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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "bkey_on_stack.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/module.h>
  29 #include <linux/posix_acl.h>
  30 #include <linux/random.h>
  31 #include <linux/statfs.h>
  32 #include <linux/xattr.h>
  33
  34 static struct kmem_cache *bch2_inode_cache;
  35
  36 static void bch2_vfs_inode_init(struct bch_fs *,
  37                                 struct bch_inode_info *,
  38                                 struct bch_inode_unpacked *);
  39
  40 static void journal_seq_copy(struct bch_inode_info *dst,
  41                              u64 journal_seq)
  42 {
  43         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  44
  45         do {
  46                 old = v;
  47
  48                 if (old >= journal_seq)
  49                         break;
  50         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  51 }
  52
  53 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
  54 {
  55         BUG_ON(atomic_long_read(&lock->v) == 0);
  56
  57         if (atomic_long_sub_return_release(i, &lock->v) == 0)
  58                 wake_up_all(&lock->wait);
  59 }
  60
  61 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
  62 {
  63         long v = atomic_long_read(&lock->v), old;
  64
  65         do {
  66                 old = v;
  67
  68                 if (i > 0 ? v < 0 : v > 0)
  69                         return false;
  70         } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
  71                                         old, old + i)) != old);
  72         return true;
  73 }
  74
  75 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
  76 {
  77         wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
  78 }
  79
  80 void bch2_pagecache_add_put(struct pagecache_lock *lock)
  81 {
  82         __pagecache_lock_put(lock, 1);
  83 }
  84
  85 void bch2_pagecache_add_get(struct pagecache_lock *lock)
  86 {
  87         __pagecache_lock_get(lock, 1);
  88 }
  89
  90 void bch2_pagecache_block_put(struct pagecache_lock *lock)
  91 {
  92         __pagecache_lock_put(lock, -1);
  93 }
  94
  95 void bch2_pagecache_block_get(struct pagecache_lock *lock)
  96 {
  97         __pagecache_lock_get(lock, -1);
  98 }
  99
 100 void bch2_inode_update_after_write(struct bch_fs *c,
 101                                    struct bch_inode_info *inode,
 102                                    struct bch_inode_unpacked *bi,
 103                                    unsigned fields)
 104 {
 105         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
 106         i_uid_write(&inode->v, bi->bi_uid);
 107         i_gid_write(&inode->v, bi->bi_gid);
 108         inode->v.i_mode = bi->bi_mode;
 109
 110         if (fields & ATTR_ATIME)
 111                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
 112         if (fields & ATTR_MTIME)
 113                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
 114         if (fields & ATTR_CTIME)
 115                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
 116
 117         inode->ei_inode         = *bi;
 118
 119         bch2_inode_flags_to_vfs(inode);
 120 }
 121
 122 int __must_check bch2_write_inode(struct bch_fs *c,
 123                                   struct bch_inode_info *inode,
 124                                   inode_set_fn set,
 125                                   void *p, unsigned fields)
 126 {
 127         struct btree_trans trans;
 128         struct btree_iter *iter;
 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         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 137                                BTREE_ITER_INTENT);
 138         ret   = PTR_ERR_OR_ZERO(iter) ?:
 139                 (set ? set(inode, &inode_u, p) : 0) ?:
 140                 bch2_inode_write(&trans, iter, &inode_u) ?:
 141                 bch2_trans_commit(&trans, NULL,
 142                                   &inode->ei_journal_seq,
 143                                   BTREE_INSERT_NOUNLOCK|
 144                                   BTREE_INSERT_NOFAIL);
 145
 146         /*
 147          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 148          * this is important for inode updates via bchfs_write_index_update
 149          */
 150         if (!ret)
 151                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 152
 153         bch2_trans_iter_put(&trans, iter);
 154
 155         if (ret == -EINTR)
 156                 goto retry;
 157
 158         bch2_trans_exit(&trans);
 159         return ret < 0 ? ret : 0;
 160 }
 161
 162 int bch2_fs_quota_transfer(struct bch_fs *c,
 163                            struct bch_inode_info *inode,
 164                            struct bch_qid new_qid,
 165                            unsigned qtypes,
 166                            enum quota_acct_mode mode)
 167 {
 168         unsigned i;
 169         int ret;
 170
 171         qtypes &= enabled_qtypes(c);
 172
 173         for (i = 0; i < QTYP_NR; i++)
 174                 if (new_qid.q[i] == inode->ei_qid.q[i])
 175                         qtypes &= ~(1U << i);
 176
 177         if (!qtypes)
 178                 return 0;
 179
 180         mutex_lock(&inode->ei_quota_lock);
 181
 182         ret = bch2_quota_transfer(c, qtypes, new_qid,
 183                                   inode->ei_qid,
 184                                   inode->v.i_blocks +
 185                                   inode->ei_quota_reserved,
 186                                   mode);
 187         if (!ret)
 188                 for (i = 0; i < QTYP_NR; i++)
 189                         if (qtypes & (1 << i))
 190                                 inode->ei_qid.q[i] = new_qid.q[i];
 191
 192         mutex_unlock(&inode->ei_quota_lock);
 193
 194         return ret;
 195 }
 196
 197 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 198 {
 199         struct bch_inode_unpacked inode_u;
 200         struct bch_inode_info *inode;
 201         int ret;
 202
 203         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 204         if (unlikely(!inode))
 205                 return ERR_PTR(-ENOMEM);
 206         if (!(inode->v.i_state & I_NEW))
 207                 return &inode->v;
 208
 209         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 210         if (ret) {
 211                 iget_failed(&inode->v);
 212                 return ERR_PTR(ret);
 213         }
 214
 215         bch2_vfs_inode_init(c, inode, &inode_u);
 216
 217         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 218
 219         unlock_new_inode(&inode->v);
 220
 221         return &inode->v;
 222 }
 223
 224 static struct bch_inode_info *
 225 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 226               umode_t mode, dev_t rdev, bool tmpfile)
 227 {
 228         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 229         struct user_namespace *ns = dir->v.i_sb->s_user_ns;
 230         struct btree_trans trans;
 231         struct bch_inode_unpacked dir_u;
 232         struct bch_inode_info *inode, *old;
 233         struct bch_inode_unpacked inode_u;
 234         struct posix_acl *default_acl = NULL, *acl = NULL;
 235         u64 journal_seq = 0;
 236         int ret;
 237
 238         /*
 239          * preallocate acls + vfs inode before btree transaction, so that
 240          * nothing can fail after the transaction succeeds:
 241          */
 242 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 243         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 244         if (ret)
 245                 return ERR_PTR(ret);
 246 #endif
 247         inode = to_bch_ei(new_inode(c->vfs_sb));
 248         if (unlikely(!inode)) {
 249                 inode = ERR_PTR(-ENOMEM);
 250                 goto err;
 251         }
 252
 253         bch2_inode_init_early(c, &inode_u);
 254
 255         if (!tmpfile)
 256                 mutex_lock(&dir->ei_update_lock);
 257
 258         bch2_trans_init(&trans, c, 8, 1024);
 259 retry:
 260         bch2_trans_begin(&trans);
 261
 262         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 263                                   !tmpfile ? &dentry->d_name : NULL,
 264                                   from_kuid(ns, current_fsuid()),
 265                                   from_kgid(ns, current_fsgid()),
 266                                   mode, rdev,
 267                                   default_acl, acl) ?:
 268                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 269                                 KEY_TYPE_QUOTA_PREALLOC);
 270         if (unlikely(ret))
 271                 goto err_before_quota;
 272
 273         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 274                                   BTREE_INSERT_NOUNLOCK);
 275         if (unlikely(ret)) {
 276                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 277                                 KEY_TYPE_QUOTA_WARN);
 278 err_before_quota:
 279                 if (ret == -EINTR)
 280                         goto retry;
 281                 goto err_trans;
 282         }
 283
 284         if (!tmpfile) {
 285                 bch2_inode_update_after_write(c, dir, &dir_u,
 286                                               ATTR_MTIME|ATTR_CTIME);
 287                 journal_seq_copy(dir, journal_seq);
 288                 mutex_unlock(&dir->ei_update_lock);
 289         }
 290
 291         bch2_vfs_inode_init(c, inode, &inode_u);
 292         journal_seq_copy(inode, journal_seq);
 293
 294         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 295         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 296
 297         /*
 298          * we must insert the new inode into the inode cache before calling
 299          * bch2_trans_exit() and dropping locks, else we could race with another
 300          * thread pulling the inode in and modifying it:
 301          */
 302
 303         old = to_bch_ei(insert_inode_locked2(&inode->v));
 304         if (unlikely(old)) {
 305                 /*
 306                  * We raced, another process pulled the new inode into cache
 307                  * before us:
 308                  */
 309                 journal_seq_copy(old, journal_seq);
 310                 make_bad_inode(&inode->v);
 311                 iput(&inode->v);
 312
 313                 inode = old;
 314         } else {
 315                 /*
 316                  * we really don't want insert_inode_locked2() to be setting
 317                  * I_NEW...
 318                  */
 319                 unlock_new_inode(&inode->v);
 320         }
 321
 322         bch2_trans_exit(&trans);
 323 err:
 324         posix_acl_release(default_acl);
 325         posix_acl_release(acl);
 326         return inode;
 327 err_trans:
 328         if (!tmpfile)
 329                 mutex_unlock(&dir->ei_update_lock);
 330
 331         bch2_trans_exit(&trans);
 332         make_bad_inode(&inode->v);
 333         iput(&inode->v);
 334         inode = ERR_PTR(ret);
 335         goto err;
 336 }
 337
 338 /* methods */
 339
 340 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 341                                   unsigned int flags)
 342 {
 343         struct bch_fs *c = vdir->i_sb->s_fs_info;
 344         struct bch_inode_info *dir = to_bch_ei(vdir);
 345         struct inode *vinode = NULL;
 346         u64 inum;
 347
 348         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 349                                   &dir->ei_str_hash,
 350                                   &dentry->d_name);
 351
 352         if (inum)
 353                 vinode = bch2_vfs_inode_get(c, inum);
 354
 355         return d_splice_alias(vinode, dentry);
 356 }
 357
 358 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 359                       umode_t mode, dev_t rdev)
 360 {
 361         struct bch_inode_info *inode =
 362                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 363
 364         if (IS_ERR(inode))
 365                 return PTR_ERR(inode);
 366
 367         d_instantiate(dentry, &inode->v);
 368         return 0;
 369 }
 370
 371 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 372                        umode_t mode, bool excl)
 373 {
 374         return bch2_mknod(vdir, dentry, mode|S_IFREG, 0);
 375 }
 376
 377 static int __bch2_link(struct bch_fs *c,
 378                        struct bch_inode_info *inode,
 379                        struct bch_inode_info *dir,
 380                        struct dentry *dentry)
 381 {
 382         struct btree_trans trans;
 383         struct bch_inode_unpacked dir_u, inode_u;
 384         int ret;
 385
 386         mutex_lock(&inode->ei_update_lock);
 387         bch2_trans_init(&trans, c, 4, 1024);
 388
 389         do {
 390                 bch2_trans_begin(&trans);
 391                 ret   = bch2_link_trans(&trans,
 392                                         dir->v.i_ino,
 393                                         inode->v.i_ino, &dir_u, &inode_u,
 394                                         &dentry->d_name) ?:
 395                         bch2_trans_commit(&trans, NULL,
 396                                         &inode->ei_journal_seq,
 397                                         BTREE_INSERT_NOUNLOCK);
 398         } while (ret == -EINTR);
 399
 400         if (likely(!ret)) {
 401                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 402
 403                 journal_seq_copy(inode, dir->ei_journal_seq);
 404                 bch2_inode_update_after_write(c, dir, &dir_u,
 405                                               ATTR_MTIME|ATTR_CTIME);
 406                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 407         }
 408
 409         bch2_trans_exit(&trans);
 410         mutex_unlock(&inode->ei_update_lock);
 411         return ret;
 412 }
 413
 414 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 415                      struct dentry *dentry)
 416 {
 417         struct bch_fs *c = vdir->i_sb->s_fs_info;
 418         struct bch_inode_info *dir = to_bch_ei(vdir);
 419         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 420         int ret;
 421
 422         lockdep_assert_held(&inode->v.i_rwsem);
 423
 424         ret = __bch2_link(c, inode, dir, dentry);
 425         if (unlikely(ret))
 426                 return ret;
 427
 428         ihold(&inode->v);
 429         d_instantiate(dentry, &inode->v);
 430         return 0;
 431 }
 432
 433 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 434 {
 435         struct bch_fs *c = vdir->i_sb->s_fs_info;
 436         struct bch_inode_info *dir = to_bch_ei(vdir);
 437         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 438         struct bch_inode_unpacked dir_u, inode_u;
 439         struct btree_trans trans;
 440         int ret;
 441
 442         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 443         bch2_trans_init(&trans, c, 4, 1024);
 444
 445         do {
 446                 bch2_trans_begin(&trans);
 447
 448                 ret   = bch2_unlink_trans(&trans,
 449                                           dir->v.i_ino, &dir_u,
 450                                           &inode_u, &dentry->d_name) ?:
 451                         bch2_trans_commit(&trans, NULL,
 452                                           &dir->ei_journal_seq,
 453                                           BTREE_INSERT_NOUNLOCK|
 454                                           BTREE_INSERT_NOFAIL);
 455         } while (ret == -EINTR);
 456
 457         if (likely(!ret)) {
 458                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 459
 460                 journal_seq_copy(inode, dir->ei_journal_seq);
 461                 bch2_inode_update_after_write(c, dir, &dir_u,
 462                                               ATTR_MTIME|ATTR_CTIME);
 463                 bch2_inode_update_after_write(c, inode, &inode_u,
 464                                               ATTR_MTIME);
 465         }
 466
 467         bch2_trans_exit(&trans);
 468         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 469
 470         return ret;
 471 }
 472
 473 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 474                         const char *symname)
 475 {
 476         struct bch_fs *c = vdir->i_sb->s_fs_info;
 477         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 478         int ret;
 479
 480         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 481         if (unlikely(IS_ERR(inode)))
 482                 return PTR_ERR(inode);
 483
 484         inode_lock(&inode->v);
 485         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 486         inode_unlock(&inode->v);
 487
 488         if (unlikely(ret))
 489                 goto err;
 490
 491         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 492         if (unlikely(ret))
 493                 goto err;
 494
 495         journal_seq_copy(dir, inode->ei_journal_seq);
 496
 497         ret = __bch2_link(c, inode, dir, dentry);
 498         if (unlikely(ret))
 499                 goto err;
 500
 501         d_instantiate(dentry, &inode->v);
 502         return 0;
 503 err:
 504         iput(&inode->v);
 505         return ret;
 506 }
 507
 508 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 509 {
 510         return bch2_mknod(vdir, dentry, mode|S_IFDIR, 0);
 511 }
 512
 513 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 514                         struct inode *dst_vdir, struct dentry *dst_dentry,
 515                         unsigned flags)
 516 {
 517         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 518         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 519         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 520         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 521         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 522         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 523         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 524         struct btree_trans trans;
 525         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 526                 ? BCH_RENAME_EXCHANGE
 527                 : dst_dentry->d_inode
 528                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 529         u64 journal_seq = 0;
 530         int ret;
 531
 532         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 533                 return -EINVAL;
 534
 535         if (mode == BCH_RENAME_OVERWRITE) {
 536                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 537                                                    0, LLONG_MAX);
 538                 if (ret)
 539                         return ret;
 540         }
 541
 542         bch2_trans_init(&trans, c, 8, 2048);
 543
 544         bch2_lock_inodes(INODE_UPDATE_LOCK,
 545                          src_dir,
 546                          dst_dir,
 547                          src_inode,
 548                          dst_inode);
 549
 550         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 551                 ret = bch2_fs_quota_transfer(c, src_inode,
 552                                              dst_dir->ei_qid,
 553                                              1 << QTYP_PRJ,
 554                                              KEY_TYPE_QUOTA_PREALLOC);
 555                 if (ret)
 556                         goto err;
 557         }
 558
 559         if (mode == BCH_RENAME_EXCHANGE &&
 560             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 561                 ret = bch2_fs_quota_transfer(c, dst_inode,
 562                                              src_dir->ei_qid,
 563                                              1 << QTYP_PRJ,
 564                                              KEY_TYPE_QUOTA_PREALLOC);
 565                 if (ret)
 566                         goto err;
 567         }
 568
 569 retry:
 570         bch2_trans_begin(&trans);
 571         ret   = bch2_rename_trans(&trans,
 572                                   src_dir->v.i_ino, &src_dir_u,
 573                                   dst_dir->v.i_ino, &dst_dir_u,
 574                                   &src_inode_u,
 575                                   &dst_inode_u,
 576                                   &src_dentry->d_name,
 577                                   &dst_dentry->d_name,
 578                                   mode) ?:
 579                 bch2_trans_commit(&trans, NULL,
 580                                   &journal_seq,
 581                                   BTREE_INSERT_NOUNLOCK);
 582         if (ret == -EINTR)
 583                 goto retry;
 584         if (unlikely(ret))
 585                 goto err;
 586
 587         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 588         BUG_ON(dst_inode &&
 589                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 590
 591         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 592                                       ATTR_MTIME|ATTR_CTIME);
 593         journal_seq_copy(src_dir, journal_seq);
 594
 595         if (src_dir != dst_dir) {
 596                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 597                                               ATTR_MTIME|ATTR_CTIME);
 598                 journal_seq_copy(dst_dir, journal_seq);
 599         }
 600
 601         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 602                                       ATTR_CTIME);
 603         journal_seq_copy(src_inode, journal_seq);
 604
 605         if (dst_inode) {
 606                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 607                                               ATTR_CTIME);
 608                 journal_seq_copy(dst_inode, journal_seq);
 609         }
 610 err:
 611         bch2_trans_exit(&trans);
 612
 613         bch2_fs_quota_transfer(c, src_inode,
 614                                bch_qid(&src_inode->ei_inode),
 615                                1 << QTYP_PRJ,
 616                                KEY_TYPE_QUOTA_NOCHECK);
 617         if (dst_inode)
 618                 bch2_fs_quota_transfer(c, dst_inode,
 619                                        bch_qid(&dst_inode->ei_inode),
 620                                        1 << QTYP_PRJ,
 621                                        KEY_TYPE_QUOTA_NOCHECK);
 622
 623         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 624                            src_dir,
 625                            dst_dir,
 626                            src_inode,
 627                            dst_inode);
 628
 629         return ret;
 630 }
 631
 632 void bch2_setattr_copy(struct bch_inode_info *inode,
 633                        struct bch_inode_unpacked *bi,
 634                        struct iattr *attr)
 635 {
 636         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 637         unsigned int ia_valid = attr->ia_valid;
 638
 639         if (ia_valid & ATTR_UID)
 640                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
 641         if (ia_valid & ATTR_GID)
 642                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
 643
 644         if (ia_valid & ATTR_ATIME)
 645                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 646         if (ia_valid & ATTR_MTIME)
 647                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 648         if (ia_valid & ATTR_CTIME)
 649                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 650
 651         if (ia_valid & ATTR_MODE) {
 652                 umode_t mode = attr->ia_mode;
 653                 kgid_t gid = ia_valid & ATTR_GID
 654                         ? attr->ia_gid
 655                         : inode->v.i_gid;
 656
 657                 if (!in_group_p(gid) &&
 658                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 659                         mode &= ~S_ISGID;
 660                 bi->bi_mode = mode;
 661         }
 662 }
 663
 664 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
 665                                 struct iattr *attr)
 666 {
 667         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 668         struct bch_qid qid;
 669         struct btree_trans trans;
 670         struct btree_iter *inode_iter;
 671         struct bch_inode_unpacked inode_u;
 672         struct posix_acl *acl = NULL;
 673         int ret;
 674
 675         mutex_lock(&inode->ei_update_lock);
 676
 677         qid = inode->ei_qid;
 678
 679         if (attr->ia_valid & ATTR_UID)
 680                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 681
 682         if (attr->ia_valid & ATTR_GID)
 683                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 684
 685         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 686                                      KEY_TYPE_QUOTA_PREALLOC);
 687         if (ret)
 688                 goto err;
 689
 690         bch2_trans_init(&trans, c, 0, 0);
 691 retry:
 692         bch2_trans_begin(&trans);
 693         kfree(acl);
 694         acl = NULL;
 695
 696         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 697                                      BTREE_ITER_INTENT);
 698         ret = PTR_ERR_OR_ZERO(inode_iter);
 699         if (ret)
 700                 goto btree_err;
 701
 702         bch2_setattr_copy(inode, &inode_u, attr);
 703
 704         if (attr->ia_valid & ATTR_MODE) {
 705                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 706                 if (ret)
 707                         goto btree_err;
 708         }
 709
 710         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 711                 bch2_trans_commit(&trans, NULL,
 712                                   &inode->ei_journal_seq,
 713                                   BTREE_INSERT_NOUNLOCK|
 714                                   BTREE_INSERT_NOFAIL);
 715 btree_err:
 716         if (ret == -EINTR)
 717                 goto retry;
 718         if (unlikely(ret))
 719                 goto err_trans;
 720
 721         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 722
 723         if (acl)
 724                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 725 err_trans:
 726         bch2_trans_exit(&trans);
 727 err:
 728         mutex_unlock(&inode->ei_update_lock);
 729
 730         return ret;
 731 }
 732
 733 static int bch2_getattr(const struct path *path, struct kstat *stat,
 734                         u32 request_mask, unsigned query_flags)
 735 {
 736         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 737         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 738
 739         stat->dev       = inode->v.i_sb->s_dev;
 740         stat->ino       = inode->v.i_ino;
 741         stat->mode      = inode->v.i_mode;
 742         stat->nlink     = inode->v.i_nlink;
 743         stat->uid       = inode->v.i_uid;
 744         stat->gid       = inode->v.i_gid;
 745         stat->rdev      = inode->v.i_rdev;
 746         stat->size      = i_size_read(&inode->v);
 747         stat->atime     = inode->v.i_atime;
 748         stat->mtime     = inode->v.i_mtime;
 749         stat->ctime     = inode->v.i_ctime;
 750         stat->blksize   = block_bytes(c);
 751         stat->blocks    = inode->v.i_blocks;
 752
 753         if (request_mask & STATX_BTIME) {
 754                 stat->result_mask |= STATX_BTIME;
 755                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 756         }
 757
 758         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 759                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 760         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 761
 762         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 763                 stat->attributes |= STATX_ATTR_APPEND;
 764         stat->attributes_mask    |= STATX_ATTR_APPEND;
 765
 766         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 767                 stat->attributes |= STATX_ATTR_NODUMP;
 768         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 769
 770         return 0;
 771 }
 772
 773 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 774 {
 775         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 776         int ret;
 777
 778         lockdep_assert_held(&inode->v.i_rwsem);
 779
 780         ret = setattr_prepare(dentry, iattr);
 781         if (ret)
 782                 return ret;
 783
 784         return iattr->ia_valid & ATTR_SIZE
 785                 ? bch2_truncate(inode, iattr)
 786                 : bch2_setattr_nonsize(inode, iattr);
 787 }
 788
 789 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
 790 {
 791         struct bch_inode_info *inode =
 792                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
 793
 794         if (IS_ERR(inode))
 795                 return PTR_ERR(inode);
 796
 797         d_mark_tmpfile(dentry, &inode->v);
 798         d_instantiate(dentry, &inode->v);
 799         return 0;
 800 }
 801
 802 static int bch2_fill_extent(struct bch_fs *c,
 803                             struct fiemap_extent_info *info,
 804                             struct bkey_s_c k, unsigned flags)
 805 {
 806         if (bkey_extent_is_data(k.k)) {
 807                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 808                 const union bch_extent_entry *entry;
 809                 struct extent_ptr_decoded p;
 810                 int ret;
 811
 812                 if (k.k->type == KEY_TYPE_reflink_v)
 813                         flags |= FIEMAP_EXTENT_SHARED;
 814
 815                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 816                         int flags2 = 0;
 817                         u64 offset = p.ptr.offset;
 818
 819                         if (p.crc.compression_type)
 820                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 821                         else
 822                                 offset += p.crc.offset;
 823
 824                         if ((offset & (c->opts.block_size - 1)) ||
 825                             (k.k->size & (c->opts.block_size - 1)))
 826                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 827
 828                         ret = fiemap_fill_next_extent(info,
 829                                                 bkey_start_offset(k.k) << 9,
 830                                                 offset << 9,
 831                                                 k.k->size << 9, flags|flags2);
 832                         if (ret)
 833                                 return ret;
 834                 }
 835
 836                 return 0;
 837         } else if (k.k->type == KEY_TYPE_reservation) {
 838                 return fiemap_fill_next_extent(info,
 839                                                bkey_start_offset(k.k) << 9,
 840                                                0, k.k->size << 9,
 841                                                flags|
 842                                                FIEMAP_EXTENT_DELALLOC|
 843                                                FIEMAP_EXTENT_UNWRITTEN);
 844         } else {
 845                 BUG();
 846         }
 847 }
 848
 849 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 850                        u64 start, u64 len)
 851 {
 852         struct bch_fs *c = vinode->i_sb->s_fs_info;
 853         struct bch_inode_info *ei = to_bch_ei(vinode);
 854         struct btree_trans trans;
 855         struct btree_iter *iter;
 856         struct bkey_s_c k;
 857         struct bkey_on_stack cur, prev;
 858         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 859         unsigned offset_into_extent, sectors;
 860         bool have_extent = false;
 861         int ret = 0;
 862
 863         if (start + len < start)
 864                 return -EINVAL;
 865
 866         bkey_on_stack_init(&cur);
 867         bkey_on_stack_init(&prev);
 868         bch2_trans_init(&trans, c, 0, 0);
 869
 870         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 871                                    POS(ei->v.i_ino, start >> 9), 0);
 872 retry:
 873         while ((k = bch2_btree_iter_peek(iter)).k &&
 874                !(ret = bkey_err(k)) &&
 875                bkey_cmp(iter->pos, end) < 0) {
 876                 if (!bkey_extent_is_data(k.k) &&
 877                     k.k->type != KEY_TYPE_reservation) {
 878                         bch2_btree_iter_next(iter);
 879                         continue;
 880                 }
 881
 882                 bkey_on_stack_realloc(&cur, c, k.k->u64s);
 883                 bkey_on_stack_realloc(&prev, c, k.k->u64s);
 884                 bkey_reassemble(cur.k, k);
 885                 k = bkey_i_to_s_c(cur.k);
 886
 887                 offset_into_extent      = iter->pos.offset -
 888                         bkey_start_offset(k.k);
 889                 sectors                 = k.k->size - offset_into_extent;
 890
 891                 ret = bch2_read_indirect_extent(&trans,
 892                                         &offset_into_extent, &cur);
 893                 if (ret)
 894                         break;
 895
 896                 sectors = min(sectors, k.k->size - offset_into_extent);
 897
 898                 if (offset_into_extent)
 899                         bch2_cut_front(POS(k.k->p.inode,
 900                                            bkey_start_offset(k.k) +
 901                                            offset_into_extent),
 902                                        cur.k);
 903                 bch2_key_resize(&cur.k->k, sectors);
 904                 cur.k->k.p = iter->pos;
 905                 cur.k->k.p.offset += cur.k->k.size;
 906
 907                 if (have_extent) {
 908                         ret = bch2_fill_extent(c, info,
 909                                         bkey_i_to_s_c(prev.k), 0);
 910                         if (ret)
 911                                 break;
 912                 }
 913
 914                 bkey_copy(prev.k, cur.k);
 915                 have_extent = true;
 916
 917                 if (k.k->type == KEY_TYPE_reflink_v)
 918                         bch2_btree_iter_set_pos(iter, k.k->p);
 919                 else
 920                         bch2_btree_iter_next(iter);
 921         }
 922
 923         if (ret == -EINTR)
 924                 goto retry;
 925
 926         if (!ret && have_extent)
 927                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 928                                        FIEMAP_EXTENT_LAST);
 929
 930         ret = bch2_trans_exit(&trans) ?: ret;
 931         bkey_on_stack_exit(&cur, c);
 932         bkey_on_stack_exit(&prev, c);
 933         return ret < 0 ? ret : 0;
 934 }
 935
 936 static const struct vm_operations_struct bch_vm_ops = {
 937         .fault          = bch2_page_fault,
 938         .map_pages      = filemap_map_pages,
 939         .page_mkwrite   = bch2_page_mkwrite,
 940 };
 941
 942 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 943 {
 944         file_accessed(file);
 945
 946         vma->vm_ops = &bch_vm_ops;
 947         return 0;
 948 }
 949
 950 /* Directories: */
 951
 952 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 953 {
 954         return generic_file_llseek_size(file, offset, whence,
 955                                         S64_MAX, S64_MAX);
 956 }
 957
 958 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 959 {
 960         struct bch_inode_info *inode = file_bch_inode(file);
 961         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 962
 963         if (!dir_emit_dots(file, ctx))
 964                 return 0;
 965
 966         return bch2_readdir(c, inode->v.i_ino, ctx);
 967 }
 968
 969 static const struct file_operations bch_file_operations = {
 970         .llseek         = bch2_llseek,
 971         .read_iter      = bch2_read_iter,
 972         .write_iter     = bch2_write_iter,
 973         .mmap           = bch2_mmap,
 974         .open           = generic_file_open,
 975         .fsync          = bch2_fsync,
 976         .splice_read    = generic_file_splice_read,
 977         /*
 978          * Broken, on v5.3:
 979         .splice_write   = iter_file_splice_write,
 980         */
 981         .fallocate      = bch2_fallocate_dispatch,
 982         .unlocked_ioctl = bch2_fs_file_ioctl,
 983 #ifdef CONFIG_COMPAT
 984         .compat_ioctl   = bch2_compat_fs_ioctl,
 985 #endif
 986         .remap_file_range = bch2_remap_file_range,
 987 };
 988
 989 static const struct inode_operations bch_file_inode_operations = {
 990         .getattr        = bch2_getattr,
 991         .setattr        = bch2_setattr,
 992         .fiemap         = bch2_fiemap,
 993         .listxattr      = bch2_xattr_list,
 994 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 995         .get_acl        = bch2_get_acl,
 996         .set_acl        = bch2_set_acl,
 997 #endif
 998 };
 999
1000 static const struct inode_operations bch_dir_inode_operations = {
1001         .lookup         = bch2_lookup,
1002         .create         = bch2_create,
1003         .link           = bch2_link,
1004         .unlink         = bch2_unlink,
1005         .symlink        = bch2_symlink,
1006         .mkdir          = bch2_mkdir,
1007         .rmdir          = bch2_unlink,
1008         .mknod          = bch2_mknod,
1009         .rename         = bch2_rename2,
1010         .getattr        = bch2_getattr,
1011         .setattr        = bch2_setattr,
1012         .tmpfile        = bch2_tmpfile,
1013         .listxattr      = bch2_xattr_list,
1014 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1015         .get_acl        = bch2_get_acl,
1016         .set_acl        = bch2_set_acl,
1017 #endif
1018 };
1019
1020 static const struct file_operations bch_dir_file_operations = {
1021         .llseek         = bch2_dir_llseek,
1022         .read           = generic_read_dir,
1023         .iterate_shared = bch2_vfs_readdir,
1024         .fsync          = bch2_fsync,
1025         .unlocked_ioctl = bch2_fs_file_ioctl,
1026 #ifdef CONFIG_COMPAT
1027         .compat_ioctl   = bch2_compat_fs_ioctl,
1028 #endif
1029 };
1030
1031 static const struct inode_operations bch_symlink_inode_operations = {
1032         .get_link       = page_get_link,
1033         .getattr        = bch2_getattr,
1034         .setattr        = bch2_setattr,
1035         .listxattr      = bch2_xattr_list,
1036 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1037         .get_acl        = bch2_get_acl,
1038         .set_acl        = bch2_set_acl,
1039 #endif
1040 };
1041
1042 static const struct inode_operations bch_special_inode_operations = {
1043         .getattr        = bch2_getattr,
1044         .setattr        = bch2_setattr,
1045         .listxattr      = bch2_xattr_list,
1046 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1047         .get_acl        = bch2_get_acl,
1048         .set_acl        = bch2_set_acl,
1049 #endif
1050 };
1051
1052 static const struct address_space_operations bch_address_space_operations = {
1053         .writepage      = bch2_writepage,
1054         .readpage       = bch2_readpage,
1055         .writepages     = bch2_writepages,
1056         .readpages      = bch2_readpages,
1057         .set_page_dirty = __set_page_dirty_nobuffers,
1058         .write_begin    = bch2_write_begin,
1059         .write_end      = bch2_write_end,
1060         .invalidatepage = bch2_invalidatepage,
1061         .releasepage    = bch2_releasepage,
1062         .direct_IO      = noop_direct_IO,
1063 #ifdef CONFIG_MIGRATION
1064         .migratepage    = bch2_migrate_page,
1065 #endif
1066         .error_remove_page = generic_error_remove_page,
1067 };
1068
1069 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1070                 u64 ino, u32 generation)
1071 {
1072         struct bch_fs *c = sb->s_fs_info;
1073         struct inode *vinode;
1074
1075         if (ino < BCACHEFS_ROOT_INO)
1076                 return ERR_PTR(-ESTALE);
1077
1078         vinode = bch2_vfs_inode_get(c, ino);
1079         if (IS_ERR(vinode))
1080                 return ERR_CAST(vinode);
1081         if (generation && vinode->i_generation != generation) {
1082                 /* we didn't find the right inode.. */
1083                 iput(vinode);
1084                 return ERR_PTR(-ESTALE);
1085         }
1086         return vinode;
1087 }
1088
1089 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1090                 int fh_len, int fh_type)
1091 {
1092         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1093                                     bch2_nfs_get_inode);
1094 }
1095
1096 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1097                 int fh_len, int fh_type)
1098 {
1099         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1100                                     bch2_nfs_get_inode);
1101 }
1102
1103 static const struct export_operations bch_export_ops = {
1104         .fh_to_dentry   = bch2_fh_to_dentry,
1105         .fh_to_parent   = bch2_fh_to_parent,
1106         //.get_parent   = bch2_get_parent,
1107 };
1108
1109 static void bch2_vfs_inode_init(struct bch_fs *c,
1110                                 struct bch_inode_info *inode,
1111                                 struct bch_inode_unpacked *bi)
1112 {
1113         bch2_inode_update_after_write(c, inode, bi, ~0);
1114
1115         inode->v.i_blocks       = bi->bi_sectors;
1116         inode->v.i_ino          = bi->bi_inum;
1117         inode->v.i_rdev         = bi->bi_dev;
1118         inode->v.i_generation   = bi->bi_generation;
1119         inode->v.i_size         = bi->bi_size;
1120
1121         inode->ei_journal_seq   = 0;
1122         inode->ei_quota_reserved = 0;
1123         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1124         inode->ei_qid           = bch_qid(bi);
1125
1126         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1127
1128         switch (inode->v.i_mode & S_IFMT) {
1129         case S_IFREG:
1130                 inode->v.i_op   = &bch_file_inode_operations;
1131                 inode->v.i_fop  = &bch_file_operations;
1132                 break;
1133         case S_IFDIR:
1134                 inode->v.i_op   = &bch_dir_inode_operations;
1135                 inode->v.i_fop  = &bch_dir_file_operations;
1136                 break;
1137         case S_IFLNK:
1138                 inode_nohighmem(&inode->v);
1139                 inode->v.i_op   = &bch_symlink_inode_operations;
1140                 break;
1141         default:
1142                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1143                 inode->v.i_op   = &bch_special_inode_operations;
1144                 break;
1145         }
1146 }
1147
1148 static struct inode *bch2_alloc_inode(struct super_block *sb)
1149 {
1150         struct bch_inode_info *inode;
1151
1152         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1153         if (!inode)
1154                 return NULL;
1155
1156         inode_init_once(&inode->v);
1157         mutex_init(&inode->ei_update_lock);
1158         pagecache_lock_init(&inode->ei_pagecache_lock);
1159         mutex_init(&inode->ei_quota_lock);
1160         inode->ei_journal_seq = 0;
1161
1162         return &inode->v;
1163 }
1164
1165 static void bch2_i_callback(struct rcu_head *head)
1166 {
1167         struct inode *vinode = container_of(head, struct inode, i_rcu);
1168         struct bch_inode_info *inode = to_bch_ei(vinode);
1169
1170         kmem_cache_free(bch2_inode_cache, inode);
1171 }
1172
1173 static void bch2_destroy_inode(struct inode *vinode)
1174 {
1175         call_rcu(&vinode->i_rcu, bch2_i_callback);
1176 }
1177
1178 static int inode_update_times_fn(struct bch_inode_info *inode,
1179                                  struct bch_inode_unpacked *bi,
1180                                  void *p)
1181 {
1182         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1183
1184         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1185         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1186         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1187
1188         return 0;
1189 }
1190
1191 static int bch2_vfs_write_inode(struct inode *vinode,
1192                                 struct writeback_control *wbc)
1193 {
1194         struct bch_fs *c = vinode->i_sb->s_fs_info;
1195         struct bch_inode_info *inode = to_bch_ei(vinode);
1196         int ret;
1197
1198         mutex_lock(&inode->ei_update_lock);
1199         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1200                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1201         mutex_unlock(&inode->ei_update_lock);
1202
1203         return ret;
1204 }
1205
1206 static void bch2_evict_inode(struct inode *vinode)
1207 {
1208         struct bch_fs *c = vinode->i_sb->s_fs_info;
1209         struct bch_inode_info *inode = to_bch_ei(vinode);
1210
1211         truncate_inode_pages_final(&inode->v.i_data);
1212
1213         clear_inode(&inode->v);
1214
1215         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1216
1217         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1218                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1219                                 KEY_TYPE_QUOTA_WARN);
1220                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1221                                 KEY_TYPE_QUOTA_WARN);
1222                 bch2_inode_rm(c, inode->v.i_ino);
1223         }
1224 }
1225
1226 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1227 {
1228         struct super_block *sb = dentry->d_sb;
1229         struct bch_fs *c = sb->s_fs_info;
1230         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1231         unsigned shift = sb->s_blocksize_bits - 9;
1232         u64 fsid;
1233
1234         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1235         buf->f_bsize    = sb->s_blocksize;
1236         buf->f_blocks   = usage.capacity >> shift;
1237         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1238         buf->f_bavail   = buf->f_bfree;
1239         buf->f_files    = usage.nr_inodes;
1240         buf->f_ffree    = U64_MAX;
1241
1242         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1243                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1244         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1245         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1246         buf->f_namelen  = BCH_NAME_MAX;
1247
1248         return 0;
1249 }
1250
1251 static int bch2_sync_fs(struct super_block *sb, int wait)
1252 {
1253         struct bch_fs *c = sb->s_fs_info;
1254
1255         if (c->opts.journal_flush_disabled)
1256                 return 0;
1257
1258         if (!wait) {
1259                 bch2_journal_flush_async(&c->journal, NULL);
1260                 return 0;
1261         }
1262
1263         return bch2_journal_flush(&c->journal);
1264 }
1265
1266 static struct bch_fs *bch2_path_to_fs(const char *dev)
1267 {
1268         struct bch_fs *c;
1269         struct block_device *bdev = lookup_bdev(dev);
1270
1271         if (IS_ERR(bdev))
1272                 return ERR_CAST(bdev);
1273
1274         c = bch2_bdev_to_fs(bdev);
1275         bdput(bdev);
1276         return c ?: ERR_PTR(-ENOENT);
1277 }
1278
1279 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1280                                                unsigned nr_devs, struct bch_opts opts)
1281 {
1282         struct bch_fs *c, *c1, *c2;
1283         size_t i;
1284
1285         if (!nr_devs)
1286                 return ERR_PTR(-EINVAL);
1287
1288         c = bch2_fs_open(devs, nr_devs, opts);
1289
1290         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1291                 /*
1292                  * Already open?
1293                  * Look up each block device, make sure they all belong to a
1294                  * filesystem and they all belong to the _same_ filesystem
1295                  */
1296
1297                 c1 = bch2_path_to_fs(devs[0]);
1298                 if (IS_ERR(c1))
1299                         return c;
1300
1301                 for (i = 1; i < nr_devs; i++) {
1302                         c2 = bch2_path_to_fs(devs[i]);
1303                         if (!IS_ERR(c2))
1304                                 closure_put(&c2->cl);
1305
1306                         if (c1 != c2) {
1307                                 closure_put(&c1->cl);
1308                                 return c;
1309                         }
1310                 }
1311
1312                 c = c1;
1313         }
1314
1315         if (IS_ERR(c))
1316                 return c;
1317
1318         mutex_lock(&c->state_lock);
1319
1320         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1321                 mutex_unlock(&c->state_lock);
1322                 closure_put(&c->cl);
1323                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1324                 return ERR_PTR(-EINVAL);
1325         }
1326
1327         mutex_unlock(&c->state_lock);
1328
1329         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1330         return c;
1331 }
1332
1333 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1334                                              struct bch_opts opts)
1335 {
1336         char *dev_name = NULL, **devs = NULL, *s;
1337         struct bch_fs *c = ERR_PTR(-ENOMEM);
1338         size_t i, nr_devs = 0;
1339
1340         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1341         if (!dev_name)
1342                 goto err;
1343
1344         for (s = dev_name; s; s = strchr(s + 1, ':'))
1345                 nr_devs++;
1346
1347         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1348         if (!devs)
1349                 goto err;
1350
1351         for (i = 0, s = dev_name;
1352              s;
1353              (s = strchr(s, ':')) && (*s++ = '\0'))
1354                 devs[i++] = s;
1355
1356         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1357 err:
1358         kfree(devs);
1359         kfree(dev_name);
1360         return c;
1361 }
1362
1363 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1364 {
1365         struct bch_fs *c = sb->s_fs_info;
1366         struct bch_opts opts = bch2_opts_empty();
1367         int ret;
1368
1369         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1370
1371         ret = bch2_parse_mount_opts(&opts, data);
1372         if (ret)
1373                 return ret;
1374
1375         if (opts.read_only != c->opts.read_only) {
1376                 mutex_lock(&c->state_lock);
1377
1378                 if (opts.read_only) {
1379                         bch2_fs_read_only(c);
1380
1381                         sb->s_flags |= SB_RDONLY;
1382                 } else {
1383                         ret = bch2_fs_read_write(c);
1384                         if (ret) {
1385                                 bch_err(c, "error going rw: %i", ret);
1386                                 mutex_unlock(&c->state_lock);
1387                                 return -EINVAL;
1388                         }
1389
1390                         sb->s_flags &= ~SB_RDONLY;
1391                 }
1392
1393                 c->opts.read_only = opts.read_only;
1394
1395                 mutex_unlock(&c->state_lock);
1396         }
1397
1398         if (opts.errors >= 0)
1399                 c->opts.errors = opts.errors;
1400
1401         return ret;
1402 }
1403
1404 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1405 {
1406         struct bch_fs *c = root->d_sb->s_fs_info;
1407         enum bch_opt_id i;
1408         char buf[512];
1409
1410         for (i = 0; i < bch2_opts_nr; i++) {
1411                 const struct bch_option *opt = &bch2_opt_table[i];
1412                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1413
1414                 if (!(opt->mode & OPT_MOUNT))
1415                         continue;
1416
1417                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1418                         continue;
1419
1420                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1421                                  OPT_SHOW_MOUNT_STYLE);
1422                 seq_putc(seq, ',');
1423                 seq_puts(seq, buf);
1424         }
1425
1426         return 0;
1427
1428 }
1429
1430 static const struct super_operations bch_super_operations = {
1431         .alloc_inode    = bch2_alloc_inode,
1432         .destroy_inode  = bch2_destroy_inode,
1433         .write_inode    = bch2_vfs_write_inode,
1434         .evict_inode    = bch2_evict_inode,
1435         .sync_fs        = bch2_sync_fs,
1436         .statfs         = bch2_statfs,
1437         .show_options   = bch2_show_options,
1438         .remount_fs     = bch2_remount,
1439 #if 0
1440         .put_super      = bch2_put_super,
1441         .freeze_fs      = bch2_freeze,
1442         .unfreeze_fs    = bch2_unfreeze,
1443 #endif
1444 };
1445
1446 static int bch2_test_super(struct super_block *s, void *data)
1447 {
1448         return s->s_fs_info == data;
1449 }
1450
1451 static int bch2_set_super(struct super_block *s, void *data)
1452 {
1453         s->s_fs_info = data;
1454         return 0;
1455 }
1456
1457 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1458                                  int flags, const char *dev_name, void *data)
1459 {
1460         struct bch_fs *c;
1461         struct bch_dev *ca;
1462         struct super_block *sb;
1463         struct inode *vinode;
1464         struct bch_opts opts = bch2_opts_empty();
1465         unsigned i;
1466         int ret;
1467
1468         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1469
1470         ret = bch2_parse_mount_opts(&opts, data);
1471         if (ret)
1472                 return ERR_PTR(ret);
1473
1474         c = bch2_open_as_blockdevs(dev_name, opts);
1475         if (IS_ERR(c))
1476                 return ERR_CAST(c);
1477
1478         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1479         if (IS_ERR(sb)) {
1480                 closure_put(&c->cl);
1481                 return ERR_CAST(sb);
1482         }
1483
1484         BUG_ON(sb->s_fs_info != c);
1485
1486         if (sb->s_root) {
1487                 closure_put(&c->cl);
1488
1489                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1490                         ret = -EBUSY;
1491                         goto err_put_super;
1492                 }
1493                 goto out;
1494         }
1495
1496         sb->s_blocksize         = block_bytes(c);
1497         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1498         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1499         sb->s_op                = &bch_super_operations;
1500         sb->s_export_op         = &bch_export_ops;
1501 #ifdef CONFIG_BCACHEFS_QUOTA
1502         sb->s_qcop              = &bch2_quotactl_operations;
1503         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1504 #endif
1505         sb->s_xattr             = bch2_xattr_handlers;
1506         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1507         sb->s_time_gran         = c->sb.time_precision;
1508         c->vfs_sb               = sb;
1509         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1510
1511         ret = super_setup_bdi(sb);
1512         if (ret)
1513                 goto err_put_super;
1514
1515         sb->s_bdi->congested_fn         = bch2_congested;
1516         sb->s_bdi->congested_data       = c;
1517         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1518
1519         for_each_online_member(ca, c, i) {
1520                 struct block_device *bdev = ca->disk_sb.bdev;
1521
1522                 /* XXX: create an anonymous device for multi device filesystems */
1523                 sb->s_bdev      = bdev;
1524                 sb->s_dev       = bdev->bd_dev;
1525                 percpu_ref_put(&ca->io_ref);
1526                 break;
1527         }
1528
1529 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1530         if (c->opts.acl)
1531                 sb->s_flags     |= SB_POSIXACL;
1532 #endif
1533
1534         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1535         if (IS_ERR(vinode)) {
1536                 bch_err(c, "error mounting: error getting root inode %i",
1537                         (int) PTR_ERR(vinode));
1538                 ret = PTR_ERR(vinode);
1539                 goto err_put_super;
1540         }
1541
1542         sb->s_root = d_make_root(vinode);
1543         if (!sb->s_root) {
1544                 bch_err(c, "error mounting: error allocating root dentry");
1545                 ret = -ENOMEM;
1546                 goto err_put_super;
1547         }
1548
1549         sb->s_flags |= SB_ACTIVE;
1550 out:
1551         return dget(sb->s_root);
1552
1553 err_put_super:
1554         deactivate_locked_super(sb);
1555         return ERR_PTR(ret);
1556 }
1557
1558 static void bch2_kill_sb(struct super_block *sb)
1559 {
1560         struct bch_fs *c = sb->s_fs_info;
1561
1562         generic_shutdown_super(sb);
1563
1564         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1565                 bch2_fs_stop(c);
1566         else
1567                 closure_put(&c->cl);
1568 }
1569
1570 static struct file_system_type bcache_fs_type = {
1571         .owner          = THIS_MODULE,
1572         .name           = "bcachefs",
1573         .mount          = bch2_mount,
1574         .kill_sb        = bch2_kill_sb,
1575         .fs_flags       = FS_REQUIRES_DEV,
1576 };
1577
1578 MODULE_ALIAS_FS("bcachefs");
1579
1580 void bch2_vfs_exit(void)
1581 {
1582         unregister_filesystem(&bcache_fs_type);
1583         if (bch2_inode_cache)
1584                 kmem_cache_destroy(bch2_inode_cache);
1585 }
1586
1587 int __init bch2_vfs_init(void)
1588 {
1589         int ret = -ENOMEM;
1590
1591         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1592         if (!bch2_inode_cache)
1593                 goto err;
1594
1595         ret = register_filesystem(&bcache_fs_type);
1596         if (ret)
1597                 goto err;
1598
1599         return 0;
1600 err:
1601         bch2_vfs_exit();
1602         return ret;
1603 }
1604
1605 #endif /* NO_BCACHEFS_FS */