1 // SPDX-License-Identifier: GPL-2.0
6 #include "bkey_on_stack.h"
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/xattr.h>
35 static struct kmem_cache *bch2_inode_cache;
37 static void bch2_vfs_inode_init(struct bch_fs *,
38 struct bch_inode_info *,
39 struct bch_inode_unpacked *);
41 static void journal_seq_copy(struct bch_fs *c,
42 struct bch_inode_info *dst,
46 * atomic64_cmpxchg has a fallback for archs that don't support it,
49 atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
50 u64 old, v = READ_ONCE(dst->ei_journal_seq);
55 if (old >= journal_seq)
57 } while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
59 bch2_journal_set_has_inum(&c->journal, dst->v.i_ino, journal_seq);
62 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
64 BUG_ON(atomic_long_read(&lock->v) == 0);
66 if (atomic_long_sub_return_release(i, &lock->v) == 0)
67 wake_up_all(&lock->wait);
70 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
72 long v = atomic_long_read(&lock->v), old;
77 if (i > 0 ? v < 0 : v > 0)
79 } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
80 old, old + i)) != old);
84 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
86 wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
89 void bch2_pagecache_add_put(struct pagecache_lock *lock)
91 __pagecache_lock_put(lock, 1);
94 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
96 return __pagecache_lock_tryget(lock, 1);
99 void bch2_pagecache_add_get(struct pagecache_lock *lock)
101 __pagecache_lock_get(lock, 1);
104 void bch2_pagecache_block_put(struct pagecache_lock *lock)
106 __pagecache_lock_put(lock, -1);
109 void bch2_pagecache_block_get(struct pagecache_lock *lock)
111 __pagecache_lock_get(lock, -1);
114 void bch2_inode_update_after_write(struct bch_fs *c,
115 struct bch_inode_info *inode,
116 struct bch_inode_unpacked *bi,
119 set_nlink(&inode->v, bch2_inode_nlink_get(bi));
120 i_uid_write(&inode->v, bi->bi_uid);
121 i_gid_write(&inode->v, bi->bi_gid);
122 inode->v.i_mode = bi->bi_mode;
124 if (fields & ATTR_ATIME)
125 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
126 if (fields & ATTR_MTIME)
127 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
128 if (fields & ATTR_CTIME)
129 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
131 inode->ei_inode = *bi;
133 bch2_inode_flags_to_vfs(inode);
136 int __must_check bch2_write_inode(struct bch_fs *c,
137 struct bch_inode_info *inode,
139 void *p, unsigned fields)
141 struct btree_trans trans;
142 struct btree_iter *iter;
143 struct bch_inode_unpacked inode_u;
146 bch2_trans_init(&trans, c, 0, 0);
148 bch2_trans_begin(&trans);
150 iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
152 ret = PTR_ERR_OR_ZERO(iter) ?:
153 (set ? set(inode, &inode_u, p) : 0) ?:
154 bch2_inode_write(&trans, iter, &inode_u) ?:
155 bch2_trans_commit(&trans, NULL,
156 &inode->ei_journal_seq,
157 BTREE_INSERT_NOUNLOCK|
158 BTREE_INSERT_NOFAIL);
161 * the btree node lock protects inode->ei_inode, not ei_update_lock;
162 * this is important for inode updates via bchfs_write_index_update
165 bch2_inode_update_after_write(c, inode, &inode_u, fields);
167 bch2_trans_iter_put(&trans, iter);
172 bch2_trans_exit(&trans);
173 return ret < 0 ? ret : 0;
176 int bch2_fs_quota_transfer(struct bch_fs *c,
177 struct bch_inode_info *inode,
178 struct bch_qid new_qid,
180 enum quota_acct_mode mode)
185 qtypes &= enabled_qtypes(c);
187 for (i = 0; i < QTYP_NR; i++)
188 if (new_qid.q[i] == inode->ei_qid.q[i])
189 qtypes &= ~(1U << i);
194 mutex_lock(&inode->ei_quota_lock);
196 ret = bch2_quota_transfer(c, qtypes, new_qid,
199 inode->ei_quota_reserved,
202 for (i = 0; i < QTYP_NR; i++)
203 if (qtypes & (1 << i))
204 inode->ei_qid.q[i] = new_qid.q[i];
206 mutex_unlock(&inode->ei_quota_lock);
211 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
213 struct bch_inode_unpacked inode_u;
214 struct bch_inode_info *inode;
217 inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
218 if (unlikely(!inode))
219 return ERR_PTR(-ENOMEM);
220 if (!(inode->v.i_state & I_NEW))
223 ret = bch2_inode_find_by_inum(c, inum, &inode_u);
225 iget_failed(&inode->v);
229 bch2_vfs_inode_init(c, inode, &inode_u);
231 inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
233 unlock_new_inode(&inode->v);
238 static int inum_test(struct inode *inode, void *p)
240 unsigned long *ino = p;
242 return *ino == inode->i_ino;
245 static struct bch_inode_info *
246 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
247 umode_t mode, dev_t rdev, bool tmpfile)
249 struct bch_fs *c = dir->v.i_sb->s_fs_info;
250 struct user_namespace *ns = dir->v.i_sb->s_user_ns;
251 struct btree_trans trans;
252 struct bch_inode_unpacked dir_u;
253 struct bch_inode_info *inode, *old;
254 struct bch_inode_unpacked inode_u;
255 struct posix_acl *default_acl = NULL, *acl = NULL;
260 * preallocate acls + vfs inode before btree transaction, so that
261 * nothing can fail after the transaction succeeds:
263 #ifdef CONFIG_BCACHEFS_POSIX_ACL
264 ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
268 inode = to_bch_ei(new_inode(c->vfs_sb));
269 if (unlikely(!inode)) {
270 inode = ERR_PTR(-ENOMEM);
274 bch2_inode_init_early(c, &inode_u);
277 mutex_lock(&dir->ei_update_lock);
279 bch2_trans_init(&trans, c, 8,
280 2048 + (!tmpfile ? dentry->d_name.len : 0));
282 bch2_trans_begin(&trans);
284 ret = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
285 !tmpfile ? &dentry->d_name : NULL,
286 from_kuid(ns, current_fsuid()),
287 from_kgid(ns, current_fsgid()),
290 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
291 KEY_TYPE_QUOTA_PREALLOC);
293 goto err_before_quota;
295 ret = bch2_trans_commit(&trans, NULL, &journal_seq,
296 BTREE_INSERT_NOUNLOCK);
298 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
299 KEY_TYPE_QUOTA_WARN);
307 bch2_inode_update_after_write(c, dir, &dir_u,
308 ATTR_MTIME|ATTR_CTIME);
309 journal_seq_copy(c, dir, journal_seq);
310 mutex_unlock(&dir->ei_update_lock);
313 bch2_vfs_inode_init(c, inode, &inode_u);
314 journal_seq_copy(c, inode, journal_seq);
316 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
317 set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
320 * we must insert the new inode into the inode cache before calling
321 * bch2_trans_exit() and dropping locks, else we could race with another
322 * thread pulling the inode in and modifying it:
325 inode->v.i_state |= I_CREATING;
326 old = to_bch_ei(inode_insert5(&inode->v, inode->v.i_ino,
327 inum_test, NULL, &inode->v.i_ino));
330 if (unlikely(old != inode)) {
332 * We raced, another process pulled the new inode into cache
335 journal_seq_copy(c, old, journal_seq);
336 make_bad_inode(&inode->v);
342 * we really don't want insert_inode_locked2() to be setting
345 unlock_new_inode(&inode->v);
348 bch2_trans_exit(&trans);
350 posix_acl_release(default_acl);
351 posix_acl_release(acl);
355 mutex_unlock(&dir->ei_update_lock);
357 bch2_trans_exit(&trans);
358 make_bad_inode(&inode->v);
360 inode = ERR_PTR(ret);
366 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
369 struct bch_fs *c = vdir->i_sb->s_fs_info;
370 struct bch_inode_info *dir = to_bch_ei(vdir);
371 struct inode *vinode = NULL;
374 inum = bch2_dirent_lookup(c, dir->v.i_ino,
379 vinode = bch2_vfs_inode_get(c, inum);
381 return d_splice_alias(vinode, dentry);
384 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
385 umode_t mode, dev_t rdev)
387 struct bch_inode_info *inode =
388 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
391 return PTR_ERR(inode);
393 d_instantiate(dentry, &inode->v);
397 static int bch2_create(struct inode *vdir, struct dentry *dentry,
398 umode_t mode, bool excl)
400 return bch2_mknod(vdir, dentry, mode|S_IFREG, 0);
403 static int __bch2_link(struct bch_fs *c,
404 struct bch_inode_info *inode,
405 struct bch_inode_info *dir,
406 struct dentry *dentry)
408 struct btree_trans trans;
409 struct bch_inode_unpacked dir_u, inode_u;
412 mutex_lock(&inode->ei_update_lock);
413 bch2_trans_init(&trans, c, 4, 1024);
416 bch2_trans_begin(&trans);
417 ret = bch2_link_trans(&trans,
419 inode->v.i_ino, &dir_u, &inode_u,
421 bch2_trans_commit(&trans, NULL,
422 &inode->ei_journal_seq,
423 BTREE_INSERT_NOUNLOCK);
424 } while (ret == -EINTR);
427 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
429 journal_seq_copy(c, inode, dir->ei_journal_seq);
430 bch2_inode_update_after_write(c, dir, &dir_u,
431 ATTR_MTIME|ATTR_CTIME);
432 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
435 bch2_trans_exit(&trans);
436 mutex_unlock(&inode->ei_update_lock);
440 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
441 struct dentry *dentry)
443 struct bch_fs *c = vdir->i_sb->s_fs_info;
444 struct bch_inode_info *dir = to_bch_ei(vdir);
445 struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
448 lockdep_assert_held(&inode->v.i_rwsem);
450 ret = __bch2_link(c, inode, dir, dentry);
455 d_instantiate(dentry, &inode->v);
459 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
461 struct bch_fs *c = vdir->i_sb->s_fs_info;
462 struct bch_inode_info *dir = to_bch_ei(vdir);
463 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
464 struct bch_inode_unpacked dir_u, inode_u;
465 struct btree_trans trans;
468 bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
469 bch2_trans_init(&trans, c, 4, 1024);
472 bch2_trans_begin(&trans);
474 ret = bch2_unlink_trans(&trans,
475 dir->v.i_ino, &dir_u,
476 &inode_u, &dentry->d_name) ?:
477 bch2_trans_commit(&trans, NULL,
478 &dir->ei_journal_seq,
479 BTREE_INSERT_NOUNLOCK|
480 BTREE_INSERT_NOFAIL);
481 } while (ret == -EINTR);
484 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
486 journal_seq_copy(c, inode, dir->ei_journal_seq);
487 bch2_inode_update_after_write(c, dir, &dir_u,
488 ATTR_MTIME|ATTR_CTIME);
489 bch2_inode_update_after_write(c, inode, &inode_u,
493 bch2_trans_exit(&trans);
494 bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
499 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
502 struct bch_fs *c = vdir->i_sb->s_fs_info;
503 struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
506 inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
507 if (unlikely(IS_ERR(inode)))
508 return PTR_ERR(inode);
510 inode_lock(&inode->v);
511 ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
512 inode_unlock(&inode->v);
517 ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
521 journal_seq_copy(c, dir, inode->ei_journal_seq);
523 ret = __bch2_link(c, inode, dir, dentry);
527 d_instantiate(dentry, &inode->v);
534 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
536 return bch2_mknod(vdir, dentry, mode|S_IFDIR, 0);
539 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
540 struct inode *dst_vdir, struct dentry *dst_dentry,
543 struct bch_fs *c = src_vdir->i_sb->s_fs_info;
544 struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
545 struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
546 struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
547 struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
548 struct bch_inode_unpacked dst_dir_u, src_dir_u;
549 struct bch_inode_unpacked src_inode_u, dst_inode_u;
550 struct btree_trans trans;
551 enum bch_rename_mode mode = flags & RENAME_EXCHANGE
552 ? BCH_RENAME_EXCHANGE
553 : dst_dentry->d_inode
554 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
558 if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
561 if (mode == BCH_RENAME_OVERWRITE) {
562 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
568 bch2_trans_init(&trans, c, 8, 2048);
570 bch2_lock_inodes(INODE_UPDATE_LOCK,
576 if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
577 ret = bch2_fs_quota_transfer(c, src_inode,
580 KEY_TYPE_QUOTA_PREALLOC);
585 if (mode == BCH_RENAME_EXCHANGE &&
586 inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
587 ret = bch2_fs_quota_transfer(c, dst_inode,
590 KEY_TYPE_QUOTA_PREALLOC);
596 bch2_trans_begin(&trans);
597 ret = bch2_rename_trans(&trans,
598 src_dir->v.i_ino, &src_dir_u,
599 dst_dir->v.i_ino, &dst_dir_u,
605 bch2_trans_commit(&trans, NULL,
607 BTREE_INSERT_NOUNLOCK);
613 BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
615 dst_inode->v.i_ino != dst_inode_u.bi_inum);
617 bch2_inode_update_after_write(c, src_dir, &src_dir_u,
618 ATTR_MTIME|ATTR_CTIME);
619 journal_seq_copy(c, src_dir, journal_seq);
621 if (src_dir != dst_dir) {
622 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
623 ATTR_MTIME|ATTR_CTIME);
624 journal_seq_copy(c, dst_dir, journal_seq);
627 bch2_inode_update_after_write(c, src_inode, &src_inode_u,
629 journal_seq_copy(c, src_inode, journal_seq);
632 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
634 journal_seq_copy(c, dst_inode, journal_seq);
637 bch2_trans_exit(&trans);
639 bch2_fs_quota_transfer(c, src_inode,
640 bch_qid(&src_inode->ei_inode),
642 KEY_TYPE_QUOTA_NOCHECK);
644 bch2_fs_quota_transfer(c, dst_inode,
645 bch_qid(&dst_inode->ei_inode),
647 KEY_TYPE_QUOTA_NOCHECK);
649 bch2_unlock_inodes(INODE_UPDATE_LOCK,
658 void bch2_setattr_copy(struct bch_inode_info *inode,
659 struct bch_inode_unpacked *bi,
662 struct bch_fs *c = inode->v.i_sb->s_fs_info;
663 unsigned int ia_valid = attr->ia_valid;
665 if (ia_valid & ATTR_UID)
666 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
667 if (ia_valid & ATTR_GID)
668 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
670 if (ia_valid & ATTR_ATIME)
671 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
672 if (ia_valid & ATTR_MTIME)
673 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
674 if (ia_valid & ATTR_CTIME)
675 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
677 if (ia_valid & ATTR_MODE) {
678 umode_t mode = attr->ia_mode;
679 kgid_t gid = ia_valid & ATTR_GID
683 if (!in_group_p(gid) &&
684 !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
690 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
693 struct bch_fs *c = inode->v.i_sb->s_fs_info;
695 struct btree_trans trans;
696 struct btree_iter *inode_iter;
697 struct bch_inode_unpacked inode_u;
698 struct posix_acl *acl = NULL;
701 mutex_lock(&inode->ei_update_lock);
705 if (attr->ia_valid & ATTR_UID)
706 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
708 if (attr->ia_valid & ATTR_GID)
709 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
711 ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
712 KEY_TYPE_QUOTA_PREALLOC);
716 bch2_trans_init(&trans, c, 0, 0);
718 bch2_trans_begin(&trans);
722 inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
724 ret = PTR_ERR_OR_ZERO(inode_iter);
728 bch2_setattr_copy(inode, &inode_u, attr);
730 if (attr->ia_valid & ATTR_MODE) {
731 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
736 ret = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
737 bch2_trans_commit(&trans, NULL,
738 &inode->ei_journal_seq,
739 BTREE_INSERT_NOUNLOCK|
740 BTREE_INSERT_NOFAIL);
747 bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
750 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
752 bch2_trans_exit(&trans);
754 mutex_unlock(&inode->ei_update_lock);
759 static int bch2_getattr(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 dentry *dentry, struct iattr *iattr)
801 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
804 lockdep_assert_held(&inode->v.i_rwsem);
806 ret = setattr_prepare(dentry, iattr);
810 return iattr->ia_valid & ATTR_SIZE
811 ? bch2_truncate(inode, iattr)
812 : bch2_setattr_nonsize(inode, iattr);
815 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
817 struct bch_inode_info *inode =
818 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
821 return PTR_ERR(inode);
823 d_mark_tmpfile(dentry, &inode->v);
824 d_instantiate(dentry, &inode->v);
828 static int bch2_fill_extent(struct bch_fs *c,
829 struct fiemap_extent_info *info,
830 struct bkey_s_c k, unsigned flags)
832 if (bkey_extent_is_direct_data(k.k)) {
833 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
834 const union bch_extent_entry *entry;
835 struct extent_ptr_decoded p;
838 if (k.k->type == KEY_TYPE_reflink_v)
839 flags |= FIEMAP_EXTENT_SHARED;
841 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
843 u64 offset = p.ptr.offset;
845 if (p.crc.compression_type)
846 flags2 |= FIEMAP_EXTENT_ENCODED;
848 offset += p.crc.offset;
850 if ((offset & (c->opts.block_size - 1)) ||
851 (k.k->size & (c->opts.block_size - 1)))
852 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
854 ret = fiemap_fill_next_extent(info,
855 bkey_start_offset(k.k) << 9,
857 k.k->size << 9, flags|flags2);
863 } else if (bkey_extent_is_inline_data(k.k)) {
864 return fiemap_fill_next_extent(info,
865 bkey_start_offset(k.k) << 9,
868 FIEMAP_EXTENT_DATA_INLINE);
869 } else if (k.k->type == KEY_TYPE_reservation) {
870 return fiemap_fill_next_extent(info,
871 bkey_start_offset(k.k) << 9,
874 FIEMAP_EXTENT_DELALLOC|
875 FIEMAP_EXTENT_UNWRITTEN);
881 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
884 struct bch_fs *c = vinode->i_sb->s_fs_info;
885 struct bch_inode_info *ei = to_bch_ei(vinode);
886 struct btree_trans trans;
887 struct btree_iter *iter;
889 struct bkey_on_stack cur, prev;
890 struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
891 unsigned offset_into_extent, sectors;
892 bool have_extent = false;
895 ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
899 if (start + len < start)
902 bkey_on_stack_init(&cur);
903 bkey_on_stack_init(&prev);
904 bch2_trans_init(&trans, c, 0, 0);
906 iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
907 POS(ei->v.i_ino, start >> 9), 0);
909 while ((k = bch2_btree_iter_peek(iter)).k &&
910 !(ret = bkey_err(k)) &&
911 bkey_cmp(iter->pos, end) < 0) {
912 if (!bkey_extent_is_data(k.k) &&
913 k.k->type != KEY_TYPE_reservation) {
914 bch2_btree_iter_next(iter);
918 offset_into_extent = iter->pos.offset -
919 bkey_start_offset(k.k);
920 sectors = k.k->size - offset_into_extent;
922 bkey_on_stack_reassemble(&cur, c, k);
924 ret = bch2_read_indirect_extent(&trans,
925 &offset_into_extent, &cur);
929 k = bkey_i_to_s_c(cur.k);
930 bkey_on_stack_realloc(&prev, c, k.k->u64s);
932 sectors = min(sectors, k.k->size - offset_into_extent);
934 bch2_cut_front(POS(k.k->p.inode,
935 bkey_start_offset(k.k) +
938 bch2_key_resize(&cur.k->k, sectors);
939 cur.k->k.p = iter->pos;
940 cur.k->k.p.offset += cur.k->k.size;
943 ret = bch2_fill_extent(c, info,
944 bkey_i_to_s_c(prev.k), 0);
949 bkey_copy(prev.k, cur.k);
952 bch2_btree_iter_set_pos(iter,
953 POS(iter->pos.inode, iter->pos.offset + sectors));
959 if (!ret && have_extent)
960 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
963 ret = bch2_trans_exit(&trans) ?: ret;
964 bkey_on_stack_exit(&cur, c);
965 bkey_on_stack_exit(&prev, c);
966 return ret < 0 ? ret : 0;
969 static const struct vm_operations_struct bch_vm_ops = {
970 .fault = bch2_page_fault,
971 .map_pages = filemap_map_pages,
972 .page_mkwrite = bch2_page_mkwrite,
975 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
979 vma->vm_ops = &bch_vm_ops;
985 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
987 return generic_file_llseek_size(file, offset, whence,
991 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
993 struct bch_inode_info *inode = file_bch_inode(file);
994 struct bch_fs *c = inode->v.i_sb->s_fs_info;
996 if (!dir_emit_dots(file, ctx))
999 return bch2_readdir(c, inode->v.i_ino, ctx);
1002 static const struct file_operations bch_file_operations = {
1003 .llseek = bch2_llseek,
1004 .read_iter = bch2_read_iter,
1005 .write_iter = bch2_write_iter,
1007 .open = generic_file_open,
1008 .fsync = bch2_fsync,
1009 .splice_read = generic_file_splice_read,
1012 .splice_write = iter_file_splice_write,
1014 .fallocate = bch2_fallocate_dispatch,
1015 .unlocked_ioctl = bch2_fs_file_ioctl,
1016 #ifdef CONFIG_COMPAT
1017 .compat_ioctl = bch2_compat_fs_ioctl,
1019 .remap_file_range = bch2_remap_file_range,
1022 static const struct inode_operations bch_file_inode_operations = {
1023 .getattr = bch2_getattr,
1024 .setattr = bch2_setattr,
1025 .fiemap = bch2_fiemap,
1026 .listxattr = bch2_xattr_list,
1027 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1028 .get_acl = bch2_get_acl,
1029 .set_acl = bch2_set_acl,
1033 static const struct inode_operations bch_dir_inode_operations = {
1034 .lookup = bch2_lookup,
1035 .create = bch2_create,
1037 .unlink = bch2_unlink,
1038 .symlink = bch2_symlink,
1039 .mkdir = bch2_mkdir,
1040 .rmdir = bch2_unlink,
1041 .mknod = bch2_mknod,
1042 .rename = bch2_rename2,
1043 .getattr = bch2_getattr,
1044 .setattr = bch2_setattr,
1045 .tmpfile = bch2_tmpfile,
1046 .listxattr = bch2_xattr_list,
1047 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1048 .get_acl = bch2_get_acl,
1049 .set_acl = bch2_set_acl,
1053 static const struct file_operations bch_dir_file_operations = {
1054 .llseek = bch2_dir_llseek,
1055 .read = generic_read_dir,
1056 .iterate_shared = bch2_vfs_readdir,
1057 .fsync = bch2_fsync,
1058 .unlocked_ioctl = bch2_fs_file_ioctl,
1059 #ifdef CONFIG_COMPAT
1060 .compat_ioctl = bch2_compat_fs_ioctl,
1064 static const struct inode_operations bch_symlink_inode_operations = {
1065 .get_link = page_get_link,
1066 .getattr = bch2_getattr,
1067 .setattr = bch2_setattr,
1068 .listxattr = bch2_xattr_list,
1069 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1070 .get_acl = bch2_get_acl,
1071 .set_acl = bch2_set_acl,
1075 static const struct inode_operations bch_special_inode_operations = {
1076 .getattr = bch2_getattr,
1077 .setattr = bch2_setattr,
1078 .listxattr = bch2_xattr_list,
1079 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1080 .get_acl = bch2_get_acl,
1081 .set_acl = bch2_set_acl,
1085 static const struct address_space_operations bch_address_space_operations = {
1086 .writepage = bch2_writepage,
1087 .readpage = bch2_readpage,
1088 .writepages = bch2_writepages,
1089 .readahead = bch2_readahead,
1090 .set_page_dirty = __set_page_dirty_nobuffers,
1091 .write_begin = bch2_write_begin,
1092 .write_end = bch2_write_end,
1093 .invalidatepage = bch2_invalidatepage,
1094 .releasepage = bch2_releasepage,
1095 .direct_IO = noop_direct_IO,
1096 #ifdef CONFIG_MIGRATION
1097 .migratepage = bch2_migrate_page,
1099 .error_remove_page = generic_error_remove_page,
1102 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1103 u64 ino, u32 generation)
1105 struct bch_fs *c = sb->s_fs_info;
1106 struct inode *vinode;
1108 if (ino < BCACHEFS_ROOT_INO)
1109 return ERR_PTR(-ESTALE);
1111 vinode = bch2_vfs_inode_get(c, ino);
1113 return ERR_CAST(vinode);
1114 if (generation && vinode->i_generation != generation) {
1115 /* we didn't find the right inode.. */
1117 return ERR_PTR(-ESTALE);
1122 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1123 int fh_len, int fh_type)
1125 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1126 bch2_nfs_get_inode);
1129 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1130 int fh_len, int fh_type)
1132 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1133 bch2_nfs_get_inode);
1136 static const struct export_operations bch_export_ops = {
1137 .fh_to_dentry = bch2_fh_to_dentry,
1138 .fh_to_parent = bch2_fh_to_parent,
1139 //.get_parent = bch2_get_parent,
1142 static void bch2_vfs_inode_init(struct bch_fs *c,
1143 struct bch_inode_info *inode,
1144 struct bch_inode_unpacked *bi)
1146 bch2_inode_update_after_write(c, inode, bi, ~0);
1148 inode->v.i_blocks = bi->bi_sectors;
1149 inode->v.i_ino = bi->bi_inum;
1150 inode->v.i_rdev = bi->bi_dev;
1151 inode->v.i_generation = bi->bi_generation;
1152 inode->v.i_size = bi->bi_size;
1154 inode->ei_journal_seq = 0;
1155 inode->ei_quota_reserved = 0;
1156 inode->ei_str_hash = bch2_hash_info_init(c, bi);
1157 inode->ei_qid = bch_qid(bi);
1159 inode->v.i_mapping->a_ops = &bch_address_space_operations;
1161 switch (inode->v.i_mode & S_IFMT) {
1163 inode->v.i_op = &bch_file_inode_operations;
1164 inode->v.i_fop = &bch_file_operations;
1167 inode->v.i_op = &bch_dir_inode_operations;
1168 inode->v.i_fop = &bch_dir_file_operations;
1171 inode_nohighmem(&inode->v);
1172 inode->v.i_op = &bch_symlink_inode_operations;
1175 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1176 inode->v.i_op = &bch_special_inode_operations;
1181 static struct inode *bch2_alloc_inode(struct super_block *sb)
1183 struct bch_inode_info *inode;
1185 inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1189 inode_init_once(&inode->v);
1190 mutex_init(&inode->ei_update_lock);
1191 pagecache_lock_init(&inode->ei_pagecache_lock);
1192 mutex_init(&inode->ei_quota_lock);
1193 inode->ei_journal_seq = 0;
1198 static void bch2_i_callback(struct rcu_head *head)
1200 struct inode *vinode = container_of(head, struct inode, i_rcu);
1201 struct bch_inode_info *inode = to_bch_ei(vinode);
1203 kmem_cache_free(bch2_inode_cache, inode);
1206 static void bch2_destroy_inode(struct inode *vinode)
1208 call_rcu(&vinode->i_rcu, bch2_i_callback);
1211 static int inode_update_times_fn(struct bch_inode_info *inode,
1212 struct bch_inode_unpacked *bi,
1215 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1217 bi->bi_atime = timespec_to_bch2_time(c, inode->v.i_atime);
1218 bi->bi_mtime = timespec_to_bch2_time(c, inode->v.i_mtime);
1219 bi->bi_ctime = timespec_to_bch2_time(c, inode->v.i_ctime);
1224 static int bch2_vfs_write_inode(struct inode *vinode,
1225 struct writeback_control *wbc)
1227 struct bch_fs *c = vinode->i_sb->s_fs_info;
1228 struct bch_inode_info *inode = to_bch_ei(vinode);
1231 mutex_lock(&inode->ei_update_lock);
1232 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1233 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1234 mutex_unlock(&inode->ei_update_lock);
1239 static void bch2_evict_inode(struct inode *vinode)
1241 struct bch_fs *c = vinode->i_sb->s_fs_info;
1242 struct bch_inode_info *inode = to_bch_ei(vinode);
1244 truncate_inode_pages_final(&inode->v.i_data);
1246 clear_inode(&inode->v);
1248 BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1250 if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1251 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1252 KEY_TYPE_QUOTA_WARN);
1253 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1254 KEY_TYPE_QUOTA_WARN);
1255 bch2_inode_rm(c, inode->v.i_ino);
1259 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1261 struct super_block *sb = dentry->d_sb;
1262 struct bch_fs *c = sb->s_fs_info;
1263 struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1264 unsigned shift = sb->s_blocksize_bits - 9;
1266 * this assumes inodes take up 64 bytes, which is a decent average
1269 u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1272 buf->f_type = BCACHEFS_STATFS_MAGIC;
1273 buf->f_bsize = sb->s_blocksize;
1274 buf->f_blocks = usage.capacity >> shift;
1275 buf->f_bfree = (usage.capacity - usage.used) >> shift;
1276 buf->f_bavail = buf->f_bfree;
1278 buf->f_files = usage.nr_inodes + avail_inodes;
1279 buf->f_ffree = avail_inodes;
1281 fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1282 le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1283 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1284 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1285 buf->f_namelen = BCH_NAME_MAX;
1290 static int bch2_sync_fs(struct super_block *sb, int wait)
1292 struct bch_fs *c = sb->s_fs_info;
1294 if (c->opts.journal_flush_disabled)
1298 bch2_journal_flush_async(&c->journal, NULL);
1302 return bch2_journal_flush(&c->journal);
1305 static struct bch_fs *bch2_path_to_fs(const char *dev)
1308 struct block_device *bdev = lookup_bdev(dev);
1311 return ERR_CAST(bdev);
1313 c = bch2_bdev_to_fs(bdev);
1316 closure_put(&c->cl);
1317 return c ?: ERR_PTR(-ENOENT);
1320 static char **split_devs(const char *_dev_name, unsigned *nr)
1322 char *dev_name = NULL, **devs = NULL, *s;
1323 size_t i, nr_devs = 0;
1325 dev_name = kstrdup(_dev_name, GFP_KERNEL);
1329 for (s = dev_name; s; s = strchr(s + 1, ':'))
1332 devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1338 for (i = 0, s = dev_name;
1340 (s = strchr(s, ':')) && (*s++ = '\0'))
1347 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1349 struct bch_fs *c = sb->s_fs_info;
1350 struct bch_opts opts = bch2_opts_empty();
1353 opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1355 ret = bch2_parse_mount_opts(c, &opts, data);
1359 if (opts.read_only != c->opts.read_only) {
1360 down_write(&c->state_lock);
1362 if (opts.read_only) {
1363 bch2_fs_read_only(c);
1365 sb->s_flags |= SB_RDONLY;
1367 ret = bch2_fs_read_write(c);
1369 bch_err(c, "error going rw: %i", ret);
1370 up_write(&c->state_lock);
1374 sb->s_flags &= ~SB_RDONLY;
1377 c->opts.read_only = opts.read_only;
1379 up_write(&c->state_lock);
1382 if (opts.errors >= 0)
1383 c->opts.errors = opts.errors;
1388 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1390 struct bch_fs *c = root->d_sb->s_fs_info;
1395 for_each_online_member(ca, c, i) {
1399 seq_puts(seq, "/dev/");
1400 seq_puts(seq, ca->name);
1406 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1408 struct bch_fs *c = root->d_sb->s_fs_info;
1412 for (i = 0; i < bch2_opts_nr; i++) {
1413 const struct bch_option *opt = &bch2_opt_table[i];
1414 u64 v = bch2_opt_get_by_id(&c->opts, i);
1416 if (!(opt->mode & OPT_MOUNT))
1419 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1422 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1423 OPT_SHOW_MOUNT_STYLE);
1431 static void bch2_put_super(struct super_block *sb)
1433 struct bch_fs *c = sb->s_fs_info;
1438 static const struct super_operations bch_super_operations = {
1439 .alloc_inode = bch2_alloc_inode,
1440 .destroy_inode = bch2_destroy_inode,
1441 .write_inode = bch2_vfs_write_inode,
1442 .evict_inode = bch2_evict_inode,
1443 .sync_fs = bch2_sync_fs,
1444 .statfs = bch2_statfs,
1445 .show_devname = bch2_show_devname,
1446 .show_options = bch2_show_options,
1447 .remount_fs = bch2_remount,
1448 .put_super = bch2_put_super,
1450 .freeze_fs = bch2_freeze,
1451 .unfreeze_fs = bch2_unfreeze,
1455 static int bch2_set_super(struct super_block *s, void *data)
1457 s->s_fs_info = data;
1461 static int bch2_noset_super(struct super_block *s, void *data)
1466 static int bch2_test_super(struct super_block *s, void *data)
1468 struct bch_fs *c = s->s_fs_info;
1469 struct bch_fs **devs = data;
1475 for (i = 0; devs[i]; i++)
1481 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1482 int flags, const char *dev_name, void *data)
1486 struct super_block *sb;
1487 struct inode *vinode;
1488 struct bch_opts opts = bch2_opts_empty();
1490 struct bch_fs **devs_to_fs = NULL;
1491 unsigned i, nr_devs;
1494 opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1496 ret = bch2_parse_mount_opts(NULL, &opts, data);
1498 return ERR_PTR(ret);
1500 devs = split_devs(dev_name, &nr_devs);
1502 return ERR_PTR(-ENOMEM);
1504 devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1506 sb = ERR_PTR(-ENOMEM);
1510 for (i = 0; i < nr_devs; i++)
1511 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1513 sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1514 flags|SB_NOSEC, devs_to_fs);
1518 c = bch2_fs_open(devs, nr_devs, opts);
1524 /* Some options can't be parsed until after the fs is started: */
1525 ret = bch2_parse_mount_opts(c, &opts, data);
1532 bch2_opts_apply(&c->opts, opts);
1534 sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1543 return ERR_CAST(sb);
1548 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1555 sb->s_blocksize = block_bytes(c);
1556 sb->s_blocksize_bits = ilog2(block_bytes(c));
1557 sb->s_maxbytes = MAX_LFS_FILESIZE;
1558 sb->s_op = &bch_super_operations;
1559 sb->s_export_op = &bch_export_ops;
1560 #ifdef CONFIG_BCACHEFS_QUOTA
1561 sb->s_qcop = &bch2_quotactl_operations;
1562 sb->s_quota_types = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1564 sb->s_xattr = bch2_xattr_handlers;
1565 sb->s_magic = BCACHEFS_STATFS_MAGIC;
1566 sb->s_time_gran = c->sb.time_precision;
1568 strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1570 ret = super_setup_bdi(sb);
1574 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1576 for_each_online_member(ca, c, i) {
1577 struct block_device *bdev = ca->disk_sb.bdev;
1579 /* XXX: create an anonymous device for multi device filesystems */
1581 sb->s_dev = bdev->bd_dev;
1582 percpu_ref_put(&ca->io_ref);
1586 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1588 sb->s_flags |= SB_POSIXACL;
1591 vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1592 if (IS_ERR(vinode)) {
1593 bch_err(c, "error mounting: error getting root inode %i",
1594 (int) PTR_ERR(vinode));
1595 ret = PTR_ERR(vinode);
1599 sb->s_root = d_make_root(vinode);
1601 bch_err(c, "error mounting: error allocating root dentry");
1606 sb->s_flags |= SB_ACTIVE;
1608 return dget(sb->s_root);
1611 deactivate_locked_super(sb);
1612 return ERR_PTR(ret);
1615 static void bch2_kill_sb(struct super_block *sb)
1617 struct bch_fs *c = sb->s_fs_info;
1619 generic_shutdown_super(sb);
1623 static struct file_system_type bcache_fs_type = {
1624 .owner = THIS_MODULE,
1626 .mount = bch2_mount,
1627 .kill_sb = bch2_kill_sb,
1628 .fs_flags = FS_REQUIRES_DEV,
1631 MODULE_ALIAS_FS("bcachefs");
1633 void bch2_vfs_exit(void)
1635 unregister_filesystem(&bcache_fs_type);
1636 if (bch2_inode_cache)
1637 kmem_cache_destroy(bch2_inode_cache);
1640 int __init bch2_vfs_init(void)
1644 bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1645 if (!bch2_inode_cache)
1648 ret = register_filesystem(&bcache_fs_type);
1658 #endif /* NO_BCACHEFS_FS */