1 // SPDX-License-Identifier: GPL-2.0
7 #include "btree_update.h"
14 #include "fs-common.h"
17 #include "fs-io-buffered.h"
18 #include "fs-io-direct.h"
19 #include "fs-io-pagecache.h"
30 #include <linux/aio.h>
31 #include <linux/backing-dev.h>
32 #include <linux/exportfs.h>
33 #include <linux/fiemap.h>
34 #include <linux/module.h>
35 #include <linux/pagemap.h>
36 #include <linux/posix_acl.h>
37 #include <linux/random.h>
38 #include <linux/seq_file.h>
39 #include <linux/statfs.h>
40 #include <linux/string.h>
41 #include <linux/xattr.h>
43 static struct kmem_cache *bch2_inode_cache;
45 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
46 struct bch_inode_info *,
47 struct bch_inode_unpacked *,
48 struct bch_subvolume *);
50 void bch2_inode_update_after_write(struct btree_trans *trans,
51 struct bch_inode_info *inode,
52 struct bch_inode_unpacked *bi,
55 struct bch_fs *c = trans->c;
57 BUG_ON(bi->bi_inum != inode->v.i_ino);
59 bch2_assert_pos_locked(trans, BTREE_ID_inodes,
61 c->opts.inodes_use_key_cache);
63 set_nlink(&inode->v, bch2_inode_nlink_get(bi));
64 i_uid_write(&inode->v, bi->bi_uid);
65 i_gid_write(&inode->v, bi->bi_gid);
66 inode->v.i_mode = bi->bi_mode;
68 if (fields & ATTR_ATIME)
69 inode_set_atime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_atime));
70 if (fields & ATTR_MTIME)
71 inode_set_mtime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_mtime));
72 if (fields & ATTR_CTIME)
73 inode_set_ctime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_ctime));
75 inode->ei_inode = *bi;
77 bch2_inode_flags_to_vfs(inode);
80 int __must_check bch2_write_inode(struct bch_fs *c,
81 struct bch_inode_info *inode,
83 void *p, unsigned fields)
85 struct btree_trans *trans = bch2_trans_get(c);
86 struct btree_iter iter = { NULL };
87 struct bch_inode_unpacked inode_u;
90 bch2_trans_begin(trans);
92 ret = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
94 (set ? set(trans, inode, &inode_u, p) : 0) ?:
95 bch2_inode_write(trans, &iter, &inode_u) ?:
96 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
99 * the btree node lock protects inode->ei_inode, not ei_update_lock;
100 * this is important for inode updates via bchfs_write_index_update
103 bch2_inode_update_after_write(trans, inode, &inode_u, fields);
105 bch2_trans_iter_exit(trans, &iter);
107 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
110 bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
111 "inode %u:%llu not found when updating",
112 inode_inum(inode).subvol,
113 inode_inum(inode).inum);
115 bch2_trans_put(trans);
116 return ret < 0 ? ret : 0;
119 int bch2_fs_quota_transfer(struct bch_fs *c,
120 struct bch_inode_info *inode,
121 struct bch_qid new_qid,
123 enum quota_acct_mode mode)
128 qtypes &= enabled_qtypes(c);
130 for (i = 0; i < QTYP_NR; i++)
131 if (new_qid.q[i] == inode->ei_qid.q[i])
132 qtypes &= ~(1U << i);
137 mutex_lock(&inode->ei_quota_lock);
139 ret = bch2_quota_transfer(c, qtypes, new_qid,
142 inode->ei_quota_reserved,
145 for (i = 0; i < QTYP_NR; i++)
146 if (qtypes & (1 << i))
147 inode->ei_qid.q[i] = new_qid.q[i];
149 mutex_unlock(&inode->ei_quota_lock);
154 static int bch2_iget5_test(struct inode *vinode, void *p)
156 struct bch_inode_info *inode = to_bch_ei(vinode);
157 subvol_inum *inum = p;
159 return inode->ei_subvol == inum->subvol &&
160 inode->ei_inode.bi_inum == inum->inum;
163 static int bch2_iget5_set(struct inode *vinode, void *p)
165 struct bch_inode_info *inode = to_bch_ei(vinode);
166 subvol_inum *inum = p;
168 inode->v.i_ino = inum->inum;
169 inode->ei_subvol = inum->subvol;
170 inode->ei_inode.bi_inum = inum->inum;
174 static unsigned bch2_inode_hash(subvol_inum inum)
176 return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
179 static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_inode_info *inode)
181 subvol_inum inum = inode_inum(inode);
182 struct bch_inode_info *old = to_bch_ei(inode_insert5(&inode->v,
183 bch2_inode_hash(inum),
189 if (unlikely(old != inode)) {
190 discard_new_inode(&inode->v);
193 mutex_lock(&c->vfs_inodes_lock);
194 list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
195 mutex_unlock(&c->vfs_inodes_lock);
197 * we really don't want insert_inode_locked2() to be setting
200 unlock_new_inode(&inode->v);
206 #define memalloc_flags_do(_flags, _do) \
208 unsigned _saved_flags = memalloc_flags_save(_flags); \
209 typeof(_do) _ret = _do; \
210 memalloc_noreclaim_restore(_saved_flags); \
215 * Allocate a new inode, dropping/retaking btree locks if necessary:
217 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
219 struct bch_fs *c = trans->c;
221 struct bch_inode_info *inode =
222 memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
223 to_bch_ei(new_inode(c->vfs_sb)));
225 if (unlikely(!inode)) {
226 int ret = drop_locks_do(trans, (inode = to_bch_ei(new_inode(c->vfs_sb))) ? 0 : -ENOMEM);
228 discard_new_inode(&inode->v);
236 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
238 struct bch_inode_info *inode =
239 to_bch_ei(ilookup5_nowait(c->vfs_sb,
240 bch2_inode_hash(inum),
246 struct btree_trans *trans = bch2_trans_get(c);
248 struct bch_inode_unpacked inode_u;
249 struct bch_subvolume subvol;
250 int ret = lockrestart_do(trans,
251 bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
252 bch2_inode_find_by_inum_trans(trans, inum, &inode_u)) ?:
253 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
255 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
256 inode = bch2_inode_insert(c, inode);
258 bch2_trans_put(trans);
260 return ret ? ERR_PTR(ret) : &inode->v;
263 struct bch_inode_info *
264 __bch2_create(struct mnt_idmap *idmap,
265 struct bch_inode_info *dir, struct dentry *dentry,
266 umode_t mode, dev_t rdev, subvol_inum snapshot_src,
269 struct bch_fs *c = dir->v.i_sb->s_fs_info;
270 struct btree_trans *trans;
271 struct bch_inode_unpacked dir_u;
272 struct bch_inode_info *inode;
273 struct bch_inode_unpacked inode_u;
274 struct posix_acl *default_acl = NULL, *acl = NULL;
276 struct bch_subvolume subvol;
281 * preallocate acls + vfs inode before btree transaction, so that
282 * nothing can fail after the transaction succeeds:
284 #ifdef CONFIG_BCACHEFS_POSIX_ACL
285 ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
289 inode = to_bch_ei(new_inode(c->vfs_sb));
290 if (unlikely(!inode)) {
291 inode = ERR_PTR(-ENOMEM);
295 bch2_inode_init_early(c, &inode_u);
297 if (!(flags & BCH_CREATE_TMPFILE))
298 mutex_lock(&dir->ei_update_lock);
300 trans = bch2_trans_get(c);
302 bch2_trans_begin(trans);
304 ret = bch2_subvol_is_ro_trans(trans, dir->ei_subvol) ?:
305 bch2_create_trans(trans,
306 inode_inum(dir), &dir_u, &inode_u,
307 !(flags & BCH_CREATE_TMPFILE)
308 ? &dentry->d_name : NULL,
309 from_kuid(i_user_ns(&dir->v), current_fsuid()),
310 from_kgid(i_user_ns(&dir->v), current_fsgid()),
312 default_acl, acl, snapshot_src, flags) ?:
313 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
314 KEY_TYPE_QUOTA_PREALLOC);
316 goto err_before_quota;
318 inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
319 inum.inum = inode_u.bi_inum;
321 ret = bch2_subvolume_get(trans, inum.subvol, true,
322 BTREE_ITER_WITH_UPDATES, &subvol) ?:
323 bch2_trans_commit(trans, NULL, &journal_seq, 0);
325 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
326 KEY_TYPE_QUOTA_WARN);
328 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
333 if (!(flags & BCH_CREATE_TMPFILE)) {
334 bch2_inode_update_after_write(trans, dir, &dir_u,
335 ATTR_MTIME|ATTR_CTIME);
336 mutex_unlock(&dir->ei_update_lock);
339 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
341 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
342 set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
345 * we must insert the new inode into the inode cache before calling
346 * bch2_trans_exit() and dropping locks, else we could race with another
347 * thread pulling the inode in and modifying it:
349 inode = bch2_inode_insert(c, inode);
350 bch2_trans_put(trans);
352 posix_acl_release(default_acl);
353 posix_acl_release(acl);
356 if (!(flags & BCH_CREATE_TMPFILE))
357 mutex_unlock(&dir->ei_update_lock);
359 bch2_trans_put(trans);
360 make_bad_inode(&inode->v);
362 inode = ERR_PTR(ret);
368 static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
369 subvol_inum dir, struct bch_hash_info *dir_hash_info,
370 const struct qstr *name)
372 struct bch_fs *c = trans->c;
373 struct btree_iter dirent_iter = {};
374 subvol_inum inum = {};
376 int ret = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
377 dir_hash_info, dir, name, 0);
381 struct bkey_s_c k = bch2_btree_iter_peek_slot(&dirent_iter);
386 ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
392 struct bch_inode_info *inode =
393 to_bch_ei(ilookup5_nowait(c->vfs_sb,
394 bch2_inode_hash(inum),
400 struct bch_subvolume subvol;
401 struct bch_inode_unpacked inode_u;
402 ret = bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
403 bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
404 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
405 if (bch2_err_matches(ret, ENOENT)) {
406 struct printbuf buf = PRINTBUF;
408 bch2_bkey_val_to_text(&buf, c, k);
409 bch_err(c, "%s points to missing inode", buf.buf);
415 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
416 inode = bch2_inode_insert(c, inode);
418 bch2_trans_iter_exit(trans, &dirent_iter);
421 inode = ERR_PTR(ret);
425 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
428 struct bch_fs *c = vdir->i_sb->s_fs_info;
429 struct bch_inode_info *dir = to_bch_ei(vdir);
430 struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
432 struct bch_inode_info *inode;
433 bch2_trans_do(c, NULL, NULL, 0,
434 PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
435 &hash, &dentry->d_name)));
439 return d_splice_alias(&inode->v, dentry);
442 static int bch2_mknod(struct mnt_idmap *idmap,
443 struct inode *vdir, struct dentry *dentry,
444 umode_t mode, dev_t rdev)
446 struct bch_inode_info *inode =
447 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
448 (subvol_inum) { 0 }, 0);
451 return bch2_err_class(PTR_ERR(inode));
453 d_instantiate(dentry, &inode->v);
457 static int bch2_create(struct mnt_idmap *idmap,
458 struct inode *vdir, struct dentry *dentry,
459 umode_t mode, bool excl)
461 return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
464 static int __bch2_link(struct bch_fs *c,
465 struct bch_inode_info *inode,
466 struct bch_inode_info *dir,
467 struct dentry *dentry)
469 struct btree_trans *trans = bch2_trans_get(c);
470 struct bch_inode_unpacked dir_u, inode_u;
473 mutex_lock(&inode->ei_update_lock);
475 ret = commit_do(trans, NULL, NULL, 0,
476 bch2_link_trans(trans,
477 inode_inum(dir), &dir_u,
478 inode_inum(inode), &inode_u,
482 bch2_inode_update_after_write(trans, dir, &dir_u,
483 ATTR_MTIME|ATTR_CTIME);
484 bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
487 bch2_trans_put(trans);
488 mutex_unlock(&inode->ei_update_lock);
492 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
493 struct dentry *dentry)
495 struct bch_fs *c = vdir->i_sb->s_fs_info;
496 struct bch_inode_info *dir = to_bch_ei(vdir);
497 struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
500 lockdep_assert_held(&inode->v.i_rwsem);
502 ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
503 bch2_subvol_is_ro(c, inode->ei_subvol) ?:
504 __bch2_link(c, inode, dir, dentry);
509 d_instantiate(dentry, &inode->v);
513 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
514 bool deleting_snapshot)
516 struct bch_fs *c = vdir->i_sb->s_fs_info;
517 struct bch_inode_info *dir = to_bch_ei(vdir);
518 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
519 struct bch_inode_unpacked dir_u, inode_u;
520 struct btree_trans *trans = bch2_trans_get(c);
523 bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
525 ret = commit_do(trans, NULL, NULL,
526 BCH_TRANS_COMMIT_no_enospc,
527 bch2_unlink_trans(trans,
528 inode_inum(dir), &dir_u,
529 &inode_u, &dentry->d_name,
534 bch2_inode_update_after_write(trans, dir, &dir_u,
535 ATTR_MTIME|ATTR_CTIME);
536 bch2_inode_update_after_write(trans, inode, &inode_u,
539 if (inode_u.bi_subvol) {
541 * Subvolume deletion is asynchronous, but we still want to tell
542 * the VFS that it's been deleted here:
544 set_nlink(&inode->v, 0);
547 bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
548 bch2_trans_put(trans);
553 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
555 struct bch_inode_info *dir= to_bch_ei(vdir);
556 struct bch_fs *c = dir->v.i_sb->s_fs_info;
558 return bch2_subvol_is_ro(c, dir->ei_subvol) ?:
559 __bch2_unlink(vdir, dentry, false);
562 static int bch2_symlink(struct mnt_idmap *idmap,
563 struct inode *vdir, struct dentry *dentry,
566 struct bch_fs *c = vdir->i_sb->s_fs_info;
567 struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
570 inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
571 (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
573 return bch2_err_class(PTR_ERR(inode));
575 inode_lock(&inode->v);
576 ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
577 inode_unlock(&inode->v);
582 ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
586 ret = __bch2_link(c, inode, dir, dentry);
590 d_instantiate(dentry, &inode->v);
597 static int bch2_mkdir(struct mnt_idmap *idmap,
598 struct inode *vdir, struct dentry *dentry, umode_t mode)
600 return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
603 static int bch2_rename2(struct mnt_idmap *idmap,
604 struct inode *src_vdir, struct dentry *src_dentry,
605 struct inode *dst_vdir, struct dentry *dst_dentry,
608 struct bch_fs *c = src_vdir->i_sb->s_fs_info;
609 struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
610 struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
611 struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
612 struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
613 struct bch_inode_unpacked dst_dir_u, src_dir_u;
614 struct bch_inode_unpacked src_inode_u, dst_inode_u;
615 struct btree_trans *trans;
616 enum bch_rename_mode mode = flags & RENAME_EXCHANGE
617 ? BCH_RENAME_EXCHANGE
618 : dst_dentry->d_inode
619 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
622 if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
625 if (mode == BCH_RENAME_OVERWRITE) {
626 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
632 trans = bch2_trans_get(c);
634 bch2_lock_inodes(INODE_UPDATE_LOCK,
640 ret = bch2_subvol_is_ro_trans(trans, src_dir->ei_subvol) ?:
641 bch2_subvol_is_ro_trans(trans, dst_dir->ei_subvol);
645 if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
646 ret = bch2_fs_quota_transfer(c, src_inode,
649 KEY_TYPE_QUOTA_PREALLOC);
654 if (mode == BCH_RENAME_EXCHANGE &&
655 inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
656 ret = bch2_fs_quota_transfer(c, dst_inode,
659 KEY_TYPE_QUOTA_PREALLOC);
664 ret = commit_do(trans, NULL, NULL, 0,
665 bch2_rename_trans(trans,
666 inode_inum(src_dir), &src_dir_u,
667 inode_inum(dst_dir), &dst_dir_u,
676 BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
678 dst_inode->v.i_ino != dst_inode_u.bi_inum);
680 bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
681 ATTR_MTIME|ATTR_CTIME);
683 if (src_dir != dst_dir)
684 bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
685 ATTR_MTIME|ATTR_CTIME);
687 bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
691 bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
694 bch2_trans_put(trans);
696 bch2_fs_quota_transfer(c, src_inode,
697 bch_qid(&src_inode->ei_inode),
699 KEY_TYPE_QUOTA_NOCHECK);
701 bch2_fs_quota_transfer(c, dst_inode,
702 bch_qid(&dst_inode->ei_inode),
704 KEY_TYPE_QUOTA_NOCHECK);
706 bch2_unlock_inodes(INODE_UPDATE_LOCK,
715 static void bch2_setattr_copy(struct mnt_idmap *idmap,
716 struct bch_inode_info *inode,
717 struct bch_inode_unpacked *bi,
720 struct bch_fs *c = inode->v.i_sb->s_fs_info;
721 unsigned int ia_valid = attr->ia_valid;
723 if (ia_valid & ATTR_UID)
724 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
725 if (ia_valid & ATTR_GID)
726 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
728 if (ia_valid & ATTR_SIZE)
729 bi->bi_size = attr->ia_size;
731 if (ia_valid & ATTR_ATIME)
732 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
733 if (ia_valid & ATTR_MTIME)
734 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
735 if (ia_valid & ATTR_CTIME)
736 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
738 if (ia_valid & ATTR_MODE) {
739 umode_t mode = attr->ia_mode;
740 kgid_t gid = ia_valid & ATTR_GID
744 if (!in_group_p(gid) &&
745 !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
751 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
752 struct bch_inode_info *inode,
755 struct bch_fs *c = inode->v.i_sb->s_fs_info;
757 struct btree_trans *trans;
758 struct btree_iter inode_iter = { NULL };
759 struct bch_inode_unpacked inode_u;
760 struct posix_acl *acl = NULL;
763 mutex_lock(&inode->ei_update_lock);
767 if (attr->ia_valid & ATTR_UID)
768 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
770 if (attr->ia_valid & ATTR_GID)
771 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
773 ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
774 KEY_TYPE_QUOTA_PREALLOC);
778 trans = bch2_trans_get(c);
780 bch2_trans_begin(trans);
784 ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
789 bch2_setattr_copy(idmap, inode, &inode_u, attr);
791 if (attr->ia_valid & ATTR_MODE) {
792 ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
793 inode_u.bi_mode, &acl);
798 ret = bch2_inode_write(trans, &inode_iter, &inode_u) ?:
799 bch2_trans_commit(trans, NULL, NULL,
800 BCH_TRANS_COMMIT_no_enospc);
802 bch2_trans_iter_exit(trans, &inode_iter);
804 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
809 bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
812 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
814 bch2_trans_put(trans);
816 mutex_unlock(&inode->ei_update_lock);
818 return bch2_err_class(ret);
821 static int bch2_getattr(struct mnt_idmap *idmap,
822 const struct path *path, struct kstat *stat,
823 u32 request_mask, unsigned query_flags)
825 struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
826 struct bch_fs *c = inode->v.i_sb->s_fs_info;
828 stat->dev = inode->v.i_sb->s_dev;
829 stat->ino = inode->v.i_ino;
830 stat->mode = inode->v.i_mode;
831 stat->nlink = inode->v.i_nlink;
832 stat->uid = inode->v.i_uid;
833 stat->gid = inode->v.i_gid;
834 stat->rdev = inode->v.i_rdev;
835 stat->size = i_size_read(&inode->v);
836 stat->atime = inode_get_atime(&inode->v);
837 stat->mtime = inode_get_mtime(&inode->v);
838 stat->ctime = inode_get_ctime(&inode->v);
839 stat->blksize = block_bytes(c);
840 stat->blocks = inode->v.i_blocks;
842 if (request_mask & STATX_BTIME) {
843 stat->result_mask |= STATX_BTIME;
844 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
847 if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
848 stat->attributes |= STATX_ATTR_IMMUTABLE;
849 stat->attributes_mask |= STATX_ATTR_IMMUTABLE;
851 if (inode->ei_inode.bi_flags & BCH_INODE_append)
852 stat->attributes |= STATX_ATTR_APPEND;
853 stat->attributes_mask |= STATX_ATTR_APPEND;
855 if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
856 stat->attributes |= STATX_ATTR_NODUMP;
857 stat->attributes_mask |= STATX_ATTR_NODUMP;
862 static int bch2_setattr(struct mnt_idmap *idmap,
863 struct dentry *dentry, struct iattr *iattr)
865 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
866 struct bch_fs *c = inode->v.i_sb->s_fs_info;
869 lockdep_assert_held(&inode->v.i_rwsem);
871 ret = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
872 setattr_prepare(idmap, dentry, iattr);
876 return iattr->ia_valid & ATTR_SIZE
877 ? bchfs_truncate(idmap, inode, iattr)
878 : bch2_setattr_nonsize(idmap, inode, iattr);
881 static int bch2_tmpfile(struct mnt_idmap *idmap,
882 struct inode *vdir, struct file *file, umode_t mode)
884 struct bch_inode_info *inode =
885 __bch2_create(idmap, to_bch_ei(vdir),
886 file->f_path.dentry, mode, 0,
887 (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
890 return bch2_err_class(PTR_ERR(inode));
892 d_mark_tmpfile(file, &inode->v);
893 d_instantiate(file->f_path.dentry, &inode->v);
894 return finish_open_simple(file, 0);
897 static int bch2_fill_extent(struct bch_fs *c,
898 struct fiemap_extent_info *info,
899 struct bkey_s_c k, unsigned flags)
901 if (bkey_extent_is_direct_data(k.k)) {
902 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
903 const union bch_extent_entry *entry;
904 struct extent_ptr_decoded p;
907 if (k.k->type == KEY_TYPE_reflink_v)
908 flags |= FIEMAP_EXTENT_SHARED;
910 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
912 u64 offset = p.ptr.offset;
915 flags2 |= FIEMAP_EXTENT_UNWRITTEN;
917 if (p.crc.compression_type)
918 flags2 |= FIEMAP_EXTENT_ENCODED;
920 offset += p.crc.offset;
922 if ((offset & (block_sectors(c) - 1)) ||
923 (k.k->size & (block_sectors(c) - 1)))
924 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
926 ret = fiemap_fill_next_extent(info,
927 bkey_start_offset(k.k) << 9,
929 k.k->size << 9, flags|flags2);
935 } else if (bkey_extent_is_inline_data(k.k)) {
936 return fiemap_fill_next_extent(info,
937 bkey_start_offset(k.k) << 9,
940 FIEMAP_EXTENT_DATA_INLINE);
941 } else if (k.k->type == KEY_TYPE_reservation) {
942 return fiemap_fill_next_extent(info,
943 bkey_start_offset(k.k) << 9,
946 FIEMAP_EXTENT_DELALLOC|
947 FIEMAP_EXTENT_UNWRITTEN);
953 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
956 struct bch_fs *c = vinode->i_sb->s_fs_info;
957 struct bch_inode_info *ei = to_bch_ei(vinode);
958 struct btree_trans *trans;
959 struct btree_iter iter;
961 struct bkey_buf cur, prev;
962 struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
963 unsigned offset_into_extent, sectors;
964 bool have_extent = false;
968 ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
972 if (start + len < start)
977 bch2_bkey_buf_init(&cur);
978 bch2_bkey_buf_init(&prev);
979 trans = bch2_trans_get(c);
981 bch2_trans_begin(trans);
983 ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
987 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
988 SPOS(ei->v.i_ino, start, snapshot), 0);
990 while (!(ret = btree_trans_too_many_iters(trans)) &&
991 (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
992 !(ret = bkey_err(k))) {
993 enum btree_id data_btree = BTREE_ID_extents;
995 if (!bkey_extent_is_data(k.k) &&
996 k.k->type != KEY_TYPE_reservation) {
997 bch2_btree_iter_advance(&iter);
1001 offset_into_extent = iter.pos.offset -
1002 bkey_start_offset(k.k);
1003 sectors = k.k->size - offset_into_extent;
1005 bch2_bkey_buf_reassemble(&cur, c, k);
1007 ret = bch2_read_indirect_extent(trans, &data_btree,
1008 &offset_into_extent, &cur);
1012 k = bkey_i_to_s_c(cur.k);
1013 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
1015 sectors = min(sectors, k.k->size - offset_into_extent);
1017 bch2_cut_front(POS(k.k->p.inode,
1018 bkey_start_offset(k.k) +
1019 offset_into_extent),
1021 bch2_key_resize(&cur.k->k, sectors);
1022 cur.k->k.p = iter.pos;
1023 cur.k->k.p.offset += cur.k->k.size;
1026 bch2_trans_unlock(trans);
1027 ret = bch2_fill_extent(c, info,
1028 bkey_i_to_s_c(prev.k), 0);
1033 bkey_copy(prev.k, cur.k);
1036 bch2_btree_iter_set_pos(&iter,
1037 POS(iter.pos.inode, iter.pos.offset + sectors));
1039 start = iter.pos.offset;
1040 bch2_trans_iter_exit(trans, &iter);
1042 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1045 if (!ret && have_extent) {
1046 bch2_trans_unlock(trans);
1047 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
1048 FIEMAP_EXTENT_LAST);
1051 bch2_trans_put(trans);
1052 bch2_bkey_buf_exit(&cur, c);
1053 bch2_bkey_buf_exit(&prev, c);
1054 return ret < 0 ? ret : 0;
1057 static const struct vm_operations_struct bch_vm_ops = {
1058 .fault = bch2_page_fault,
1059 .map_pages = filemap_map_pages,
1060 .page_mkwrite = bch2_page_mkwrite,
1063 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1065 file_accessed(file);
1067 vma->vm_ops = &bch_vm_ops;
1073 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1075 return generic_file_llseek_size(file, offset, whence,
1079 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1081 struct bch_inode_info *inode = file_bch_inode(file);
1082 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1084 if (!dir_emit_dots(file, ctx))
1087 int ret = bch2_readdir(c, inode_inum(inode), ctx);
1090 return bch2_err_class(ret);
1093 static int bch2_open(struct inode *vinode, struct file *file)
1095 if (file->f_flags & (O_WRONLY|O_RDWR)) {
1096 struct bch_inode_info *inode = to_bch_ei(vinode);
1097 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1099 int ret = bch2_subvol_is_ro(c, inode->ei_subvol);
1104 return generic_file_open(vinode, file);
1107 static const struct file_operations bch_file_operations = {
1109 .llseek = bch2_llseek,
1110 .read_iter = bch2_read_iter,
1111 .write_iter = bch2_write_iter,
1113 .fsync = bch2_fsync,
1114 .splice_read = filemap_splice_read,
1115 .splice_write = iter_file_splice_write,
1116 .fallocate = bch2_fallocate_dispatch,
1117 .unlocked_ioctl = bch2_fs_file_ioctl,
1118 #ifdef CONFIG_COMPAT
1119 .compat_ioctl = bch2_compat_fs_ioctl,
1121 .remap_file_range = bch2_remap_file_range,
1124 static const struct inode_operations bch_file_inode_operations = {
1125 .getattr = bch2_getattr,
1126 .setattr = bch2_setattr,
1127 .fiemap = bch2_fiemap,
1128 .listxattr = bch2_xattr_list,
1129 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1130 .get_acl = bch2_get_acl,
1131 .set_acl = bch2_set_acl,
1135 static const struct inode_operations bch_dir_inode_operations = {
1136 .lookup = bch2_lookup,
1137 .create = bch2_create,
1139 .unlink = bch2_unlink,
1140 .symlink = bch2_symlink,
1141 .mkdir = bch2_mkdir,
1142 .rmdir = bch2_unlink,
1143 .mknod = bch2_mknod,
1144 .rename = bch2_rename2,
1145 .getattr = bch2_getattr,
1146 .setattr = bch2_setattr,
1147 .tmpfile = bch2_tmpfile,
1148 .listxattr = bch2_xattr_list,
1149 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1150 .get_acl = bch2_get_acl,
1151 .set_acl = bch2_set_acl,
1155 static const struct file_operations bch_dir_file_operations = {
1156 .llseek = bch2_dir_llseek,
1157 .read = generic_read_dir,
1158 .iterate_shared = bch2_vfs_readdir,
1159 .fsync = bch2_fsync,
1160 .unlocked_ioctl = bch2_fs_file_ioctl,
1161 #ifdef CONFIG_COMPAT
1162 .compat_ioctl = bch2_compat_fs_ioctl,
1166 static const struct inode_operations bch_symlink_inode_operations = {
1167 .get_link = page_get_link,
1168 .getattr = bch2_getattr,
1169 .setattr = bch2_setattr,
1170 .listxattr = bch2_xattr_list,
1171 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1172 .get_acl = bch2_get_acl,
1173 .set_acl = bch2_set_acl,
1177 static const struct inode_operations bch_special_inode_operations = {
1178 .getattr = bch2_getattr,
1179 .setattr = bch2_setattr,
1180 .listxattr = bch2_xattr_list,
1181 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1182 .get_acl = bch2_get_acl,
1183 .set_acl = bch2_set_acl,
1187 static const struct address_space_operations bch_address_space_operations = {
1188 .read_folio = bch2_read_folio,
1189 .writepages = bch2_writepages,
1190 .readahead = bch2_readahead,
1191 .dirty_folio = filemap_dirty_folio,
1192 .write_begin = bch2_write_begin,
1193 .write_end = bch2_write_end,
1194 .invalidate_folio = bch2_invalidate_folio,
1195 .release_folio = bch2_release_folio,
1196 .direct_IO = noop_direct_IO,
1197 #ifdef CONFIG_MIGRATION
1198 .migrate_folio = filemap_migrate_folio,
1200 .error_remove_folio = generic_error_remove_folio,
1203 struct bcachefs_fid {
1209 struct bcachefs_fid_with_parent {
1210 struct bcachefs_fid fid;
1211 struct bcachefs_fid dir;
1214 static int bcachefs_fid_valid(int fh_len, int fh_type)
1217 case FILEID_BCACHEFS_WITHOUT_PARENT:
1218 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1219 case FILEID_BCACHEFS_WITH_PARENT:
1220 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1226 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1228 return (struct bcachefs_fid) {
1229 .inum = inode->ei_inode.bi_inum,
1230 .subvol = inode->ei_subvol,
1231 .gen = inode->ei_inode.bi_generation,
1235 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1238 struct bch_inode_info *inode = to_bch_ei(vinode);
1239 struct bch_inode_info *dir = to_bch_ei(vdir);
1242 if (!S_ISDIR(inode->v.i_mode) && dir) {
1243 struct bcachefs_fid_with_parent *fid = (void *) fh;
1245 min_len = sizeof(*fid) / sizeof(u32);
1246 if (*len < min_len) {
1248 return FILEID_INVALID;
1251 fid->fid = bch2_inode_to_fid(inode);
1252 fid->dir = bch2_inode_to_fid(dir);
1255 return FILEID_BCACHEFS_WITH_PARENT;
1257 struct bcachefs_fid *fid = (void *) fh;
1259 min_len = sizeof(*fid) / sizeof(u32);
1260 if (*len < min_len) {
1262 return FILEID_INVALID;
1264 *fid = bch2_inode_to_fid(inode);
1267 return FILEID_BCACHEFS_WITHOUT_PARENT;
1271 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1272 struct bcachefs_fid fid)
1274 struct bch_fs *c = sb->s_fs_info;
1275 struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1276 .subvol = fid.subvol,
1279 if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1281 vinode = ERR_PTR(-ESTALE);
1286 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1287 int fh_len, int fh_type)
1289 struct bcachefs_fid *fid = (void *) _fid;
1291 if (!bcachefs_fid_valid(fh_len, fh_type))
1294 return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1297 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1298 int fh_len, int fh_type)
1300 struct bcachefs_fid_with_parent *fid = (void *) _fid;
1302 if (!bcachefs_fid_valid(fh_len, fh_type) ||
1303 fh_type != FILEID_BCACHEFS_WITH_PARENT)
1306 return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1309 static struct dentry *bch2_get_parent(struct dentry *child)
1311 struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1312 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1313 subvol_inum parent_inum = {
1314 .subvol = inode->ei_inode.bi_parent_subvol ?:
1316 .inum = inode->ei_inode.bi_dir,
1319 return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1322 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1324 struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1325 struct bch_inode_info *dir = to_bch_ei(parent->d_inode);
1326 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1327 struct btree_trans *trans;
1328 struct btree_iter iter1;
1329 struct btree_iter iter2;
1331 struct bkey_s_c_dirent d;
1332 struct bch_inode_unpacked inode_u;
1335 struct qstr dirent_name;
1336 unsigned name_len = 0;
1339 if (!S_ISDIR(dir->v.i_mode))
1342 trans = bch2_trans_get(c);
1344 bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1345 POS(dir->ei_inode.bi_inum, 0), 0);
1346 bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1347 POS(dir->ei_inode.bi_inum, 0), 0);
1349 bch2_trans_begin(trans);
1351 ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1355 bch2_btree_iter_set_snapshot(&iter1, snapshot);
1356 bch2_btree_iter_set_snapshot(&iter2, snapshot);
1358 ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1362 if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1363 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1365 k = bch2_btree_iter_peek_slot(&iter1);
1370 if (k.k->type != KEY_TYPE_dirent) {
1371 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1375 d = bkey_s_c_to_dirent(k);
1376 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1378 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1382 if (target.subvol == inode->ei_subvol &&
1383 target.inum == inode->ei_inode.bi_inum)
1387 * File with multiple hardlinks and our backref is to the wrong
1388 * directory - linear search:
1390 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1391 if (k.k->p.inode > dir->ei_inode.bi_inum)
1394 if (k.k->type != KEY_TYPE_dirent)
1397 d = bkey_s_c_to_dirent(k);
1398 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1404 if (target.subvol == inode->ei_subvol &&
1405 target.inum == inode->ei_inode.bi_inum)
1413 dirent_name = bch2_dirent_get_name(d);
1415 name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1416 memcpy(name, dirent_name.name, name_len);
1417 name[name_len] = '\0';
1419 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1422 bch2_trans_iter_exit(trans, &iter1);
1423 bch2_trans_iter_exit(trans, &iter2);
1424 bch2_trans_put(trans);
1429 static const struct export_operations bch_export_ops = {
1430 .encode_fh = bch2_encode_fh,
1431 .fh_to_dentry = bch2_fh_to_dentry,
1432 .fh_to_parent = bch2_fh_to_parent,
1433 .get_parent = bch2_get_parent,
1434 .get_name = bch2_get_name,
1437 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1438 struct bch_inode_info *inode,
1439 struct bch_inode_unpacked *bi,
1440 struct bch_subvolume *subvol)
1442 bch2_iget5_set(&inode->v, &inum);
1443 bch2_inode_update_after_write(trans, inode, bi, ~0);
1445 if (BCH_SUBVOLUME_SNAP(subvol))
1446 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1448 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1450 inode->v.i_blocks = bi->bi_sectors;
1451 inode->v.i_ino = bi->bi_inum;
1452 inode->v.i_rdev = bi->bi_dev;
1453 inode->v.i_generation = bi->bi_generation;
1454 inode->v.i_size = bi->bi_size;
1456 inode->ei_flags = 0;
1457 inode->ei_quota_reserved = 0;
1458 inode->ei_qid = bch_qid(bi);
1459 inode->ei_subvol = inum.subvol;
1461 inode->v.i_mapping->a_ops = &bch_address_space_operations;
1463 switch (inode->v.i_mode & S_IFMT) {
1465 inode->v.i_op = &bch_file_inode_operations;
1466 inode->v.i_fop = &bch_file_operations;
1469 inode->v.i_op = &bch_dir_inode_operations;
1470 inode->v.i_fop = &bch_dir_file_operations;
1473 inode_nohighmem(&inode->v);
1474 inode->v.i_op = &bch_symlink_inode_operations;
1477 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1478 inode->v.i_op = &bch_special_inode_operations;
1482 mapping_set_large_folios(inode->v.i_mapping);
1485 static struct inode *bch2_alloc_inode(struct super_block *sb)
1487 struct bch_inode_info *inode;
1489 inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1493 inode_init_once(&inode->v);
1494 mutex_init(&inode->ei_update_lock);
1495 two_state_lock_init(&inode->ei_pagecache_lock);
1496 INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
1497 mutex_init(&inode->ei_quota_lock);
1502 static void bch2_i_callback(struct rcu_head *head)
1504 struct inode *vinode = container_of(head, struct inode, i_rcu);
1505 struct bch_inode_info *inode = to_bch_ei(vinode);
1507 kmem_cache_free(bch2_inode_cache, inode);
1510 static void bch2_destroy_inode(struct inode *vinode)
1512 call_rcu(&vinode->i_rcu, bch2_i_callback);
1515 static int inode_update_times_fn(struct btree_trans *trans,
1516 struct bch_inode_info *inode,
1517 struct bch_inode_unpacked *bi,
1520 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1522 bi->bi_atime = timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1523 bi->bi_mtime = timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1524 bi->bi_ctime = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1529 static int bch2_vfs_write_inode(struct inode *vinode,
1530 struct writeback_control *wbc)
1532 struct bch_fs *c = vinode->i_sb->s_fs_info;
1533 struct bch_inode_info *inode = to_bch_ei(vinode);
1536 mutex_lock(&inode->ei_update_lock);
1537 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1538 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1539 mutex_unlock(&inode->ei_update_lock);
1541 return bch2_err_class(ret);
1544 static void bch2_evict_inode(struct inode *vinode)
1546 struct bch_fs *c = vinode->i_sb->s_fs_info;
1547 struct bch_inode_info *inode = to_bch_ei(vinode);
1549 truncate_inode_pages_final(&inode->v.i_data);
1551 clear_inode(&inode->v);
1553 BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1555 if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1556 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1557 KEY_TYPE_QUOTA_WARN);
1558 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1559 KEY_TYPE_QUOTA_WARN);
1560 bch2_inode_rm(c, inode_inum(inode));
1563 mutex_lock(&c->vfs_inodes_lock);
1564 list_del_init(&inode->ei_vfs_inode_list);
1565 mutex_unlock(&c->vfs_inodes_lock);
1568 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1570 struct bch_inode_info *inode;
1571 DARRAY(struct bch_inode_info *) grabbed;
1572 bool clean_pass = false, this_pass_clean;
1575 * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1576 * be pruned with d_mark_dontcache().
1578 * Once we've had a clean pass where we didn't find any inodes without
1579 * I_DONTCACHE, we wait for them to be freed:
1582 darray_init(&grabbed);
1583 darray_make_room(&grabbed, 1024);
1586 this_pass_clean = true;
1588 mutex_lock(&c->vfs_inodes_lock);
1589 list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1590 if (!snapshot_list_has_id(s, inode->ei_subvol))
1593 if (!(inode->v.i_state & I_DONTCACHE) &&
1594 !(inode->v.i_state & I_FREEING) &&
1596 this_pass_clean = false;
1598 if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1602 } else if (clean_pass && this_pass_clean) {
1603 wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1604 DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1606 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1607 mutex_unlock(&c->vfs_inodes_lock);
1610 finish_wait(wq, &wait.wq_entry);
1614 mutex_unlock(&c->vfs_inodes_lock);
1616 darray_for_each(grabbed, i) {
1618 d_mark_dontcache(&inode->v);
1619 d_prune_aliases(&inode->v);
1624 if (!clean_pass || !this_pass_clean) {
1625 clean_pass = this_pass_clean;
1629 darray_exit(&grabbed);
1632 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1634 struct super_block *sb = dentry->d_sb;
1635 struct bch_fs *c = sb->s_fs_info;
1636 struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1637 unsigned shift = sb->s_blocksize_bits - 9;
1639 * this assumes inodes take up 64 bytes, which is a decent average
1642 u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1644 buf->f_type = BCACHEFS_STATFS_MAGIC;
1645 buf->f_bsize = sb->s_blocksize;
1646 buf->f_blocks = usage.capacity >> shift;
1647 buf->f_bfree = usage.free >> shift;
1648 buf->f_bavail = avail_factor(usage.free) >> shift;
1650 buf->f_files = usage.nr_inodes + avail_inodes;
1651 buf->f_ffree = avail_inodes;
1653 buf->f_fsid = uuid_to_fsid(c->sb.user_uuid.b);
1654 buf->f_namelen = BCH_NAME_MAX;
1659 static int bch2_sync_fs(struct super_block *sb, int wait)
1661 struct bch_fs *c = sb->s_fs_info;
1664 if (c->opts.journal_flush_disabled)
1668 bch2_journal_flush_async(&c->journal, NULL);
1672 ret = bch2_journal_flush(&c->journal);
1673 return bch2_err_class(ret);
1676 static struct bch_fs *bch2_path_to_fs(const char *path)
1682 ret = lookup_bdev(path, &dev);
1684 return ERR_PTR(ret);
1686 c = bch2_dev_to_fs(dev);
1688 closure_put(&c->cl);
1689 return c ?: ERR_PTR(-ENOENT);
1692 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1694 struct bch_fs *c = sb->s_fs_info;
1695 struct bch_opts opts = bch2_opts_empty();
1698 ret = bch2_parse_mount_opts(c, &opts, data);
1702 opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1704 if (opts.read_only != c->opts.read_only) {
1705 down_write(&c->state_lock);
1707 if (opts.read_only) {
1708 bch2_fs_read_only(c);
1710 sb->s_flags |= SB_RDONLY;
1712 ret = bch2_fs_read_write(c);
1714 bch_err(c, "error going rw: %i", ret);
1715 up_write(&c->state_lock);
1720 sb->s_flags &= ~SB_RDONLY;
1723 c->opts.read_only = opts.read_only;
1725 up_write(&c->state_lock);
1728 if (opt_defined(opts, errors))
1729 c->opts.errors = opts.errors;
1731 return bch2_err_class(ret);
1734 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1736 struct bch_fs *c = root->d_sb->s_fs_info;
1739 for_each_online_member(c, ca) {
1743 seq_puts(seq, ca->disk_sb.sb_name);
1749 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1751 struct bch_fs *c = root->d_sb->s_fs_info;
1753 struct printbuf buf = PRINTBUF;
1756 for (i = 0; i < bch2_opts_nr; i++) {
1757 const struct bch_option *opt = &bch2_opt_table[i];
1758 u64 v = bch2_opt_get_by_id(&c->opts, i);
1760 if (!(opt->flags & OPT_MOUNT))
1763 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1766 printbuf_reset(&buf);
1767 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1768 OPT_SHOW_MOUNT_STYLE);
1770 seq_puts(seq, buf.buf);
1773 if (buf.allocation_failure)
1775 printbuf_exit(&buf);
1779 static void bch2_put_super(struct super_block *sb)
1781 struct bch_fs *c = sb->s_fs_info;
1787 * bcachefs doesn't currently integrate intwrite freeze protection but the
1788 * internal write references serve the same purpose. Therefore reuse the
1789 * read-only transition code to perform the quiesce. The caveat is that we don't
1790 * currently have the ability to block tasks that want a write reference while
1791 * the superblock is frozen. This is fine for now, but we should either add
1792 * blocking support or find a way to integrate sb_start_intwrite() and friends.
1794 static int bch2_freeze(struct super_block *sb)
1796 struct bch_fs *c = sb->s_fs_info;
1798 down_write(&c->state_lock);
1799 bch2_fs_read_only(c);
1800 up_write(&c->state_lock);
1804 static int bch2_unfreeze(struct super_block *sb)
1806 struct bch_fs *c = sb->s_fs_info;
1809 if (test_bit(BCH_FS_emergency_ro, &c->flags))
1812 down_write(&c->state_lock);
1813 ret = bch2_fs_read_write(c);
1814 up_write(&c->state_lock);
1818 static const struct super_operations bch_super_operations = {
1819 .alloc_inode = bch2_alloc_inode,
1820 .destroy_inode = bch2_destroy_inode,
1821 .write_inode = bch2_vfs_write_inode,
1822 .evict_inode = bch2_evict_inode,
1823 .sync_fs = bch2_sync_fs,
1824 .statfs = bch2_statfs,
1825 .show_devname = bch2_show_devname,
1826 .show_options = bch2_show_options,
1827 .remount_fs = bch2_remount,
1828 .put_super = bch2_put_super,
1829 .freeze_fs = bch2_freeze,
1830 .unfreeze_fs = bch2_unfreeze,
1833 static int bch2_set_super(struct super_block *s, void *data)
1835 s->s_fs_info = data;
1839 static int bch2_noset_super(struct super_block *s, void *data)
1844 typedef DARRAY(struct bch_fs *) darray_fs;
1846 static int bch2_test_super(struct super_block *s, void *data)
1848 struct bch_fs *c = s->s_fs_info;
1849 darray_fs *d = data;
1854 darray_for_each(*d, i)
1860 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1861 int flags, const char *dev_name, void *data)
1864 struct super_block *sb;
1865 struct inode *vinode;
1866 struct bch_opts opts = bch2_opts_empty();
1869 opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1871 ret = bch2_parse_mount_opts(NULL, &opts, data);
1873 return ERR_PTR(ret);
1875 if (!dev_name || strlen(dev_name) == 0)
1876 return ERR_PTR(-EINVAL);
1879 ret = bch2_split_devs(dev_name, &devs);
1881 return ERR_PTR(ret);
1883 darray_fs devs_to_fs = {};
1884 darray_for_each(devs, i) {
1885 ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
1892 sb = sget(fs_type, bch2_test_super, bch2_noset_super, flags|SB_NOSEC, &devs_to_fs);
1896 c = bch2_fs_open(devs.data, devs.nr, opts);
1902 /* Some options can't be parsed until after the fs is started: */
1903 ret = bch2_parse_mount_opts(c, &opts, data);
1910 bch2_opts_apply(&c->opts, opts);
1912 sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1916 darray_exit(&devs_to_fs);
1917 bch2_darray_str_exit(&devs);
1921 ret = bch2_err_class(ret);
1922 return ERR_PTR(ret);
1928 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1935 sb->s_blocksize = block_bytes(c);
1936 sb->s_blocksize_bits = ilog2(block_bytes(c));
1937 sb->s_maxbytes = MAX_LFS_FILESIZE;
1938 sb->s_op = &bch_super_operations;
1939 sb->s_export_op = &bch_export_ops;
1940 #ifdef CONFIG_BCACHEFS_QUOTA
1941 sb->s_qcop = &bch2_quotactl_operations;
1942 sb->s_quota_types = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1944 sb->s_xattr = bch2_xattr_handlers;
1945 sb->s_magic = BCACHEFS_STATFS_MAGIC;
1946 sb->s_time_gran = c->sb.nsec_per_time_unit;
1947 sb->s_time_min = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1948 sb->s_time_max = div_s64(S64_MAX, c->sb.time_units_per_sec);
1949 sb->s_uuid = c->sb.user_uuid;
1951 strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1953 ret = super_setup_bdi(sb);
1957 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1959 for_each_online_member(c, ca) {
1960 struct block_device *bdev = ca->disk_sb.bdev;
1962 /* XXX: create an anonymous device for multi device filesystems */
1964 sb->s_dev = bdev->bd_dev;
1965 percpu_ref_put(&ca->io_ref);
1971 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1973 sb->s_flags |= SB_POSIXACL;
1976 sb->s_shrink->seeks = 0;
1978 vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
1979 ret = PTR_ERR_OR_ZERO(vinode);
1980 bch_err_msg(c, ret, "mounting: error getting root inode");
1984 sb->s_root = d_make_root(vinode);
1986 bch_err(c, "error mounting: error allocating root dentry");
1991 sb->s_flags |= SB_ACTIVE;
1993 return dget(sb->s_root);
1996 deactivate_locked_super(sb);
1997 return ERR_PTR(bch2_err_class(ret));
2000 static void bch2_kill_sb(struct super_block *sb)
2002 struct bch_fs *c = sb->s_fs_info;
2004 generic_shutdown_super(sb);
2008 static struct file_system_type bcache_fs_type = {
2009 .owner = THIS_MODULE,
2011 .mount = bch2_mount,
2012 .kill_sb = bch2_kill_sb,
2013 .fs_flags = FS_REQUIRES_DEV,
2016 MODULE_ALIAS_FS("bcachefs");
2018 void bch2_vfs_exit(void)
2020 unregister_filesystem(&bcache_fs_type);
2021 kmem_cache_destroy(bch2_inode_cache);
2024 int __init bch2_vfs_init(void)
2028 bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
2029 if (!bch2_inode_cache)
2032 ret = register_filesystem(&bcache_fs_type);
2042 #endif /* NO_BCACHEFS_FS */