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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "chardev.h"
   9 #include "dirent.h"
  10 #include "extents.h"
  11 #include "fs.h"
  12 #include "fs-io.h"
  13 #include "fs-ioctl.h"
  14 #include "fsck.h"
  15 #include "inode.h"
  16 #include "io.h"
  17 #include "journal.h"
  18 #include "keylist.h"
  19 #include "quota.h"
  20 #include "super.h"
  21 #include "xattr.h"
  22
  23 #include <linux/aio.h>
  24 #include <linux/backing-dev.h>
  25 #include <linux/exportfs.h>
  26 #include <linux/module.h>
  27 #include <linux/posix_acl.h>
  28 #include <linux/random.h>
  29 #include <linux/statfs.h>
  30 #include <linux/xattr.h>
  31
  32 static struct kmem_cache *bch2_inode_cache;
  33
  34 static void bch2_vfs_inode_init(struct bch_fs *,
  35                                 struct bch_inode_info *,
  36                                 struct bch_inode_unpacked *);
  37
  38 static void journal_seq_copy(struct bch_inode_info *dst,
  39                              u64 journal_seq)
  40 {
  41         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  42
  43         do {
  44                 old = v;
  45
  46                 if (old >= journal_seq)
  47                         break;
  48         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  49 }
  50
  51 /*
  52  * I_SIZE_DIRTY requires special handling:
  53  *
  54  * To the recovery code, the flag means that there is stale data past i_size
  55  * that needs to be deleted; it's used for implementing atomic appends and
  56  * truncates.
  57  *
  58  * On append, we set I_SIZE_DIRTY before doing the write, then after the write
  59  * we clear I_SIZE_DIRTY atomically with updating i_size to the new larger size
  60  * that exposes the data we just wrote.
  61  *
  62  * On truncate, it's the reverse: We set I_SIZE_DIRTY atomically with setting
  63  * i_size to the new smaller size, then we delete the data that we just made
  64  * invisible, and then we clear I_SIZE_DIRTY.
  65  *
  66  * Because there can be multiple appends in flight at a time, we need a refcount
  67  * (i_size_dirty_count) instead of manipulating the flag directly. Nonzero
  68  * refcount means I_SIZE_DIRTY is set, zero means it's cleared.
  69  *
  70  * Because write_inode() can be called at any time, i_size_dirty_count means
  71  * something different to the runtime code - it means to write_inode() "don't
  72  * update i_size yet".
  73  *
  74  * We don't clear I_SIZE_DIRTY directly, we let write_inode() clear it when
  75  * i_size_dirty_count is zero - but the reverse is not true, I_SIZE_DIRTY must
  76  * be set explicitly.
  77  */
  78
  79 void bch2_inode_update_after_write(struct bch_fs *c,
  80                                    struct bch_inode_info *inode,
  81                                    struct bch_inode_unpacked *bi,
  82                                    unsigned fields)
  83 {
  84         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
  85         i_uid_write(&inode->v, bi->bi_uid);
  86         i_gid_write(&inode->v, bi->bi_gid);
  87         inode->v.i_mode = bi->bi_mode;
  88
  89         if (fields & ATTR_ATIME)
  90                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
  91         if (fields & ATTR_MTIME)
  92                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
  93         if (fields & ATTR_CTIME)
  94                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
  95
  96         inode->ei_inode         = *bi;
  97
  98         bch2_inode_flags_to_vfs(inode);
  99 }
 100
 101 int __must_check bch2_write_inode_trans(struct btree_trans *trans,
 102                                 struct bch_inode_info *inode,
 103                                 struct bch_inode_unpacked *inode_u,
 104                                 inode_set_fn set,
 105                                 void *p)
 106 {
 107         struct btree_iter *iter = NULL;
 108         int ret = 0;
 109
 110         iter = bch2_inode_peek(trans, inode_u, inode->v.i_ino,
 111                                BTREE_ITER_INTENT);
 112         ret = PTR_ERR_OR_ZERO(iter);
 113         if (ret)
 114                 return ret;
 115
 116         ret = set ? set(inode, inode_u, p) : 0;
 117         if (ret)
 118                 return ret;
 119
 120         return bch2_inode_write(trans, iter, inode_u);
 121 }
 122
 123 int __must_check bch2_write_inode(struct bch_fs *c,
 124                                   struct bch_inode_info *inode,
 125                                   inode_set_fn set,
 126                                   void *p, unsigned fields)
 127 {
 128         struct btree_trans trans;
 129         struct bch_inode_unpacked inode_u;
 130         int ret;
 131
 132         bch2_trans_init(&trans, c, 0, 0);
 133 retry:
 134         bch2_trans_begin(&trans);
 135
 136         ret = bch2_write_inode_trans(&trans, inode, &inode_u, set, p) ?:
 137                 bch2_trans_commit(&trans, NULL,
 138                                   &inode->ei_journal_seq,
 139                                   BTREE_INSERT_ATOMIC|
 140                                   BTREE_INSERT_NOUNLOCK|
 141                                   BTREE_INSERT_NOFAIL);
 142         if (ret == -EINTR)
 143                 goto retry;
 144
 145         /*
 146          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 147          * this is important for inode updates via bchfs_write_index_update
 148          */
 149         if (!ret)
 150                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 151
 152         bch2_trans_exit(&trans);
 153         return ret < 0 ? ret : 0;
 154 }
 155
 156 int bch2_fs_quota_transfer(struct bch_fs *c,
 157                            struct bch_inode_info *inode,
 158                            struct bch_qid new_qid,
 159                            unsigned qtypes,
 160                            enum quota_acct_mode mode)
 161 {
 162         unsigned i;
 163         int ret;
 164
 165         qtypes &= enabled_qtypes(c);
 166
 167         for (i = 0; i < QTYP_NR; i++)
 168                 if (new_qid.q[i] == inode->ei_qid.q[i])
 169                         qtypes &= ~(1U << i);
 170
 171         if (!qtypes)
 172                 return 0;
 173
 174         mutex_lock(&inode->ei_quota_lock);
 175
 176         ret = bch2_quota_transfer(c, qtypes, new_qid,
 177                                   inode->ei_qid,
 178                                   inode->v.i_blocks +
 179                                   inode->ei_quota_reserved,
 180                                   mode);
 181         if (!ret)
 182                 for (i = 0; i < QTYP_NR; i++)
 183                         if (qtypes & (1 << i))
 184                                 inode->ei_qid.q[i] = new_qid.q[i];
 185
 186         mutex_unlock(&inode->ei_quota_lock);
 187
 188         return ret;
 189 }
 190
 191 int bch2_reinherit_attrs_fn(struct bch_inode_info *inode,
 192                             struct bch_inode_unpacked *bi,
 193                             void *p)
 194 {
 195         struct bch_inode_info *dir = p;
 196         u64 src, dst;
 197         unsigned id;
 198         int ret = 1;
 199
 200         for (id = 0; id < Inode_opt_nr; id++) {
 201                 if (bi->bi_fields_set & (1 << id))
 202                         continue;
 203
 204                 src = bch2_inode_opt_get(&dir->ei_inode, id);
 205                 dst = bch2_inode_opt_get(bi, id);
 206
 207                 if (src == dst)
 208                         continue;
 209
 210                 bch2_inode_opt_set(bi, id, src);
 211                 ret = 0;
 212         }
 213
 214         return ret;
 215 }
 216
 217 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 218 {
 219         struct bch_inode_unpacked inode_u;
 220         struct bch_inode_info *inode;
 221         int ret;
 222
 223         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 224         if (unlikely(!inode))
 225                 return ERR_PTR(-ENOMEM);
 226         if (!(inode->v.i_state & I_NEW))
 227                 return &inode->v;
 228
 229         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 230         if (ret) {
 231                 iget_failed(&inode->v);
 232                 return ERR_PTR(ret);
 233         }
 234
 235         bch2_vfs_inode_init(c, inode, &inode_u);
 236
 237         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 238
 239         unlock_new_inode(&inode->v);
 240
 241         return &inode->v;
 242 }
 243
 244 static void bch2_inode_init_owner(struct bch_inode_unpacked *inode_u,
 245                                   const struct inode *dir, umode_t mode)
 246 {
 247         kuid_t uid = current_fsuid();
 248         kgid_t gid;
 249
 250         if (dir && dir->i_mode & S_ISGID) {
 251                 gid = dir->i_gid;
 252                 if (S_ISDIR(mode))
 253                         mode |= S_ISGID;
 254         } else
 255                 gid = current_fsgid();
 256
 257         inode_u->bi_uid         = from_kuid(dir->i_sb->s_user_ns, uid);
 258         inode_u->bi_gid         = from_kgid(dir->i_sb->s_user_ns, gid);
 259         inode_u->bi_mode        = mode;
 260 }
 261
 262 static int inode_update_for_create_fn(struct bch_inode_info *inode,
 263                                       struct bch_inode_unpacked *bi,
 264                                       void *p)
 265 {
 266         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 267         struct bch_inode_unpacked *new_inode = p;
 268
 269         bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
 270
 271         if (S_ISDIR(new_inode->bi_mode))
 272                 bi->bi_nlink++;
 273
 274         return 0;
 275 }
 276
 277 static struct bch_inode_info *
 278 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 279               umode_t mode, dev_t rdev, bool tmpfile)
 280 {
 281         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 282         struct btree_trans trans;
 283         struct bch_inode_unpacked dir_u;
 284         struct bch_inode_info *inode, *old;
 285         struct bch_inode_unpacked inode_u;
 286         struct bch_hash_info hash_info;
 287         struct posix_acl *default_acl = NULL, *acl = NULL;
 288         u64 journal_seq = 0;
 289         int ret;
 290
 291         bch2_inode_init(c, &inode_u, 0, 0, 0, rdev, &dir->ei_inode);
 292         bch2_inode_init_owner(&inode_u, &dir->v, mode);
 293
 294         hash_info = bch2_hash_info_init(c, &inode_u);
 295
 296         if (tmpfile)
 297                 inode_u.bi_flags |= BCH_INODE_UNLINKED;
 298
 299         ret = bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 300                               KEY_TYPE_QUOTA_PREALLOC);
 301         if (ret)
 302                 return ERR_PTR(ret);
 303
 304 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 305         ret = posix_acl_create(&dir->v, &inode_u.bi_mode, &default_acl, &acl);
 306         if (ret)
 307                 goto err;
 308 #endif
 309
 310         /*
 311          * preallocate vfs inode before btree transaction, so that nothing can
 312          * fail after the transaction succeeds:
 313          */
 314         inode = to_bch_ei(new_inode(c->vfs_sb));
 315         if (unlikely(!inode)) {
 316                 ret = -ENOMEM;
 317                 goto err;
 318         }
 319
 320         if (!tmpfile)
 321                 mutex_lock(&dir->ei_update_lock);
 322
 323         bch2_trans_init(&trans, c, 8, 1024);
 324 retry:
 325         bch2_trans_begin(&trans);
 326
 327         ret   = __bch2_inode_create(&trans, &inode_u,
 328                                     BLOCKDEV_INODE_MAX, 0,
 329                                     &c->unused_inode_hint) ?:
 330                 (default_acl
 331                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 332                                       default_acl, ACL_TYPE_DEFAULT)
 333                  : 0) ?:
 334                 (acl
 335                  ? bch2_set_acl_trans(&trans, &inode_u, &hash_info,
 336                                       acl, ACL_TYPE_ACCESS)
 337                  : 0) ?:
 338                 (!tmpfile
 339                  ? __bch2_dirent_create(&trans, dir->v.i_ino,
 340                                         &dir->ei_str_hash,
 341                                         mode_to_type(mode),
 342                                         &dentry->d_name,
 343                                         inode_u.bi_inum,
 344                                         BCH_HASH_SET_MUST_CREATE)
 345                 : 0) ?:
 346                 (!tmpfile
 347                  ? bch2_write_inode_trans(&trans, dir, &dir_u,
 348                                           inode_update_for_create_fn,
 349                                           &inode_u)
 350                  : 0) ?:
 351                 bch2_trans_commit(&trans, NULL,
 352                                   &journal_seq,
 353                                   BTREE_INSERT_ATOMIC|
 354                                   BTREE_INSERT_NOUNLOCK);
 355         if (ret == -EINTR)
 356                 goto retry;
 357         if (unlikely(ret))
 358                 goto err_trans;
 359
 360         if (!tmpfile) {
 361                 bch2_inode_update_after_write(c, dir, &dir_u,
 362                                               ATTR_MTIME|ATTR_CTIME);
 363                 journal_seq_copy(dir, journal_seq);
 364                 mutex_unlock(&dir->ei_update_lock);
 365         }
 366
 367         bch2_vfs_inode_init(c, inode, &inode_u);
 368         journal_seq_copy(inode, journal_seq);
 369
 370         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 371         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 372
 373         /*
 374          * we must insert the new inode into the inode cache before calling
 375          * bch2_trans_exit() and dropping locks, else we could race with another
 376          * thread pulling the inode in and modifying it:
 377          */
 378
 379         old = to_bch_ei(insert_inode_locked2(&inode->v));
 380         if (unlikely(old)) {
 381                 /*
 382                  * We raced, another process pulled the new inode into cache
 383                  * before us:
 384                  */
 385                 old->ei_journal_seq = inode->ei_journal_seq;
 386                 make_bad_inode(&inode->v);
 387                 iput(&inode->v);
 388
 389                 inode = old;
 390         } else {
 391                 /*
 392                  * we really don't want insert_inode_locked2() to be setting
 393                  * I_NEW...
 394                  */
 395                 unlock_new_inode(&inode->v);
 396         }
 397
 398         bch2_trans_exit(&trans);
 399 out:
 400         posix_acl_release(default_acl);
 401         posix_acl_release(acl);
 402         return inode;
 403 err_trans:
 404         if (!tmpfile)
 405                 mutex_unlock(&dir->ei_update_lock);
 406
 407         bch2_trans_exit(&trans);
 408         make_bad_inode(&inode->v);
 409         iput(&inode->v);
 410 err:
 411         bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1, KEY_TYPE_QUOTA_WARN);
 412         inode = ERR_PTR(ret);
 413         goto out;
 414 }
 415
 416 /* methods */
 417
 418 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 419                                   unsigned int flags)
 420 {
 421         struct bch_fs *c = vdir->i_sb->s_fs_info;
 422         struct bch_inode_info *dir = to_bch_ei(vdir);
 423         struct inode *vinode = NULL;
 424         u64 inum;
 425
 426         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 427                                   &dir->ei_str_hash,
 428                                   &dentry->d_name);
 429
 430         if (inum)
 431                 vinode = bch2_vfs_inode_get(c, inum);
 432
 433         return d_splice_alias(vinode, dentry);
 434 }
 435
 436 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 437                        umode_t mode, bool excl)
 438 {
 439         struct bch_inode_info *inode =
 440                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0, false);
 441
 442         if (IS_ERR(inode))
 443                 return PTR_ERR(inode);
 444
 445         d_instantiate(dentry, &inode->v);
 446         return 0;
 447 }
 448
 449 static int __bch2_link(struct bch_fs *c,
 450                        struct bch_inode_info *inode,
 451                        struct bch_inode_info *dir,
 452                        struct dentry *dentry)
 453 {
 454         struct btree_trans trans;
 455         struct btree_iter *inode_iter;
 456         struct bch_inode_unpacked inode_u;
 457         int ret;
 458
 459         mutex_lock(&inode->ei_update_lock);
 460         bch2_trans_init(&trans, c, 4, 1024);
 461 retry:
 462         bch2_trans_begin(&trans);
 463         ret   = __bch2_dirent_create(&trans, dir->v.i_ino,
 464                                      &dir->ei_str_hash,
 465                                      mode_to_type(inode->v.i_mode),
 466                                      &dentry->d_name,
 467                                      inode->v.i_ino,
 468                                      BCH_HASH_SET_MUST_CREATE);
 469         if (ret)
 470                 goto err;
 471
 472         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 473                                      BTREE_ITER_INTENT);
 474         ret = PTR_ERR_OR_ZERO(inode_iter);
 475         if (ret)
 476                 goto err;
 477
 478         inode_u.bi_ctime = bch2_current_time(c);
 479         bch2_inode_nlink_inc(&inode_u);
 480
 481         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 482                 bch2_trans_commit(&trans, NULL,
 483                                 &inode->ei_journal_seq,
 484                                 BTREE_INSERT_ATOMIC|
 485                                 BTREE_INSERT_NOUNLOCK);
 486 err:
 487         if (ret == -EINTR)
 488                 goto retry;
 489
 490         if (likely(!ret))
 491                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 492
 493         bch2_trans_exit(&trans);
 494         mutex_unlock(&inode->ei_update_lock);
 495         return ret;
 496 }
 497
 498 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 499                      struct dentry *dentry)
 500 {
 501         struct bch_fs *c = vdir->i_sb->s_fs_info;
 502         struct bch_inode_info *dir = to_bch_ei(vdir);
 503         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 504         int ret;
 505
 506         lockdep_assert_held(&inode->v.i_rwsem);
 507
 508         ret = __bch2_link(c, inode, dir, dentry);
 509         if (unlikely(ret))
 510                 return ret;
 511
 512         ihold(&inode->v);
 513         d_instantiate(dentry, &inode->v);
 514         return 0;
 515 }
 516
 517 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 518 {
 519         struct bch_fs *c = vdir->i_sb->s_fs_info;
 520         struct bch_inode_info *dir = to_bch_ei(vdir);
 521         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 522         struct btree_iter *dir_iter, *inode_iter;
 523         struct bch_inode_unpacked dir_u, inode_u;
 524         struct btree_trans trans;
 525         int ret;
 526
 527         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 528         bch2_trans_init(&trans, c, 4, 1024);
 529 retry:
 530         bch2_trans_begin(&trans);
 531
 532         ret   = __bch2_dirent_delete(&trans, dir->v.i_ino,
 533                                      &dir->ei_str_hash,
 534                                      &dentry->d_name);
 535         if (ret)
 536                 goto btree_err;
 537
 538         dir_iter = bch2_inode_peek(&trans, &dir_u, dir->v.i_ino,
 539                                    BTREE_ITER_INTENT);
 540         ret = PTR_ERR_OR_ZERO(dir_iter);
 541         if (ret)
 542                 goto btree_err;
 543
 544         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 545                                      BTREE_ITER_INTENT);
 546         ret = PTR_ERR_OR_ZERO(inode_iter);
 547         if (ret)
 548                 goto btree_err;
 549
 550         dir_u.bi_mtime = dir_u.bi_ctime = inode_u.bi_ctime =
 551                 bch2_current_time(c);
 552
 553         dir_u.bi_nlink -= S_ISDIR(inode_u.bi_mode);
 554         bch2_inode_nlink_dec(&inode_u);
 555
 556         ret =   bch2_inode_write(&trans, dir_iter, &dir_u) ?:
 557                 bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 558                 bch2_trans_commit(&trans, NULL,
 559                                   &dir->ei_journal_seq,
 560                                   BTREE_INSERT_ATOMIC|
 561                                   BTREE_INSERT_NOUNLOCK|
 562                                   BTREE_INSERT_NOFAIL);
 563 btree_err:
 564         if (ret == -EINTR)
 565                 goto retry;
 566         if (ret)
 567                 goto err;
 568
 569         journal_seq_copy(inode, dir->ei_journal_seq);
 570
 571         bch2_inode_update_after_write(c, dir, &dir_u,
 572                                       ATTR_MTIME|ATTR_CTIME);
 573         bch2_inode_update_after_write(c, inode, &inode_u,
 574                                       ATTR_MTIME);
 575 err:
 576         bch2_trans_exit(&trans);
 577         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 578
 579         return ret;
 580 }
 581
 582 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 583                         const char *symname)
 584 {
 585         struct bch_fs *c = vdir->i_sb->s_fs_info;
 586         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 587         int ret;
 588
 589         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 590         if (unlikely(IS_ERR(inode)))
 591                 return PTR_ERR(inode);
 592
 593         inode_lock(&inode->v);
 594         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 595         inode_unlock(&inode->v);
 596
 597         if (unlikely(ret))
 598                 goto err;
 599
 600         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 601         if (unlikely(ret))
 602                 goto err;
 603
 604         journal_seq_copy(dir, inode->ei_journal_seq);
 605
 606         ret = __bch2_link(c, inode, dir, dentry);
 607         if (unlikely(ret))
 608                 goto err;
 609
 610         d_instantiate(dentry, &inode->v);
 611         return 0;
 612 err:
 613         iput(&inode->v);
 614         return ret;
 615 }
 616
 617 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 618 {
 619         struct bch_inode_info *inode =
 620                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFDIR, 0, false);
 621
 622         if (IS_ERR(inode))
 623                 return PTR_ERR(inode);
 624
 625         d_instantiate(dentry, &inode->v);
 626         return 0;
 627 }
 628
 629 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 630 {
 631         struct bch_fs *c = vdir->i_sb->s_fs_info;
 632
 633         if (bch2_empty_dir(c, dentry->d_inode->i_ino))
 634                 return -ENOTEMPTY;
 635
 636         return bch2_unlink(vdir, dentry);
 637 }
 638
 639 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 640                       umode_t mode, dev_t rdev)
 641 {
 642         struct bch_inode_info *inode =
 643                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 644
 645         if (IS_ERR(inode))
 646                 return PTR_ERR(inode);
 647
 648         d_instantiate(dentry, &inode->v);
 649         return 0;
 650 }
 651
 652 struct rename_info {
 653         u64                     now;
 654         struct bch_inode_info   *src_dir;
 655         struct bch_inode_info   *dst_dir;
 656         struct bch_inode_info   *src_inode;
 657         struct bch_inode_info   *dst_inode;
 658         enum bch_rename_mode    mode;
 659 };
 660
 661 static int inode_update_for_rename_fn(struct bch_inode_info *inode,
 662                                       struct bch_inode_unpacked *bi,
 663                                       void *p)
 664 {
 665         struct rename_info *info = p;
 666         int ret;
 667
 668         if (inode == info->src_dir) {
 669                 bi->bi_nlink -= S_ISDIR(info->src_inode->v.i_mode);
 670                 bi->bi_nlink += info->dst_inode &&
 671                         S_ISDIR(info->dst_inode->v.i_mode) &&
 672                         info->mode == BCH_RENAME_EXCHANGE;
 673         }
 674
 675         if (inode == info->dst_dir) {
 676                 bi->bi_nlink += S_ISDIR(info->src_inode->v.i_mode);
 677                 bi->bi_nlink -= info->dst_inode &&
 678                         S_ISDIR(info->dst_inode->v.i_mode);
 679         }
 680
 681         if (inode == info->src_inode) {
 682                 ret = bch2_reinherit_attrs_fn(inode, bi, info->dst_dir);
 683
 684                 BUG_ON(!ret && S_ISDIR(info->src_inode->v.i_mode));
 685         }
 686
 687         if (inode == info->dst_inode &&
 688             info->mode == BCH_RENAME_EXCHANGE) {
 689                 ret = bch2_reinherit_attrs_fn(inode, bi, info->src_dir);
 690
 691                 BUG_ON(!ret && S_ISDIR(info->dst_inode->v.i_mode));
 692         }
 693
 694         if (inode == info->dst_inode &&
 695             info->mode == BCH_RENAME_OVERWRITE) {
 696                 BUG_ON(bi->bi_nlink &&
 697                        S_ISDIR(info->dst_inode->v.i_mode));
 698
 699                 bch2_inode_nlink_dec(bi);
 700         }
 701
 702         if (inode == info->src_dir ||
 703             inode == info->dst_dir)
 704                 bi->bi_mtime = info->now;
 705         bi->bi_ctime = info->now;
 706
 707         return 0;
 708 }
 709
 710 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 711                         struct inode *dst_vdir, struct dentry *dst_dentry,
 712                         unsigned flags)
 713 {
 714         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 715         struct rename_info i = {
 716                 .src_dir        = to_bch_ei(src_vdir),
 717                 .dst_dir        = to_bch_ei(dst_vdir),
 718                 .src_inode      = to_bch_ei(src_dentry->d_inode),
 719                 .dst_inode      = to_bch_ei(dst_dentry->d_inode),
 720                 .mode           = flags & RENAME_EXCHANGE
 721                                 ? BCH_RENAME_EXCHANGE
 722                         : dst_dentry->d_inode
 723                                 ? BCH_RENAME_OVERWRITE : BCH_RENAME,
 724         };
 725         struct btree_trans trans;
 726         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 727         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 728         u64 journal_seq = 0;
 729         int ret;
 730
 731         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 732                 return -EINVAL;
 733
 734         if (i.mode == BCH_RENAME_OVERWRITE) {
 735                 if (S_ISDIR(i.src_inode->v.i_mode) !=
 736                     S_ISDIR(i.dst_inode->v.i_mode))
 737                         return -ENOTDIR;
 738
 739                 if (S_ISDIR(i.src_inode->v.i_mode) &&
 740                     bch2_empty_dir(c, i.dst_inode->v.i_ino))
 741                         return -ENOTEMPTY;
 742
 743                 ret = filemap_write_and_wait_range(i.src_inode->v.i_mapping,
 744                                                    0, LLONG_MAX);
 745                 if (ret)
 746                         return ret;
 747         }
 748
 749         bch2_trans_init(&trans, c, 8, 2048);
 750
 751         bch2_lock_inodes(INODE_UPDATE_LOCK,
 752                          i.src_dir,
 753                          i.dst_dir,
 754                          i.src_inode,
 755                          i.dst_inode);
 756
 757         if (S_ISDIR(i.src_inode->v.i_mode) &&
 758             inode_attrs_changing(i.dst_dir, i.src_inode)) {
 759                 ret = -EXDEV;
 760                 goto err;
 761         }
 762
 763         if (i.mode == BCH_RENAME_EXCHANGE &&
 764             S_ISDIR(i.dst_inode->v.i_mode) &&
 765             inode_attrs_changing(i.src_dir, i.dst_inode)) {
 766                 ret = -EXDEV;
 767                 goto err;
 768         }
 769
 770         if (inode_attr_changing(i.dst_dir, i.src_inode, Inode_opt_project)) {
 771                 ret = bch2_fs_quota_transfer(c, i.src_inode,
 772                                              i.dst_dir->ei_qid,
 773                                              1 << QTYP_PRJ,
 774                                              KEY_TYPE_QUOTA_PREALLOC);
 775                 if (ret)
 776                         goto err;
 777         }
 778
 779         if (i.mode == BCH_RENAME_EXCHANGE &&
 780             inode_attr_changing(i.src_dir, i.dst_inode, Inode_opt_project)) {
 781                 ret = bch2_fs_quota_transfer(c, i.dst_inode,
 782                                              i.src_dir->ei_qid,
 783                                              1 << QTYP_PRJ,
 784                                              KEY_TYPE_QUOTA_PREALLOC);
 785                 if (ret)
 786                         goto err;
 787         }
 788
 789 retry:
 790         bch2_trans_begin(&trans);
 791         i.now = bch2_current_time(c);
 792
 793         ret   = bch2_dirent_rename(&trans,
 794                                    i.src_dir, &src_dentry->d_name,
 795                                    i.dst_dir, &dst_dentry->d_name,
 796                                    i.mode) ?:
 797                 bch2_write_inode_trans(&trans, i.src_dir, &src_dir_u,
 798                                        inode_update_for_rename_fn, &i) ?:
 799                 (i.src_dir != i.dst_dir
 800                  ? bch2_write_inode_trans(&trans, i.dst_dir, &dst_dir_u,
 801                                        inode_update_for_rename_fn, &i)
 802                  : 0 ) ?:
 803                 bch2_write_inode_trans(&trans, i.src_inode, &src_inode_u,
 804                                        inode_update_for_rename_fn, &i) ?:
 805                 (i.dst_inode
 806                  ? bch2_write_inode_trans(&trans, i.dst_inode, &dst_inode_u,
 807                                        inode_update_for_rename_fn, &i)
 808                  : 0 ) ?:
 809                 bch2_trans_commit(&trans, NULL,
 810                                   &journal_seq,
 811                                   BTREE_INSERT_ATOMIC|
 812                                   BTREE_INSERT_NOUNLOCK);
 813         if (ret == -EINTR)
 814                 goto retry;
 815         if (unlikely(ret))
 816                 goto err;
 817
 818         bch2_inode_update_after_write(c, i.src_dir, &src_dir_u,
 819                                       ATTR_MTIME|ATTR_CTIME);
 820         journal_seq_copy(i.src_dir, journal_seq);
 821
 822         if (i.src_dir != i.dst_dir) {
 823                 bch2_inode_update_after_write(c, i.dst_dir, &dst_dir_u,
 824                                               ATTR_MTIME|ATTR_CTIME);
 825                 journal_seq_copy(i.dst_dir, journal_seq);
 826         }
 827
 828         journal_seq_copy(i.src_inode, journal_seq);
 829         if (i.dst_inode)
 830                 journal_seq_copy(i.dst_inode, journal_seq);
 831
 832         bch2_inode_update_after_write(c, i.src_inode, &src_inode_u,
 833                                       ATTR_CTIME);
 834         if (i.dst_inode)
 835                 bch2_inode_update_after_write(c, i.dst_inode, &dst_inode_u,
 836                                               ATTR_CTIME);
 837 err:
 838         bch2_trans_exit(&trans);
 839
 840         bch2_fs_quota_transfer(c, i.src_inode,
 841                                bch_qid(&i.src_inode->ei_inode),
 842                                1 << QTYP_PRJ,
 843                                KEY_TYPE_QUOTA_NOCHECK);
 844         if (i.dst_inode)
 845                 bch2_fs_quota_transfer(c, i.dst_inode,
 846                                        bch_qid(&i.dst_inode->ei_inode),
 847                                        1 << QTYP_PRJ,
 848                                        KEY_TYPE_QUOTA_NOCHECK);
 849
 850         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 851                            i.src_dir,
 852                            i.dst_dir,
 853                            i.src_inode,
 854                            i.dst_inode);
 855
 856         return ret;
 857 }
 858
 859 void bch2_setattr_copy(struct bch_inode_info *inode,
 860                        struct bch_inode_unpacked *bi,
 861                        struct iattr *attr)
 862 {
 863         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 864         unsigned int ia_valid = attr->ia_valid;
 865
 866         if (ia_valid & ATTR_UID)
 867                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
 868         if (ia_valid & ATTR_GID)
 869                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
 870
 871         if (ia_valid & ATTR_ATIME)
 872                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 873         if (ia_valid & ATTR_MTIME)
 874                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 875         if (ia_valid & ATTR_CTIME)
 876                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 877
 878         if (ia_valid & ATTR_MODE) {
 879                 umode_t mode = attr->ia_mode;
 880                 kgid_t gid = ia_valid & ATTR_GID
 881                         ? attr->ia_gid
 882                         : inode->v.i_gid;
 883
 884                 if (!in_group_p(gid) &&
 885                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 886                         mode &= ~S_ISGID;
 887                 bi->bi_mode = mode;
 888         }
 889 }
 890
 891 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
 892                                 struct iattr *attr)
 893 {
 894         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 895         struct bch_qid qid;
 896         struct btree_trans trans;
 897         struct btree_iter *inode_iter;
 898         struct bch_inode_unpacked inode_u;
 899         struct posix_acl *acl = NULL;
 900         int ret;
 901
 902         mutex_lock(&inode->ei_update_lock);
 903
 904         qid = inode->ei_qid;
 905
 906         if (attr->ia_valid & ATTR_UID)
 907                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 908
 909         if (attr->ia_valid & ATTR_GID)
 910                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 911
 912         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 913                                      KEY_TYPE_QUOTA_PREALLOC);
 914         if (ret)
 915                 goto err;
 916
 917         bch2_trans_init(&trans, c, 0, 0);
 918 retry:
 919         bch2_trans_begin(&trans);
 920         kfree(acl);
 921         acl = NULL;
 922
 923         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 924                                      BTREE_ITER_INTENT);
 925         ret = PTR_ERR_OR_ZERO(inode_iter);
 926         if (ret)
 927                 goto btree_err;
 928
 929         bch2_setattr_copy(inode, &inode_u, attr);
 930
 931         if (attr->ia_valid & ATTR_MODE) {
 932                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 933                 if (ret)
 934                         goto btree_err;
 935         }
 936
 937         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 938                 bch2_trans_commit(&trans, NULL,
 939                                   &inode->ei_journal_seq,
 940                                   BTREE_INSERT_ATOMIC|
 941                                   BTREE_INSERT_NOUNLOCK|
 942                                   BTREE_INSERT_NOFAIL);
 943 btree_err:
 944         if (ret == -EINTR)
 945                 goto retry;
 946         if (unlikely(ret))
 947                 goto err_trans;
 948
 949         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 950
 951         if (acl)
 952                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 953 err_trans:
 954         bch2_trans_exit(&trans);
 955 err:
 956         mutex_unlock(&inode->ei_update_lock);
 957
 958         return ret;
 959 }
 960
 961 static int bch2_getattr(const struct path *path, struct kstat *stat,
 962                         u32 request_mask, unsigned query_flags)
 963 {
 964         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 965         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 966
 967         stat->dev       = inode->v.i_sb->s_dev;
 968         stat->ino       = inode->v.i_ino;
 969         stat->mode      = inode->v.i_mode;
 970         stat->nlink     = inode->v.i_nlink;
 971         stat->uid       = inode->v.i_uid;
 972         stat->gid       = inode->v.i_gid;
 973         stat->rdev      = inode->v.i_rdev;
 974         stat->size      = i_size_read(&inode->v);
 975         stat->atime     = inode->v.i_atime;
 976         stat->mtime     = inode->v.i_mtime;
 977         stat->ctime     = inode->v.i_ctime;
 978         stat->blksize   = block_bytes(c);
 979         stat->blocks    = inode->v.i_blocks;
 980
 981         if (request_mask & STATX_BTIME) {
 982                 stat->result_mask |= STATX_BTIME;
 983                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 984         }
 985
 986         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 987                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 988         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 989                 stat->attributes |= STATX_ATTR_APPEND;
 990         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 991                 stat->attributes |= STATX_ATTR_NODUMP;
 992
 993         return 0;
 994 }
 995
 996 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 997 {
 998         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 999         int ret;
1000
1001         lockdep_assert_held(&inode->v.i_rwsem);
1002
1003         ret = setattr_prepare(dentry, iattr);
1004         if (ret)
1005                 return ret;
1006
1007         return iattr->ia_valid & ATTR_SIZE
1008                 ? bch2_truncate(inode, iattr)
1009                 : bch2_setattr_nonsize(inode, iattr);
1010 }
1011
1012 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
1013 {
1014         struct bch_inode_info *inode =
1015                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
1016
1017         if (IS_ERR(inode))
1018                 return PTR_ERR(inode);
1019
1020         d_mark_tmpfile(dentry, &inode->v);
1021         d_instantiate(dentry, &inode->v);
1022         return 0;
1023 }
1024
1025 static int bch2_fill_extent(struct bch_fs *c,
1026                             struct fiemap_extent_info *info,
1027                             struct bkey_s_c k, unsigned flags)
1028 {
1029         if (bkey_extent_is_data(k.k)) {
1030                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1031                 const union bch_extent_entry *entry;
1032                 struct extent_ptr_decoded p;
1033                 int ret;
1034
1035                 if (k.k->type == KEY_TYPE_reflink_v)
1036                         flags |= FIEMAP_EXTENT_SHARED;
1037
1038                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1039                         int flags2 = 0;
1040                         u64 offset = p.ptr.offset;
1041
1042                         if (p.crc.compression_type)
1043                                 flags2 |= FIEMAP_EXTENT_ENCODED;
1044                         else
1045                                 offset += p.crc.offset;
1046
1047                         if ((offset & (c->opts.block_size - 1)) ||
1048                             (k.k->size & (c->opts.block_size - 1)))
1049                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
1050
1051                         ret = fiemap_fill_next_extent(info,
1052                                                 bkey_start_offset(k.k) << 9,
1053                                                 offset << 9,
1054                                                 k.k->size << 9, flags|flags2);
1055                         if (ret)
1056                                 return ret;
1057                 }
1058
1059                 return 0;
1060         } else if (k.k->type == KEY_TYPE_reservation) {
1061                 return fiemap_fill_next_extent(info,
1062                                                bkey_start_offset(k.k) << 9,
1063                                                0, k.k->size << 9,
1064                                                flags|
1065                                                FIEMAP_EXTENT_DELALLOC|
1066                                                FIEMAP_EXTENT_UNWRITTEN);
1067         } else {
1068                 BUG();
1069         }
1070 }
1071
1072 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
1073                        u64 start, u64 len)
1074 {
1075         struct bch_fs *c = vinode->i_sb->s_fs_info;
1076         struct bch_inode_info *ei = to_bch_ei(vinode);
1077         struct btree_trans trans;
1078         struct btree_iter *iter;
1079         struct bkey_s_c k;
1080         BKEY_PADDED(k) cur, prev;
1081         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
1082         unsigned offset_into_extent, sectors;
1083         bool have_extent = false;
1084         int ret = 0;
1085
1086         if (start + len < start)
1087                 return -EINVAL;
1088
1089         bch2_trans_init(&trans, c, 0, 0);
1090
1091         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
1092                                    POS(ei->v.i_ino, start >> 9), 0);
1093 retry:
1094         while ((k = bch2_btree_iter_peek(iter)).k &&
1095                !(ret = bkey_err(k)) &&
1096                bkey_cmp(iter->pos, end) < 0) {
1097                 if (!bkey_extent_is_data(k.k) &&
1098                     k.k->type != KEY_TYPE_reservation) {
1099                         bch2_btree_iter_next(iter);
1100                         continue;
1101                 }
1102
1103                 bkey_reassemble(&cur.k, k);
1104                 k = bkey_i_to_s_c(&cur.k);
1105
1106                 offset_into_extent      = iter->pos.offset -
1107                         bkey_start_offset(k.k);
1108                 sectors                 = k.k->size - offset_into_extent;
1109
1110                 ret = bch2_read_indirect_extent(&trans,
1111                                         &offset_into_extent, &cur.k);
1112                 if (ret)
1113                         break;
1114
1115                 sectors = min(sectors, k.k->size - offset_into_extent);
1116
1117                 if (offset_into_extent)
1118                         bch2_cut_front(POS(k.k->p.inode,
1119                                            bkey_start_offset(k.k) +
1120                                            offset_into_extent),
1121                                        &cur.k);
1122                 bch2_key_resize(&cur.k.k, sectors);
1123                 cur.k.k.p = iter->pos;
1124                 cur.k.k.p.offset += cur.k.k.size;
1125
1126                 if (have_extent) {
1127                         ret = bch2_fill_extent(c, info,
1128                                         bkey_i_to_s_c(&prev.k), 0);
1129                         if (ret)
1130                                 break;
1131                 }
1132
1133                 bkey_copy(&prev.k, &cur.k);
1134                 have_extent = true;
1135
1136                 if (k.k->type == KEY_TYPE_reflink_v)
1137                         bch2_btree_iter_set_pos(iter, k.k->p);
1138                 else
1139                         bch2_btree_iter_next(iter);
1140         }
1141
1142         if (ret == -EINTR)
1143                 goto retry;
1144
1145         if (!ret && have_extent)
1146                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(&prev.k),
1147                                        FIEMAP_EXTENT_LAST);
1148
1149         ret = bch2_trans_exit(&trans) ?: ret;
1150         return ret < 0 ? ret : 0;
1151 }
1152
1153 static const struct vm_operations_struct bch_vm_ops = {
1154         .fault          = filemap_fault,
1155         .map_pages      = filemap_map_pages,
1156         .page_mkwrite   = bch2_page_mkwrite,
1157 };
1158
1159 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1160 {
1161         file_accessed(file);
1162
1163         vma->vm_ops = &bch_vm_ops;
1164         return 0;
1165 }
1166
1167 /* Directories: */
1168
1169 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1170 {
1171         return generic_file_llseek_size(file, offset, whence,
1172                                         S64_MAX, S64_MAX);
1173 }
1174
1175 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1176 {
1177         struct bch_fs *c = file_inode(file)->i_sb->s_fs_info;
1178
1179         return bch2_readdir(c, file, ctx);
1180 }
1181
1182 static const struct file_operations bch_file_operations = {
1183         .llseek         = bch2_llseek,
1184         .read_iter      = generic_file_read_iter,
1185         .write_iter     = bch2_write_iter,
1186         .mmap           = bch2_mmap,
1187         .open           = generic_file_open,
1188         .fsync          = bch2_fsync,
1189         .splice_read    = generic_file_splice_read,
1190         .splice_write   = iter_file_splice_write,
1191         .fallocate      = bch2_fallocate_dispatch,
1192         .unlocked_ioctl = bch2_fs_file_ioctl,
1193 #ifdef CONFIG_COMPAT
1194         .compat_ioctl   = bch2_compat_fs_ioctl,
1195 #endif
1196         .remap_file_range = bch2_remap_file_range,
1197 };
1198
1199 static const struct inode_operations bch_file_inode_operations = {
1200         .getattr        = bch2_getattr,
1201         .setattr        = bch2_setattr,
1202         .fiemap         = bch2_fiemap,
1203         .listxattr      = bch2_xattr_list,
1204 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1205         .get_acl        = bch2_get_acl,
1206         .set_acl        = bch2_set_acl,
1207 #endif
1208 };
1209
1210 static const struct inode_operations bch_dir_inode_operations = {
1211         .lookup         = bch2_lookup,
1212         .create         = bch2_create,
1213         .link           = bch2_link,
1214         .unlink         = bch2_unlink,
1215         .symlink        = bch2_symlink,
1216         .mkdir          = bch2_mkdir,
1217         .rmdir          = bch2_rmdir,
1218         .mknod          = bch2_mknod,
1219         .rename         = bch2_rename2,
1220         .getattr        = bch2_getattr,
1221         .setattr        = bch2_setattr,
1222         .tmpfile        = bch2_tmpfile,
1223         .listxattr      = bch2_xattr_list,
1224 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1225         .get_acl        = bch2_get_acl,
1226         .set_acl        = bch2_set_acl,
1227 #endif
1228 };
1229
1230 static const struct file_operations bch_dir_file_operations = {
1231         .llseek         = bch2_dir_llseek,
1232         .read           = generic_read_dir,
1233         .iterate        = bch2_vfs_readdir,
1234         .fsync          = bch2_fsync,
1235         .unlocked_ioctl = bch2_fs_file_ioctl,
1236 #ifdef CONFIG_COMPAT
1237         .compat_ioctl   = bch2_compat_fs_ioctl,
1238 #endif
1239 };
1240
1241 static const struct inode_operations bch_symlink_inode_operations = {
1242         .get_link       = page_get_link,
1243         .getattr        = bch2_getattr,
1244         .setattr        = bch2_setattr,
1245         .listxattr      = bch2_xattr_list,
1246 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1247         .get_acl        = bch2_get_acl,
1248         .set_acl        = bch2_set_acl,
1249 #endif
1250 };
1251
1252 static const struct inode_operations bch_special_inode_operations = {
1253         .getattr        = bch2_getattr,
1254         .setattr        = bch2_setattr,
1255         .listxattr      = bch2_xattr_list,
1256 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1257         .get_acl        = bch2_get_acl,
1258         .set_acl        = bch2_set_acl,
1259 #endif
1260 };
1261
1262 static const struct address_space_operations bch_address_space_operations = {
1263         .writepage      = bch2_writepage,
1264         .readpage       = bch2_readpage,
1265         .writepages     = bch2_writepages,
1266         .readpages      = bch2_readpages,
1267         .set_page_dirty = __set_page_dirty_nobuffers,
1268         .write_begin    = bch2_write_begin,
1269         .write_end      = bch2_write_end,
1270         .invalidatepage = bch2_invalidatepage,
1271         .releasepage    = bch2_releasepage,
1272         .direct_IO      = bch2_direct_IO,
1273 #ifdef CONFIG_MIGRATION
1274         .migratepage    = bch2_migrate_page,
1275 #endif
1276         .error_remove_page = generic_error_remove_page,
1277 };
1278
1279 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1280                 u64 ino, u32 generation)
1281 {
1282         struct bch_fs *c = sb->s_fs_info;
1283         struct inode *vinode;
1284
1285         if (ino < BCACHEFS_ROOT_INO)
1286                 return ERR_PTR(-ESTALE);
1287
1288         vinode = bch2_vfs_inode_get(c, ino);
1289         if (IS_ERR(vinode))
1290                 return ERR_CAST(vinode);
1291         if (generation && vinode->i_generation != generation) {
1292                 /* we didn't find the right inode.. */
1293                 iput(vinode);
1294                 return ERR_PTR(-ESTALE);
1295         }
1296         return vinode;
1297 }
1298
1299 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1300                 int fh_len, int fh_type)
1301 {
1302         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1303                                     bch2_nfs_get_inode);
1304 }
1305
1306 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1307                 int fh_len, int fh_type)
1308 {
1309         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1310                                     bch2_nfs_get_inode);
1311 }
1312
1313 static const struct export_operations bch_export_ops = {
1314         .fh_to_dentry   = bch2_fh_to_dentry,
1315         .fh_to_parent   = bch2_fh_to_parent,
1316         //.get_parent   = bch2_get_parent,
1317 };
1318
1319 static void bch2_vfs_inode_init(struct bch_fs *c,
1320                                 struct bch_inode_info *inode,
1321                                 struct bch_inode_unpacked *bi)
1322 {
1323         bch2_inode_update_after_write(c, inode, bi, ~0);
1324
1325         inode->v.i_blocks       = bi->bi_sectors;
1326         inode->v.i_ino          = bi->bi_inum;
1327         inode->v.i_rdev         = bi->bi_dev;
1328         inode->v.i_generation   = bi->bi_generation;
1329         inode->v.i_size         = bi->bi_size;
1330
1331         inode->ei_journal_seq   = 0;
1332         inode->ei_quota_reserved = 0;
1333         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1334         inode->ei_qid           = bch_qid(bi);
1335
1336         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1337
1338         switch (inode->v.i_mode & S_IFMT) {
1339         case S_IFREG:
1340                 inode->v.i_op   = &bch_file_inode_operations;
1341                 inode->v.i_fop  = &bch_file_operations;
1342                 break;
1343         case S_IFDIR:
1344                 inode->v.i_op   = &bch_dir_inode_operations;
1345                 inode->v.i_fop  = &bch_dir_file_operations;
1346                 break;
1347         case S_IFLNK:
1348                 inode_nohighmem(&inode->v);
1349                 inode->v.i_op   = &bch_symlink_inode_operations;
1350                 break;
1351         default:
1352                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1353                 inode->v.i_op   = &bch_special_inode_operations;
1354                 break;
1355         }
1356 }
1357
1358 static struct inode *bch2_alloc_inode(struct super_block *sb)
1359 {
1360         struct bch_inode_info *inode;
1361
1362         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1363         if (!inode)
1364                 return NULL;
1365
1366         inode_init_once(&inode->v);
1367         mutex_init(&inode->ei_update_lock);
1368         mutex_init(&inode->ei_quota_lock);
1369         inode->ei_journal_seq = 0;
1370
1371         return &inode->v;
1372 }
1373
1374 static void bch2_i_callback(struct rcu_head *head)
1375 {
1376         struct inode *vinode = container_of(head, struct inode, i_rcu);
1377         struct bch_inode_info *inode = to_bch_ei(vinode);
1378
1379         kmem_cache_free(bch2_inode_cache, inode);
1380 }
1381
1382 static void bch2_destroy_inode(struct inode *vinode)
1383 {
1384         call_rcu(&vinode->i_rcu, bch2_i_callback);
1385 }
1386
1387 static int inode_update_times_fn(struct bch_inode_info *inode,
1388                                  struct bch_inode_unpacked *bi,
1389                                  void *p)
1390 {
1391         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1392
1393         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1394         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1395         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1396
1397         return 0;
1398 }
1399
1400 static int bch2_vfs_write_inode(struct inode *vinode,
1401                                 struct writeback_control *wbc)
1402 {
1403         struct bch_fs *c = vinode->i_sb->s_fs_info;
1404         struct bch_inode_info *inode = to_bch_ei(vinode);
1405         int ret;
1406
1407         mutex_lock(&inode->ei_update_lock);
1408         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1409                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1410         mutex_unlock(&inode->ei_update_lock);
1411
1412         return ret;
1413 }
1414
1415 static void bch2_evict_inode(struct inode *vinode)
1416 {
1417         struct bch_fs *c = vinode->i_sb->s_fs_info;
1418         struct bch_inode_info *inode = to_bch_ei(vinode);
1419
1420         truncate_inode_pages_final(&inode->v.i_data);
1421
1422         clear_inode(&inode->v);
1423
1424         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1425
1426         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1427                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1428                                 KEY_TYPE_QUOTA_WARN);
1429                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1430                                 KEY_TYPE_QUOTA_WARN);
1431                 bch2_inode_rm(c, inode->v.i_ino);
1432         }
1433 }
1434
1435 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1436 {
1437         struct super_block *sb = dentry->d_sb;
1438         struct bch_fs *c = sb->s_fs_info;
1439         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1440         unsigned shift = sb->s_blocksize_bits - 9;
1441         u64 fsid;
1442
1443         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1444         buf->f_bsize    = sb->s_blocksize;
1445         buf->f_blocks   = usage.capacity >> shift;
1446         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1447         buf->f_bavail   = buf->f_bfree;
1448         buf->f_files    = usage.nr_inodes;
1449         buf->f_ffree    = U64_MAX;
1450
1451         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1452                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1453         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1454         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1455         buf->f_namelen  = BCH_NAME_MAX;
1456
1457         return 0;
1458 }
1459
1460 static int bch2_sync_fs(struct super_block *sb, int wait)
1461 {
1462         struct bch_fs *c = sb->s_fs_info;
1463
1464         if (c->opts.journal_flush_disabled)
1465                 return 0;
1466
1467         if (!wait) {
1468                 bch2_journal_flush_async(&c->journal, NULL);
1469                 return 0;
1470         }
1471
1472         return bch2_journal_flush(&c->journal);
1473 }
1474
1475 static struct bch_fs *bch2_path_to_fs(const char *dev)
1476 {
1477         struct bch_fs *c;
1478         struct block_device *bdev = lookup_bdev(dev);
1479
1480         if (IS_ERR(bdev))
1481                 return ERR_CAST(bdev);
1482
1483         c = bch2_bdev_to_fs(bdev);
1484         bdput(bdev);
1485         return c ?: ERR_PTR(-ENOENT);
1486 }
1487
1488 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1489                                                unsigned nr_devs, struct bch_opts opts)
1490 {
1491         struct bch_fs *c, *c1, *c2;
1492         size_t i;
1493
1494         if (!nr_devs)
1495                 return ERR_PTR(-EINVAL);
1496
1497         c = bch2_fs_open(devs, nr_devs, opts);
1498
1499         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1500                 /*
1501                  * Already open?
1502                  * Look up each block device, make sure they all belong to a
1503                  * filesystem and they all belong to the _same_ filesystem
1504                  */
1505
1506                 c1 = bch2_path_to_fs(devs[0]);
1507                 if (IS_ERR(c1))
1508                         return c;
1509
1510                 for (i = 1; i < nr_devs; i++) {
1511                         c2 = bch2_path_to_fs(devs[i]);
1512                         if (!IS_ERR(c2))
1513                                 closure_put(&c2->cl);
1514
1515                         if (c1 != c2) {
1516                                 closure_put(&c1->cl);
1517                                 return c;
1518                         }
1519                 }
1520
1521                 c = c1;
1522         }
1523
1524         if (IS_ERR(c))
1525                 return c;
1526
1527         mutex_lock(&c->state_lock);
1528
1529         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1530                 mutex_unlock(&c->state_lock);
1531                 closure_put(&c->cl);
1532                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1533                 return ERR_PTR(-EINVAL);
1534         }
1535
1536         mutex_unlock(&c->state_lock);
1537
1538         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1539         return c;
1540 }
1541
1542 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1543                                              struct bch_opts opts)
1544 {
1545         char *dev_name = NULL, **devs = NULL, *s;
1546         struct bch_fs *c = ERR_PTR(-ENOMEM);
1547         size_t i, nr_devs = 0;
1548
1549         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1550         if (!dev_name)
1551                 goto err;
1552
1553         for (s = dev_name; s; s = strchr(s + 1, ':'))
1554                 nr_devs++;
1555
1556         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1557         if (!devs)
1558                 goto err;
1559
1560         for (i = 0, s = dev_name;
1561              s;
1562              (s = strchr(s, ':')) && (*s++ = '\0'))
1563                 devs[i++] = s;
1564
1565         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1566 err:
1567         kfree(devs);
1568         kfree(dev_name);
1569         return c;
1570 }
1571
1572 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1573 {
1574         struct bch_fs *c = sb->s_fs_info;
1575         struct bch_opts opts = bch2_opts_empty();
1576         int ret;
1577
1578         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1579
1580         ret = bch2_parse_mount_opts(&opts, data);
1581         if (ret)
1582                 return ret;
1583
1584         if (opts.read_only != c->opts.read_only) {
1585                 mutex_lock(&c->state_lock);
1586
1587                 if (opts.read_only) {
1588                         bch2_fs_read_only(c);
1589
1590                         sb->s_flags |= SB_RDONLY;
1591                 } else {
1592                         ret = bch2_fs_read_write(c);
1593                         if (ret) {
1594                                 bch_err(c, "error going rw: %i", ret);
1595                                 mutex_unlock(&c->state_lock);
1596                                 return -EINVAL;
1597                         }
1598
1599                         sb->s_flags &= ~SB_RDONLY;
1600                 }
1601
1602                 c->opts.read_only = opts.read_only;
1603
1604                 mutex_unlock(&c->state_lock);
1605         }
1606
1607         if (opts.errors >= 0)
1608                 c->opts.errors = opts.errors;
1609
1610         return ret;
1611 }
1612
1613 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1614 {
1615         struct bch_fs *c = root->d_sb->s_fs_info;
1616         enum bch_opt_id i;
1617         char buf[512];
1618
1619         for (i = 0; i < bch2_opts_nr; i++) {
1620                 const struct bch_option *opt = &bch2_opt_table[i];
1621                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1622
1623                 if (!(opt->mode & OPT_MOUNT))
1624                         continue;
1625
1626                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1627                         continue;
1628
1629                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1630                                  OPT_SHOW_MOUNT_STYLE);
1631                 seq_putc(seq, ',');
1632                 seq_puts(seq, buf);
1633         }
1634
1635         return 0;
1636
1637 }
1638
1639 static const struct super_operations bch_super_operations = {
1640         .alloc_inode    = bch2_alloc_inode,
1641         .destroy_inode  = bch2_destroy_inode,
1642         .write_inode    = bch2_vfs_write_inode,
1643         .evict_inode    = bch2_evict_inode,
1644         .sync_fs        = bch2_sync_fs,
1645         .statfs         = bch2_statfs,
1646         .show_options   = bch2_show_options,
1647         .remount_fs     = bch2_remount,
1648 #if 0
1649         .put_super      = bch2_put_super,
1650         .freeze_fs      = bch2_freeze,
1651         .unfreeze_fs    = bch2_unfreeze,
1652 #endif
1653 };
1654
1655 static int bch2_test_super(struct super_block *s, void *data)
1656 {
1657         return s->s_fs_info == data;
1658 }
1659
1660 static int bch2_set_super(struct super_block *s, void *data)
1661 {
1662         s->s_fs_info = data;
1663         return 0;
1664 }
1665
1666 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1667                                  int flags, const char *dev_name, void *data)
1668 {
1669         struct bch_fs *c;
1670         struct bch_dev *ca;
1671         struct super_block *sb;
1672         struct inode *vinode;
1673         struct bch_opts opts = bch2_opts_empty();
1674         unsigned i;
1675         int ret;
1676
1677         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1678
1679         ret = bch2_parse_mount_opts(&opts, data);
1680         if (ret)
1681                 return ERR_PTR(ret);
1682
1683         c = bch2_open_as_blockdevs(dev_name, opts);
1684         if (IS_ERR(c))
1685                 return ERR_CAST(c);
1686
1687         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1688         if (IS_ERR(sb)) {
1689                 closure_put(&c->cl);
1690                 return ERR_CAST(sb);
1691         }
1692
1693         BUG_ON(sb->s_fs_info != c);
1694
1695         if (sb->s_root) {
1696                 closure_put(&c->cl);
1697
1698                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1699                         ret = -EBUSY;
1700                         goto err_put_super;
1701                 }
1702                 goto out;
1703         }
1704
1705         sb->s_blocksize         = block_bytes(c);
1706         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1707         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1708         sb->s_op                = &bch_super_operations;
1709         sb->s_export_op         = &bch_export_ops;
1710 #ifdef CONFIG_BCACHEFS_QUOTA
1711         sb->s_qcop              = &bch2_quotactl_operations;
1712         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1713 #endif
1714         sb->s_xattr             = bch2_xattr_handlers;
1715         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1716         sb->s_time_gran         = c->sb.time_precision;
1717         c->vfs_sb               = sb;
1718         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1719
1720         ret = super_setup_bdi(sb);
1721         if (ret)
1722                 goto err_put_super;
1723
1724         sb->s_bdi->congested_fn         = bch2_congested;
1725         sb->s_bdi->congested_data       = c;
1726         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1727
1728         for_each_online_member(ca, c, i) {
1729                 struct block_device *bdev = ca->disk_sb.bdev;
1730
1731                 /* XXX: create an anonymous device for multi device filesystems */
1732                 sb->s_bdev      = bdev;
1733                 sb->s_dev       = bdev->bd_dev;
1734                 percpu_ref_put(&ca->io_ref);
1735                 break;
1736         }
1737
1738 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1739         if (c->opts.acl)
1740                 sb->s_flags     |= SB_POSIXACL;
1741 #endif
1742
1743         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1744         if (IS_ERR(vinode)) {
1745                 bch_err(c, "error mounting: error getting root inode %i",
1746                         (int) PTR_ERR(vinode));
1747                 ret = PTR_ERR(vinode);
1748                 goto err_put_super;
1749         }
1750
1751         sb->s_root = d_make_root(vinode);
1752         if (!sb->s_root) {
1753                 bch_err(c, "error mounting: error allocating root dentry");
1754                 ret = -ENOMEM;
1755                 goto err_put_super;
1756         }
1757
1758         sb->s_flags |= SB_ACTIVE;
1759 out:
1760         return dget(sb->s_root);
1761
1762 err_put_super:
1763         deactivate_locked_super(sb);
1764         return ERR_PTR(ret);
1765 }
1766
1767 static void bch2_kill_sb(struct super_block *sb)
1768 {
1769         struct bch_fs *c = sb->s_fs_info;
1770
1771         generic_shutdown_super(sb);
1772
1773         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1774                 bch2_fs_stop(c);
1775         else
1776                 closure_put(&c->cl);
1777 }
1778
1779 static struct file_system_type bcache_fs_type = {
1780         .owner          = THIS_MODULE,
1781         .name           = "bcachefs",
1782         .mount          = bch2_mount,
1783         .kill_sb        = bch2_kill_sb,
1784         .fs_flags       = FS_REQUIRES_DEV,
1785 };
1786
1787 MODULE_ALIAS_FS("bcachefs");
1788
1789 void bch2_vfs_exit(void)
1790 {
1791         unregister_filesystem(&bcache_fs_type);
1792         if (bch2_inode_cache)
1793                 kmem_cache_destroy(bch2_inode_cache);
1794 }
1795
1796 int __init bch2_vfs_init(void)
1797 {
1798         int ret = -ENOMEM;
1799
1800         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1801         if (!bch2_inode_cache)
1802                 goto err;
1803
1804         ret = register_filesystem(&bcache_fs_type);
1805         if (ret)
1806                 goto err;
1807
1808         return 0;
1809 err:
1810         bch2_vfs_exit();
1811         return ret;
1812 }
1813
1814 #endif /* NO_BCACHEFS_FS */