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_ATOMIC|
 144                                   BTREE_INSERT_NOUNLOCK|
 145                                   BTREE_INSERT_NOFAIL);
 146         if (ret == -EINTR)
 147                 goto retry;
 148
 149         /*
 150          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 151          * this is important for inode updates via bchfs_write_index_update
 152          */
 153         if (!ret)
 154                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 155
 156         bch2_trans_exit(&trans);
 157         return ret < 0 ? ret : 0;
 158 }
 159
 160 int bch2_fs_quota_transfer(struct bch_fs *c,
 161                            struct bch_inode_info *inode,
 162                            struct bch_qid new_qid,
 163                            unsigned qtypes,
 164                            enum quota_acct_mode mode)
 165 {
 166         unsigned i;
 167         int ret;
 168
 169         qtypes &= enabled_qtypes(c);
 170
 171         for (i = 0; i < QTYP_NR; i++)
 172                 if (new_qid.q[i] == inode->ei_qid.q[i])
 173                         qtypes &= ~(1U << i);
 174
 175         if (!qtypes)
 176                 return 0;
 177
 178         mutex_lock(&inode->ei_quota_lock);
 179
 180         ret = bch2_quota_transfer(c, qtypes, new_qid,
 181                                   inode->ei_qid,
 182                                   inode->v.i_blocks +
 183                                   inode->ei_quota_reserved,
 184                                   mode);
 185         if (!ret)
 186                 for (i = 0; i < QTYP_NR; i++)
 187                         if (qtypes & (1 << i))
 188                                 inode->ei_qid.q[i] = new_qid.q[i];
 189
 190         mutex_unlock(&inode->ei_quota_lock);
 191
 192         return ret;
 193 }
 194
 195 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 196 {
 197         struct bch_inode_unpacked inode_u;
 198         struct bch_inode_info *inode;
 199         int ret;
 200
 201         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 202         if (unlikely(!inode))
 203                 return ERR_PTR(-ENOMEM);
 204         if (!(inode->v.i_state & I_NEW))
 205                 return &inode->v;
 206
 207         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 208         if (ret) {
 209                 iget_failed(&inode->v);
 210                 return ERR_PTR(ret);
 211         }
 212
 213         bch2_vfs_inode_init(c, inode, &inode_u);
 214
 215         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 216
 217         unlock_new_inode(&inode->v);
 218
 219         return &inode->v;
 220 }
 221
 222 static struct bch_inode_info *
 223 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 224               umode_t mode, dev_t rdev, bool tmpfile)
 225 {
 226         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 227         struct user_namespace *ns = dir->v.i_sb->s_user_ns;
 228         struct btree_trans trans;
 229         struct bch_inode_unpacked dir_u;
 230         struct bch_inode_info *inode, *old;
 231         struct bch_inode_unpacked inode_u;
 232         struct posix_acl *default_acl = NULL, *acl = NULL;
 233         u64 journal_seq = 0;
 234         int ret;
 235
 236         /*
 237          * preallocate acls + vfs inode before btree transaction, so that
 238          * nothing can fail after the transaction succeeds:
 239          */
 240 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 241         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 242         if (ret)
 243                 return ERR_PTR(ret);
 244 #endif
 245         inode = to_bch_ei(new_inode(c->vfs_sb));
 246         if (unlikely(!inode)) {
 247                 inode = ERR_PTR(-ENOMEM);
 248                 goto err;
 249         }
 250
 251         bch2_inode_init_early(c, &inode_u);
 252
 253         if (!tmpfile)
 254                 mutex_lock(&dir->ei_update_lock);
 255
 256         bch2_trans_init(&trans, c, 8, 1024);
 257 retry:
 258         bch2_trans_begin(&trans);
 259
 260         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 261                                   !tmpfile ? &dentry->d_name : NULL,
 262                                   from_kuid(ns, current_fsuid()),
 263                                   from_kgid(ns, current_fsgid()),
 264                                   mode, rdev,
 265                                   default_acl, acl) ?:
 266                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 267                                 KEY_TYPE_QUOTA_PREALLOC);
 268         if (unlikely(ret))
 269                 goto err_before_quota;
 270
 271         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 272                                   BTREE_INSERT_ATOMIC|
 273                                   BTREE_INSERT_NOUNLOCK);
 274         if (unlikely(ret)) {
 275                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 276                                 KEY_TYPE_QUOTA_WARN);
 277 err_before_quota:
 278                 if (ret == -EINTR)
 279                         goto retry;
 280                 goto err_trans;
 281         }
 282
 283         if (!tmpfile) {
 284                 bch2_inode_update_after_write(c, dir, &dir_u,
 285                                               ATTR_MTIME|ATTR_CTIME);
 286                 journal_seq_copy(dir, journal_seq);
 287                 mutex_unlock(&dir->ei_update_lock);
 288         }
 289
 290         bch2_vfs_inode_init(c, inode, &inode_u);
 291         journal_seq_copy(inode, journal_seq);
 292
 293         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 294         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 295
 296         /*
 297          * we must insert the new inode into the inode cache before calling
 298          * bch2_trans_exit() and dropping locks, else we could race with another
 299          * thread pulling the inode in and modifying it:
 300          */
 301
 302         old = to_bch_ei(insert_inode_locked2(&inode->v));
 303         if (unlikely(old)) {
 304                 /*
 305                  * We raced, another process pulled the new inode into cache
 306                  * before us:
 307                  */
 308                 journal_seq_copy(old, journal_seq);
 309                 make_bad_inode(&inode->v);
 310                 iput(&inode->v);
 311
 312                 inode = old;
 313         } else {
 314                 /*
 315                  * we really don't want insert_inode_locked2() to be setting
 316                  * I_NEW...
 317                  */
 318                 unlock_new_inode(&inode->v);
 319         }
 320
 321         bch2_trans_exit(&trans);
 322 err:
 323         posix_acl_release(default_acl);
 324         posix_acl_release(acl);
 325         return inode;
 326 err_trans:
 327         if (!tmpfile)
 328                 mutex_unlock(&dir->ei_update_lock);
 329
 330         bch2_trans_exit(&trans);
 331         make_bad_inode(&inode->v);
 332         iput(&inode->v);
 333         inode = ERR_PTR(ret);
 334         goto err;
 335 }
 336
 337 /* methods */
 338
 339 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 340                                   unsigned int flags)
 341 {
 342         struct bch_fs *c = vdir->i_sb->s_fs_info;
 343         struct bch_inode_info *dir = to_bch_ei(vdir);
 344         struct inode *vinode = NULL;
 345         u64 inum;
 346
 347         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 348                                   &dir->ei_str_hash,
 349                                   &dentry->d_name);
 350
 351         if (inum)
 352                 vinode = bch2_vfs_inode_get(c, inum);
 353
 354         return d_splice_alias(vinode, dentry);
 355 }
 356
 357 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 358                       umode_t mode, dev_t rdev)
 359 {
 360         struct bch_inode_info *inode =
 361                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 362
 363         if (IS_ERR(inode))
 364                 return PTR_ERR(inode);
 365
 366         d_instantiate(dentry, &inode->v);
 367         return 0;
 368 }
 369
 370 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 371                        umode_t mode, bool excl)
 372 {
 373         return bch2_mknod(vdir, dentry, mode|S_IFREG, 0);
 374 }
 375
 376 static int __bch2_link(struct bch_fs *c,
 377                        struct bch_inode_info *inode,
 378                        struct bch_inode_info *dir,
 379                        struct dentry *dentry)
 380 {
 381         struct btree_trans trans;
 382         struct bch_inode_unpacked dir_u, inode_u;
 383         int ret;
 384
 385         mutex_lock(&inode->ei_update_lock);
 386         bch2_trans_init(&trans, c, 4, 1024);
 387
 388         do {
 389                 bch2_trans_begin(&trans);
 390                 ret   = bch2_link_trans(&trans,
 391                                         dir->v.i_ino,
 392                                         inode->v.i_ino, &dir_u, &inode_u,
 393                                         &dentry->d_name) ?:
 394                         bch2_trans_commit(&trans, NULL,
 395                                         &inode->ei_journal_seq,
 396                                         BTREE_INSERT_ATOMIC|
 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_ATOMIC|
 454                                           BTREE_INSERT_NOUNLOCK|
 455                                           BTREE_INSERT_NOFAIL);
 456         } while (ret == -EINTR);
 457
 458         if (likely(!ret)) {
 459                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 460
 461                 journal_seq_copy(inode, dir->ei_journal_seq);
 462                 bch2_inode_update_after_write(c, dir, &dir_u,
 463                                               ATTR_MTIME|ATTR_CTIME);
 464                 bch2_inode_update_after_write(c, inode, &inode_u,
 465                                               ATTR_MTIME);
 466         }
 467
 468         bch2_trans_exit(&trans);
 469         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 470
 471         return ret;
 472 }
 473
 474 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 475                         const char *symname)
 476 {
 477         struct bch_fs *c = vdir->i_sb->s_fs_info;
 478         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 479         int ret;
 480
 481         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 482         if (unlikely(IS_ERR(inode)))
 483                 return PTR_ERR(inode);
 484
 485         inode_lock(&inode->v);
 486         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 487         inode_unlock(&inode->v);
 488
 489         if (unlikely(ret))
 490                 goto err;
 491
 492         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 493         if (unlikely(ret))
 494                 goto err;
 495
 496         journal_seq_copy(dir, inode->ei_journal_seq);
 497
 498         ret = __bch2_link(c, inode, dir, dentry);
 499         if (unlikely(ret))
 500                 goto err;
 501
 502         d_instantiate(dentry, &inode->v);
 503         return 0;
 504 err:
 505         iput(&inode->v);
 506         return ret;
 507 }
 508
 509 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 510 {
 511         return bch2_mknod(vdir, dentry, mode|S_IFDIR, 0);
 512 }
 513
 514 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 515                         struct inode *dst_vdir, struct dentry *dst_dentry,
 516                         unsigned flags)
 517 {
 518         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 519         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 520         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 521         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 522         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 523         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 524         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 525         struct btree_trans trans;
 526         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 527                 ? BCH_RENAME_EXCHANGE
 528                 : dst_dentry->d_inode
 529                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 530         u64 journal_seq = 0;
 531         int ret;
 532
 533         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 534                 return -EINVAL;
 535
 536         if (mode == BCH_RENAME_OVERWRITE) {
 537                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 538                                                    0, LLONG_MAX);
 539                 if (ret)
 540                         return ret;
 541         }
 542
 543         bch2_trans_init(&trans, c, 8, 2048);
 544
 545         bch2_lock_inodes(INODE_UPDATE_LOCK,
 546                          src_dir,
 547                          dst_dir,
 548                          src_inode,
 549                          dst_inode);
 550
 551         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 552                 ret = bch2_fs_quota_transfer(c, src_inode,
 553                                              dst_dir->ei_qid,
 554                                              1 << QTYP_PRJ,
 555                                              KEY_TYPE_QUOTA_PREALLOC);
 556                 if (ret)
 557                         goto err;
 558         }
 559
 560         if (mode == BCH_RENAME_EXCHANGE &&
 561             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 562                 ret = bch2_fs_quota_transfer(c, dst_inode,
 563                                              src_dir->ei_qid,
 564                                              1 << QTYP_PRJ,
 565                                              KEY_TYPE_QUOTA_PREALLOC);
 566                 if (ret)
 567                         goto err;
 568         }
 569
 570 retry:
 571         bch2_trans_begin(&trans);
 572         ret   = bch2_rename_trans(&trans,
 573                                   src_dir->v.i_ino, &src_dir_u,
 574                                   dst_dir->v.i_ino, &dst_dir_u,
 575                                   &src_inode_u,
 576                                   &dst_inode_u,
 577                                   &src_dentry->d_name,
 578                                   &dst_dentry->d_name,
 579                                   mode) ?:
 580                 bch2_trans_commit(&trans, NULL,
 581                                   &journal_seq,
 582                                   BTREE_INSERT_ATOMIC|
 583                                   BTREE_INSERT_NOUNLOCK);
 584         if (ret == -EINTR)
 585                 goto retry;
 586         if (unlikely(ret))
 587                 goto err;
 588
 589         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 590         BUG_ON(dst_inode &&
 591                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 592
 593         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 594                                       ATTR_MTIME|ATTR_CTIME);
 595         journal_seq_copy(src_dir, journal_seq);
 596
 597         if (src_dir != dst_dir) {
 598                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 599                                               ATTR_MTIME|ATTR_CTIME);
 600                 journal_seq_copy(dst_dir, journal_seq);
 601         }
 602
 603         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 604                                       ATTR_CTIME);
 605         journal_seq_copy(src_inode, journal_seq);
 606
 607         if (dst_inode) {
 608                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 609                                               ATTR_CTIME);
 610                 journal_seq_copy(dst_inode, journal_seq);
 611         }
 612 err:
 613         bch2_trans_exit(&trans);
 614
 615         bch2_fs_quota_transfer(c, src_inode,
 616                                bch_qid(&src_inode->ei_inode),
 617                                1 << QTYP_PRJ,
 618                                KEY_TYPE_QUOTA_NOCHECK);
 619         if (dst_inode)
 620                 bch2_fs_quota_transfer(c, dst_inode,
 621                                        bch_qid(&dst_inode->ei_inode),
 622                                        1 << QTYP_PRJ,
 623                                        KEY_TYPE_QUOTA_NOCHECK);
 624
 625         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 626                            src_dir,
 627                            dst_dir,
 628                            src_inode,
 629                            dst_inode);
 630
 631         return ret;
 632 }
 633
 634 void bch2_setattr_copy(struct bch_inode_info *inode,
 635                        struct bch_inode_unpacked *bi,
 636                        struct iattr *attr)
 637 {
 638         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 639         unsigned int ia_valid = attr->ia_valid;
 640
 641         if (ia_valid & ATTR_UID)
 642                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
 643         if (ia_valid & ATTR_GID)
 644                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
 645
 646         if (ia_valid & ATTR_ATIME)
 647                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 648         if (ia_valid & ATTR_MTIME)
 649                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 650         if (ia_valid & ATTR_CTIME)
 651                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 652
 653         if (ia_valid & ATTR_MODE) {
 654                 umode_t mode = attr->ia_mode;
 655                 kgid_t gid = ia_valid & ATTR_GID
 656                         ? attr->ia_gid
 657                         : inode->v.i_gid;
 658
 659                 if (!in_group_p(gid) &&
 660                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 661                         mode &= ~S_ISGID;
 662                 bi->bi_mode = mode;
 663         }
 664 }
 665
 666 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
 667                                 struct iattr *attr)
 668 {
 669         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 670         struct bch_qid qid;
 671         struct btree_trans trans;
 672         struct btree_iter *inode_iter;
 673         struct bch_inode_unpacked inode_u;
 674         struct posix_acl *acl = NULL;
 675         int ret;
 676
 677         mutex_lock(&inode->ei_update_lock);
 678
 679         qid = inode->ei_qid;
 680
 681         if (attr->ia_valid & ATTR_UID)
 682                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 683
 684         if (attr->ia_valid & ATTR_GID)
 685                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 686
 687         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 688                                      KEY_TYPE_QUOTA_PREALLOC);
 689         if (ret)
 690                 goto err;
 691
 692         bch2_trans_init(&trans, c, 0, 0);
 693 retry:
 694         bch2_trans_begin(&trans);
 695         kfree(acl);
 696         acl = NULL;
 697
 698         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 699                                      BTREE_ITER_INTENT);
 700         ret = PTR_ERR_OR_ZERO(inode_iter);
 701         if (ret)
 702                 goto btree_err;
 703
 704         bch2_setattr_copy(inode, &inode_u, attr);
 705
 706         if (attr->ia_valid & ATTR_MODE) {
 707                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 708                 if (ret)
 709                         goto btree_err;
 710         }
 711
 712         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 713                 bch2_trans_commit(&trans, NULL,
 714                                   &inode->ei_journal_seq,
 715                                   BTREE_INSERT_ATOMIC|
 716                                   BTREE_INSERT_NOUNLOCK|
 717                                   BTREE_INSERT_NOFAIL);
 718 btree_err:
 719         if (ret == -EINTR)
 720                 goto retry;
 721         if (unlikely(ret))
 722                 goto err_trans;
 723
 724         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 725
 726         if (acl)
 727                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 728 err_trans:
 729         bch2_trans_exit(&trans);
 730 err:
 731         mutex_unlock(&inode->ei_update_lock);
 732
 733         return ret;
 734 }
 735
 736 static int bch2_getattr(const struct path *path, struct kstat *stat,
 737                         u32 request_mask, unsigned query_flags)
 738 {
 739         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 740         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 741
 742         stat->dev       = inode->v.i_sb->s_dev;
 743         stat->ino       = inode->v.i_ino;
 744         stat->mode      = inode->v.i_mode;
 745         stat->nlink     = inode->v.i_nlink;
 746         stat->uid       = inode->v.i_uid;
 747         stat->gid       = inode->v.i_gid;
 748         stat->rdev      = inode->v.i_rdev;
 749         stat->size      = i_size_read(&inode->v);
 750         stat->atime     = inode->v.i_atime;
 751         stat->mtime     = inode->v.i_mtime;
 752         stat->ctime     = inode->v.i_ctime;
 753         stat->blksize   = block_bytes(c);
 754         stat->blocks    = inode->v.i_blocks;
 755
 756         if (request_mask & STATX_BTIME) {
 757                 stat->result_mask |= STATX_BTIME;
 758                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 759         }
 760
 761         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 762                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 763         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
 764
 765         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 766                 stat->attributes |= STATX_ATTR_APPEND;
 767         stat->attributes_mask    |= STATX_ATTR_APPEND;
 768
 769         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 770                 stat->attributes |= STATX_ATTR_NODUMP;
 771         stat->attributes_mask    |= STATX_ATTR_NODUMP;
 772
 773         return 0;
 774 }
 775
 776 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 777 {
 778         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 779         int ret;
 780
 781         lockdep_assert_held(&inode->v.i_rwsem);
 782
 783         ret = setattr_prepare(dentry, iattr);
 784         if (ret)
 785                 return ret;
 786
 787         return iattr->ia_valid & ATTR_SIZE
 788                 ? bch2_truncate(inode, iattr)
 789                 : bch2_setattr_nonsize(inode, iattr);
 790 }
 791
 792 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
 793 {
 794         struct bch_inode_info *inode =
 795                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
 796
 797         if (IS_ERR(inode))
 798                 return PTR_ERR(inode);
 799
 800         d_mark_tmpfile(dentry, &inode->v);
 801         d_instantiate(dentry, &inode->v);
 802         return 0;
 803 }
 804
 805 static int bch2_fill_extent(struct bch_fs *c,
 806                             struct fiemap_extent_info *info,
 807                             struct bkey_s_c k, unsigned flags)
 808 {
 809         if (bkey_extent_is_data(k.k)) {
 810                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 811                 const union bch_extent_entry *entry;
 812                 struct extent_ptr_decoded p;
 813                 int ret;
 814
 815                 if (k.k->type == KEY_TYPE_reflink_v)
 816                         flags |= FIEMAP_EXTENT_SHARED;
 817
 818                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 819                         int flags2 = 0;
 820                         u64 offset = p.ptr.offset;
 821
 822                         if (p.crc.compression_type)
 823                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 824                         else
 825                                 offset += p.crc.offset;
 826
 827                         if ((offset & (c->opts.block_size - 1)) ||
 828                             (k.k->size & (c->opts.block_size - 1)))
 829                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 830
 831                         ret = fiemap_fill_next_extent(info,
 832                                                 bkey_start_offset(k.k) << 9,
 833                                                 offset << 9,
 834                                                 k.k->size << 9, flags|flags2);
 835                         if (ret)
 836                                 return ret;
 837                 }
 838
 839                 return 0;
 840         } else if (k.k->type == KEY_TYPE_reservation) {
 841                 return fiemap_fill_next_extent(info,
 842                                                bkey_start_offset(k.k) << 9,
 843                                                0, k.k->size << 9,
 844                                                flags|
 845                                                FIEMAP_EXTENT_DELALLOC|
 846                                                FIEMAP_EXTENT_UNWRITTEN);
 847         } else {
 848                 BUG();
 849         }
 850 }
 851
 852 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 853                        u64 start, u64 len)
 854 {
 855         struct bch_fs *c = vinode->i_sb->s_fs_info;
 856         struct bch_inode_info *ei = to_bch_ei(vinode);
 857         struct btree_trans trans;
 858         struct btree_iter *iter;
 859         struct bkey_s_c k;
 860         struct bkey_on_stack cur, prev;
 861         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 862         unsigned offset_into_extent, sectors;
 863         bool have_extent = false;
 864         int ret = 0;
 865
 866         if (start + len < start)
 867                 return -EINVAL;
 868
 869         bkey_on_stack_init(&cur);
 870         bkey_on_stack_init(&prev);
 871         bch2_trans_init(&trans, c, 0, 0);
 872
 873         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 874                                    POS(ei->v.i_ino, start >> 9), 0);
 875 retry:
 876         while ((k = bch2_btree_iter_peek(iter)).k &&
 877                !(ret = bkey_err(k)) &&
 878                bkey_cmp(iter->pos, end) < 0) {
 879                 if (!bkey_extent_is_data(k.k) &&
 880                     k.k->type != KEY_TYPE_reservation) {
 881                         bch2_btree_iter_next(iter);
 882                         continue;
 883                 }
 884
 885                 bkey_on_stack_realloc(&cur, c, k.k->u64s);
 886                 bkey_on_stack_realloc(&prev, c, k.k->u64s);
 887                 bkey_reassemble(cur.k, k);
 888                 k = bkey_i_to_s_c(cur.k);
 889
 890                 offset_into_extent      = iter->pos.offset -
 891                         bkey_start_offset(k.k);
 892                 sectors                 = k.k->size - offset_into_extent;
 893
 894                 ret = bch2_read_indirect_extent(&trans,
 895                                         &offset_into_extent, cur.k);
 896                 if (ret)
 897                         break;
 898
 899                 sectors = min(sectors, k.k->size - offset_into_extent);
 900
 901                 if (offset_into_extent)
 902                         bch2_cut_front(POS(k.k->p.inode,
 903                                            bkey_start_offset(k.k) +
 904                                            offset_into_extent),
 905                                        cur.k);
 906                 bch2_key_resize(&cur.k->k, sectors);
 907                 cur.k->k.p = iter->pos;
 908                 cur.k->k.p.offset += cur.k->k.size;
 909
 910                 if (have_extent) {
 911                         ret = bch2_fill_extent(c, info,
 912                                         bkey_i_to_s_c(prev.k), 0);
 913                         if (ret)
 914                                 break;
 915                 }
 916
 917                 bkey_copy(prev.k, cur.k);
 918                 have_extent = true;
 919
 920                 if (k.k->type == KEY_TYPE_reflink_v)
 921                         bch2_btree_iter_set_pos(iter, k.k->p);
 922                 else
 923                         bch2_btree_iter_next(iter);
 924         }
 925
 926         if (ret == -EINTR)
 927                 goto retry;
 928
 929         if (!ret && have_extent)
 930                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 931                                        FIEMAP_EXTENT_LAST);
 932
 933         ret = bch2_trans_exit(&trans) ?: ret;
 934         bkey_on_stack_exit(&cur, c);
 935         bkey_on_stack_exit(&prev, c);
 936         return ret < 0 ? ret : 0;
 937 }
 938
 939 static const struct vm_operations_struct bch_vm_ops = {
 940         .fault          = bch2_page_fault,
 941         .map_pages      = filemap_map_pages,
 942         .page_mkwrite   = bch2_page_mkwrite,
 943 };
 944
 945 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 946 {
 947         file_accessed(file);
 948
 949         vma->vm_ops = &bch_vm_ops;
 950         return 0;
 951 }
 952
 953 /* Directories: */
 954
 955 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 956 {
 957         return generic_file_llseek_size(file, offset, whence,
 958                                         S64_MAX, S64_MAX);
 959 }
 960
 961 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 962 {
 963         struct bch_inode_info *inode = file_bch_inode(file);
 964         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 965
 966         if (!dir_emit_dots(file, ctx))
 967                 return 0;
 968
 969         return bch2_readdir(c, inode->v.i_ino, ctx);
 970 }
 971
 972 static const struct file_operations bch_file_operations = {
 973         .llseek         = bch2_llseek,
 974         .read_iter      = bch2_read_iter,
 975         .write_iter     = bch2_write_iter,
 976         .mmap           = bch2_mmap,
 977         .open           = generic_file_open,
 978         .fsync          = bch2_fsync,
 979         .splice_read    = generic_file_splice_read,
 980         /*
 981          * Broken, on v5.3:
 982         .splice_write   = iter_file_splice_write,
 983         */
 984         .fallocate      = bch2_fallocate_dispatch,
 985         .unlocked_ioctl = bch2_fs_file_ioctl,
 986 #ifdef CONFIG_COMPAT
 987         .compat_ioctl   = bch2_compat_fs_ioctl,
 988 #endif
 989         .remap_file_range = bch2_remap_file_range,
 990 };
 991
 992 static const struct inode_operations bch_file_inode_operations = {
 993         .getattr        = bch2_getattr,
 994         .setattr        = bch2_setattr,
 995         .fiemap         = bch2_fiemap,
 996         .listxattr      = bch2_xattr_list,
 997 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 998         .get_acl        = bch2_get_acl,
 999         .set_acl        = bch2_set_acl,
1000 #endif
1001 };
1002
1003 static const struct inode_operations bch_dir_inode_operations = {
1004         .lookup         = bch2_lookup,
1005         .create         = bch2_create,
1006         .link           = bch2_link,
1007         .unlink         = bch2_unlink,
1008         .symlink        = bch2_symlink,
1009         .mkdir          = bch2_mkdir,
1010         .rmdir          = bch2_unlink,
1011         .mknod          = bch2_mknod,
1012         .rename         = bch2_rename2,
1013         .getattr        = bch2_getattr,
1014         .setattr        = bch2_setattr,
1015         .tmpfile        = bch2_tmpfile,
1016         .listxattr      = bch2_xattr_list,
1017 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1018         .get_acl        = bch2_get_acl,
1019         .set_acl        = bch2_set_acl,
1020 #endif
1021 };
1022
1023 static const struct file_operations bch_dir_file_operations = {
1024         .llseek         = bch2_dir_llseek,
1025         .read           = generic_read_dir,
1026         .iterate_shared = bch2_vfs_readdir,
1027         .fsync          = bch2_fsync,
1028         .unlocked_ioctl = bch2_fs_file_ioctl,
1029 #ifdef CONFIG_COMPAT
1030         .compat_ioctl   = bch2_compat_fs_ioctl,
1031 #endif
1032 };
1033
1034 static const struct inode_operations bch_symlink_inode_operations = {
1035         .get_link       = page_get_link,
1036         .getattr        = bch2_getattr,
1037         .setattr        = bch2_setattr,
1038         .listxattr      = bch2_xattr_list,
1039 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1040         .get_acl        = bch2_get_acl,
1041         .set_acl        = bch2_set_acl,
1042 #endif
1043 };
1044
1045 static const struct inode_operations bch_special_inode_operations = {
1046         .getattr        = bch2_getattr,
1047         .setattr        = bch2_setattr,
1048         .listxattr      = bch2_xattr_list,
1049 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1050         .get_acl        = bch2_get_acl,
1051         .set_acl        = bch2_set_acl,
1052 #endif
1053 };
1054
1055 static const struct address_space_operations bch_address_space_operations = {
1056         .writepage      = bch2_writepage,
1057         .readpage       = bch2_readpage,
1058         .writepages     = bch2_writepages,
1059         .readpages      = bch2_readpages,
1060         .set_page_dirty = __set_page_dirty_nobuffers,
1061         .write_begin    = bch2_write_begin,
1062         .write_end      = bch2_write_end,
1063         .invalidatepage = bch2_invalidatepage,
1064         .releasepage    = bch2_releasepage,
1065         .direct_IO      = noop_direct_IO,
1066 #ifdef CONFIG_MIGRATION
1067         .migratepage    = bch2_migrate_page,
1068 #endif
1069         .error_remove_page = generic_error_remove_page,
1070 };
1071
1072 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1073                 u64 ino, u32 generation)
1074 {
1075         struct bch_fs *c = sb->s_fs_info;
1076         struct inode *vinode;
1077
1078         if (ino < BCACHEFS_ROOT_INO)
1079                 return ERR_PTR(-ESTALE);
1080
1081         vinode = bch2_vfs_inode_get(c, ino);
1082         if (IS_ERR(vinode))
1083                 return ERR_CAST(vinode);
1084         if (generation && vinode->i_generation != generation) {
1085                 /* we didn't find the right inode.. */
1086                 iput(vinode);
1087                 return ERR_PTR(-ESTALE);
1088         }
1089         return vinode;
1090 }
1091
1092 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1093                 int fh_len, int fh_type)
1094 {
1095         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1096                                     bch2_nfs_get_inode);
1097 }
1098
1099 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1100                 int fh_len, int fh_type)
1101 {
1102         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1103                                     bch2_nfs_get_inode);
1104 }
1105
1106 static const struct export_operations bch_export_ops = {
1107         .fh_to_dentry   = bch2_fh_to_dentry,
1108         .fh_to_parent   = bch2_fh_to_parent,
1109         //.get_parent   = bch2_get_parent,
1110 };
1111
1112 static void bch2_vfs_inode_init(struct bch_fs *c,
1113                                 struct bch_inode_info *inode,
1114                                 struct bch_inode_unpacked *bi)
1115 {
1116         bch2_inode_update_after_write(c, inode, bi, ~0);
1117
1118         inode->v.i_blocks       = bi->bi_sectors;
1119         inode->v.i_ino          = bi->bi_inum;
1120         inode->v.i_rdev         = bi->bi_dev;
1121         inode->v.i_generation   = bi->bi_generation;
1122         inode->v.i_size         = bi->bi_size;
1123
1124         inode->ei_journal_seq   = 0;
1125         inode->ei_quota_reserved = 0;
1126         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1127         inode->ei_qid           = bch_qid(bi);
1128
1129         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1130
1131         switch (inode->v.i_mode & S_IFMT) {
1132         case S_IFREG:
1133                 inode->v.i_op   = &bch_file_inode_operations;
1134                 inode->v.i_fop  = &bch_file_operations;
1135                 break;
1136         case S_IFDIR:
1137                 inode->v.i_op   = &bch_dir_inode_operations;
1138                 inode->v.i_fop  = &bch_dir_file_operations;
1139                 break;
1140         case S_IFLNK:
1141                 inode_nohighmem(&inode->v);
1142                 inode->v.i_op   = &bch_symlink_inode_operations;
1143                 break;
1144         default:
1145                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1146                 inode->v.i_op   = &bch_special_inode_operations;
1147                 break;
1148         }
1149 }
1150
1151 static struct inode *bch2_alloc_inode(struct super_block *sb)
1152 {
1153         struct bch_inode_info *inode;
1154
1155         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1156         if (!inode)
1157                 return NULL;
1158
1159         inode_init_once(&inode->v);
1160         mutex_init(&inode->ei_update_lock);
1161         pagecache_lock_init(&inode->ei_pagecache_lock);
1162         mutex_init(&inode->ei_quota_lock);
1163         inode->ei_journal_seq = 0;
1164
1165         return &inode->v;
1166 }
1167
1168 static void bch2_i_callback(struct rcu_head *head)
1169 {
1170         struct inode *vinode = container_of(head, struct inode, i_rcu);
1171         struct bch_inode_info *inode = to_bch_ei(vinode);
1172
1173         kmem_cache_free(bch2_inode_cache, inode);
1174 }
1175
1176 static void bch2_destroy_inode(struct inode *vinode)
1177 {
1178         call_rcu(&vinode->i_rcu, bch2_i_callback);
1179 }
1180
1181 static int inode_update_times_fn(struct bch_inode_info *inode,
1182                                  struct bch_inode_unpacked *bi,
1183                                  void *p)
1184 {
1185         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1186
1187         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1188         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1189         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1190
1191         return 0;
1192 }
1193
1194 static int bch2_vfs_write_inode(struct inode *vinode,
1195                                 struct writeback_control *wbc)
1196 {
1197         struct bch_fs *c = vinode->i_sb->s_fs_info;
1198         struct bch_inode_info *inode = to_bch_ei(vinode);
1199         int ret;
1200
1201         mutex_lock(&inode->ei_update_lock);
1202         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1203                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1204         mutex_unlock(&inode->ei_update_lock);
1205
1206         return ret;
1207 }
1208
1209 static void bch2_evict_inode(struct inode *vinode)
1210 {
1211         struct bch_fs *c = vinode->i_sb->s_fs_info;
1212         struct bch_inode_info *inode = to_bch_ei(vinode);
1213
1214         truncate_inode_pages_final(&inode->v.i_data);
1215
1216         clear_inode(&inode->v);
1217
1218         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1219
1220         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1221                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1222                                 KEY_TYPE_QUOTA_WARN);
1223                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1224                                 KEY_TYPE_QUOTA_WARN);
1225                 bch2_inode_rm(c, inode->v.i_ino);
1226         }
1227 }
1228
1229 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1230 {
1231         struct super_block *sb = dentry->d_sb;
1232         struct bch_fs *c = sb->s_fs_info;
1233         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1234         unsigned shift = sb->s_blocksize_bits - 9;
1235         u64 fsid;
1236
1237         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1238         buf->f_bsize    = sb->s_blocksize;
1239         buf->f_blocks   = usage.capacity >> shift;
1240         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1241         buf->f_bavail   = buf->f_bfree;
1242         buf->f_files    = usage.nr_inodes;
1243         buf->f_ffree    = U64_MAX;
1244
1245         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1246                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1247         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1248         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1249         buf->f_namelen  = BCH_NAME_MAX;
1250
1251         return 0;
1252 }
1253
1254 static int bch2_sync_fs(struct super_block *sb, int wait)
1255 {
1256         struct bch_fs *c = sb->s_fs_info;
1257
1258         if (c->opts.journal_flush_disabled)
1259                 return 0;
1260
1261         if (!wait) {
1262                 bch2_journal_flush_async(&c->journal, NULL);
1263                 return 0;
1264         }
1265
1266         return bch2_journal_flush(&c->journal);
1267 }
1268
1269 static struct bch_fs *bch2_path_to_fs(const char *dev)
1270 {
1271         struct bch_fs *c;
1272         struct block_device *bdev = lookup_bdev(dev);
1273
1274         if (IS_ERR(bdev))
1275                 return ERR_CAST(bdev);
1276
1277         c = bch2_bdev_to_fs(bdev);
1278         bdput(bdev);
1279         return c ?: ERR_PTR(-ENOENT);
1280 }
1281
1282 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1283                                                unsigned nr_devs, struct bch_opts opts)
1284 {
1285         struct bch_fs *c, *c1, *c2;
1286         size_t i;
1287
1288         if (!nr_devs)
1289                 return ERR_PTR(-EINVAL);
1290
1291         c = bch2_fs_open(devs, nr_devs, opts);
1292
1293         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1294                 /*
1295                  * Already open?
1296                  * Look up each block device, make sure they all belong to a
1297                  * filesystem and they all belong to the _same_ filesystem
1298                  */
1299
1300                 c1 = bch2_path_to_fs(devs[0]);
1301                 if (IS_ERR(c1))
1302                         return c;
1303
1304                 for (i = 1; i < nr_devs; i++) {
1305                         c2 = bch2_path_to_fs(devs[i]);
1306                         if (!IS_ERR(c2))
1307                                 closure_put(&c2->cl);
1308
1309                         if (c1 != c2) {
1310                                 closure_put(&c1->cl);
1311                                 return c;
1312                         }
1313                 }
1314
1315                 c = c1;
1316         }
1317
1318         if (IS_ERR(c))
1319                 return c;
1320
1321         mutex_lock(&c->state_lock);
1322
1323         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1324                 mutex_unlock(&c->state_lock);
1325                 closure_put(&c->cl);
1326                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1327                 return ERR_PTR(-EINVAL);
1328         }
1329
1330         mutex_unlock(&c->state_lock);
1331
1332         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1333         return c;
1334 }
1335
1336 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1337                                              struct bch_opts opts)
1338 {
1339         char *dev_name = NULL, **devs = NULL, *s;
1340         struct bch_fs *c = ERR_PTR(-ENOMEM);
1341         size_t i, nr_devs = 0;
1342
1343         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1344         if (!dev_name)
1345                 goto err;
1346
1347         for (s = dev_name; s; s = strchr(s + 1, ':'))
1348                 nr_devs++;
1349
1350         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1351         if (!devs)
1352                 goto err;
1353
1354         for (i = 0, s = dev_name;
1355              s;
1356              (s = strchr(s, ':')) && (*s++ = '\0'))
1357                 devs[i++] = s;
1358
1359         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1360 err:
1361         kfree(devs);
1362         kfree(dev_name);
1363         return c;
1364 }
1365
1366 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1367 {
1368         struct bch_fs *c = sb->s_fs_info;
1369         struct bch_opts opts = bch2_opts_empty();
1370         int ret;
1371
1372         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1373
1374         ret = bch2_parse_mount_opts(&opts, data);
1375         if (ret)
1376                 return ret;
1377
1378         if (opts.read_only != c->opts.read_only) {
1379                 mutex_lock(&c->state_lock);
1380
1381                 if (opts.read_only) {
1382                         bch2_fs_read_only(c);
1383
1384                         sb->s_flags |= SB_RDONLY;
1385                 } else {
1386                         ret = bch2_fs_read_write(c);
1387                         if (ret) {
1388                                 bch_err(c, "error going rw: %i", ret);
1389                                 mutex_unlock(&c->state_lock);
1390                                 return -EINVAL;
1391                         }
1392
1393                         sb->s_flags &= ~SB_RDONLY;
1394                 }
1395
1396                 c->opts.read_only = opts.read_only;
1397
1398                 mutex_unlock(&c->state_lock);
1399         }
1400
1401         if (opts.errors >= 0)
1402                 c->opts.errors = opts.errors;
1403
1404         return ret;
1405 }
1406
1407 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1408 {
1409         struct bch_fs *c = root->d_sb->s_fs_info;
1410         enum bch_opt_id i;
1411         char buf[512];
1412
1413         for (i = 0; i < bch2_opts_nr; i++) {
1414                 const struct bch_option *opt = &bch2_opt_table[i];
1415                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1416
1417                 if (!(opt->mode & OPT_MOUNT))
1418                         continue;
1419
1420                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1421                         continue;
1422
1423                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1424                                  OPT_SHOW_MOUNT_STYLE);
1425                 seq_putc(seq, ',');
1426                 seq_puts(seq, buf);
1427         }
1428
1429         return 0;
1430
1431 }
1432
1433 static const struct super_operations bch_super_operations = {
1434         .alloc_inode    = bch2_alloc_inode,
1435         .destroy_inode  = bch2_destroy_inode,
1436         .write_inode    = bch2_vfs_write_inode,
1437         .evict_inode    = bch2_evict_inode,
1438         .sync_fs        = bch2_sync_fs,
1439         .statfs         = bch2_statfs,
1440         .show_options   = bch2_show_options,
1441         .remount_fs     = bch2_remount,
1442 #if 0
1443         .put_super      = bch2_put_super,
1444         .freeze_fs      = bch2_freeze,
1445         .unfreeze_fs    = bch2_unfreeze,
1446 #endif
1447 };
1448
1449 static int bch2_test_super(struct super_block *s, void *data)
1450 {
1451         return s->s_fs_info == data;
1452 }
1453
1454 static int bch2_set_super(struct super_block *s, void *data)
1455 {
1456         s->s_fs_info = data;
1457         return 0;
1458 }
1459
1460 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1461                                  int flags, const char *dev_name, void *data)
1462 {
1463         struct bch_fs *c;
1464         struct bch_dev *ca;
1465         struct super_block *sb;
1466         struct inode *vinode;
1467         struct bch_opts opts = bch2_opts_empty();
1468         unsigned i;
1469         int ret;
1470
1471         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1472
1473         ret = bch2_parse_mount_opts(&opts, data);
1474         if (ret)
1475                 return ERR_PTR(ret);
1476
1477         c = bch2_open_as_blockdevs(dev_name, opts);
1478         if (IS_ERR(c))
1479                 return ERR_CAST(c);
1480
1481         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1482         if (IS_ERR(sb)) {
1483                 closure_put(&c->cl);
1484                 return ERR_CAST(sb);
1485         }
1486
1487         BUG_ON(sb->s_fs_info != c);
1488
1489         if (sb->s_root) {
1490                 closure_put(&c->cl);
1491
1492                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1493                         ret = -EBUSY;
1494                         goto err_put_super;
1495                 }
1496                 goto out;
1497         }
1498
1499         sb->s_blocksize         = block_bytes(c);
1500         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1501         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1502         sb->s_op                = &bch_super_operations;
1503         sb->s_export_op         = &bch_export_ops;
1504 #ifdef CONFIG_BCACHEFS_QUOTA
1505         sb->s_qcop              = &bch2_quotactl_operations;
1506         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1507 #endif
1508         sb->s_xattr             = bch2_xattr_handlers;
1509         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1510         sb->s_time_gran         = c->sb.time_precision;
1511         c->vfs_sb               = sb;
1512         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1513
1514         ret = super_setup_bdi(sb);
1515         if (ret)
1516                 goto err_put_super;
1517
1518         sb->s_bdi->congested_fn         = bch2_congested;
1519         sb->s_bdi->congested_data       = c;
1520         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1521
1522         for_each_online_member(ca, c, i) {
1523                 struct block_device *bdev = ca->disk_sb.bdev;
1524
1525                 /* XXX: create an anonymous device for multi device filesystems */
1526                 sb->s_bdev      = bdev;
1527                 sb->s_dev       = bdev->bd_dev;
1528                 percpu_ref_put(&ca->io_ref);
1529                 break;
1530         }
1531
1532 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1533         if (c->opts.acl)
1534                 sb->s_flags     |= SB_POSIXACL;
1535 #endif
1536
1537         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1538         if (IS_ERR(vinode)) {
1539                 bch_err(c, "error mounting: error getting root inode %i",
1540                         (int) PTR_ERR(vinode));
1541                 ret = PTR_ERR(vinode);
1542                 goto err_put_super;
1543         }
1544
1545         sb->s_root = d_make_root(vinode);
1546         if (!sb->s_root) {
1547                 bch_err(c, "error mounting: error allocating root dentry");
1548                 ret = -ENOMEM;
1549                 goto err_put_super;
1550         }
1551
1552         sb->s_flags |= SB_ACTIVE;
1553 out:
1554         return dget(sb->s_root);
1555
1556 err_put_super:
1557         deactivate_locked_super(sb);
1558         return ERR_PTR(ret);
1559 }
1560
1561 static void bch2_kill_sb(struct super_block *sb)
1562 {
1563         struct bch_fs *c = sb->s_fs_info;
1564
1565         generic_shutdown_super(sb);
1566
1567         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1568                 bch2_fs_stop(c);
1569         else
1570                 closure_put(&c->cl);
1571 }
1572
1573 static struct file_system_type bcache_fs_type = {
1574         .owner          = THIS_MODULE,
1575         .name           = "bcachefs",
1576         .mount          = bch2_mount,
1577         .kill_sb        = bch2_kill_sb,
1578         .fs_flags       = FS_REQUIRES_DEV,
1579 };
1580
1581 MODULE_ALIAS_FS("bcachefs");
1582
1583 void bch2_vfs_exit(void)
1584 {
1585         unregister_filesystem(&bcache_fs_type);
1586         if (bch2_inode_cache)
1587                 kmem_cache_destroy(bch2_inode_cache);
1588 }
1589
1590 int __init bch2_vfs_init(void)
1591 {
1592         int ret = -ENOMEM;
1593
1594         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1595         if (!bch2_inode_cache)
1596                 goto err;
1597
1598         ret = register_filesystem(&bcache_fs_type);
1599         if (ret)
1600                 goto err;
1601
1602         return 0;
1603 err:
1604         bch2_vfs_exit();
1605         return ret;
1606 }
1607
1608 #endif /* NO_BCACHEFS_FS */