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/module.h>
29 #include <linux/posix_acl.h>
30 #include <linux/random.h>
31 #include <linux/statfs.h>
32 #include <linux/xattr.h>
34 static struct kmem_cache *bch2_inode_cache;
36 static void bch2_vfs_inode_init(struct bch_fs *,
37 struct bch_inode_info *,
38 struct bch_inode_unpacked *);
40 static void journal_seq_copy(struct bch_inode_info *dst,
43 u64 old, v = READ_ONCE(dst->ei_journal_seq);
48 if (old >= journal_seq)
50 } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
53 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
55 BUG_ON(atomic_long_read(&lock->v) == 0);
57 if (atomic_long_sub_return_release(i, &lock->v) == 0)
58 wake_up_all(&lock->wait);
61 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
63 long v = atomic_long_read(&lock->v), old;
68 if (i > 0 ? v < 0 : v > 0)
70 } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
71 old, old + i)) != old);
75 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
77 wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
80 void bch2_pagecache_add_put(struct pagecache_lock *lock)
82 __pagecache_lock_put(lock, 1);
85 void bch2_pagecache_add_get(struct pagecache_lock *lock)
87 __pagecache_lock_get(lock, 1);
90 void bch2_pagecache_block_put(struct pagecache_lock *lock)
92 __pagecache_lock_put(lock, -1);
95 void bch2_pagecache_block_get(struct pagecache_lock *lock)
97 __pagecache_lock_get(lock, -1);
100 void bch2_inode_update_after_write(struct bch_fs *c,
101 struct bch_inode_info *inode,
102 struct bch_inode_unpacked *bi,
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;
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);
117 inode->ei_inode = *bi;
119 bch2_inode_flags_to_vfs(inode);
122 int __must_check bch2_write_inode(struct bch_fs *c,
123 struct bch_inode_info *inode,
125 void *p, unsigned fields)
127 struct btree_trans trans;
128 struct btree_iter *iter;
129 struct bch_inode_unpacked inode_u;
132 bch2_trans_init(&trans, c, 0, 0);
134 bch2_trans_begin(&trans);
136 iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
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,
144 BTREE_INSERT_NOUNLOCK|
145 BTREE_INSERT_NOFAIL);
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
154 bch2_inode_update_after_write(c, inode, &inode_u, fields);
156 bch2_trans_exit(&trans);
157 return ret < 0 ? ret : 0;
160 int bch2_fs_quota_transfer(struct bch_fs *c,
161 struct bch_inode_info *inode,
162 struct bch_qid new_qid,
164 enum quota_acct_mode mode)
169 qtypes &= enabled_qtypes(c);
171 for (i = 0; i < QTYP_NR; i++)
172 if (new_qid.q[i] == inode->ei_qid.q[i])
173 qtypes &= ~(1U << i);
178 mutex_lock(&inode->ei_quota_lock);
180 ret = bch2_quota_transfer(c, qtypes, new_qid,
183 inode->ei_quota_reserved,
186 for (i = 0; i < QTYP_NR; i++)
187 if (qtypes & (1 << i))
188 inode->ei_qid.q[i] = new_qid.q[i];
190 mutex_unlock(&inode->ei_quota_lock);
195 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
197 struct bch_inode_unpacked inode_u;
198 struct bch_inode_info *inode;
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))
207 ret = bch2_inode_find_by_inum(c, inum, &inode_u);
209 iget_failed(&inode->v);
213 bch2_vfs_inode_init(c, inode, &inode_u);
215 inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
217 unlock_new_inode(&inode->v);
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)
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;
237 * preallocate acls + vfs inode before btree transaction, so that
238 * nothing can fail after the transaction succeeds:
240 #ifdef CONFIG_BCACHEFS_POSIX_ACL
241 ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
245 inode = to_bch_ei(new_inode(c->vfs_sb));
246 if (unlikely(!inode)) {
247 inode = ERR_PTR(-ENOMEM);
251 bch2_inode_init_early(c, &inode_u);
254 mutex_lock(&dir->ei_update_lock);
256 bch2_trans_init(&trans, c, 8, 1024);
258 bch2_trans_begin(&trans);
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()),
266 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
267 KEY_TYPE_QUOTA_PREALLOC);
269 goto err_before_quota;
271 ret = bch2_trans_commit(&trans, NULL, &journal_seq,
273 BTREE_INSERT_NOUNLOCK);
275 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
276 KEY_TYPE_QUOTA_WARN);
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);
290 bch2_vfs_inode_init(c, inode, &inode_u);
291 journal_seq_copy(inode, journal_seq);
293 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
294 set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
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:
302 old = to_bch_ei(insert_inode_locked2(&inode->v));
305 * We raced, another process pulled the new inode into cache
308 journal_seq_copy(old, journal_seq);
309 make_bad_inode(&inode->v);
315 * we really don't want insert_inode_locked2() to be setting
318 unlock_new_inode(&inode->v);
321 bch2_trans_exit(&trans);
323 posix_acl_release(default_acl);
324 posix_acl_release(acl);
328 mutex_unlock(&dir->ei_update_lock);
330 bch2_trans_exit(&trans);
331 make_bad_inode(&inode->v);
333 inode = ERR_PTR(ret);
339 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
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;
347 inum = bch2_dirent_lookup(c, dir->v.i_ino,
352 vinode = bch2_vfs_inode_get(c, inum);
354 return d_splice_alias(vinode, dentry);
357 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
358 umode_t mode, dev_t rdev)
360 struct bch_inode_info *inode =
361 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
364 return PTR_ERR(inode);
366 d_instantiate(dentry, &inode->v);
370 static int bch2_create(struct inode *vdir, struct dentry *dentry,
371 umode_t mode, bool excl)
373 return bch2_mknod(vdir, dentry, mode|S_IFREG, 0);
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)
381 struct btree_trans trans;
382 struct bch_inode_unpacked inode_u;
385 mutex_lock(&inode->ei_update_lock);
386 bch2_trans_init(&trans, c, 4, 1024);
389 bch2_trans_begin(&trans);
390 ret = bch2_link_trans(&trans,
392 inode->v.i_ino, &inode_u,
394 bch2_trans_commit(&trans, NULL,
395 &inode->ei_journal_seq,
397 BTREE_INSERT_NOUNLOCK);
398 } while (ret == -EINTR);
401 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
403 bch2_trans_exit(&trans);
404 mutex_unlock(&inode->ei_update_lock);
408 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
409 struct dentry *dentry)
411 struct bch_fs *c = vdir->i_sb->s_fs_info;
412 struct bch_inode_info *dir = to_bch_ei(vdir);
413 struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
416 lockdep_assert_held(&inode->v.i_rwsem);
418 ret = __bch2_link(c, inode, dir, dentry);
423 d_instantiate(dentry, &inode->v);
427 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
429 struct bch_fs *c = vdir->i_sb->s_fs_info;
430 struct bch_inode_info *dir = to_bch_ei(vdir);
431 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
432 struct bch_inode_unpacked dir_u, inode_u;
433 struct btree_trans trans;
436 bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
437 bch2_trans_init(&trans, c, 4, 1024);
440 bch2_trans_begin(&trans);
442 ret = bch2_unlink_trans(&trans,
443 dir->v.i_ino, &dir_u,
444 &inode_u, &dentry->d_name) ?:
445 bch2_trans_commit(&trans, NULL,
446 &dir->ei_journal_seq,
448 BTREE_INSERT_NOUNLOCK|
449 BTREE_INSERT_NOFAIL);
450 } while (ret == -EINTR);
453 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
455 journal_seq_copy(inode, dir->ei_journal_seq);
456 bch2_inode_update_after_write(c, dir, &dir_u,
457 ATTR_MTIME|ATTR_CTIME);
458 bch2_inode_update_after_write(c, inode, &inode_u,
462 bch2_trans_exit(&trans);
463 bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
468 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
471 struct bch_fs *c = vdir->i_sb->s_fs_info;
472 struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
475 inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
476 if (unlikely(IS_ERR(inode)))
477 return PTR_ERR(inode);
479 inode_lock(&inode->v);
480 ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
481 inode_unlock(&inode->v);
486 ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
490 journal_seq_copy(dir, inode->ei_journal_seq);
492 ret = __bch2_link(c, inode, dir, dentry);
496 d_instantiate(dentry, &inode->v);
503 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
505 return bch2_mknod(vdir, dentry, mode|S_IFDIR, 0);
508 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
509 struct inode *dst_vdir, struct dentry *dst_dentry,
512 struct bch_fs *c = src_vdir->i_sb->s_fs_info;
513 struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
514 struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
515 struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
516 struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
517 struct bch_inode_unpacked dst_dir_u, src_dir_u;
518 struct bch_inode_unpacked src_inode_u, dst_inode_u;
519 struct btree_trans trans;
520 enum bch_rename_mode mode = flags & RENAME_EXCHANGE
521 ? BCH_RENAME_EXCHANGE
522 : dst_dentry->d_inode
523 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
527 if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
530 if (mode == BCH_RENAME_OVERWRITE) {
531 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
537 bch2_trans_init(&trans, c, 8, 2048);
539 bch2_lock_inodes(INODE_UPDATE_LOCK,
545 if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
546 ret = bch2_fs_quota_transfer(c, src_inode,
549 KEY_TYPE_QUOTA_PREALLOC);
554 if (mode == BCH_RENAME_EXCHANGE &&
555 inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
556 ret = bch2_fs_quota_transfer(c, dst_inode,
559 KEY_TYPE_QUOTA_PREALLOC);
565 bch2_trans_begin(&trans);
566 ret = bch2_rename_trans(&trans,
567 src_dir->v.i_ino, &src_dir_u,
568 dst_dir->v.i_ino, &dst_dir_u,
574 bch2_trans_commit(&trans, NULL,
577 BTREE_INSERT_NOUNLOCK);
583 BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
585 dst_inode->v.i_ino != dst_inode_u.bi_inum);
587 bch2_inode_update_after_write(c, src_dir, &src_dir_u,
588 ATTR_MTIME|ATTR_CTIME);
589 journal_seq_copy(src_dir, journal_seq);
591 if (src_dir != dst_dir) {
592 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
593 ATTR_MTIME|ATTR_CTIME);
594 journal_seq_copy(dst_dir, journal_seq);
597 bch2_inode_update_after_write(c, src_inode, &src_inode_u,
599 journal_seq_copy(src_inode, journal_seq);
602 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
604 journal_seq_copy(dst_inode, journal_seq);
607 bch2_trans_exit(&trans);
609 bch2_fs_quota_transfer(c, src_inode,
610 bch_qid(&src_inode->ei_inode),
612 KEY_TYPE_QUOTA_NOCHECK);
614 bch2_fs_quota_transfer(c, dst_inode,
615 bch_qid(&dst_inode->ei_inode),
617 KEY_TYPE_QUOTA_NOCHECK);
619 bch2_unlock_inodes(INODE_UPDATE_LOCK,
628 void bch2_setattr_copy(struct bch_inode_info *inode,
629 struct bch_inode_unpacked *bi,
632 struct bch_fs *c = inode->v.i_sb->s_fs_info;
633 unsigned int ia_valid = attr->ia_valid;
635 if (ia_valid & ATTR_UID)
636 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
637 if (ia_valid & ATTR_GID)
638 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
640 if (ia_valid & ATTR_ATIME)
641 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
642 if (ia_valid & ATTR_MTIME)
643 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
644 if (ia_valid & ATTR_CTIME)
645 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
647 if (ia_valid & ATTR_MODE) {
648 umode_t mode = attr->ia_mode;
649 kgid_t gid = ia_valid & ATTR_GID
653 if (!in_group_p(gid) &&
654 !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
660 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
663 struct bch_fs *c = inode->v.i_sb->s_fs_info;
665 struct btree_trans trans;
666 struct btree_iter *inode_iter;
667 struct bch_inode_unpacked inode_u;
668 struct posix_acl *acl = NULL;
671 mutex_lock(&inode->ei_update_lock);
675 if (attr->ia_valid & ATTR_UID)
676 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
678 if (attr->ia_valid & ATTR_GID)
679 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
681 ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
682 KEY_TYPE_QUOTA_PREALLOC);
686 bch2_trans_init(&trans, c, 0, 0);
688 bch2_trans_begin(&trans);
692 inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
694 ret = PTR_ERR_OR_ZERO(inode_iter);
698 bch2_setattr_copy(inode, &inode_u, attr);
700 if (attr->ia_valid & ATTR_MODE) {
701 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
706 ret = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
707 bch2_trans_commit(&trans, NULL,
708 &inode->ei_journal_seq,
710 BTREE_INSERT_NOUNLOCK|
711 BTREE_INSERT_NOFAIL);
718 bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
721 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
723 bch2_trans_exit(&trans);
725 mutex_unlock(&inode->ei_update_lock);
730 static int bch2_getattr(const struct path *path, struct kstat *stat,
731 u32 request_mask, unsigned query_flags)
733 struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
734 struct bch_fs *c = inode->v.i_sb->s_fs_info;
736 stat->dev = inode->v.i_sb->s_dev;
737 stat->ino = inode->v.i_ino;
738 stat->mode = inode->v.i_mode;
739 stat->nlink = inode->v.i_nlink;
740 stat->uid = inode->v.i_uid;
741 stat->gid = inode->v.i_gid;
742 stat->rdev = inode->v.i_rdev;
743 stat->size = i_size_read(&inode->v);
744 stat->atime = inode->v.i_atime;
745 stat->mtime = inode->v.i_mtime;
746 stat->ctime = inode->v.i_ctime;
747 stat->blksize = block_bytes(c);
748 stat->blocks = inode->v.i_blocks;
750 if (request_mask & STATX_BTIME) {
751 stat->result_mask |= STATX_BTIME;
752 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
755 if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
756 stat->attributes |= STATX_ATTR_IMMUTABLE;
757 stat->attributes_mask |= STATX_ATTR_IMMUTABLE;
759 if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
760 stat->attributes |= STATX_ATTR_APPEND;
761 stat->attributes_mask |= STATX_ATTR_APPEND;
763 if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
764 stat->attributes |= STATX_ATTR_NODUMP;
765 stat->attributes_mask |= STATX_ATTR_NODUMP;
770 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
772 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
775 lockdep_assert_held(&inode->v.i_rwsem);
777 ret = setattr_prepare(dentry, iattr);
781 return iattr->ia_valid & ATTR_SIZE
782 ? bch2_truncate(inode, iattr)
783 : bch2_setattr_nonsize(inode, iattr);
786 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
788 struct bch_inode_info *inode =
789 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
792 return PTR_ERR(inode);
794 d_mark_tmpfile(dentry, &inode->v);
795 d_instantiate(dentry, &inode->v);
799 static int bch2_fill_extent(struct bch_fs *c,
800 struct fiemap_extent_info *info,
801 struct bkey_s_c k, unsigned flags)
803 if (bkey_extent_is_data(k.k)) {
804 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
805 const union bch_extent_entry *entry;
806 struct extent_ptr_decoded p;
809 if (k.k->type == KEY_TYPE_reflink_v)
810 flags |= FIEMAP_EXTENT_SHARED;
812 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
814 u64 offset = p.ptr.offset;
816 if (p.crc.compression_type)
817 flags2 |= FIEMAP_EXTENT_ENCODED;
819 offset += p.crc.offset;
821 if ((offset & (c->opts.block_size - 1)) ||
822 (k.k->size & (c->opts.block_size - 1)))
823 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
825 ret = fiemap_fill_next_extent(info,
826 bkey_start_offset(k.k) << 9,
828 k.k->size << 9, flags|flags2);
834 } else if (k.k->type == KEY_TYPE_reservation) {
835 return fiemap_fill_next_extent(info,
836 bkey_start_offset(k.k) << 9,
839 FIEMAP_EXTENT_DELALLOC|
840 FIEMAP_EXTENT_UNWRITTEN);
846 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
849 struct bch_fs *c = vinode->i_sb->s_fs_info;
850 struct bch_inode_info *ei = to_bch_ei(vinode);
851 struct btree_trans trans;
852 struct btree_iter *iter;
854 struct bkey_on_stack cur, prev;
855 struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
856 unsigned offset_into_extent, sectors;
857 bool have_extent = false;
860 if (start + len < start)
863 bkey_on_stack_init(&cur);
864 bkey_on_stack_init(&prev);
865 bch2_trans_init(&trans, c, 0, 0);
867 iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
868 POS(ei->v.i_ino, start >> 9), 0);
870 while ((k = bch2_btree_iter_peek(iter)).k &&
871 !(ret = bkey_err(k)) &&
872 bkey_cmp(iter->pos, end) < 0) {
873 if (!bkey_extent_is_data(k.k) &&
874 k.k->type != KEY_TYPE_reservation) {
875 bch2_btree_iter_next(iter);
879 bkey_on_stack_realloc(&cur, c, k.k->u64s);
880 bkey_on_stack_realloc(&prev, c, k.k->u64s);
881 bkey_reassemble(cur.k, k);
882 k = bkey_i_to_s_c(cur.k);
884 offset_into_extent = iter->pos.offset -
885 bkey_start_offset(k.k);
886 sectors = k.k->size - offset_into_extent;
888 ret = bch2_read_indirect_extent(&trans,
889 &offset_into_extent, cur.k);
893 sectors = min(sectors, k.k->size - offset_into_extent);
895 if (offset_into_extent)
896 bch2_cut_front(POS(k.k->p.inode,
897 bkey_start_offset(k.k) +
900 bch2_key_resize(&cur.k->k, sectors);
901 cur.k->k.p = iter->pos;
902 cur.k->k.p.offset += cur.k->k.size;
905 ret = bch2_fill_extent(c, info,
906 bkey_i_to_s_c(prev.k), 0);
911 bkey_copy(prev.k, cur.k);
914 if (k.k->type == KEY_TYPE_reflink_v)
915 bch2_btree_iter_set_pos(iter, k.k->p);
917 bch2_btree_iter_next(iter);
923 if (!ret && have_extent)
924 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
927 ret = bch2_trans_exit(&trans) ?: ret;
928 bkey_on_stack_exit(&cur, c);
929 bkey_on_stack_exit(&prev, c);
930 return ret < 0 ? ret : 0;
933 static const struct vm_operations_struct bch_vm_ops = {
934 .fault = bch2_page_fault,
935 .map_pages = filemap_map_pages,
936 .page_mkwrite = bch2_page_mkwrite,
939 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
943 vma->vm_ops = &bch_vm_ops;
949 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
951 return generic_file_llseek_size(file, offset, whence,
955 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
957 struct bch_inode_info *inode = file_bch_inode(file);
958 struct bch_fs *c = inode->v.i_sb->s_fs_info;
960 if (!dir_emit_dots(file, ctx))
963 return bch2_readdir(c, inode->v.i_ino, ctx);
966 static const struct file_operations bch_file_operations = {
967 .llseek = bch2_llseek,
968 .read_iter = bch2_read_iter,
969 .write_iter = bch2_write_iter,
971 .open = generic_file_open,
973 .splice_read = generic_file_splice_read,
974 .splice_write = iter_file_splice_write,
975 .fallocate = bch2_fallocate_dispatch,
976 .unlocked_ioctl = bch2_fs_file_ioctl,
978 .compat_ioctl = bch2_compat_fs_ioctl,
980 .remap_file_range = bch2_remap_file_range,
983 static const struct inode_operations bch_file_inode_operations = {
984 .getattr = bch2_getattr,
985 .setattr = bch2_setattr,
986 .fiemap = bch2_fiemap,
987 .listxattr = bch2_xattr_list,
988 #ifdef CONFIG_BCACHEFS_POSIX_ACL
989 .get_acl = bch2_get_acl,
990 .set_acl = bch2_set_acl,
994 static const struct inode_operations bch_dir_inode_operations = {
995 .lookup = bch2_lookup,
996 .create = bch2_create,
998 .unlink = bch2_unlink,
999 .symlink = bch2_symlink,
1000 .mkdir = bch2_mkdir,
1001 .rmdir = bch2_unlink,
1002 .mknod = bch2_mknod,
1003 .rename = bch2_rename2,
1004 .getattr = bch2_getattr,
1005 .setattr = bch2_setattr,
1006 .tmpfile = bch2_tmpfile,
1007 .listxattr = bch2_xattr_list,
1008 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1009 .get_acl = bch2_get_acl,
1010 .set_acl = bch2_set_acl,
1014 static const struct file_operations bch_dir_file_operations = {
1015 .llseek = bch2_dir_llseek,
1016 .read = generic_read_dir,
1017 .iterate_shared = bch2_vfs_readdir,
1018 .fsync = bch2_fsync,
1019 .unlocked_ioctl = bch2_fs_file_ioctl,
1020 #ifdef CONFIG_COMPAT
1021 .compat_ioctl = bch2_compat_fs_ioctl,
1025 static const struct inode_operations bch_symlink_inode_operations = {
1026 .get_link = page_get_link,
1027 .getattr = bch2_getattr,
1028 .setattr = bch2_setattr,
1029 .listxattr = bch2_xattr_list,
1030 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1031 .get_acl = bch2_get_acl,
1032 .set_acl = bch2_set_acl,
1036 static const struct inode_operations bch_special_inode_operations = {
1037 .getattr = bch2_getattr,
1038 .setattr = bch2_setattr,
1039 .listxattr = bch2_xattr_list,
1040 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1041 .get_acl = bch2_get_acl,
1042 .set_acl = bch2_set_acl,
1046 static const struct address_space_operations bch_address_space_operations = {
1047 .writepage = bch2_writepage,
1048 .readpage = bch2_readpage,
1049 .writepages = bch2_writepages,
1050 .readpages = bch2_readpages,
1051 .set_page_dirty = __set_page_dirty_nobuffers,
1052 .write_begin = bch2_write_begin,
1053 .write_end = bch2_write_end,
1054 .invalidatepage = bch2_invalidatepage,
1055 .releasepage = bch2_releasepage,
1056 .direct_IO = noop_direct_IO,
1057 #ifdef CONFIG_MIGRATION
1058 .migratepage = bch2_migrate_page,
1060 .error_remove_page = generic_error_remove_page,
1063 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1064 u64 ino, u32 generation)
1066 struct bch_fs *c = sb->s_fs_info;
1067 struct inode *vinode;
1069 if (ino < BCACHEFS_ROOT_INO)
1070 return ERR_PTR(-ESTALE);
1072 vinode = bch2_vfs_inode_get(c, ino);
1074 return ERR_CAST(vinode);
1075 if (generation && vinode->i_generation != generation) {
1076 /* we didn't find the right inode.. */
1078 return ERR_PTR(-ESTALE);
1083 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1084 int fh_len, int fh_type)
1086 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1087 bch2_nfs_get_inode);
1090 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1091 int fh_len, int fh_type)
1093 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1094 bch2_nfs_get_inode);
1097 static const struct export_operations bch_export_ops = {
1098 .fh_to_dentry = bch2_fh_to_dentry,
1099 .fh_to_parent = bch2_fh_to_parent,
1100 //.get_parent = bch2_get_parent,
1103 static void bch2_vfs_inode_init(struct bch_fs *c,
1104 struct bch_inode_info *inode,
1105 struct bch_inode_unpacked *bi)
1107 bch2_inode_update_after_write(c, inode, bi, ~0);
1109 inode->v.i_blocks = bi->bi_sectors;
1110 inode->v.i_ino = bi->bi_inum;
1111 inode->v.i_rdev = bi->bi_dev;
1112 inode->v.i_generation = bi->bi_generation;
1113 inode->v.i_size = bi->bi_size;
1115 inode->ei_journal_seq = 0;
1116 inode->ei_quota_reserved = 0;
1117 inode->ei_str_hash = bch2_hash_info_init(c, bi);
1118 inode->ei_qid = bch_qid(bi);
1120 inode->v.i_mapping->a_ops = &bch_address_space_operations;
1122 switch (inode->v.i_mode & S_IFMT) {
1124 inode->v.i_op = &bch_file_inode_operations;
1125 inode->v.i_fop = &bch_file_operations;
1128 inode->v.i_op = &bch_dir_inode_operations;
1129 inode->v.i_fop = &bch_dir_file_operations;
1132 inode_nohighmem(&inode->v);
1133 inode->v.i_op = &bch_symlink_inode_operations;
1136 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1137 inode->v.i_op = &bch_special_inode_operations;
1142 static struct inode *bch2_alloc_inode(struct super_block *sb)
1144 struct bch_inode_info *inode;
1146 inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1150 inode_init_once(&inode->v);
1151 mutex_init(&inode->ei_update_lock);
1152 pagecache_lock_init(&inode->ei_pagecache_lock);
1153 mutex_init(&inode->ei_quota_lock);
1154 inode->ei_journal_seq = 0;
1159 static void bch2_i_callback(struct rcu_head *head)
1161 struct inode *vinode = container_of(head, struct inode, i_rcu);
1162 struct bch_inode_info *inode = to_bch_ei(vinode);
1164 kmem_cache_free(bch2_inode_cache, inode);
1167 static void bch2_destroy_inode(struct inode *vinode)
1169 call_rcu(&vinode->i_rcu, bch2_i_callback);
1172 static int inode_update_times_fn(struct bch_inode_info *inode,
1173 struct bch_inode_unpacked *bi,
1176 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1178 bi->bi_atime = timespec_to_bch2_time(c, inode->v.i_atime);
1179 bi->bi_mtime = timespec_to_bch2_time(c, inode->v.i_mtime);
1180 bi->bi_ctime = timespec_to_bch2_time(c, inode->v.i_ctime);
1185 static int bch2_vfs_write_inode(struct inode *vinode,
1186 struct writeback_control *wbc)
1188 struct bch_fs *c = vinode->i_sb->s_fs_info;
1189 struct bch_inode_info *inode = to_bch_ei(vinode);
1192 mutex_lock(&inode->ei_update_lock);
1193 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1194 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1195 mutex_unlock(&inode->ei_update_lock);
1200 static void bch2_evict_inode(struct inode *vinode)
1202 struct bch_fs *c = vinode->i_sb->s_fs_info;
1203 struct bch_inode_info *inode = to_bch_ei(vinode);
1205 truncate_inode_pages_final(&inode->v.i_data);
1207 clear_inode(&inode->v);
1209 BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1211 if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1212 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1213 KEY_TYPE_QUOTA_WARN);
1214 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1215 KEY_TYPE_QUOTA_WARN);
1216 bch2_inode_rm(c, inode->v.i_ino);
1220 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1222 struct super_block *sb = dentry->d_sb;
1223 struct bch_fs *c = sb->s_fs_info;
1224 struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1225 unsigned shift = sb->s_blocksize_bits - 9;
1228 buf->f_type = BCACHEFS_STATFS_MAGIC;
1229 buf->f_bsize = sb->s_blocksize;
1230 buf->f_blocks = usage.capacity >> shift;
1231 buf->f_bfree = (usage.capacity - usage.used) >> shift;
1232 buf->f_bavail = buf->f_bfree;
1233 buf->f_files = usage.nr_inodes;
1234 buf->f_ffree = U64_MAX;
1236 fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1237 le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1238 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1239 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1240 buf->f_namelen = BCH_NAME_MAX;
1245 static int bch2_sync_fs(struct super_block *sb, int wait)
1247 struct bch_fs *c = sb->s_fs_info;
1249 if (c->opts.journal_flush_disabled)
1253 bch2_journal_flush_async(&c->journal, NULL);
1257 return bch2_journal_flush(&c->journal);
1260 static struct bch_fs *bch2_path_to_fs(const char *dev)
1263 struct block_device *bdev = lookup_bdev(dev);
1266 return ERR_CAST(bdev);
1268 c = bch2_bdev_to_fs(bdev);
1270 return c ?: ERR_PTR(-ENOENT);
1273 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1274 unsigned nr_devs, struct bch_opts opts)
1276 struct bch_fs *c, *c1, *c2;
1280 return ERR_PTR(-EINVAL);
1282 c = bch2_fs_open(devs, nr_devs, opts);
1284 if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1287 * Look up each block device, make sure they all belong to a
1288 * filesystem and they all belong to the _same_ filesystem
1291 c1 = bch2_path_to_fs(devs[0]);
1295 for (i = 1; i < nr_devs; i++) {
1296 c2 = bch2_path_to_fs(devs[i]);
1298 closure_put(&c2->cl);
1301 closure_put(&c1->cl);
1312 mutex_lock(&c->state_lock);
1314 if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1315 mutex_unlock(&c->state_lock);
1316 closure_put(&c->cl);
1317 pr_err("err mounting %s: incomplete filesystem", dev_name);
1318 return ERR_PTR(-EINVAL);
1321 mutex_unlock(&c->state_lock);
1323 set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1327 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1328 struct bch_opts opts)
1330 char *dev_name = NULL, **devs = NULL, *s;
1331 struct bch_fs *c = ERR_PTR(-ENOMEM);
1332 size_t i, nr_devs = 0;
1334 dev_name = kstrdup(_dev_name, GFP_KERNEL);
1338 for (s = dev_name; s; s = strchr(s + 1, ':'))
1341 devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1345 for (i = 0, s = dev_name;
1347 (s = strchr(s, ':')) && (*s++ = '\0'))
1350 c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1357 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1359 struct bch_fs *c = sb->s_fs_info;
1360 struct bch_opts opts = bch2_opts_empty();
1363 opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1365 ret = bch2_parse_mount_opts(&opts, data);
1369 if (opts.read_only != c->opts.read_only) {
1370 mutex_lock(&c->state_lock);
1372 if (opts.read_only) {
1373 bch2_fs_read_only(c);
1375 sb->s_flags |= SB_RDONLY;
1377 ret = bch2_fs_read_write(c);
1379 bch_err(c, "error going rw: %i", ret);
1380 mutex_unlock(&c->state_lock);
1384 sb->s_flags &= ~SB_RDONLY;
1387 c->opts.read_only = opts.read_only;
1389 mutex_unlock(&c->state_lock);
1392 if (opts.errors >= 0)
1393 c->opts.errors = opts.errors;
1398 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1400 struct bch_fs *c = root->d_sb->s_fs_info;
1404 for (i = 0; i < bch2_opts_nr; i++) {
1405 const struct bch_option *opt = &bch2_opt_table[i];
1406 u64 v = bch2_opt_get_by_id(&c->opts, i);
1408 if (!(opt->mode & OPT_MOUNT))
1411 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1414 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1415 OPT_SHOW_MOUNT_STYLE);
1424 static const struct super_operations bch_super_operations = {
1425 .alloc_inode = bch2_alloc_inode,
1426 .destroy_inode = bch2_destroy_inode,
1427 .write_inode = bch2_vfs_write_inode,
1428 .evict_inode = bch2_evict_inode,
1429 .sync_fs = bch2_sync_fs,
1430 .statfs = bch2_statfs,
1431 .show_options = bch2_show_options,
1432 .remount_fs = bch2_remount,
1434 .put_super = bch2_put_super,
1435 .freeze_fs = bch2_freeze,
1436 .unfreeze_fs = bch2_unfreeze,
1440 static int bch2_test_super(struct super_block *s, void *data)
1442 return s->s_fs_info == data;
1445 static int bch2_set_super(struct super_block *s, void *data)
1447 s->s_fs_info = data;
1451 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1452 int flags, const char *dev_name, void *data)
1456 struct super_block *sb;
1457 struct inode *vinode;
1458 struct bch_opts opts = bch2_opts_empty();
1462 opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1464 ret = bch2_parse_mount_opts(&opts, data);
1466 return ERR_PTR(ret);
1468 c = bch2_open_as_blockdevs(dev_name, opts);
1472 sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1474 closure_put(&c->cl);
1475 return ERR_CAST(sb);
1478 BUG_ON(sb->s_fs_info != c);
1481 closure_put(&c->cl);
1483 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1490 sb->s_blocksize = block_bytes(c);
1491 sb->s_blocksize_bits = ilog2(block_bytes(c));
1492 sb->s_maxbytes = MAX_LFS_FILESIZE;
1493 sb->s_op = &bch_super_operations;
1494 sb->s_export_op = &bch_export_ops;
1495 #ifdef CONFIG_BCACHEFS_QUOTA
1496 sb->s_qcop = &bch2_quotactl_operations;
1497 sb->s_quota_types = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1499 sb->s_xattr = bch2_xattr_handlers;
1500 sb->s_magic = BCACHEFS_STATFS_MAGIC;
1501 sb->s_time_gran = c->sb.time_precision;
1503 strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1505 ret = super_setup_bdi(sb);
1509 sb->s_bdi->congested_fn = bch2_congested;
1510 sb->s_bdi->congested_data = c;
1511 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1513 for_each_online_member(ca, c, i) {
1514 struct block_device *bdev = ca->disk_sb.bdev;
1516 /* XXX: create an anonymous device for multi device filesystems */
1518 sb->s_dev = bdev->bd_dev;
1519 percpu_ref_put(&ca->io_ref);
1523 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1525 sb->s_flags |= SB_POSIXACL;
1528 vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1529 if (IS_ERR(vinode)) {
1530 bch_err(c, "error mounting: error getting root inode %i",
1531 (int) PTR_ERR(vinode));
1532 ret = PTR_ERR(vinode);
1536 sb->s_root = d_make_root(vinode);
1538 bch_err(c, "error mounting: error allocating root dentry");
1543 sb->s_flags |= SB_ACTIVE;
1545 return dget(sb->s_root);
1548 deactivate_locked_super(sb);
1549 return ERR_PTR(ret);
1552 static void bch2_kill_sb(struct super_block *sb)
1554 struct bch_fs *c = sb->s_fs_info;
1556 generic_shutdown_super(sb);
1558 if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1561 closure_put(&c->cl);
1564 static struct file_system_type bcache_fs_type = {
1565 .owner = THIS_MODULE,
1567 .mount = bch2_mount,
1568 .kill_sb = bch2_kill_sb,
1569 .fs_flags = FS_REQUIRES_DEV,
1572 MODULE_ALIAS_FS("bcachefs");
1574 void bch2_vfs_exit(void)
1576 unregister_filesystem(&bcache_fs_type);
1577 if (bch2_inode_cache)
1578 kmem_cache_destroy(bch2_inode_cache);
1581 int __init bch2_vfs_init(void)
1585 bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1586 if (!bch2_inode_cache)
1589 ret = register_filesystem(&bcache_fs_type);
1599 #endif /* NO_BCACHEFS_FS */