1 // SPDX-License-Identifier: GPL-2.0
7 #include "btree_update.h"
13 #include "fs-common.h"
25 #include <linux/aio.h>
26 #include <linux/backing-dev.h>
27 #include <linux/exportfs.h>
28 #include <linux/fiemap.h>
29 #include <linux/module.h>
30 #include <linux/posix_acl.h>
31 #include <linux/random.h>
32 #include <linux/statfs.h>
33 #include <linux/string.h>
34 #include <linux/xattr.h>
36 static struct kmem_cache *bch2_inode_cache;
38 static void bch2_vfs_inode_init(struct bch_fs *,
39 struct bch_inode_info *,
40 struct bch_inode_unpacked *);
42 static void journal_seq_copy(struct bch_fs *c,
43 struct bch_inode_info *dst,
47 * atomic64_cmpxchg has a fallback for archs that don't support it,
50 atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
51 u64 old, v = READ_ONCE(dst->ei_journal_seq);
56 if (old >= journal_seq)
58 } while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
60 bch2_journal_set_has_inum(&c->journal, dst->v.i_ino, journal_seq);
63 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
65 BUG_ON(atomic_long_read(&lock->v) == 0);
67 if (atomic_long_sub_return_release(i, &lock->v) == 0)
68 wake_up_all(&lock->wait);
71 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
73 long v = atomic_long_read(&lock->v), old;
78 if (i > 0 ? v < 0 : v > 0)
80 } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
81 old, old + i)) != old);
85 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
87 wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
90 void bch2_pagecache_add_put(struct pagecache_lock *lock)
92 __pagecache_lock_put(lock, 1);
95 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
97 return __pagecache_lock_tryget(lock, 1);
100 void bch2_pagecache_add_get(struct pagecache_lock *lock)
102 __pagecache_lock_get(lock, 1);
105 void bch2_pagecache_block_put(struct pagecache_lock *lock)
107 __pagecache_lock_put(lock, -1);
110 void bch2_pagecache_block_get(struct pagecache_lock *lock)
112 __pagecache_lock_get(lock, -1);
115 void bch2_inode_update_after_write(struct bch_fs *c,
116 struct bch_inode_info *inode,
117 struct bch_inode_unpacked *bi,
120 set_nlink(&inode->v, bch2_inode_nlink_get(bi));
121 i_uid_write(&inode->v, bi->bi_uid);
122 i_gid_write(&inode->v, bi->bi_gid);
123 inode->v.i_mode = bi->bi_mode;
125 if (fields & ATTR_ATIME)
126 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
127 if (fields & ATTR_MTIME)
128 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
129 if (fields & ATTR_CTIME)
130 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
132 inode->ei_inode = *bi;
134 bch2_inode_flags_to_vfs(inode);
137 int __must_check bch2_write_inode(struct bch_fs *c,
138 struct bch_inode_info *inode,
140 void *p, unsigned fields)
142 struct btree_trans trans;
143 struct btree_iter *iter;
144 struct bch_inode_unpacked inode_u;
147 bch2_trans_init(&trans, c, 0, 512);
149 bch2_trans_begin(&trans);
151 iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
153 ret = PTR_ERR_OR_ZERO(iter) ?:
154 (set ? set(inode, &inode_u, p) : 0) ?:
155 bch2_inode_write(&trans, iter, &inode_u) ?:
156 bch2_trans_commit(&trans, NULL,
157 &inode->ei_journal_seq,
158 BTREE_INSERT_NOUNLOCK|
159 BTREE_INSERT_NOFAIL);
162 * the btree node lock protects inode->ei_inode, not ei_update_lock;
163 * this is important for inode updates via bchfs_write_index_update
166 bch2_inode_update_after_write(c, inode, &inode_u, fields);
168 bch2_trans_iter_put(&trans, iter);
173 bch2_trans_exit(&trans);
174 return ret < 0 ? ret : 0;
177 int bch2_fs_quota_transfer(struct bch_fs *c,
178 struct bch_inode_info *inode,
179 struct bch_qid new_qid,
181 enum quota_acct_mode mode)
186 qtypes &= enabled_qtypes(c);
188 for (i = 0; i < QTYP_NR; i++)
189 if (new_qid.q[i] == inode->ei_qid.q[i])
190 qtypes &= ~(1U << i);
195 mutex_lock(&inode->ei_quota_lock);
197 ret = bch2_quota_transfer(c, qtypes, new_qid,
200 inode->ei_quota_reserved,
203 for (i = 0; i < QTYP_NR; i++)
204 if (qtypes & (1 << i))
205 inode->ei_qid.q[i] = new_qid.q[i];
207 mutex_unlock(&inode->ei_quota_lock);
212 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
214 struct bch_inode_unpacked inode_u;
215 struct bch_inode_info *inode;
218 inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
219 if (unlikely(!inode))
220 return ERR_PTR(-ENOMEM);
221 if (!(inode->v.i_state & I_NEW))
224 ret = bch2_inode_find_by_inum(c, inum, &inode_u);
226 iget_failed(&inode->v);
230 bch2_vfs_inode_init(c, inode, &inode_u);
232 inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
234 unlock_new_inode(&inode->v);
239 static int inum_test(struct inode *inode, void *p)
241 unsigned long *ino = p;
243 return *ino == inode->i_ino;
246 static struct bch_inode_info *
247 __bch2_create(struct user_namespace *mnt_userns,
248 struct bch_inode_info *dir, struct dentry *dentry,
249 umode_t mode, dev_t rdev, bool tmpfile)
251 struct bch_fs *c = dir->v.i_sb->s_fs_info;
252 struct btree_trans trans;
253 struct bch_inode_unpacked dir_u;
254 struct bch_inode_info *inode, *old;
255 struct bch_inode_unpacked inode_u;
256 struct posix_acl *default_acl = NULL, *acl = NULL;
261 * preallocate acls + vfs inode before btree transaction, so that
262 * nothing can fail after the transaction succeeds:
264 #ifdef CONFIG_BCACHEFS_POSIX_ACL
265 ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
269 inode = to_bch_ei(new_inode(c->vfs_sb));
270 if (unlikely(!inode)) {
271 inode = ERR_PTR(-ENOMEM);
275 bch2_inode_init_early(c, &inode_u);
278 mutex_lock(&dir->ei_update_lock);
280 bch2_trans_init(&trans, c, 8,
281 2048 + (!tmpfile ? dentry->d_name.len : 0));
283 bch2_trans_begin(&trans);
285 ret = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
286 !tmpfile ? &dentry->d_name : NULL,
287 from_kuid(mnt_userns, current_fsuid()),
288 from_kgid(mnt_userns, current_fsgid()),
291 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
292 KEY_TYPE_QUOTA_PREALLOC);
294 goto err_before_quota;
296 ret = bch2_trans_commit(&trans, NULL, &journal_seq,
297 BTREE_INSERT_NOUNLOCK);
299 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
300 KEY_TYPE_QUOTA_WARN);
308 bch2_inode_update_after_write(c, dir, &dir_u,
309 ATTR_MTIME|ATTR_CTIME);
310 journal_seq_copy(c, dir, journal_seq);
311 mutex_unlock(&dir->ei_update_lock);
314 bch2_vfs_inode_init(c, inode, &inode_u);
315 journal_seq_copy(c, inode, journal_seq);
317 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
318 set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
321 * we must insert the new inode into the inode cache before calling
322 * bch2_trans_exit() and dropping locks, else we could race with another
323 * thread pulling the inode in and modifying it:
326 inode->v.i_state |= I_CREATING;
327 old = to_bch_ei(inode_insert5(&inode->v, inode->v.i_ino,
328 inum_test, NULL, &inode->v.i_ino));
331 if (unlikely(old != inode)) {
333 * We raced, another process pulled the new inode into cache
336 journal_seq_copy(c, old, journal_seq);
337 make_bad_inode(&inode->v);
343 * we really don't want insert_inode_locked2() to be setting
346 unlock_new_inode(&inode->v);
349 bch2_trans_exit(&trans);
351 posix_acl_release(default_acl);
352 posix_acl_release(acl);
356 mutex_unlock(&dir->ei_update_lock);
358 bch2_trans_exit(&trans);
359 make_bad_inode(&inode->v);
361 inode = ERR_PTR(ret);
367 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
370 struct bch_fs *c = vdir->i_sb->s_fs_info;
371 struct bch_inode_info *dir = to_bch_ei(vdir);
372 struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
373 struct inode *vinode = NULL;
376 inum = bch2_dirent_lookup(c, dir->v.i_ino, &hash,
380 vinode = bch2_vfs_inode_get(c, inum);
382 return d_splice_alias(vinode, dentry);
385 static int bch2_mknod(struct user_namespace *mnt_userns,
386 struct inode *vdir, struct dentry *dentry,
387 umode_t mode, dev_t rdev)
389 struct bch_inode_info *inode =
390 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, rdev, false);
393 return PTR_ERR(inode);
395 d_instantiate(dentry, &inode->v);
399 static int bch2_create(struct user_namespace *mnt_userns,
400 struct inode *vdir, struct dentry *dentry,
401 umode_t mode, bool excl)
403 return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFREG, 0);
406 static int __bch2_link(struct bch_fs *c,
407 struct bch_inode_info *inode,
408 struct bch_inode_info *dir,
409 struct dentry *dentry)
411 struct btree_trans trans;
412 struct bch_inode_unpacked dir_u, inode_u;
415 mutex_lock(&inode->ei_update_lock);
416 bch2_trans_init(&trans, c, 4, 1024);
418 ret = __bch2_trans_do(&trans, NULL, &inode->ei_journal_seq,
419 BTREE_INSERT_NOUNLOCK,
420 bch2_link_trans(&trans,
422 inode->v.i_ino, &dir_u, &inode_u,
426 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
428 journal_seq_copy(c, inode, dir->ei_journal_seq);
429 bch2_inode_update_after_write(c, dir, &dir_u,
430 ATTR_MTIME|ATTR_CTIME);
431 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
434 bch2_trans_exit(&trans);
435 mutex_unlock(&inode->ei_update_lock);
439 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
440 struct dentry *dentry)
442 struct bch_fs *c = vdir->i_sb->s_fs_info;
443 struct bch_inode_info *dir = to_bch_ei(vdir);
444 struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
447 lockdep_assert_held(&inode->v.i_rwsem);
449 ret = __bch2_link(c, inode, dir, dentry);
454 d_instantiate(dentry, &inode->v);
458 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
460 struct bch_fs *c = vdir->i_sb->s_fs_info;
461 struct bch_inode_info *dir = to_bch_ei(vdir);
462 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
463 struct bch_inode_unpacked dir_u, inode_u;
464 struct btree_trans trans;
467 bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
468 bch2_trans_init(&trans, c, 4, 1024);
470 ret = __bch2_trans_do(&trans, NULL, &dir->ei_journal_seq,
471 BTREE_INSERT_NOUNLOCK|
473 bch2_unlink_trans(&trans,
474 dir->v.i_ino, &dir_u,
475 &inode_u, &dentry->d_name));
478 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
480 journal_seq_copy(c, inode, dir->ei_journal_seq);
481 bch2_inode_update_after_write(c, dir, &dir_u,
482 ATTR_MTIME|ATTR_CTIME);
483 bch2_inode_update_after_write(c, inode, &inode_u,
487 bch2_trans_exit(&trans);
488 bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
493 static int bch2_symlink(struct user_namespace *mnt_userns,
494 struct inode *vdir, struct dentry *dentry,
497 struct bch_fs *c = vdir->i_sb->s_fs_info;
498 struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
501 inode = __bch2_create(mnt_userns, dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
502 if (unlikely(IS_ERR(inode)))
503 return PTR_ERR(inode);
505 inode_lock(&inode->v);
506 ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
507 inode_unlock(&inode->v);
512 ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
516 journal_seq_copy(c, dir, inode->ei_journal_seq);
518 ret = __bch2_link(c, inode, dir, dentry);
522 d_instantiate(dentry, &inode->v);
529 static int bch2_mkdir(struct user_namespace *mnt_userns,
530 struct inode *vdir, struct dentry *dentry, umode_t mode)
532 return bch2_mknod(mnt_userns, vdir, dentry, mode|S_IFDIR, 0);
535 static int bch2_rename2(struct user_namespace *mnt_userns,
536 struct inode *src_vdir, struct dentry *src_dentry,
537 struct inode *dst_vdir, struct dentry *dst_dentry,
540 struct bch_fs *c = src_vdir->i_sb->s_fs_info;
541 struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
542 struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
543 struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
544 struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
545 struct bch_inode_unpacked dst_dir_u, src_dir_u;
546 struct bch_inode_unpacked src_inode_u, dst_inode_u;
547 struct btree_trans trans;
548 enum bch_rename_mode mode = flags & RENAME_EXCHANGE
549 ? BCH_RENAME_EXCHANGE
550 : dst_dentry->d_inode
551 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
555 if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
558 if (mode == BCH_RENAME_OVERWRITE) {
559 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
565 bch2_trans_init(&trans, c, 8, 2048);
567 bch2_lock_inodes(INODE_UPDATE_LOCK,
573 if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
574 ret = bch2_fs_quota_transfer(c, src_inode,
577 KEY_TYPE_QUOTA_PREALLOC);
582 if (mode == BCH_RENAME_EXCHANGE &&
583 inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
584 ret = bch2_fs_quota_transfer(c, dst_inode,
587 KEY_TYPE_QUOTA_PREALLOC);
592 ret = __bch2_trans_do(&trans, NULL, &journal_seq,
593 BTREE_INSERT_NOUNLOCK,
594 bch2_rename_trans(&trans,
595 src_dir->v.i_ino, &src_dir_u,
596 dst_dir->v.i_ino, &dst_dir_u,
605 BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
607 dst_inode->v.i_ino != dst_inode_u.bi_inum);
609 bch2_inode_update_after_write(c, src_dir, &src_dir_u,
610 ATTR_MTIME|ATTR_CTIME);
611 journal_seq_copy(c, src_dir, journal_seq);
613 if (src_dir != dst_dir) {
614 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
615 ATTR_MTIME|ATTR_CTIME);
616 journal_seq_copy(c, dst_dir, journal_seq);
619 bch2_inode_update_after_write(c, src_inode, &src_inode_u,
621 journal_seq_copy(c, src_inode, journal_seq);
624 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
626 journal_seq_copy(c, dst_inode, journal_seq);
629 bch2_trans_exit(&trans);
631 bch2_fs_quota_transfer(c, src_inode,
632 bch_qid(&src_inode->ei_inode),
634 KEY_TYPE_QUOTA_NOCHECK);
636 bch2_fs_quota_transfer(c, dst_inode,
637 bch_qid(&dst_inode->ei_inode),
639 KEY_TYPE_QUOTA_NOCHECK);
641 bch2_unlock_inodes(INODE_UPDATE_LOCK,
650 static void bch2_setattr_copy(struct user_namespace *mnt_userns,
651 struct bch_inode_info *inode,
652 struct bch_inode_unpacked *bi,
655 struct bch_fs *c = inode->v.i_sb->s_fs_info;
656 unsigned int ia_valid = attr->ia_valid;
658 if (ia_valid & ATTR_UID)
659 bi->bi_uid = from_kuid(mnt_userns, attr->ia_uid);
660 if (ia_valid & ATTR_GID)
661 bi->bi_gid = from_kgid(mnt_userns, attr->ia_gid);
663 if (ia_valid & ATTR_SIZE)
664 bi->bi_size = attr->ia_size;
666 if (ia_valid & ATTR_ATIME)
667 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
668 if (ia_valid & ATTR_MTIME)
669 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
670 if (ia_valid & ATTR_CTIME)
671 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
673 if (ia_valid & ATTR_MODE) {
674 umode_t mode = attr->ia_mode;
675 kgid_t gid = ia_valid & ATTR_GID
679 if (!in_group_p(gid) &&
680 !capable_wrt_inode_uidgid(mnt_userns, &inode->v, CAP_FSETID))
686 int bch2_setattr_nonsize(struct user_namespace *mnt_userns,
687 struct bch_inode_info *inode,
690 struct bch_fs *c = inode->v.i_sb->s_fs_info;
692 struct btree_trans trans;
693 struct btree_iter *inode_iter;
694 struct bch_inode_unpacked inode_u;
695 struct posix_acl *acl = NULL;
698 mutex_lock(&inode->ei_update_lock);
702 if (attr->ia_valid & ATTR_UID)
703 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
705 if (attr->ia_valid & ATTR_GID)
706 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
708 ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
709 KEY_TYPE_QUOTA_PREALLOC);
713 bch2_trans_init(&trans, c, 0, 0);
715 bch2_trans_begin(&trans);
719 inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
721 ret = PTR_ERR_OR_ZERO(inode_iter);
725 bch2_setattr_copy(mnt_userns, inode, &inode_u, attr);
727 if (attr->ia_valid & ATTR_MODE) {
728 ret = bch2_acl_chmod(&trans, &inode_u, inode_u.bi_mode, &acl);
733 ret = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
734 bch2_trans_commit(&trans, NULL,
735 &inode->ei_journal_seq,
736 BTREE_INSERT_NOUNLOCK|
737 BTREE_INSERT_NOFAIL);
739 bch2_trans_iter_put(&trans, inode_iter);
746 bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
749 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
751 bch2_trans_exit(&trans);
753 mutex_unlock(&inode->ei_update_lock);
758 static int bch2_getattr(struct user_namespace *mnt_userns,
759 const struct path *path, struct kstat *stat,
760 u32 request_mask, unsigned query_flags)
762 struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
763 struct bch_fs *c = inode->v.i_sb->s_fs_info;
765 stat->dev = inode->v.i_sb->s_dev;
766 stat->ino = inode->v.i_ino;
767 stat->mode = inode->v.i_mode;
768 stat->nlink = inode->v.i_nlink;
769 stat->uid = inode->v.i_uid;
770 stat->gid = inode->v.i_gid;
771 stat->rdev = inode->v.i_rdev;
772 stat->size = i_size_read(&inode->v);
773 stat->atime = inode->v.i_atime;
774 stat->mtime = inode->v.i_mtime;
775 stat->ctime = inode->v.i_ctime;
776 stat->blksize = block_bytes(c);
777 stat->blocks = inode->v.i_blocks;
779 if (request_mask & STATX_BTIME) {
780 stat->result_mask |= STATX_BTIME;
781 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
784 if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
785 stat->attributes |= STATX_ATTR_IMMUTABLE;
786 stat->attributes_mask |= STATX_ATTR_IMMUTABLE;
788 if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
789 stat->attributes |= STATX_ATTR_APPEND;
790 stat->attributes_mask |= STATX_ATTR_APPEND;
792 if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
793 stat->attributes |= STATX_ATTR_NODUMP;
794 stat->attributes_mask |= STATX_ATTR_NODUMP;
799 static int bch2_setattr(struct user_namespace *mnt_userns,
800 struct dentry *dentry, struct iattr *iattr)
802 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
805 lockdep_assert_held(&inode->v.i_rwsem);
807 ret = setattr_prepare(mnt_userns, dentry, iattr);
811 return iattr->ia_valid & ATTR_SIZE
812 ? bch2_truncate(mnt_userns, inode, iattr)
813 : bch2_setattr_nonsize(mnt_userns, inode, iattr);
816 static int bch2_tmpfile(struct user_namespace *mnt_userns,
817 struct inode *vdir, struct dentry *dentry, umode_t mode)
819 struct bch_inode_info *inode =
820 __bch2_create(mnt_userns, to_bch_ei(vdir), dentry, mode, 0, true);
823 return PTR_ERR(inode);
825 d_mark_tmpfile(dentry, &inode->v);
826 d_instantiate(dentry, &inode->v);
830 static int bch2_fill_extent(struct bch_fs *c,
831 struct fiemap_extent_info *info,
832 struct bkey_s_c k, unsigned flags)
834 if (bkey_extent_is_direct_data(k.k)) {
835 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
836 const union bch_extent_entry *entry;
837 struct extent_ptr_decoded p;
840 if (k.k->type == KEY_TYPE_reflink_v)
841 flags |= FIEMAP_EXTENT_SHARED;
843 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
845 u64 offset = p.ptr.offset;
847 if (p.crc.compression_type)
848 flags2 |= FIEMAP_EXTENT_ENCODED;
850 offset += p.crc.offset;
852 if ((offset & (c->opts.block_size - 1)) ||
853 (k.k->size & (c->opts.block_size - 1)))
854 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
856 ret = fiemap_fill_next_extent(info,
857 bkey_start_offset(k.k) << 9,
859 k.k->size << 9, flags|flags2);
865 } else if (bkey_extent_is_inline_data(k.k)) {
866 return fiemap_fill_next_extent(info,
867 bkey_start_offset(k.k) << 9,
870 FIEMAP_EXTENT_DATA_INLINE);
871 } else if (k.k->type == KEY_TYPE_reservation) {
872 return fiemap_fill_next_extent(info,
873 bkey_start_offset(k.k) << 9,
876 FIEMAP_EXTENT_DELALLOC|
877 FIEMAP_EXTENT_UNWRITTEN);
883 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
886 struct bch_fs *c = vinode->i_sb->s_fs_info;
887 struct bch_inode_info *ei = to_bch_ei(vinode);
888 struct btree_trans trans;
889 struct btree_iter *iter;
891 struct bkey_buf cur, prev;
892 struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
893 unsigned offset_into_extent, sectors;
894 bool have_extent = false;
897 ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
901 if (start + len < start)
904 bch2_bkey_buf_init(&cur);
905 bch2_bkey_buf_init(&prev);
906 bch2_trans_init(&trans, c, 0, 0);
908 iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
909 POS(ei->v.i_ino, start >> 9), 0);
911 while ((k = bch2_btree_iter_peek(iter)).k &&
912 !(ret = bkey_err(k)) &&
913 bkey_cmp(iter->pos, end) < 0) {
914 enum btree_id data_btree = BTREE_ID_extents;
916 if (!bkey_extent_is_data(k.k) &&
917 k.k->type != KEY_TYPE_reservation) {
918 bch2_btree_iter_advance(iter);
922 offset_into_extent = iter->pos.offset -
923 bkey_start_offset(k.k);
924 sectors = k.k->size - offset_into_extent;
926 bch2_bkey_buf_reassemble(&cur, c, k);
928 ret = bch2_read_indirect_extent(&trans, &data_btree,
929 &offset_into_extent, &cur);
933 k = bkey_i_to_s_c(cur.k);
934 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
936 sectors = min(sectors, k.k->size - offset_into_extent);
938 bch2_cut_front(POS(k.k->p.inode,
939 bkey_start_offset(k.k) +
942 bch2_key_resize(&cur.k->k, sectors);
943 cur.k->k.p = iter->pos;
944 cur.k->k.p.offset += cur.k->k.size;
947 ret = bch2_fill_extent(c, info,
948 bkey_i_to_s_c(prev.k), 0);
953 bkey_copy(prev.k, cur.k);
956 bch2_btree_iter_set_pos(iter,
957 POS(iter->pos.inode, iter->pos.offset + sectors));
963 if (!ret && have_extent)
964 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
967 bch2_trans_iter_put(&trans, iter);
968 ret = bch2_trans_exit(&trans) ?: ret;
969 bch2_bkey_buf_exit(&cur, c);
970 bch2_bkey_buf_exit(&prev, c);
971 return ret < 0 ? ret : 0;
974 static const struct vm_operations_struct bch_vm_ops = {
975 .fault = bch2_page_fault,
976 .map_pages = filemap_map_pages,
977 .page_mkwrite = bch2_page_mkwrite,
980 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
984 vma->vm_ops = &bch_vm_ops;
990 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
992 return generic_file_llseek_size(file, offset, whence,
996 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
998 struct bch_inode_info *inode = file_bch_inode(file);
999 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1001 if (!dir_emit_dots(file, ctx))
1004 return bch2_readdir(c, inode->v.i_ino, ctx);
1007 static const struct file_operations bch_file_operations = {
1008 .llseek = bch2_llseek,
1009 .read_iter = bch2_read_iter,
1010 .write_iter = bch2_write_iter,
1012 .open = generic_file_open,
1013 .fsync = bch2_fsync,
1014 .splice_read = generic_file_splice_read,
1015 .splice_write = iter_file_splice_write,
1016 .fallocate = bch2_fallocate_dispatch,
1017 .unlocked_ioctl = bch2_fs_file_ioctl,
1018 #ifdef CONFIG_COMPAT
1019 .compat_ioctl = bch2_compat_fs_ioctl,
1021 .remap_file_range = bch2_remap_file_range,
1024 static const struct inode_operations bch_file_inode_operations = {
1025 .getattr = bch2_getattr,
1026 .setattr = bch2_setattr,
1027 .fiemap = bch2_fiemap,
1028 .listxattr = bch2_xattr_list,
1029 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1030 .get_acl = bch2_get_acl,
1031 .set_acl = bch2_set_acl,
1035 static const struct inode_operations bch_dir_inode_operations = {
1036 .lookup = bch2_lookup,
1037 .create = bch2_create,
1039 .unlink = bch2_unlink,
1040 .symlink = bch2_symlink,
1041 .mkdir = bch2_mkdir,
1042 .rmdir = bch2_unlink,
1043 .mknod = bch2_mknod,
1044 .rename = bch2_rename2,
1045 .getattr = bch2_getattr,
1046 .setattr = bch2_setattr,
1047 .tmpfile = bch2_tmpfile,
1048 .listxattr = bch2_xattr_list,
1049 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1050 .get_acl = bch2_get_acl,
1051 .set_acl = bch2_set_acl,
1055 static const struct file_operations bch_dir_file_operations = {
1056 .llseek = bch2_dir_llseek,
1057 .read = generic_read_dir,
1058 .iterate_shared = bch2_vfs_readdir,
1059 .fsync = bch2_fsync,
1060 .unlocked_ioctl = bch2_fs_file_ioctl,
1061 #ifdef CONFIG_COMPAT
1062 .compat_ioctl = bch2_compat_fs_ioctl,
1066 static const struct inode_operations bch_symlink_inode_operations = {
1067 .get_link = page_get_link,
1068 .getattr = bch2_getattr,
1069 .setattr = bch2_setattr,
1070 .listxattr = bch2_xattr_list,
1071 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1072 .get_acl = bch2_get_acl,
1073 .set_acl = bch2_set_acl,
1077 static const struct inode_operations bch_special_inode_operations = {
1078 .getattr = bch2_getattr,
1079 .setattr = bch2_setattr,
1080 .listxattr = bch2_xattr_list,
1081 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1082 .get_acl = bch2_get_acl,
1083 .set_acl = bch2_set_acl,
1087 static const struct address_space_operations bch_address_space_operations = {
1088 .writepage = bch2_writepage,
1089 .readpage = bch2_readpage,
1090 .writepages = bch2_writepages,
1091 .readahead = bch2_readahead,
1092 .set_page_dirty = __set_page_dirty_nobuffers,
1093 .write_begin = bch2_write_begin,
1094 .write_end = bch2_write_end,
1095 .invalidatepage = bch2_invalidatepage,
1096 .releasepage = bch2_releasepage,
1097 .direct_IO = noop_direct_IO,
1098 #ifdef CONFIG_MIGRATION
1099 .migratepage = bch2_migrate_page,
1101 .error_remove_page = generic_error_remove_page,
1104 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1105 u64 ino, u32 generation)
1107 struct bch_fs *c = sb->s_fs_info;
1108 struct inode *vinode;
1110 if (ino < BCACHEFS_ROOT_INO)
1111 return ERR_PTR(-ESTALE);
1113 vinode = bch2_vfs_inode_get(c, ino);
1115 return ERR_CAST(vinode);
1116 if (generation && vinode->i_generation != generation) {
1117 /* we didn't find the right inode.. */
1119 return ERR_PTR(-ESTALE);
1124 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1125 int fh_len, int fh_type)
1127 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1128 bch2_nfs_get_inode);
1131 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1132 int fh_len, int fh_type)
1134 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1135 bch2_nfs_get_inode);
1138 static const struct export_operations bch_export_ops = {
1139 .fh_to_dentry = bch2_fh_to_dentry,
1140 .fh_to_parent = bch2_fh_to_parent,
1141 //.get_parent = bch2_get_parent,
1144 static void bch2_vfs_inode_init(struct bch_fs *c,
1145 struct bch_inode_info *inode,
1146 struct bch_inode_unpacked *bi)
1148 bch2_inode_update_after_write(c, inode, bi, ~0);
1150 inode->v.i_blocks = bi->bi_sectors;
1151 inode->v.i_ino = bi->bi_inum;
1152 inode->v.i_rdev = bi->bi_dev;
1153 inode->v.i_generation = bi->bi_generation;
1154 inode->v.i_size = bi->bi_size;
1156 inode->ei_flags = 0;
1157 inode->ei_journal_seq = 0;
1158 inode->ei_quota_reserved = 0;
1159 inode->ei_qid = bch_qid(bi);
1161 inode->v.i_mapping->a_ops = &bch_address_space_operations;
1163 switch (inode->v.i_mode & S_IFMT) {
1165 inode->v.i_op = &bch_file_inode_operations;
1166 inode->v.i_fop = &bch_file_operations;
1169 inode->v.i_op = &bch_dir_inode_operations;
1170 inode->v.i_fop = &bch_dir_file_operations;
1173 inode_nohighmem(&inode->v);
1174 inode->v.i_op = &bch_symlink_inode_operations;
1177 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1178 inode->v.i_op = &bch_special_inode_operations;
1183 static struct inode *bch2_alloc_inode(struct super_block *sb)
1185 struct bch_inode_info *inode;
1187 inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1191 inode_init_once(&inode->v);
1192 mutex_init(&inode->ei_update_lock);
1193 pagecache_lock_init(&inode->ei_pagecache_lock);
1194 mutex_init(&inode->ei_quota_lock);
1195 inode->ei_journal_seq = 0;
1200 static void bch2_i_callback(struct rcu_head *head)
1202 struct inode *vinode = container_of(head, struct inode, i_rcu);
1203 struct bch_inode_info *inode = to_bch_ei(vinode);
1205 kmem_cache_free(bch2_inode_cache, inode);
1208 static void bch2_destroy_inode(struct inode *vinode)
1210 call_rcu(&vinode->i_rcu, bch2_i_callback);
1213 static int inode_update_times_fn(struct bch_inode_info *inode,
1214 struct bch_inode_unpacked *bi,
1217 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1219 bi->bi_atime = timespec_to_bch2_time(c, inode->v.i_atime);
1220 bi->bi_mtime = timespec_to_bch2_time(c, inode->v.i_mtime);
1221 bi->bi_ctime = timespec_to_bch2_time(c, inode->v.i_ctime);
1226 static int bch2_vfs_write_inode(struct inode *vinode,
1227 struct writeback_control *wbc)
1229 struct bch_fs *c = vinode->i_sb->s_fs_info;
1230 struct bch_inode_info *inode = to_bch_ei(vinode);
1233 mutex_lock(&inode->ei_update_lock);
1234 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1235 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1236 mutex_unlock(&inode->ei_update_lock);
1241 static void bch2_evict_inode(struct inode *vinode)
1243 struct bch_fs *c = vinode->i_sb->s_fs_info;
1244 struct bch_inode_info *inode = to_bch_ei(vinode);
1246 truncate_inode_pages_final(&inode->v.i_data);
1248 clear_inode(&inode->v);
1250 BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1252 if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1253 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1254 KEY_TYPE_QUOTA_WARN);
1255 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1256 KEY_TYPE_QUOTA_WARN);
1257 bch2_inode_rm(c, inode->v.i_ino, true);
1261 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1263 struct super_block *sb = dentry->d_sb;
1264 struct bch_fs *c = sb->s_fs_info;
1265 struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1266 unsigned shift = sb->s_blocksize_bits - 9;
1268 * this assumes inodes take up 64 bytes, which is a decent average
1271 u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1274 buf->f_type = BCACHEFS_STATFS_MAGIC;
1275 buf->f_bsize = sb->s_blocksize;
1276 buf->f_blocks = usage.capacity >> shift;
1277 buf->f_bfree = avail_factor(usage.free) >> shift;
1278 buf->f_bavail = buf->f_bfree;
1280 buf->f_files = usage.nr_inodes + avail_inodes;
1281 buf->f_ffree = avail_inodes;
1283 fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1284 le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1285 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1286 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1287 buf->f_namelen = BCH_NAME_MAX;
1292 static int bch2_sync_fs(struct super_block *sb, int wait)
1294 struct bch_fs *c = sb->s_fs_info;
1296 if (c->opts.journal_flush_disabled)
1300 bch2_journal_flush_async(&c->journal, NULL);
1304 return bch2_journal_flush(&c->journal);
1307 static struct bch_fs *bch2_path_to_fs(const char *path)
1313 ret = lookup_bdev(path, &dev);
1315 return ERR_PTR(ret);
1317 c = bch2_dev_to_fs(dev);
1319 closure_put(&c->cl);
1320 return c ?: ERR_PTR(-ENOENT);
1323 static char **split_devs(const char *_dev_name, unsigned *nr)
1325 char *dev_name = NULL, **devs = NULL, *s;
1326 size_t i, nr_devs = 0;
1328 dev_name = kstrdup(_dev_name, GFP_KERNEL);
1332 for (s = dev_name; s; s = strchr(s + 1, ':'))
1335 devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1341 for (i = 0, s = dev_name;
1343 (s = strchr(s, ':')) && (*s++ = '\0'))
1350 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1352 struct bch_fs *c = sb->s_fs_info;
1353 struct bch_opts opts = bch2_opts_empty();
1356 opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1358 ret = bch2_parse_mount_opts(c, &opts, data);
1362 if (opts.read_only != c->opts.read_only) {
1363 down_write(&c->state_lock);
1365 if (opts.read_only) {
1366 bch2_fs_read_only(c);
1368 sb->s_flags |= SB_RDONLY;
1370 ret = bch2_fs_read_write(c);
1372 bch_err(c, "error going rw: %i", ret);
1373 up_write(&c->state_lock);
1377 sb->s_flags &= ~SB_RDONLY;
1380 c->opts.read_only = opts.read_only;
1382 up_write(&c->state_lock);
1385 if (opts.errors >= 0)
1386 c->opts.errors = opts.errors;
1391 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1393 struct bch_fs *c = root->d_sb->s_fs_info;
1398 for_each_online_member(ca, c, i) {
1402 seq_puts(seq, "/dev/");
1403 seq_puts(seq, ca->name);
1409 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1411 struct bch_fs *c = root->d_sb->s_fs_info;
1415 for (i = 0; i < bch2_opts_nr; i++) {
1416 const struct bch_option *opt = &bch2_opt_table[i];
1417 u64 v = bch2_opt_get_by_id(&c->opts, i);
1419 if (!(opt->mode & OPT_MOUNT))
1422 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1425 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1426 OPT_SHOW_MOUNT_STYLE);
1434 static void bch2_put_super(struct super_block *sb)
1436 struct bch_fs *c = sb->s_fs_info;
1441 static const struct super_operations bch_super_operations = {
1442 .alloc_inode = bch2_alloc_inode,
1443 .destroy_inode = bch2_destroy_inode,
1444 .write_inode = bch2_vfs_write_inode,
1445 .evict_inode = bch2_evict_inode,
1446 .sync_fs = bch2_sync_fs,
1447 .statfs = bch2_statfs,
1448 .show_devname = bch2_show_devname,
1449 .show_options = bch2_show_options,
1450 .remount_fs = bch2_remount,
1451 .put_super = bch2_put_super,
1453 .freeze_fs = bch2_freeze,
1454 .unfreeze_fs = bch2_unfreeze,
1458 static int bch2_set_super(struct super_block *s, void *data)
1460 s->s_fs_info = data;
1464 static int bch2_noset_super(struct super_block *s, void *data)
1469 static int bch2_test_super(struct super_block *s, void *data)
1471 struct bch_fs *c = s->s_fs_info;
1472 struct bch_fs **devs = data;
1478 for (i = 0; devs[i]; i++)
1484 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1485 int flags, const char *dev_name, void *data)
1489 struct super_block *sb;
1490 struct inode *vinode;
1491 struct bch_opts opts = bch2_opts_empty();
1493 struct bch_fs **devs_to_fs = NULL;
1494 unsigned i, nr_devs;
1497 opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1499 ret = bch2_parse_mount_opts(NULL, &opts, data);
1501 return ERR_PTR(ret);
1503 if (!dev_name || strlen(dev_name) == 0)
1504 return ERR_PTR(-EINVAL);
1506 devs = split_devs(dev_name, &nr_devs);
1508 return ERR_PTR(-ENOMEM);
1510 devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1512 sb = ERR_PTR(-ENOMEM);
1516 for (i = 0; i < nr_devs; i++)
1517 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1519 sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1520 flags|SB_NOSEC, devs_to_fs);
1524 c = bch2_fs_open(devs, nr_devs, opts);
1530 /* Some options can't be parsed until after the fs is started: */
1531 ret = bch2_parse_mount_opts(c, &opts, data);
1538 bch2_opts_apply(&c->opts, opts);
1540 sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1549 return ERR_CAST(sb);
1554 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1561 sb->s_blocksize = block_bytes(c);
1562 sb->s_blocksize_bits = ilog2(block_bytes(c));
1563 sb->s_maxbytes = MAX_LFS_FILESIZE;
1564 sb->s_op = &bch_super_operations;
1565 sb->s_export_op = &bch_export_ops;
1566 #ifdef CONFIG_BCACHEFS_QUOTA
1567 sb->s_qcop = &bch2_quotactl_operations;
1568 sb->s_quota_types = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1570 sb->s_xattr = bch2_xattr_handlers;
1571 sb->s_magic = BCACHEFS_STATFS_MAGIC;
1572 sb->s_time_gran = c->sb.nsec_per_time_unit;
1573 sb->s_time_min = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1574 sb->s_time_max = div_s64(S64_MAX, c->sb.time_units_per_sec);
1576 strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1578 ret = super_setup_bdi(sb);
1582 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1584 for_each_online_member(ca, c, i) {
1585 struct block_device *bdev = ca->disk_sb.bdev;
1587 /* XXX: create an anonymous device for multi device filesystems */
1589 sb->s_dev = bdev->bd_dev;
1590 percpu_ref_put(&ca->io_ref);
1596 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1598 sb->s_flags |= SB_POSIXACL;
1601 vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1602 if (IS_ERR(vinode)) {
1603 bch_err(c, "error mounting: error getting root inode %i",
1604 (int) PTR_ERR(vinode));
1605 ret = PTR_ERR(vinode);
1609 sb->s_root = d_make_root(vinode);
1611 bch_err(c, "error mounting: error allocating root dentry");
1616 sb->s_flags |= SB_ACTIVE;
1618 return dget(sb->s_root);
1621 deactivate_locked_super(sb);
1622 return ERR_PTR(ret);
1625 static void bch2_kill_sb(struct super_block *sb)
1627 struct bch_fs *c = sb->s_fs_info;
1629 generic_shutdown_super(sb);
1633 static struct file_system_type bcache_fs_type = {
1634 .owner = THIS_MODULE,
1636 .mount = bch2_mount,
1637 .kill_sb = bch2_kill_sb,
1638 .fs_flags = FS_REQUIRES_DEV,
1641 MODULE_ALIAS_FS("bcachefs");
1643 void bch2_vfs_exit(void)
1645 unregister_filesystem(&bcache_fs_type);
1646 if (bch2_inode_cache)
1647 kmem_cache_destroy(bch2_inode_cache);
1650 int __init bch2_vfs_init(void)
1654 bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1655 if (!bch2_inode_cache)
1658 ret = register_filesystem(&bcache_fs_type);
1668 #endif /* NO_BCACHEFS_FS */