fs/bcachefs/fs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "bkey_buf.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "chardev.h"
  10 #include "dirent.h"
  11 #include "extents.h"
  12 #include "fs.h"
  13 #include "fs-common.h"
  14 #include "fs-io.h"
  15 #include "fs-ioctl.h"
  16 #include "fsck.h"
  17 #include "inode.h"
  18 #include "io.h"
  19 #include "journal.h"
  20 #include "keylist.h"
  21 #include "quota.h"
  22 #include "super.h"
  23 #include "xattr.h"
  24
  25 #include <linux/aio.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/exportfs.h>
  28 #include <linux/fiemap.h>
  29 #include <linux/module.h>
  30 #include <linux/pagemap.h>
  31 #include <linux/posix_acl.h>
  32 #include <linux/random.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/statfs.h>
  35 #include <linux/xattr.h>
  36
  37 static struct kmem_cache *bch2_inode_cache;
  38
  39 static void bch2_vfs_inode_init(struct bch_fs *,
  40                                 struct bch_inode_info *,
  41                                 struct bch_inode_unpacked *);
  42
  43 static void journal_seq_copy(struct bch_fs *c,
  44                              struct bch_inode_info *dst,
  45                              u64 journal_seq)
  46 {
  47         /*
  48          * atomic64_cmpxchg has a fallback for archs that don't support it,
  49          * cmpxchg does not:
  50          */
  51         atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
  52         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  53
  54         do {
  55                 old = v;
  56
  57                 if (old >= journal_seq)
  58                         break;
  59         } while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
  60
  61         bch2_journal_set_has_inum(&c->journal, dst->v.i_ino, journal_seq);
  62 }
  63
  64 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
  65 {
  66         BUG_ON(atomic_long_read(&lock->v) == 0);
  67
  68         if (atomic_long_sub_return_release(i, &lock->v) == 0)
  69                 wake_up_all(&lock->wait);
  70 }
  71
  72 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
  73 {
  74         long v = atomic_long_read(&lock->v), old;
  75
  76         do {
  77                 old = v;
  78
  79                 if (i > 0 ? v < 0 : v > 0)
  80                         return false;
  81         } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
  82                                         old, old + i)) != old);
  83         return true;
  84 }
  85
  86 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
  87 {
  88         wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
  89 }
  90
  91 void bch2_pagecache_add_put(struct pagecache_lock *lock)
  92 {
  93         __pagecache_lock_put(lock, 1);
  94 }
  95
  96 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
  97 {
  98         return __pagecache_lock_tryget(lock, 1);
  99 }
 100
 101 void bch2_pagecache_add_get(struct pagecache_lock *lock)
 102 {
 103         __pagecache_lock_get(lock, 1);
 104 }
 105
 106 void bch2_pagecache_block_put(struct pagecache_lock *lock)
 107 {
 108         __pagecache_lock_put(lock, -1);
 109 }
 110
 111 void bch2_pagecache_block_get(struct pagecache_lock *lock)
 112 {
 113         __pagecache_lock_get(lock, -1);
 114 }
 115
 116 void bch2_inode_update_after_write(struct bch_fs *c,
 117                                    struct bch_inode_info *inode,
 118                                    struct bch_inode_unpacked *bi,
 119                                    unsigned fields)
 120 {
 121         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
 122         i_uid_write(&inode->v, bi->bi_uid);
 123         i_gid_write(&inode->v, bi->bi_gid);
 124         inode->v.i_mode = bi->bi_mode;
 125
 126         if (fields & ATTR_ATIME)
 127                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
 128         if (fields & ATTR_MTIME)
 129                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
 130         if (fields & ATTR_CTIME)
 131                 inode_set_ctime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_ctime));
 132
 133         inode->ei_inode         = *bi;
 134
 135         bch2_inode_flags_to_vfs(inode);
 136 }
 137
 138 int __must_check bch2_write_inode(struct bch_fs *c,
 139                                   struct bch_inode_info *inode,
 140                                   inode_set_fn set,
 141                                   void *p, unsigned fields)
 142 {
 143         struct btree_trans trans;
 144         struct btree_iter *iter;
 145         struct bch_inode_unpacked inode_u;
 146         int ret;
 147
 148         bch2_trans_init(&trans, c, 0, 0);
 149 retry:
 150         bch2_trans_begin(&trans);
 151
 152         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 153                                BTREE_ITER_INTENT);
 154         ret   = PTR_ERR_OR_ZERO(iter) ?:
 155                 (set ? set(inode, &inode_u, p) : 0) ?:
 156                 bch2_inode_write(&trans, iter, &inode_u) ?:
 157                 bch2_trans_commit(&trans, NULL,
 158                                   &inode->ei_journal_seq,
 159                                   BTREE_INSERT_NOUNLOCK|
 160                                   BTREE_INSERT_NOFAIL);
 161
 162         /*
 163          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 164          * this is important for inode updates via bchfs_write_index_update
 165          */
 166         if (!ret)
 167                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 168
 169         bch2_trans_iter_put(&trans, iter);
 170
 171         if (ret == -EINTR)
 172                 goto retry;
 173
 174         bch2_trans_exit(&trans);
 175         return ret < 0 ? ret : 0;
 176 }
 177
 178 int bch2_fs_quota_transfer(struct bch_fs *c,
 179                            struct bch_inode_info *inode,
 180                            struct bch_qid new_qid,
 181                            unsigned qtypes,
 182                            enum quota_acct_mode mode)
 183 {
 184         unsigned i;
 185         int ret;
 186
 187         qtypes &= enabled_qtypes(c);
 188
 189         for (i = 0; i < QTYP_NR; i++)
 190                 if (new_qid.q[i] == inode->ei_qid.q[i])
 191                         qtypes &= ~(1U << i);
 192
 193         if (!qtypes)
 194                 return 0;
 195
 196         mutex_lock(&inode->ei_quota_lock);
 197
 198         ret = bch2_quota_transfer(c, qtypes, new_qid,
 199                                   inode->ei_qid,
 200                                   inode->v.i_blocks +
 201                                   inode->ei_quota_reserved,
 202                                   mode);
 203         if (!ret)
 204                 for (i = 0; i < QTYP_NR; i++)
 205                         if (qtypes & (1 << i))
 206                                 inode->ei_qid.q[i] = new_qid.q[i];
 207
 208         mutex_unlock(&inode->ei_quota_lock);
 209
 210         return ret;
 211 }
 212
 213 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 214 {
 215         struct bch_inode_unpacked inode_u;
 216         struct bch_inode_info *inode;
 217         int ret;
 218
 219         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 220         if (unlikely(!inode))
 221                 return ERR_PTR(-ENOMEM);
 222         if (!(inode->v.i_state & I_NEW))
 223                 return &inode->v;
 224
 225         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 226         if (ret) {
 227                 iget_failed(&inode->v);
 228                 return ERR_PTR(ret);
 229         }
 230
 231         bch2_vfs_inode_init(c, inode, &inode_u);
 232
 233         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 234
 235         unlock_new_inode(&inode->v);
 236
 237         return &inode->v;
 238 }
 239
 240 static int inum_test(struct inode *inode, void *p)
 241 {
 242         unsigned long *ino = p;
 243
 244         return *ino == inode->i_ino;
 245 }
 246
 247 static struct bch_inode_info *
 248 __bch2_create(struct mnt_idmap *idmap,
 249               struct bch_inode_info *dir, struct dentry *dentry,
 250               umode_t mode, dev_t rdev, bool tmpfile)
 251 {
 252         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 253         struct btree_trans trans;
 254         struct bch_inode_unpacked dir_u;
 255         struct bch_inode_info *inode, *old;
 256         struct bch_inode_unpacked inode_u;
 257         struct posix_acl *default_acl = NULL, *acl = NULL;
 258         u64 journal_seq = 0;
 259         int ret;
 260
 261         /*
 262          * preallocate acls + vfs inode before btree transaction, so that
 263          * nothing can fail after the transaction succeeds:
 264          */
 265 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 266         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 267         if (ret)
 268                 return ERR_PTR(ret);
 269 #endif
 270         inode = to_bch_ei(new_inode(c->vfs_sb));
 271         if (unlikely(!inode)) {
 272                 inode = ERR_PTR(-ENOMEM);
 273                 goto err;
 274         }
 275
 276         bch2_inode_init_early(c, &inode_u);
 277
 278         if (!tmpfile)
 279                 mutex_lock(&dir->ei_update_lock);
 280
 281         bch2_trans_init(&trans, c, 8,
 282                         2048 + (!tmpfile ? dentry->d_name.len : 0));
 283 retry:
 284         bch2_trans_begin(&trans);
 285
 286         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 287                                   !tmpfile ? &dentry->d_name : NULL,
 288                                   from_kuid(i_user_ns(&dir->v), current_fsuid()),
 289                                   from_kgid(i_user_ns(&dir->v), current_fsgid()),
 290                                   mode, rdev,
 291                                   default_acl, acl) ?:
 292                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 293                                 KEY_TYPE_QUOTA_PREALLOC);
 294         if (unlikely(ret))
 295                 goto err_before_quota;
 296
 297         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 298                                   BTREE_INSERT_NOUNLOCK);
 299         if (unlikely(ret)) {
 300                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 301                                 KEY_TYPE_QUOTA_WARN);
 302 err_before_quota:
 303                 if (ret == -EINTR)
 304                         goto retry;
 305                 goto err_trans;
 306         }
 307
 308         if (!tmpfile) {
 309                 bch2_inode_update_after_write(c, dir, &dir_u,
 310                                               ATTR_MTIME|ATTR_CTIME);
 311                 journal_seq_copy(c, dir, journal_seq);
 312                 mutex_unlock(&dir->ei_update_lock);
 313         }
 314
 315         bch2_vfs_inode_init(c, inode, &inode_u);
 316         journal_seq_copy(c, inode, journal_seq);
 317
 318         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 319         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 320
 321         /*
 322          * we must insert the new inode into the inode cache before calling
 323          * bch2_trans_exit() and dropping locks, else we could race with another
 324          * thread pulling the inode in and modifying it:
 325          */
 326
 327         inode->v.i_state |= I_CREATING;
 328         old = to_bch_ei(inode_insert5(&inode->v, inode->v.i_ino,
 329                                       inum_test, NULL, &inode->v.i_ino));
 330         BUG_ON(!old);
 331
 332         if (unlikely(old != inode)) {
 333                 /*
 334                  * We raced, another process pulled the new inode into cache
 335                  * before us:
 336                  */
 337                 journal_seq_copy(c, old, journal_seq);
 338                 make_bad_inode(&inode->v);
 339                 iput(&inode->v);
 340
 341                 inode = old;
 342         } else {
 343                 /*
 344                  * we really don't want insert_inode_locked2() to be setting
 345                  * I_NEW...
 346                  */
 347                 unlock_new_inode(&inode->v);
 348         }
 349
 350         bch2_trans_exit(&trans);
 351 err:
 352         posix_acl_release(default_acl);
 353         posix_acl_release(acl);
 354         return inode;
 355 err_trans:
 356         if (!tmpfile)
 357                 mutex_unlock(&dir->ei_update_lock);
 358
 359         bch2_trans_exit(&trans);
 360         make_bad_inode(&inode->v);
 361         iput(&inode->v);
 362         inode = ERR_PTR(ret);
 363         goto err;
 364 }
 365
 366 /* methods */
 367
 368 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 369                                   unsigned int flags)
 370 {
 371         struct bch_fs *c = vdir->i_sb->s_fs_info;
 372         struct bch_inode_info *dir = to_bch_ei(vdir);
 373         struct inode *vinode = NULL;
 374         u64 inum;
 375
 376         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 377                                   &dir->ei_str_hash,
 378                                   &dentry->d_name);
 379
 380         if (inum)
 381                 vinode = bch2_vfs_inode_get(c, inum);
 382
 383         return d_splice_alias(vinode, dentry);
 384 }
 385
 386 static int bch2_mknod(struct mnt_idmap *idmap,
 387                       struct inode *vdir, struct dentry *dentry,
 388                       umode_t mode, dev_t rdev)
 389 {
 390         struct bch_inode_info *inode =
 391                 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev, false);
 392
 393         if (IS_ERR(inode))
 394                 return PTR_ERR(inode);
 395
 396         d_instantiate(dentry, &inode->v);
 397         return 0;
 398 }
 399
 400 static int bch2_create(struct mnt_idmap *idmap,
 401                        struct inode *vdir, struct dentry *dentry,
 402                        umode_t mode, bool excl)
 403 {
 404         return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
 405 }
 406
 407 static int __bch2_link(struct bch_fs *c,
 408                        struct bch_inode_info *inode,
 409                        struct bch_inode_info *dir,
 410                        struct dentry *dentry)
 411 {
 412         struct btree_trans trans;
 413         struct bch_inode_unpacked dir_u, inode_u;
 414         int ret;
 415
 416         mutex_lock(&inode->ei_update_lock);
 417         bch2_trans_init(&trans, c, 4, 1024);
 418
 419         do {
 420                 bch2_trans_begin(&trans);
 421                 ret   = bch2_link_trans(&trans,
 422                                         dir->v.i_ino,
 423                                         inode->v.i_ino, &dir_u, &inode_u,
 424                                         &dentry->d_name) ?:
 425                         bch2_trans_commit(&trans, NULL,
 426                                         &inode->ei_journal_seq,
 427                                         BTREE_INSERT_NOUNLOCK);
 428         } while (ret == -EINTR);
 429
 430         if (likely(!ret)) {
 431                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 432
 433                 journal_seq_copy(c, inode, dir->ei_journal_seq);
 434                 bch2_inode_update_after_write(c, dir, &dir_u,
 435                                               ATTR_MTIME|ATTR_CTIME);
 436                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 437         }
 438
 439         bch2_trans_exit(&trans);
 440         mutex_unlock(&inode->ei_update_lock);
 441         return ret;
 442 }
 443
 444 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 445                      struct dentry *dentry)
 446 {
 447         struct bch_fs *c = vdir->i_sb->s_fs_info;
 448         struct bch_inode_info *dir = to_bch_ei(vdir);
 449         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 450         int ret;
 451
 452         lockdep_assert_held(&inode->v.i_rwsem);
 453
 454         ret = __bch2_link(c, inode, dir, dentry);
 455         if (unlikely(ret))
 456                 return ret;
 457
 458         ihold(&inode->v);
 459         d_instantiate(dentry, &inode->v);
 460         return 0;
 461 }
 462
 463 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 464 {
 465         struct bch_fs *c = vdir->i_sb->s_fs_info;
 466         struct bch_inode_info *dir = to_bch_ei(vdir);
 467         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 468         struct bch_inode_unpacked dir_u, inode_u;
 469         struct btree_trans trans;
 470         int ret;
 471
 472         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 473         bch2_trans_init(&trans, c, 4, 1024);
 474
 475         do {
 476                 bch2_trans_begin(&trans);
 477
 478                 ret   = bch2_unlink_trans(&trans,
 479                                           dir->v.i_ino, &dir_u,
 480                                           &inode_u, &dentry->d_name) ?:
 481                         bch2_trans_commit(&trans, NULL,
 482                                           &dir->ei_journal_seq,
 483                                           BTREE_INSERT_NOUNLOCK|
 484                                           BTREE_INSERT_NOFAIL);
 485         } while (ret == -EINTR);
 486
 487         if (likely(!ret)) {
 488                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 489
 490                 journal_seq_copy(c, inode, dir->ei_journal_seq);
 491                 bch2_inode_update_after_write(c, dir, &dir_u,
 492                                               ATTR_MTIME|ATTR_CTIME);
 493                 bch2_inode_update_after_write(c, inode, &inode_u,
 494                                               ATTR_MTIME);
 495         }
 496
 497         bch2_trans_exit(&trans);
 498         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 499
 500         return ret;
 501 }
 502
 503 static int bch2_symlink(struct mnt_idmap *idmap,
 504                         struct inode *vdir, struct dentry *dentry,
 505                         const char *symname)
 506 {
 507         struct bch_fs *c = vdir->i_sb->s_fs_info;
 508         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 509         int ret;
 510
 511         inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 512         if (unlikely(IS_ERR(inode)))
 513                 return PTR_ERR(inode);
 514
 515         inode_lock(&inode->v);
 516         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 517         inode_unlock(&inode->v);
 518
 519         if (unlikely(ret))
 520                 goto err;
 521
 522         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 523         if (unlikely(ret))
 524                 goto err;
 525
 526         journal_seq_copy(c, dir, inode->ei_journal_seq);
 527
 528         ret = __bch2_link(c, inode, dir, dentry);
 529         if (unlikely(ret))
 530                 goto err;
 531
 532         d_instantiate(dentry, &inode->v);
 533         return 0;
 534 err:
 535         iput(&inode->v);
 536         return ret;
 537 }
 538
 539 static int bch2_mkdir(struct mnt_idmap *idmap,
 540                       struct inode *vdir, struct dentry *dentry, umode_t mode)
 541 {
 542         return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
 543 }
 544
 545 static int bch2_rename2(struct mnt_idmap *idmap,
 546                         struct inode *src_vdir, struct dentry *src_dentry,
 547                         struct inode *dst_vdir, struct dentry *dst_dentry,
 548                         unsigned flags)
 549 {
 550         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 551         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 552         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 553         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 554         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 555         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 556         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 557         struct btree_trans trans;
 558         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 559                 ? BCH_RENAME_EXCHANGE
 560                 : dst_dentry->d_inode
 561                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 562         u64 journal_seq = 0;
 563         int ret;
 564
 565         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 566                 return -EINVAL;
 567
 568         if (mode == BCH_RENAME_OVERWRITE) {
 569                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 570                                                    0, LLONG_MAX);
 571                 if (ret)
 572                         return ret;
 573         }
 574
 575         bch2_trans_init(&trans, c, 8, 2048);
 576
 577         bch2_lock_inodes(INODE_UPDATE_LOCK,
 578                          src_dir,
 579                          dst_dir,
 580                          src_inode,
 581                          dst_inode);
 582
 583         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 584                 ret = bch2_fs_quota_transfer(c, src_inode,
 585                                              dst_dir->ei_qid,
 586                                              1 << QTYP_PRJ,
 587                                              KEY_TYPE_QUOTA_PREALLOC);
 588                 if (ret)
 589                         goto err;
 590         }
 591
 592         if (mode == BCH_RENAME_EXCHANGE &&
 593             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 594                 ret = bch2_fs_quota_transfer(c, dst_inode,
 595                                              src_dir->ei_qid,
 596                                              1 << QTYP_PRJ,
 597                                              KEY_TYPE_QUOTA_PREALLOC);
 598                 if (ret)
 599                         goto err;
 600         }
 601
 602 retry:
 603         bch2_trans_begin(&trans);
 604         ret   = bch2_rename_trans(&trans,
 605                                   src_dir->v.i_ino, &src_dir_u,
 606                                   dst_dir->v.i_ino, &dst_dir_u,
 607                                   &src_inode_u,
 608                                   &dst_inode_u,
 609                                   &src_dentry->d_name,
 610                                   &dst_dentry->d_name,
 611                                   mode) ?:
 612                 bch2_trans_commit(&trans, NULL,
 613                                   &journal_seq,
 614                                   BTREE_INSERT_NOUNLOCK);
 615         if (ret == -EINTR)
 616                 goto retry;
 617         if (unlikely(ret))
 618                 goto err;
 619
 620         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 621         BUG_ON(dst_inode &&
 622                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 623
 624         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 625                                       ATTR_MTIME|ATTR_CTIME);
 626         journal_seq_copy(c, src_dir, journal_seq);
 627
 628         if (src_dir != dst_dir) {
 629                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 630                                               ATTR_MTIME|ATTR_CTIME);
 631                 journal_seq_copy(c, dst_dir, journal_seq);
 632         }
 633
 634         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 635                                       ATTR_CTIME);
 636         journal_seq_copy(c, src_inode, journal_seq);
 637
 638         if (dst_inode) {
 639                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 640                                               ATTR_CTIME);
 641                 journal_seq_copy(c, dst_inode, journal_seq);
 642         }
 643 err:
 644         bch2_trans_exit(&trans);
 645
 646         bch2_fs_quota_transfer(c, src_inode,
 647                                bch_qid(&src_inode->ei_inode),
 648                                1 << QTYP_PRJ,
 649                                KEY_TYPE_QUOTA_NOCHECK);
 650         if (dst_inode)
 651                 bch2_fs_quota_transfer(c, dst_inode,
 652                                        bch_qid(&dst_inode->ei_inode),
 653                                        1 << QTYP_PRJ,
 654                                        KEY_TYPE_QUOTA_NOCHECK);
 655
 656         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 657                            src_dir,
 658                            dst_dir,
 659                            src_inode,
 660                            dst_inode);
 661
 662         return ret;
 663 }
 664
 665 static void bch2_setattr_copy(struct mnt_idmap *idmap,
 666                               struct bch_inode_info *inode,
 667                               struct bch_inode_unpacked *bi,
 668                               struct iattr *attr)
 669 {
 670         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 671         unsigned int ia_valid = attr->ia_valid;
 672
 673         if (ia_valid & ATTR_UID)
 674                 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 675         if (ia_valid & ATTR_GID)
 676                 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 677
 678         if (ia_valid & ATTR_ATIME)
 679                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 680         if (ia_valid & ATTR_MTIME)
 681                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 682         if (ia_valid & ATTR_CTIME)
 683                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 684
 685         if (ia_valid & ATTR_MODE) {
 686                 umode_t mode = attr->ia_mode;
 687                 kgid_t gid = ia_valid & ATTR_GID
 688                         ? attr->ia_gid
 689                         : inode->v.i_gid;
 690
 691                 if (!in_group_p(gid) &&
 692                     !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
 693                         mode &= ~S_ISGID;
 694                 bi->bi_mode = mode;
 695         }
 696 }
 697
 698 static int bch2_setattr_nonsize(struct mnt_idmap *idmap,
 699                                 struct bch_inode_info *inode,
 700                                 struct iattr *attr)
 701 {
 702         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 703         struct bch_qid qid;
 704         struct btree_trans trans;
 705         struct btree_iter *inode_iter;
 706         struct bch_inode_unpacked inode_u;
 707         struct posix_acl *acl = NULL;
 708         int ret;
 709
 710         mutex_lock(&inode->ei_update_lock);
 711
 712         qid = inode->ei_qid;
 713
 714         if (attr->ia_valid & ATTR_UID)
 715                 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
 716
 717         if (attr->ia_valid & ATTR_GID)
 718                 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
 719
 720         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 721                                      KEY_TYPE_QUOTA_PREALLOC);
 722         if (ret)
 723                 goto err;
 724
 725         bch2_trans_init(&trans, c, 0, 0);
 726 retry:
 727         bch2_trans_begin(&trans);
 728         kfree(acl);
 729         acl = NULL;
 730
 731         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 732                                      BTREE_ITER_INTENT);
 733         ret = PTR_ERR_OR_ZERO(inode_iter);
 734         if (ret)
 735                 goto btree_err;
 736
 737         bch2_setattr_copy(idmap, inode, &inode_u, attr);
 738
 739         if (attr->ia_valid & ATTR_MODE) {
 740                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 741                 if (ret)
 742                         goto btree_err;
 743         }
 744
 745         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 746                 bch2_trans_commit(&trans, NULL,
 747                                   &inode->ei_journal_seq,
 748                                   BTREE_INSERT_NOUNLOCK|
 749                                   BTREE_INSERT_NOFAIL);
 750 btree_err:
 751         if (ret == -EINTR)
 752                 goto retry;
 753         if (unlikely(ret))
 754                 goto err_trans;
 755
 756         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 757
 758         if (acl)
 759                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 760 err_trans:
 761         bch2_trans_exit(&trans);
 762 err:
 763         mutex_unlock(&inode->ei_update_lock);
 764
 765         return ret;
 766 }
 767
 768 static int bch2_getattr(struct mnt_idmap *idmap,
 769                         const struct path *path, struct kstat *stat,
 770                         u32 request_mask, unsigned query_flags)
 771 {
 772         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 773         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 774
 775         stat->dev       = inode->v.i_sb->s_dev;
 776         stat->ino       = inode->v.i_ino;
 777         stat->mode      = inode->v.i_mode;
 778         stat->nlink     = inode->v.i_nlink;
 779         stat->uid       = inode->v.i_uid;
 780         stat->gid       = inode->v.i_gid;
 781         stat->rdev      = inode->v.i_rdev;
 782         stat->size      = i_size_read(&inode->v);
 783         stat->atime     = inode->v.i_atime;
 784         stat->mtime     = inode->v.i_mtime;
 785         stat->ctime     = inode_get_ctime(&inode->v);
 786         stat->blksize   = block_bytes(c);
 787         stat->blocks    = inode->v.i_blocks;
 788
 789         if (request_mask & STATX_BTIME) {
 790                 stat->result_mask |= STATX_BTIME;
 791                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 792         }
 793
 794         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 795                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 796         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 797
 798         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 799                 stat->attributes |= STATX_ATTR_APPEND;
 800         stat->attributes_mask    |= STATX_ATTR_APPEND;
 801
 802         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 803                 stat->attributes |= STATX_ATTR_NODUMP;
 804         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 805
 806         return 0;
 807 }
 808
 809 static int bch2_setattr(struct mnt_idmap *idmap,
 810                         struct dentry *dentry, struct iattr *iattr)
 811 {
 812         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 813         int ret;
 814
 815         lockdep_assert_held(&inode->v.i_rwsem);
 816
 817         ret = setattr_prepare(idmap, dentry, iattr);
 818         if (ret)
 819                 return ret;
 820
 821         return iattr->ia_valid & ATTR_SIZE
 822                 ? bch2_truncate(inode, iattr)
 823                 : bch2_setattr_nonsize(idmap, inode, iattr);
 824 }
 825
 826 static int bch2_tmpfile(struct mnt_idmap *idmap,
 827                         struct inode *vdir, struct file *file, umode_t mode)
 828 {
 829         struct bch_inode_info *inode =
 830                 __bch2_create(idmap, to_bch_ei(vdir),
 831                               file->f_path.dentry, mode, 0, true);
 832
 833         if (IS_ERR(inode))
 834                 return PTR_ERR(inode);
 835
 836         d_mark_tmpfile(file, &inode->v);
 837         d_instantiate(file->f_path.dentry, &inode->v);
 838         return finish_open_simple(file, 0);
 839 }
 840
 841 static int bch2_fill_extent(struct bch_fs *c,
 842                             struct fiemap_extent_info *info,
 843                             struct bkey_s_c k, unsigned flags)
 844 {
 845         if (bkey_extent_is_direct_data(k.k)) {
 846                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 847                 const union bch_extent_entry *entry;
 848                 struct extent_ptr_decoded p;
 849                 int ret;
 850
 851                 if (k.k->type == KEY_TYPE_reflink_v)
 852                         flags |= FIEMAP_EXTENT_SHARED;
 853
 854                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 855                         int flags2 = 0;
 856                         u64 offset = p.ptr.offset;
 857
 858                         if (p.crc.compression_type)
 859                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 860                         else
 861                                 offset += p.crc.offset;
 862
 863                         if ((offset & (c->opts.block_size - 1)) ||
 864                             (k.k->size & (c->opts.block_size - 1)))
 865                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 866
 867                         ret = fiemap_fill_next_extent(info,
 868                                                 bkey_start_offset(k.k) << 9,
 869                                                 offset << 9,
 870                                                 k.k->size << 9, flags|flags2);
 871                         if (ret)
 872                                 return ret;
 873                 }
 874
 875                 return 0;
 876         } else if (bkey_extent_is_inline_data(k.k)) {
 877                 return fiemap_fill_next_extent(info,
 878                                                bkey_start_offset(k.k) << 9,
 879                                                0, k.k->size << 9,
 880                                                flags|
 881                                                FIEMAP_EXTENT_DATA_INLINE);
 882         } else if (k.k->type == KEY_TYPE_reservation) {
 883                 return fiemap_fill_next_extent(info,
 884                                                bkey_start_offset(k.k) << 9,
 885                                                0, k.k->size << 9,
 886                                                flags|
 887                                                FIEMAP_EXTENT_DELALLOC|
 888                                                FIEMAP_EXTENT_UNWRITTEN);
 889         } else {
 890                 BUG();
 891         }
 892 }
 893
 894 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 895                        u64 start, u64 len)
 896 {
 897         struct bch_fs *c = vinode->i_sb->s_fs_info;
 898         struct bch_inode_info *ei = to_bch_ei(vinode);
 899         struct btree_trans trans;
 900         struct btree_iter *iter;
 901         struct bkey_s_c k;
 902         struct bkey_buf cur, prev;
 903         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 904         unsigned offset_into_extent, sectors;
 905         bool have_extent = false;
 906         int ret = 0;
 907
 908         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
 909         if (ret)
 910                 return ret;
 911
 912         if (start + len < start)
 913                 return -EINVAL;
 914
 915         bch2_bkey_buf_init(&cur);
 916         bch2_bkey_buf_init(&prev);
 917         bch2_trans_init(&trans, c, 0, 0);
 918
 919         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 920                                    POS(ei->v.i_ino, start >> 9), 0);
 921 retry:
 922         while ((k = bch2_btree_iter_peek(iter)).k &&
 923                !(ret = bkey_err(k)) &&
 924                bkey_cmp(iter->pos, end) < 0) {
 925                 if (!bkey_extent_is_data(k.k) &&
 926                     k.k->type != KEY_TYPE_reservation) {
 927                         bch2_btree_iter_next(iter);
 928                         continue;
 929                 }
 930
 931                 offset_into_extent      = iter->pos.offset -
 932                         bkey_start_offset(k.k);
 933                 sectors                 = k.k->size - offset_into_extent;
 934
 935                 bch2_bkey_buf_reassemble(&cur, c, k);
 936
 937                 ret = bch2_read_indirect_extent(&trans,
 938                                         &offset_into_extent, &cur);
 939                 if (ret)
 940                         break;
 941
 942                 k = bkey_i_to_s_c(cur.k);
 943                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 944
 945                 sectors = min(sectors, k.k->size - offset_into_extent);
 946
 947                 bch2_cut_front(POS(k.k->p.inode,
 948                                    bkey_start_offset(k.k) +
 949                                    offset_into_extent),
 950                                cur.k);
 951                 bch2_key_resize(&cur.k->k, sectors);
 952                 cur.k->k.p = iter->pos;
 953                 cur.k->k.p.offset += cur.k->k.size;
 954
 955                 if (have_extent) {
 956                         ret = bch2_fill_extent(c, info,
 957                                         bkey_i_to_s_c(prev.k), 0);
 958                         if (ret)
 959                                 break;
 960                 }
 961
 962                 bkey_copy(prev.k, cur.k);
 963                 have_extent = true;
 964
 965                 bch2_btree_iter_set_pos(iter,
 966                         POS(iter->pos.inode, iter->pos.offset + sectors));
 967         }
 968
 969         if (ret == -EINTR)
 970                 goto retry;
 971
 972         if (!ret && have_extent)
 973                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 974                                        FIEMAP_EXTENT_LAST);
 975
 976         ret = bch2_trans_exit(&trans) ?: ret;
 977         bch2_bkey_buf_exit(&cur, c);
 978         bch2_bkey_buf_exit(&prev, c);
 979         return ret < 0 ? ret : 0;
 980 }
 981
 982 static const struct vm_operations_struct bch_vm_ops = {
 983         .fault          = bch2_page_fault,
 984         .map_pages      = filemap_map_pages,
 985         .page_mkwrite   = bch2_page_mkwrite,
 986 };
 987
 988 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 989 {
 990         file_accessed(file);
 991
 992         vma->vm_ops = &bch_vm_ops;
 993         return 0;
 994 }
 995
 996 /* Directories: */
 997
 998 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 999 {
1000         return generic_file_llseek_size(file, offset, whence,
1001                                         S64_MAX, S64_MAX);
1002 }
1003
1004 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1005 {
1006         struct bch_inode_info *inode = file_bch_inode(file);
1007         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1008
1009         if (!dir_emit_dots(file, ctx))
1010                 return 0;
1011
1012         return bch2_readdir(c, inode->v.i_ino, ctx);
1013 }
1014
1015 static const struct file_operations bch_file_operations = {
1016         .llseek         = bch2_llseek,
1017         .read_iter      = bch2_read_iter,
1018         .write_iter     = bch2_write_iter,
1019         .mmap           = bch2_mmap,
1020         .open           = generic_file_open,
1021         .fsync          = bch2_fsync,
1022         .splice_read    = filemap_splice_read,
1023         .splice_write   = iter_file_splice_write,
1024         .fallocate      = bch2_fallocate_dispatch,
1025         .unlocked_ioctl = bch2_fs_file_ioctl,
1026 #ifdef CONFIG_COMPAT
1027         .compat_ioctl   = bch2_compat_fs_ioctl,
1028 #endif
1029         .remap_file_range = bch2_remap_file_range,
1030 };
1031
1032 static const struct inode_operations bch_file_inode_operations = {
1033         .getattr        = bch2_getattr,
1034         .setattr        = bch2_setattr,
1035         .fiemap         = bch2_fiemap,
1036         .listxattr      = bch2_xattr_list,
1037 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1038         .get_acl        = bch2_get_acl,
1039         .set_acl        = bch2_set_acl,
1040 #endif
1041 };
1042
1043 static const struct inode_operations bch_dir_inode_operations = {
1044         .lookup         = bch2_lookup,
1045         .create         = bch2_create,
1046         .link           = bch2_link,
1047         .unlink         = bch2_unlink,
1048         .symlink        = bch2_symlink,
1049         .mkdir          = bch2_mkdir,
1050         .rmdir          = bch2_unlink,
1051         .mknod          = bch2_mknod,
1052         .rename         = bch2_rename2,
1053         .getattr        = bch2_getattr,
1054         .setattr        = bch2_setattr,
1055         .tmpfile        = bch2_tmpfile,
1056         .listxattr      = bch2_xattr_list,
1057 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1058         .get_acl        = bch2_get_acl,
1059         .set_acl        = bch2_set_acl,
1060 #endif
1061 };
1062
1063 static const struct file_operations bch_dir_file_operations = {
1064         .llseek         = bch2_dir_llseek,
1065         .read           = generic_read_dir,
1066         .iterate_shared = bch2_vfs_readdir,
1067         .fsync          = bch2_fsync,
1068         .unlocked_ioctl = bch2_fs_file_ioctl,
1069 #ifdef CONFIG_COMPAT
1070         .compat_ioctl   = bch2_compat_fs_ioctl,
1071 #endif
1072 };
1073
1074 static const struct inode_operations bch_symlink_inode_operations = {
1075         .get_link       = page_get_link,
1076         .getattr        = bch2_getattr,
1077         .setattr        = bch2_setattr,
1078         .listxattr      = bch2_xattr_list,
1079 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1080         .get_acl        = bch2_get_acl,
1081         .set_acl        = bch2_set_acl,
1082 #endif
1083 };
1084
1085 static const struct inode_operations bch_special_inode_operations = {
1086         .getattr        = bch2_getattr,
1087         .setattr        = bch2_setattr,
1088         .listxattr      = bch2_xattr_list,
1089 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1090         .get_acl        = bch2_get_acl,
1091         .set_acl        = bch2_set_acl,
1092 #endif
1093 };
1094
1095 static const struct address_space_operations bch_address_space_operations = {
1096         .writepage      = bch2_writepage,
1097         .read_folio     = bch2_read_folio,
1098         .writepages     = bch2_writepages,
1099         .readahead      = bch2_readahead,
1100         .dirty_folio    = filemap_dirty_folio,
1101         .write_begin    = bch2_write_begin,
1102         .write_end      = bch2_write_end,
1103         .invalidate_folio = bch2_invalidate_folio,
1104         .release_folio  = bch2_release_folio,
1105         .direct_IO      = noop_direct_IO,
1106 #ifdef CONFIG_MIGRATION
1107         .migrate_folio  = filemap_migrate_folio,
1108 #endif
1109         .error_remove_page = generic_error_remove_page,
1110 };
1111
1112 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1113                 u64 ino, u32 generation)
1114 {
1115         struct bch_fs *c = sb->s_fs_info;
1116         struct inode *vinode;
1117
1118         if (ino < BCACHEFS_ROOT_INO)
1119                 return ERR_PTR(-ESTALE);
1120
1121         vinode = bch2_vfs_inode_get(c, ino);
1122         if (IS_ERR(vinode))
1123                 return ERR_CAST(vinode);
1124         if (generation && vinode->i_generation != generation) {
1125                 /* we didn't find the right inode.. */
1126                 iput(vinode);
1127                 return ERR_PTR(-ESTALE);
1128         }
1129         return vinode;
1130 }
1131
1132 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1133                 int fh_len, int fh_type)
1134 {
1135         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1136                                     bch2_nfs_get_inode);
1137 }
1138
1139 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1140                 int fh_len, int fh_type)
1141 {
1142         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1143                                     bch2_nfs_get_inode);
1144 }
1145
1146 static const struct export_operations bch_export_ops = {
1147         .fh_to_dentry   = bch2_fh_to_dentry,
1148         .fh_to_parent   = bch2_fh_to_parent,
1149         //.get_parent   = bch2_get_parent,
1150 };
1151
1152 static void bch2_vfs_inode_init(struct bch_fs *c,
1153                                 struct bch_inode_info *inode,
1154                                 struct bch_inode_unpacked *bi)
1155 {
1156         bch2_inode_update_after_write(c, inode, bi, ~0);
1157
1158         inode->v.i_blocks       = bi->bi_sectors;
1159         inode->v.i_ino          = bi->bi_inum;
1160         inode->v.i_rdev         = bi->bi_dev;
1161         inode->v.i_generation   = bi->bi_generation;
1162         inode->v.i_size         = bi->bi_size;
1163
1164         inode->ei_flags         = 0;
1165         inode->ei_journal_seq   = 0;
1166         inode->ei_quota_reserved = 0;
1167         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1168         inode->ei_qid           = bch_qid(bi);
1169
1170         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1171
1172         switch (inode->v.i_mode & S_IFMT) {
1173         case S_IFREG:
1174                 inode->v.i_op   = &bch_file_inode_operations;
1175                 inode->v.i_fop  = &bch_file_operations;
1176                 break;
1177         case S_IFDIR:
1178                 inode->v.i_op   = &bch_dir_inode_operations;
1179                 inode->v.i_fop  = &bch_dir_file_operations;
1180                 break;
1181         case S_IFLNK:
1182                 inode_nohighmem(&inode->v);
1183                 inode->v.i_op   = &bch_symlink_inode_operations;
1184                 break;
1185         default:
1186                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1187                 inode->v.i_op   = &bch_special_inode_operations;
1188                 break;
1189         }
1190 }
1191
1192 static struct inode *bch2_alloc_inode(struct super_block *sb)
1193 {
1194         struct bch_inode_info *inode;
1195
1196         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1197         if (!inode)
1198                 return NULL;
1199
1200         inode_init_once(&inode->v);
1201         mutex_init(&inode->ei_update_lock);
1202         pagecache_lock_init(&inode->ei_pagecache_lock);
1203         mutex_init(&inode->ei_quota_lock);
1204         inode->ei_journal_seq = 0;
1205
1206         return &inode->v;
1207 }
1208
1209 static void bch2_i_callback(struct rcu_head *head)
1210 {
1211         struct inode *vinode = container_of(head, struct inode, i_rcu);
1212         struct bch_inode_info *inode = to_bch_ei(vinode);
1213
1214         kmem_cache_free(bch2_inode_cache, inode);
1215 }
1216
1217 static void bch2_destroy_inode(struct inode *vinode)
1218 {
1219         call_rcu(&vinode->i_rcu, bch2_i_callback);
1220 }
1221
1222 static int inode_update_times_fn(struct bch_inode_info *inode,
1223                                  struct bch_inode_unpacked *bi,
1224                                  void *p)
1225 {
1226         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1227
1228         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1229         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1230         bi->bi_ctime    = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1231
1232         return 0;
1233 }
1234
1235 static int bch2_vfs_write_inode(struct inode *vinode,
1236                                 struct writeback_control *wbc)
1237 {
1238         struct bch_fs *c = vinode->i_sb->s_fs_info;
1239         struct bch_inode_info *inode = to_bch_ei(vinode);
1240         int ret;
1241
1242         mutex_lock(&inode->ei_update_lock);
1243         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1244                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1245         mutex_unlock(&inode->ei_update_lock);
1246
1247         return ret;
1248 }
1249
1250 static void bch2_evict_inode(struct inode *vinode)
1251 {
1252         struct bch_fs *c = vinode->i_sb->s_fs_info;
1253         struct bch_inode_info *inode = to_bch_ei(vinode);
1254
1255         truncate_inode_pages_final(&inode->v.i_data);
1256
1257         clear_inode(&inode->v);
1258
1259         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1260
1261         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1262                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1263                                 KEY_TYPE_QUOTA_WARN);
1264                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1265                                 KEY_TYPE_QUOTA_WARN);
1266                 bch2_inode_rm(c, inode->v.i_ino, true);
1267         }
1268 }
1269
1270 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1271 {
1272         struct super_block *sb = dentry->d_sb;
1273         struct bch_fs *c = sb->s_fs_info;
1274         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1275         unsigned shift = sb->s_blocksize_bits - 9;
1276         /*
1277          * this assumes inodes take up 64 bytes, which is a decent average
1278          * number:
1279          */
1280         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1281         u64 fsid;
1282
1283         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1284         buf->f_bsize    = sb->s_blocksize;
1285         buf->f_blocks   = usage.capacity >> shift;
1286         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1287         buf->f_bavail   = buf->f_bfree;
1288
1289         buf->f_files    = usage.nr_inodes + avail_inodes;
1290         buf->f_ffree    = avail_inodes;
1291
1292         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1293                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1294         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1295         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1296         buf->f_namelen  = BCH_NAME_MAX;
1297
1298         return 0;
1299 }
1300
1301 static int bch2_sync_fs(struct super_block *sb, int wait)
1302 {
1303         struct bch_fs *c = sb->s_fs_info;
1304
1305         if (c->opts.journal_flush_disabled)
1306                 return 0;
1307
1308         if (!wait) {
1309                 bch2_journal_flush_async(&c->journal, NULL);
1310                 return 0;
1311         }
1312
1313         return bch2_journal_flush(&c->journal);
1314 }
1315
1316 static struct bch_fs *bch2_path_to_fs(const char *path)
1317 {
1318         struct bch_fs *c;
1319         dev_t dev;
1320         int ret;
1321
1322         ret = lookup_bdev(path, &dev);
1323         if (ret)
1324                 return ERR_PTR(ret);
1325
1326         c = bch2_dev_to_fs(dev);
1327         if (c)
1328                 closure_put(&c->cl);
1329         return c ?: ERR_PTR(-ENOENT);
1330 }
1331
1332 static char **split_devs(const char *_dev_name, unsigned *nr)
1333 {
1334         char *dev_name = NULL, **devs = NULL, *s;
1335         size_t i, nr_devs = 0;
1336
1337         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1338         if (!dev_name)
1339                 return NULL;
1340
1341         for (s = dev_name; s; s = strchr(s + 1, ':'))
1342                 nr_devs++;
1343
1344         devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1345         if (!devs) {
1346                 kfree(dev_name);
1347                 return NULL;
1348         }
1349
1350         for (i = 0, s = dev_name;
1351              s;
1352              (s = strchr(s, ':')) && (*s++ = '\0'))
1353                 devs[i++] = s;
1354
1355         *nr = nr_devs;
1356         return devs;
1357 }
1358
1359 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1360 {
1361         struct bch_fs *c = sb->s_fs_info;
1362         struct bch_opts opts = bch2_opts_empty();
1363         int ret;
1364
1365         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1366
1367         ret = bch2_parse_mount_opts(c, &opts, data);
1368         if (ret)
1369                 return ret;
1370
1371         if (opts.read_only != c->opts.read_only) {
1372                 down_write(&c->state_lock);
1373
1374                 if (opts.read_only) {
1375                         bch2_fs_read_only(c);
1376
1377                         sb->s_flags |= SB_RDONLY;
1378                 } else {
1379                         ret = bch2_fs_read_write(c);
1380                         if (ret) {
1381                                 bch_err(c, "error going rw: %i", ret);
1382                                 up_write(&c->state_lock);
1383                                 return -EINVAL;
1384                         }
1385
1386                         sb->s_flags &= ~SB_RDONLY;
1387                 }
1388
1389                 c->opts.read_only = opts.read_only;
1390
1391                 up_write(&c->state_lock);
1392         }
1393
1394         if (opts.errors >= 0)
1395                 c->opts.errors = opts.errors;
1396
1397         return ret;
1398 }
1399
1400 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1401 {
1402         struct bch_fs *c = root->d_sb->s_fs_info;
1403         struct bch_dev *ca;
1404         unsigned i;
1405         bool first = true;
1406
1407         for_each_online_member(ca, c, i) {
1408                 if (!first)
1409                         seq_putc(seq, ':');
1410                 first = false;
1411                 seq_puts(seq, "/dev/");
1412                 seq_puts(seq, ca->name);
1413         }
1414
1415         return 0;
1416 }
1417
1418 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1419 {
1420         struct bch_fs *c = root->d_sb->s_fs_info;
1421         enum bch_opt_id i;
1422         char buf[512];
1423
1424         for (i = 0; i < bch2_opts_nr; i++) {
1425                 const struct bch_option *opt = &bch2_opt_table[i];
1426                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1427
1428                 if (!(opt->mode & OPT_MOUNT))
1429                         continue;
1430
1431                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1432                         continue;
1433
1434                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1435                                  OPT_SHOW_MOUNT_STYLE);
1436                 seq_putc(seq, ',');
1437                 seq_puts(seq, buf);
1438         }
1439
1440         return 0;
1441 }
1442
1443 static void bch2_put_super(struct super_block *sb)
1444 {
1445         struct bch_fs *c = sb->s_fs_info;
1446
1447         __bch2_fs_stop(c);
1448 }
1449
1450 static const struct super_operations bch_super_operations = {
1451         .alloc_inode    = bch2_alloc_inode,
1452         .destroy_inode  = bch2_destroy_inode,
1453         .write_inode    = bch2_vfs_write_inode,
1454         .evict_inode    = bch2_evict_inode,
1455         .sync_fs        = bch2_sync_fs,
1456         .statfs         = bch2_statfs,
1457         .show_devname   = bch2_show_devname,
1458         .show_options   = bch2_show_options,
1459         .remount_fs     = bch2_remount,
1460         .put_super      = bch2_put_super,
1461 #if 0
1462         .freeze_fs      = bch2_freeze,
1463         .unfreeze_fs    = bch2_unfreeze,
1464 #endif
1465 };
1466
1467 static int bch2_set_super(struct super_block *s, void *data)
1468 {
1469         s->s_fs_info = data;
1470         return 0;
1471 }
1472
1473 static int bch2_noset_super(struct super_block *s, void *data)
1474 {
1475         return -EBUSY;
1476 }
1477
1478 static int bch2_test_super(struct super_block *s, void *data)
1479 {
1480         struct bch_fs *c = s->s_fs_info;
1481         struct bch_fs **devs = data;
1482         unsigned i;
1483
1484         if (!c)
1485                 return false;
1486
1487         for (i = 0; devs[i]; i++)
1488                 if (c != devs[i])
1489                         return false;
1490         return true;
1491 }
1492
1493 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1494                                  int flags, const char *dev_name, void *data)
1495 {
1496         struct bch_fs *c;
1497         struct bch_dev *ca;
1498         struct super_block *sb;
1499         struct inode *vinode;
1500         struct bch_opts opts = bch2_opts_empty();
1501         char **devs;
1502         struct bch_fs **devs_to_fs = NULL;
1503         unsigned i, nr_devs;
1504         int ret;
1505
1506         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1507
1508         ret = bch2_parse_mount_opts(NULL, &opts, data);
1509         if (ret)
1510                 return ERR_PTR(ret);
1511
1512         devs = split_devs(dev_name, &nr_devs);
1513         if (!devs)
1514                 return ERR_PTR(-ENOMEM);
1515
1516         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1517         if (!devs_to_fs) {
1518                 sb = ERR_PTR(-ENOMEM);
1519                 goto got_sb;
1520         }
1521
1522         for (i = 0; i < nr_devs; i++)
1523                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1524
1525         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1526                   flags|SB_NOSEC, devs_to_fs);
1527         if (!IS_ERR(sb))
1528                 goto got_sb;
1529
1530         c = bch2_fs_open(devs, nr_devs, opts);
1531         if (IS_ERR(c)) {
1532                 sb = ERR_CAST(c);
1533                 goto got_sb;
1534         }
1535
1536         /* Some options can't be parsed until after the fs is started: */
1537         ret = bch2_parse_mount_opts(c, &opts, data);
1538         if (ret) {
1539                 bch2_fs_stop(c);
1540                 sb = ERR_PTR(ret);
1541                 goto got_sb;
1542         }
1543
1544         bch2_opts_apply(&c->opts, opts);
1545
1546         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1547         if (IS_ERR(sb))
1548                 bch2_fs_stop(c);
1549 got_sb:
1550         kfree(devs_to_fs);
1551         kfree(devs[0]);
1552         kfree(devs);
1553
1554         if (IS_ERR(sb))
1555                 return ERR_CAST(sb);
1556
1557         c = sb->s_fs_info;
1558
1559         if (sb->s_root) {
1560                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1561                         ret = -EBUSY;
1562                         goto err_put_super;
1563                 }
1564                 goto out;
1565         }
1566
1567         sb->s_blocksize         = block_bytes(c);
1568         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1569         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1570         sb->s_op                = &bch_super_operations;
1571         sb->s_export_op         = &bch_export_ops;
1572 #ifdef CONFIG_BCACHEFS_QUOTA
1573         sb->s_qcop              = &bch2_quotactl_operations;
1574         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1575 #endif
1576         sb->s_xattr             = bch2_xattr_handlers;
1577         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1578         sb->s_time_gran         = c->sb.time_precision;
1579         c->vfs_sb               = sb;
1580         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1581
1582         ret = super_setup_bdi(sb);
1583         if (ret)
1584                 goto err_put_super;
1585
1586         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1587
1588         for_each_online_member(ca, c, i) {
1589                 struct block_device *bdev = ca->disk_sb.bdev;
1590
1591                 /* XXX: create an anonymous device for multi device filesystems */
1592                 sb->s_bdev      = bdev;
1593                 sb->s_dev       = bdev->bd_dev;
1594                 percpu_ref_put(&ca->io_ref);
1595                 break;
1596         }
1597
1598 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1599         if (c->opts.acl)
1600                 sb->s_flags     |= SB_POSIXACL;
1601 #endif
1602
1603         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1604         if (IS_ERR(vinode)) {
1605                 bch_err(c, "error mounting: error getting root inode %i",
1606                         (int) PTR_ERR(vinode));
1607                 ret = PTR_ERR(vinode);
1608                 goto err_put_super;
1609         }
1610
1611         sb->s_root = d_make_root(vinode);
1612         if (!sb->s_root) {
1613                 bch_err(c, "error mounting: error allocating root dentry");
1614                 ret = -ENOMEM;
1615                 goto err_put_super;
1616         }
1617
1618         sb->s_flags |= SB_ACTIVE;
1619 out:
1620         return dget(sb->s_root);
1621
1622 err_put_super:
1623         deactivate_locked_super(sb);
1624         return ERR_PTR(ret);
1625 }
1626
1627 static void bch2_kill_sb(struct super_block *sb)
1628 {
1629         struct bch_fs *c = sb->s_fs_info;
1630
1631         generic_shutdown_super(sb);
1632         bch2_fs_free(c);
1633 }
1634
1635 static struct file_system_type bcache_fs_type = {
1636         .owner          = THIS_MODULE,
1637         .name           = "bcachefs",
1638         .mount          = bch2_mount,
1639         .kill_sb        = bch2_kill_sb,
1640         .fs_flags       = FS_REQUIRES_DEV,
1641 };
1642
1643 MODULE_ALIAS_FS("bcachefs");
1644
1645 void bch2_vfs_exit(void)
1646 {
1647         unregister_filesystem(&bcache_fs_type);
1648         if (bch2_inode_cache)
1649                 kmem_cache_destroy(bch2_inode_cache);
1650 }
1651
1652 int __init bch2_vfs_init(void)
1653 {
1654         int ret = -ENOMEM;
1655
1656         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1657         if (!bch2_inode_cache)
1658                 goto err;
1659
1660         ret = register_filesystem(&bcache_fs_type);
1661         if (ret)
1662                 goto err;
1663
1664         return 0;
1665 err:
1666         bch2_vfs_exit();
1667         return ret;
1668 }
1669
1670 #endif /* NO_BCACHEFS_FS */