_info;
 446         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 447         int ret;
 448
 449         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 450         if (unlikely(IS_ERR(inode)))
 451                 return PTR_ERR(inode);
 452
 453         inode_lock(&inode->v);
 454         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 455         inode_unlock(&inode->v);
 456
 457         if (unlikely(ret))
 458                 goto err;
 459
 460         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 461         if (unlikely(ret))
 462                 goto err;
 463
 464         journal_seq_copy(dir, inode->ei_journal_seq);
 465
 466         ret = __bch2_link(c, inode, dir, dentry);
 467         if (unlikely(ret))
 468                 goto err;
 469
 470         d_instantiate(dentry, &inode->v);
 471         return 0;
 472 err:
 473         iput(&inode->v);
 474         return ret;
 475 }
 476
 477 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 478 {
 479         struct bch_inode_info *inode =
 480                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFDIR, 0, false);
 481
 482         if (IS_ERR(inode))
 483                 return PTR_ERR(inode);
 484
 485         d_instantiate(dentry, &inode->v);
 486         return 0;
 487 }
 488
 489 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 490 {
 491         return bch2_unlink(vdir, dentry);
 492 }
 493
 494 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 495                       umode_t mode, dev_t rdev)
 496 {
 497         struct bch_inode_info *inode =
 498                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 499
 500         if (IS_ERR(inode))
 501                 return PTR_ERR(inode);
 502
 503         d_instantiate(dentry, &inode->v);
 504         return 0;
 505 }
 506
 507 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 508                         struct inode *dst_vdir, struct dentry *dst_dentry,
 509                         unsigned flags)
 510 {
 511         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 512         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 513         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 514         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 515         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 516         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 517         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 518         struct btree_trans trans;
 519         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 520                 ? BCH_RENAME_EXCHANGE
 521                 : dst_dentry->d_inode
 522                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 523         u64 journal_seq = 0;
 524         int ret;
 525
 526         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 527                 return -EINVAL;
 528
 529         if (mode == BCH_RENAME_OVERWRITE) {
 530                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 531                                                    0, LLONG_MAX);
 532                 if (ret)
 533                         return ret;
 534         }
 535
 536         bch2_trans_init(&trans, c, 8, 2048);
 537
 538         bch2_lock_inodes(INODE_UPDATE_LOCK,
 539                          src_dir,
 540                          dst_dir,
 541                          src_inode,
 542                          dst_inode);
 543
 544         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 545                 ret = bch2_fs_quota_transfer(c, src_inode,
 546                                              dst_dir->ei_qid,
 547                                              1 << QTYP_PRJ,
 548                                              KEY_TYPE_QUOTA_PREALLOC);
 549                 if (ret)
 550                         goto err;
 551         }
 552
 553         if (mode == BCH_RENAME_EXCHANGE &&
 554             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 555                 ret = bch2_fs_quota_transfer(c, dst_inode,
 556                                              src_dir->ei_qid,
 557                                              1 << QTYP_PRJ,
 558                                              KEY_TYPE_QUOTA_PREALLOC);
 559                 if (ret)
 560                         goto err;
 561         }
 562
 563 retry:
 564         bch2_trans_begin(&trans);
 565         ret   = bch2_rename_trans(&trans,
 566                                   src_dir->v.i_ino, &src_dir_u,
 567                                   dst_dir->v.i_ino, &dst_dir_u,
 568                                   &src_inode_u,
 569                                   &dst_inode_u,
 570                                   &src_dentry->d_name,
 571                                   &dst_dentry->d_name,
 572                                   mode) ?:
 573                 bch2_trans_commit(&trans, NULL,
 574                                   &journal_seq,
 575                                   BTREE_INSERT_ATOMIC|
 576                                   BTREE_INSERT_NOUNLOCK);
 577         if (ret == -EINTR)
 578                 goto retry;
 579         if (unlikely(ret))
 580                 goto err;
 581
 582         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 583         BUG_ON(dst_inode &&
 584                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 585
 586         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 587                                       ATTR_MTIME|ATTR_CTIME);
 588         journal_seq_copy(src_dir, journal_seq);
 589
 590         if (src_dir != dst_dir) {
 591                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 592                                               ATTR_MTIME|ATTR_CTIME);
 593                 journal_seq_copy(dst_dir, journal_seq);
 594         }
 595
 596         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 597                                       ATTR_CTIME);
 598         journal_seq_copy(src_inode, journal_seq);
 599
 600         if (dst_inode) {
 601                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 602                                               ATTR_CTIME);
 603                 journal_seq_copy(dst_inode, journal_seq);
 604         }
 605 err:
 606         bch2_trans_exit(&trans);
 607
 608         bch2_fs_quota_transfer(c, src_inode,
 609                                bch_qid(&src_inode->ei_inode),
 610                                1 << QTYP_PRJ,
 611                                KEY_TYPE_QUOTA_NOCHECK);
 612         if (dst_inode)
 613                 bch2_fs_quota_transfer(c, dst_inode,
 614                                        bch_qid(&dst_inode->ei_inode),
 615                                        1 << QTYP_PRJ,
 616                                        KEY_TYPE_QUOTA_NOCHECK);
 617
 618         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 619                            src_dir,
 620                            dst_dir,
 621                            src_inode,
 622                            dst_inode);
 623
 624         return ret;
 625 }
 626
 627 void bch2_setattr_copy(struct bch_inode_info *inode,
 628                        struct bch_inode_unpacked *bi,
 629                        struct iattr *attr)
 630 {
 631         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 632         unsigned int ia_valid = attr->ia_valid;
 633
 634         if (ia_valid & ATTR_UID)
 635                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
 636         if (ia_valid & ATTR_GID)
 637                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
 638
 639         if (ia_valid & ATTR_ATIME)
 640                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 641         if (ia_valid & ATTR_MTIME)
 642                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 643         if (ia_valid & ATTR_CTIME)
 644                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 645
 646         if (ia_valid & ATTR_MODE) {
 647                 umode_t mode = attr->ia_mode;
 648                 kgid_t gid = ia_valid & ATTR_GID
 649                         ? attr->ia_gid
 650                         : inode->v.i_gid;
 651
 652                 if (!in_group_p(gid) &&
 653                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 654                         mode &= ~S_ISGID;
 655                 bi->bi_mode = mode;
 656         }
 657 }
 658
 659 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
 660                                 struct iattr *attr)
 661 {
 662         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 663         struct bch_qid qid;
 664         struct btree_trans trans;
 665         struct btree_iter *inode_iter;
 666         struct bch_inode_unpacked inode_u;
 667         struct posix_acl *acl = NULL;
 668         int ret;
 669
 670         mutex_lock(&inode->ei_update_lock);
 671
 672         qid = inode->ei_qid;
 673
 674         if (attr->ia_valid & ATTR_UID)
 675                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 676
 677         if (attr->ia_valid & ATTR_GID)
 678                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 679
 680         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 681                                      KEY_TYPE_QUOTA_PREALLOC);
 682         if (ret)
 683                 goto err;
 684
 685         bch2_trans_init(&trans, c, 0, 0);
 686 retry:
 687         bch2_trans_begin(&trans);
 688         kfree(acl);
 689         acl = NULL;
 690
 691         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 692                                      BTREE_ITER_INTENT);
 693         ret = PTR_ERR_OR_ZERO(inode_iter);
 694         if (ret)
 695                 goto btree_err;
 696
 697         bch2_setattr_copy(inode, &inode_u, attr);
 698
 699         if (attr->ia_valid & ATTR_MODE) {
 700                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 701                 if (ret)
 702                         goto btree_err;
 703         }
 704
 705         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 706                 bch2_trans_commit(&trans, NULL,
 707                                   &inode->ei_journal_seq,
 708                                   BTREE_INSERT_ATOMIC|
 709                                   BTREE_INSERT_NOUNLOCK|
 710                                   BTREE_INSERT_NOFAIL);
 711 btree_err:
 712         if (ret == -EINTR)
 713                 goto retry;
 714         if (unlikely(ret))
 715                 goto err_trans;
 716
 717         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 718
 719         if (acl)
 720                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 721 err_trans:
 722         bch2_trans_exit(&trans);
 723 err:
 724         mutex_unlock(&inode->ei_update_lock);
 725
 726         return ret;
 727 }
 728
 729 static int bch2_getattr(const struct path *path, struct kstat *stat,
 730                         u32 request_mask, unsigned query_flags)
 731 {
 732         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 733         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 734
 735         stat->dev       = inode->v.i_sb->s_dev;
 736         stat->ino       = inode->v.i_ino;
 737         stat->mode      = inode->v.i_mode;
 738         stat->nlink     = inode->v.i_nlink;
 739         stat->uid       = inode->v.i_uid;
 740         stat->gid       = inode->v.i_gid;
 741         stat->rdev      = inode->v.i_rdev;
 742         stat->size      = i_size_read(&inode->v);
 743         stat->atime     = inode->v.i_atime;
 744         stat->mtime     = inode->v.i_mtime;
 745         stat->ctime     = inode->v.i_ctime;
 746         stat->blksize   = block_bytes(c);
 747         stat->blocks    = inode->v.i_blocks;
 748
 749         if (request_mask & STATX_BTIME) {
 750                 stat->result_mask |= STATX_BTIME;
 751                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 752         }
 753
 754         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 755                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 756         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 757                 stat->attributes |= STATX_ATTR_APPEND;
 758         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 759                 stat->attributes |= STATX_ATTR_NODUMP;
 760
 761         return 0;
 762 }
 763
 764 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 765 {
 766         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 767         int ret;
 768
 769         lockdep_assert_held(&inode->v.i_rwsem);
 770
 771         ret = setattr_prepare(dentry, iattr);
 772         if (ret)
 773                 return ret;
 774
 775         return iattr->ia_valid & ATTR_SIZE
 776                 ? bch2_truncate(inode, iattr)
 777                 : bch2_setattr_nonsize(inode, iattr);
 778 }
 779
 780 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
 781 {
 782         struct bch_inode_info *inode =
 783                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
 784
 785         if (IS_ERR(inode))
 786                 return PTR_ERR(inode);
 787
 788         d_mark_tmpfile(dentry, &inode->v);
 789         d_instantiate(dentry, &inode->v);
 790         return 0;
 791 }
 792
 793 static int bch2_fill_extent(struct bch_fs *c,
 794                             struct fiemap_extent_info *info,
 795                             struct bkey_s_c k, unsigned flags)
 796 {
 797         if (bkey_extent_is_data(k.k)) {
 798                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 799                 const union bch_extent_entry *entry;
 800                 struct extent_ptr_decoded p;
 801                 int ret;
 802
 803                 if (k.k->type == KEY_TYPE_reflink_v)
 804                         flags |= FIEMAP_EXTENT_SHARED;
 805
 806                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 807                         int flags2 = 0;
 808                         u64 offset = p.ptr.offset;
 809
 810                         if (p.crc.compression_type)
 811                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 812                         else
 813                                 offset += p.crc.offset;
 814
 815                         if ((offset & (c->opts.block_size - 1)) ||
 816                             (k.k->size & (c->opts.block_size - 1)))
 817                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 818
 819                         ret = fiemap_fill_next_extent(info,
 820                                                 bkey_start_offset(k.k) << 9,
 821                                                 offset << 9,
 822                                                 k.k->size << 9, flags|flags2);
 823                         if (ret)
 824                                 return ret;
 825                 }
 826
 827                 return 0;
 828         } else if (k.k->type == KEY_TYPE_reservation) {
 829                 return fiemap_fill_next_extent(info,
 830                                                bkey_start_offset(k.k) << 9,
 831                                                0, k.k->size << 9,
 832                                                flags|
 833                                                FIEMAP_EXTENT_DELALLOC|
 834                                                FIEMAP_EXTENT_UNWRITTEN);
 835         } else {
 836                 BUG();
 837         }
 838 }
 839
 840 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 841                        u64 start, u64 len)
 842 {
 843         struct bch_fs *c = vinode->i_sb->s_fs_info;
 844         struct bch_inode_info *ei = to_bch_ei(vinode);
 845         struct btree_trans trans;
 846         struct btree_iter *iter;
 847         struct bkey_s_c k;
 848         BKEY_PADDED(k) cur, prev;
 849         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 850         unsigned offset_into_extent, sectors;
 851         bool have_extent = false;
 852         int ret = 0;
 853
 854         if (start + len < start)
 855                 return -EINVAL;
 856
 857         bch2_trans_init(&trans, c, 0, 0);
 858
 859         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 860                                    POS(ei->v.i_ino, start >> 9), 0);
 861 retry:
 862         while ((k = bch2_btree_iter_peek(iter)).k &&
 863                !(ret = bkey_err(k)) &&
 864                bkey_cmp(iter->pos, end) < 0) {
 865                 if (!bkey_extent_is_data(k.k) &&
 866                     k.k->type != KEY_TYPE_reservation) {
 867                         bch2_btree_iter_next(iter);
 868                         continue;
 869                 }
 870
 871                 bkey_reassemble(&cur.k, k);
 872                 k = bkey_i_to_s_c(&cur.k);
 873
 874                 offset_into_extent      = iter->pos.offset -
 875                         bkey_start_offset(k.k);
 876                 sectors                 = k.k->size - offset_into_extent;
 877
 878                 ret = bch2_read_indirect_extent(&trans,
 879                                         &offset_into_extent, &cur.k);
 880                 if (ret)
 881                         break;
 882
 883                 sectors = min(sectors, k.k->size - offset_into_extent);
 884
 885                 if (offset_into_extent)
 886                         bch2_cut_front(POS(k.k->p.inode,
 887                                            bkey_start_offset(k.k) +
 888                                            offset_into_extent),
 889                                        &cur.k);
 890                 bch2_key_resize(&cur.k.k, sectors);
 891                 cur.k.k.p = iter->pos;
 892                 cur.k.k.p.offset += cur.k.k.size;
 893
 894                 if (have_extent) {
 895                         ret = bch2_fill_extent(c, info,
 896                                         bkey_i_to_s_c(&prev.k), 0);
 897                         if (ret)
 898                                 break;
 899                 }
 900
 901                 bkey_copy(&prev.k, &cur.k);
 902                 have_extent = true;
 903
 904                 if (k.k->type == KEY_TYPE_reflink_v)
 905                         bch2_btree_iter_set_pos(iter, k.k->p);
 906                 else
 907                         bch2_btree_iter_next(iter);
 908         }
 909
 910         if (ret == -EINTR)
 911                 goto retry;
 912
 913         if (!ret && have_extent)
 914                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(&prev.k),
 915                                        FIEMAP_EXTENT_LAST);
 916
 917         ret = bch2_trans_exit(&trans) ?: ret;
 918         return ret < 0 ? ret : 0;
 919 }
 920
 921 static const struct vm_operations_struct bch_vm_ops = {
 922         .fault          = filemap_fault,
 923         .map_pages      = filemap_map_pages,
 924         .page_mkwrite   = bch2_page_mkwrite,
 925 };
 926
 927 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 928 {
 929         file_accessed(file);
 930
 931         vma->vm_ops = &bch_vm_ops;
 932         return 0;
 933 }
 934
 935 /* Directories: */
 936
 937 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 938 {
 939         return generic_file_llseek_size(file, offset, whence,
 940                                         S64_MAX, S64_MAX);
 941 }
 942
 943 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 944 {
 945         struct bch_inode_info *inode = file_bch_inode(file);
 946         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 947
 948         if (!dir_emit_dots(file, ctx))
 949                 return 0;
 950
 951         return bch2_readdir(c, inode->v.i_ino, ctx);
 952 }
 953
 954 static const struct file_operations bch_file_operations = {
 955         .llseek         = bch2_llseek,
 956         .read_iter      = generic_file_read_iter,
 957         .write_iter     = bch2_write_iter,
 958         .mmap           = bch2_mmap,
 959         .open           = generic_file_open,
 960         .fsync          = bch2_fsync,
 961         .splice_read    = generic_file_splice_read,
 962         .splice_write   = iter_file_splice_write,
 963         .fallocate      = bch2_fallocate_dispatch,
 964         .unlocked_ioctl = bch2_fs_file_ioctl,
 965 #ifdef CONFIG_COMPAT
 966         .compat_ioctl   = bch2_compat_fs_ioctl,
 967 #endif
 968         .remap_file_range = bch2_remap_file_range,
 969 };
 970
 971 static const struct inode_operations bch_file_inode_operations = {
 972         .getattr        = bch2_getattr,
 973         .setattr        = bch2_setattr,
 974         .fiemap         = bch2_fiemap,
 975         .listxattr      = bch2_xattr_list,
 976 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 977         .get_acl        = bch2_get_acl,
 978         .set_acl        = bch2_set_acl,
 979 #endif
 980 };
 981
 982 static const struct inode_operations bch_dir_inode_operations = {
 983         .lookup         = bch2_lookup,
 984         .create         = bch2_create,
 985         .link           = bch2_link,
 986         .unlink         = bch2_unlink,
 987         .symlink        = bch2_symlink,
 988         .mkdir          = bch2_mkdir,
 989         .rmdir          = bch2_rmdir,
 990         .mknod          = bch2_mknod,
 991         .rename         = bch2_rename2,
 992         .getattr        = bch2_getattr,
 993         .setattr        = bch2_setattr,
 994         .tmpfile        = bch2_tmpfile,
 995         .listxattr      = bch2_xattr_list,
 996 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 997         .get_acl        = bch2_get_acl,
 998         .set_acl        = bch2_set_acl,
 999 #endif
1000 };
1001
1002 static const struct file_operations bch_dir_file_operations = {
1003         .llseek         = bch2_dir_llseek,
1004         .read           = generic_read_dir,
1005         .iterate        = bch2_vfs_readdir,
1006         .fsync          = bch2_fsync,
1007         .unlocked_ioctl = bch2_fs_file_ioctl,
1008 #ifdef CONFIG_COMPAT
1009         .compat_ioctl   = bch2_compat_fs_ioctl,
1010 #endif
1011 };
1012
1013 static const struct inode_operations bch_symlink_inode_operations = {
1014         .get_link       = page_get_link,
1015         .getattr        = bch2_getattr,
1016         .setattr        = bch2_setattr,
1017         .listxattr      = bch2_xattr_list,
1018 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1019         .get_acl        = bch2_get_acl,
1020         .set_acl        = bch2_set_acl,
1021 #endif
1022 };
1023
1024 static const struct inode_operations bch_special_inode_operations = {
1025         .getattr        = bch2_getattr,
1026         .setattr        = bch2_setattr,
1027         .listxattr      = bch2_xattr_list,
1028 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1029         .get_acl        = bch2_get_acl,
1030         .set_acl        = bch2_set_acl,
1031 #endif
1032 };
1033
1034 static const struct address_space_operations bch_address_space_operations = {
1035         .writepage      = bch2_writepage,
1036         .readpage       = bch2_readpage,
1037         .writepages     = bch2_writepages,
1038         .readpages      = bch2_readpages,
1039         .set_page_dirty = __set_page_dirty_nobuffers,
1040         .write_begin    = bch2_write_begin,
1041         .write_end      = bch2_write_end,
1042         .invalidatepage = bch2_invalidatepage,
1043         .releasepage    = bch2_releasepage,
1044         .direct_IO      = bch2_direct_IO,
1045 #ifdef CONFIG_MIGRATION
1046         .migratepage    = bch2_migrate_page,
1047 #endif
1048         .error_remove_page = generic_error_remove_page,
1049 };
1050
1051 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1052                 u64 ino, u32 generation)
1053 {
1054         struct bch_fs *c = sb->s_fs_info;
1055         struct inode *vinode;
1056
1057         if (ino < BCACHEFS_ROOT_INO)
1058                 return ERR_PTR(-ESTALE);
1059
1060         vinode = bch2_vfs_inode_get(c, ino);
1061         if (IS_ERR(vinode))
1062                 return ERR_CAST(vinode);
1063         if (generation && vinode->i_generation != generation) {
1064                 /* we didn't find the right inode.. */
1065                 iput(vinode);
1066                 return ERR_PTR(-ESTALE);
1067         }
1068         return vinode;
1069 }
1070
1071 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1072                 int fh_len, int fh_type)
1073 {
1074         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1075                                     bch2_nfs_get_inode);
1076 }
1077
1078 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1079                 int fh_len, int fh_type)
1080 {
1081         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1082                                     bch2_nfs_get_inode);
1083 }
1084
1085 static const struct export_operations bch_export_ops = {
1086         .fh_to_dentry   = bch2_fh_to_dentry,
1087         .fh_to_parent   = bch2_fh_to_parent,
1088         //.get_parent   = bch2_get_parent,
1089 };
1090
1091 static void bch2_vfs_inode_init(struct bch_fs *c,
1092                                 struct bch_inode_info *inode,
1093                                 struct bch_inode_unpacked *bi)
1094 {
1095         bch2_inode_update_after_write(c, inode, bi, ~0);
1096
1097         inode->v.i_blocks       = bi->bi_sectors;
1098         inode->v.i_ino          = bi->bi_inum;
1099         inode->v.i_rdev         = bi->bi_dev;
1100         inode->v.i_generation   = bi->bi_generation;
1101         inode->v.i_size         = bi->bi_size;
1102
1103         inode->ei_journal_seq   = 0;
1104         inode->ei_quota_reserved = 0;
1105         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1106         inode->ei_qid           = bch_qid(bi);
1107
1108         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1109
1110         switch (inode->v.i_mode & S_IFMT) {
1111         case S_IFREG:
1112                 inode->v.i_op   = &bch_file_inode_operations;
1113                 inode->v.i_fop  = &bch_file_operations;
1114                 break;
1115         case S_IFDIR:
1116                 inode->v.i_op   = &bch_dir_inode_operations;
1117                 inode->v.i_fop  = &bch_dir_file_operations;
1118                 break;
1119         case S_IFLNK:
1120                 inode_nohighmem(&inode->v);
1121                 inode->v.i_op   = &bch_symlink_inode_operations;
1122                 break;
1123         default:
1124                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1125                 inode->v.i_op   = &bch_special_inode_operations;
1126                 break;
1127         }
1128 }
1129
1130 static struct inode *bch2_alloc_inode(struct super_block *sb)
1131 {
1132         struct bch_inode_info *inode;
1133
1134         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1135         if (!inode)
1136                 return NULL;
1137
1138         inode_init_once(&inode->v);
1139         mutex_init(&inode->ei_update_lock);
1140         mutex_init(&inode->ei_quota_lock);
1141         inode->ei_journal_seq = 0;
1142
1143         return &inode->v;
1144 }
1145
1146 static void bch2_i_callback(struct rcu_head *head)
1147 {
1148         struct inode *vinode = container_of(head, struct inode, i_rcu);
1149         struct bch_inode_info *inode = to_bch_ei(vinode);
1150
1151         kmem_cache_free(bch2_inode_cache, inode);
1152 }
1153
1154 static void bch2_destroy_inode(struct inode *vinode)
1155 {
1156         call_rcu(&vinode->i_rcu, bch2_i_callback);
1157 }
1158
1159 static int inode_update_times_fn(struct bch_inode_info *inode,
1160                                  struct bch_inode_unpacked *bi,
1161                                  void *p)
1162 {
1163         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1164
1165         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1166         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1167         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1168
1169         return 0;
1170 }
1171
1172 static int bch2_vfs_write_inode(struct inode *vinode,
1173                                 struct writeback_control *wbc)
1174 {
1175         struct bch_fs *c = vinode->i_sb->s_fs_info;
1176         struct bch_inode_info *inode = to_bch_ei(vinode);
1177         int ret;
1178
1179         mutex_lock(&inode->ei_update_lock);
1180         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1181                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1182         mutex_unlock(&inode->ei_update_lock);
1183
1184         return ret;
1185 }
1186
1187 static void bch2_evict_inode(struct inode *vinode)
1188 {
1189         struct bch_fs *c = vinode->i_sb->s_fs_info;
1190         struct bch_inode_info *inode = to_bch_ei(vinode);
1191
1192         truncate_inode_pages_final(&inode->v.i_data);
1193
1194         clear_inode(&inode->v);
1195
1196         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1197
1198         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1199                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1200                                 KEY_TYPE_QUOTA_WARN);
1201                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1202                                 KEY_TYPE_QUOTA_WARN);
1203                 bch2_inode_rm(c, inode->v.i_ino);
1204         }
1205 }
1206
1207 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1208 {
1209         struct super_block *sb = dentry->d_sb;
1210         struct bch_fs *c = sb->s_fs_info;
1211         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1212         unsigned shift = sb->s_blocksize_bits - 9;
1213         u64 fsid;
1214
1215         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1216         buf->f_bsize    = sb->s_blocksize;
1217         buf->f_blocks   = usage.capacity >> shift;
1218         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1219         buf->f_bavail   = buf->f_bfree;
1220         buf->f_files    = usage.nr_inodes;
1221         buf->f_ffree    = U64_MAX;
1222
1223         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1224                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1225         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1226         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1227         buf->f_namelen  = BCH_NAME_MAX;
1228
1229         return 0;
1230 }
1231
1232 static int bch2_sync_fs(struct super_block *sb, int wait)
1233 {
1234         struct bch_fs *c = sb->s_fs_info;
1235
1236         if (c->opts.journal_flush_disabled)
1237                 return 0;
1238
1239         if (!wait) {
1240                 bch2_journal_flush_async(&c->journal, NULL);
1241                 return 0;
1242         }
1243
1244         return bch2_journal_flush(&c->journal);
1245 }
1246
1247 static struct bch_fs *bch2_path_to_fs(const char *dev)
1248 {
1249         struct bch_fs *c;
1250         struct block_device *bdev = lookup_bdev(dev);
1251
1252         if (IS_ERR(bdev))
1253                 return ERR_CAST(bdev);
1254
1255         c = bch2_bdev_to_fs(bdev);
1256         bdput(bdev);
1257         return c ?: ERR_PTR(-ENOENT);
1258 }
1259
1260 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1261                                                unsigned nr_devs, struct bch_opts opts)
1262 {
1263         struct bch_fs *c, *c1, *c2;
1264         size_t i;
1265
1266         if (!nr_devs)
1267                 return ERR_PTR(-EINVAL);
1268
1269         c = bch2_fs_open(devs, nr_devs, opts);
1270
1271         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1272                 /*
1273                  * Already open?
1274                  * Look up each block device, make sure they all belong to a
1275                  * filesystem and they all belong to the _same_ filesystem
1276                  */
1277
1278                 c1 = bch2_path_to_fs(devs[0]);
1279                 if (IS_ERR(c1))
1280                         return c;
1281
1282                 for (i = 1; i < nr_devs; i++) {
1283                         c2 = bch2_path_to_fs(devs[i]);
1284                         if (!IS_ERR(c2))
1285                                 closure_put(&c2->cl);
1286
1287                         if (c1 != c2) {
1288                                 closure_put(&c1->cl);
1289                                 return c;
1290                         }
1291                 }
1292
1293                 c = c1;
1294         }
1295
1296         if (IS_ERR(c))
1297                 return c;
1298
1299         mutex_lock(&c->state_lock);
1300
1301         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1302                 mutex_unlock(&c->state_lock);
1303                 closure_put(&c->cl);
1304                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1305                 return ERR_PTR(-EINVAL);
1306         }
1307
1308         mutex_unlock(&c->state_lock);
1309
1310         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1311         return c;
1312 }
1313
1314 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1315                                              struct bch_opts opts)
1316 {
1317         char *dev_name = NULL, **devs = NULL, *s;
1318         struct bch_fs *c = ERR_PTR(-ENOMEM);
1319         size_t i, nr_devs = 0;
1320
1321         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1322         if (!dev_name)
1323                 goto err;
1324
1325         for (s = dev_name; s; s = strchr(s + 1, ':'))
1326                 nr_devs++;
1327
1328         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1329         if (!devs)
1330                 goto err;
1331
1332         for (i = 0, s = dev_name;
1333              s;
1334              (s = strchr(s, ':')) && (*s++ = '\0'))
1335                 devs[i++] = s;
1336
1337         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1338 err:
1339         kfree(devs);
1340         kfree(dev_name);
1341         return c;
1342 }
1343
1344 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1345 {
1346         struct bch_fs *c = sb->s_fs_info;
1347         struct bch_opts opts = bch2_opts_empty();
1348         int ret;
1349
1350         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1351
1352         ret = bch2_parse_mount_opts(&opts, data);
1353         if (ret)
1354                 return ret;
1355
1356         if (opts.read_only != c->opts.read_only) {
1357                 mutex_lock(&c->state_lock);
1358
1359                 if (opts.read_only) {
1360                         bch2_fs_read_only(c);
1361
1362                         sb->s_flags |= SB_RDONLY;
1363                 } else {
1364                         ret = bch2_fs_read_write(c);
1365                         if (ret) {
1366                                 bch_err(c, "error going rw: %i", ret);
1367                                 mutex_unlock(&c->state_lock);
1368                                 return -EINVAL;
1369                         }
1370
1371                         sb->s_flags &= ~SB_RDONLY;
1372                 }
1373
1374                 c->opts.read_only = opts.read_only;
1375
1376                 mutex_unlock(&c->state_lock);
1377         }
1378
1379         if (opts.errors >= 0)
1380                 c->opts.errors = opts.errors;
1381
1382         return ret;
1383 }
1384
1385 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1386 {
1387         struct bch_fs *c = root->d_sb->s_fs_info;
1388         enum bch_opt_id i;
1389         char buf[512];
1390
1391         for (i = 0; i < bch2_opts_nr; i++) {
1392                 const struct bch_option *opt = &bch2_opt_table[i];
1393                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1394
1395                 if (!(opt->mode & OPT_MOUNT))
1396                         continue;
1397
1398                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1399                         continue;
1400
1401                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1402                                  OPT_SHOW_MOUNT_STYLE);
1403                 seq_putc(seq, ',');
1404                 seq_puts(seq, buf);
1405         }
1406
1407         return 0;
1408
1409 }
1410
1411 static const struct super_operations bch_super_operations = {
1412         .alloc_inode    = bch2_alloc_inode,
1413         .destroy_inode  = bch2_destroy_inode,
1414         .write_inode    = bch2_vfs_write_inode,
1415         .evict_inode    = bch2_evict_inode,
1416         .sync_fs        = bch2_sync_fs,
1417         .statfs         = bch2_statfs,
1418         .show_options   = bch2_show_options,
1419         .remount_fs     = bch2_remount,
1420 #if 0
1421         .put_super      = bch2_put_super,
1422         .freeze_fs      = bch2_freeze,
1423         .unfreeze_fs    = bch2_unfreeze,
1424 #endif
1425 };
1426
1427 static int bch2_test_super(struct super_block *s, void *data)
1428 {
1429         return s->s_fs_info == data;
1430 }
1431
1432 static int bch2_set_super(struct super_block *s, void *data)
1433 {
1434         s->s_fs_info = data;
1435         return 0;
1436 }
1437
1438 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1439                                  int flags, const char *dev_name, void *data)
1440 {
1441         struct bch_fs *c;
1442         struct bch_dev *ca;
1443         struct super_block *sb;
1444         struct inode *vinode;
1445         struct bch_opts opts = bch2_opts_empty();
1446         unsigned i;
1447         int ret;
1448
1449         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1450
1451         ret = bch2_parse_mount_opts(&opts, data);
1452         if (ret)
1453                 return ERR_PTR(ret);
1454
1455         c = bch2_open_as_blockdevs(dev_name, opts);
1456         if (IS_ERR(c))
1457                 return ERR_CAST(c);
1458
1459         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1460         if (IS_ERR(sb)) {
1461                 closure_put(&c->cl);
1462                 return ERR_CAST(sb);
1463         }
1464
1465         BUG_ON(sb->s_fs_info != c);
1466
1467         if (sb->s_root) {
1468                 closure_put(&c->cl);
1469
1470                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1471                         ret = -EBUSY;
1472                         goto err_put_super;
1473                 }
1474                 goto out;
1475         }
1476
1477         sb->s_blocksize         = block_bytes(c);
1478         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1479         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1480         sb->s_op                = &bch_super_operations;
1481         sb->s_export_op         = &bch_export_ops;
1482 #ifdef CONFIG_BCACHEFS_QUOTA
1483         sb->s_qcop              = &bch2_quotactl_operations;
1484         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1485 #endif
1486         sb->s_xattr             = bch2_xattr_handlers;
1487         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1488         sb->s_time_gran         = c->sb.time_precision;
1489         c->vfs_sb               = sb;
1490         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1491
1492         ret = super_setup_bdi(sb);
1493         if (ret)
1494                 goto err_put_super;
1495
1496         sb->s_bdi->congested_fn         = bch2_congested;
1497         sb->s_bdi->congested_data       = c;
1498         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1499
1500         for_each_online_member(ca, c, i) {
1501                 struct block_device *bdev = ca->disk_sb.bdev;
1502
1503                 /* XXX: create an anonymous device for multi device filesystems */
1504                 sb->s_bdev      = bdev;
1505                 sb->s_dev       = bdev->bd_dev;
1506                 percpu_ref_put(&ca->io_ref);
1507                 break;
1508         }
1509
1510 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1511         if (c->opts.acl)
1512                 sb->s_flags     |= SB_POSIXACL;
1513 #endif
1514
1515         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1516         if (IS_ERR(vinode)) {
1517                 bch_err(c, "error mounting: error getting root inode %i",
1518                         (int) PTR_ERR(vinode));
1519                 ret = PTR_ERR(vinode);
1520                 goto err_put_super;
1521         }
1522
1523         sb->s_root = d_make_root(vinode);
1524         if (!sb->s_root) {
1525                 bch_err(c, "error mounting: error allocating root dentry");
1526                 ret = -ENOMEM;
1527                 goto err_put_super;
1528         }
1529
1530         sb->s_flags |= SB_ACTIVE;
1531 out:
1532         return dget(sb->s_root);
1533
1534 err_put_super:
1535         deactivate_locked_super(sb);
1536         return ERR_PTR(ret);
1537 }
1538
1539 static void bch2_kill_sb(struct super_block *sb)
1540 {
1541         struct bch_fs *c = sb->s_fs_info;
1542
1543         generic_shutdown_super(sb);
1544
1545         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1546                 bch2_fs_stop(c);
1547         else
1548                 closure_put(&c->cl);
1549 }
1550
1551 static struct file_system_type bcache_fs_type = {
1552         .owner          = THIS_MODULE,
1553         .name           = "bcachefs",
1554         .mount          = bch2_mount,
1555         .kill_sb        = bch2_kill_sb,
1556         .fs_flags       = FS_REQUIRES_DEV,
1557 };
1558
1559 MODULE_ALIAS_FS("bcachefs");
1560
1561 void bch2_vfs_exit(void)
1562 {
1563         unregister_filesystem(&bcache_fs_type);
1564         if (bch2_inode_cache)
1565                 kmem_cache_destroy(bch2_inode_cache);
1566 }
1567
1568 int __init bch2_vfs_init(void)
1569 {
1570         int ret = -ENOMEM;
1571
1572         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1573         if (!bch2_inode_cache)
1574                 goto err;
1575
1576         ret = register_filesystem(&bcache_fs_type);
1577         if (ret)
1578                 goto err;
1579
1580         return 0;
1581 err:
1582         bch2_vfs_exit();
1583         return ret;
1584 }
1585
1586 #endif /* NO_BCACHEFS_FS */