xref: /linux/fs/bcachefs/fs.c (revision 4f05e82003d1c20da29fa593420b8d92e2c8d4e6)
1 // SPDX-License-Identifier: GPL-2.0
2 #ifndef NO_BCACHEFS_FS
3 
4 #include "bcachefs.h"
5 #include "acl.h"
6 #include "bkey_buf.h"
7 #include "btree_update.h"
8 #include "buckets.h"
9 #include "chardev.h"
10 #include "dirent.h"
11 #include "errcode.h"
12 #include "extents.h"
13 #include "fs.h"
14 #include "fs-common.h"
15 #include "fs-io.h"
16 #include "fs-ioctl.h"
17 #include "fs-io-buffered.h"
18 #include "fs-io-direct.h"
19 #include "fs-io-pagecache.h"
20 #include "fsck.h"
21 #include "inode.h"
22 #include "io_read.h"
23 #include "journal.h"
24 #include "keylist.h"
25 #include "quota.h"
26 #include "snapshot.h"
27 #include "super.h"
28 #include "xattr.h"
29 
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>
42 
43 static struct kmem_cache *bch2_inode_cache;
44 
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 *);
49 
50 void bch2_inode_update_after_write(struct btree_trans *trans,
51 				   struct bch_inode_info *inode,
52 				   struct bch_inode_unpacked *bi,
53 				   unsigned fields)
54 {
55 	struct bch_fs *c = trans->c;
56 
57 	BUG_ON(bi->bi_inum != inode->v.i_ino);
58 
59 	bch2_assert_pos_locked(trans, BTREE_ID_inodes,
60 			       POS(0, bi->bi_inum),
61 			       c->opts.inodes_use_key_cache);
62 
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;
67 
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));
74 
75 	inode->ei_inode		= *bi;
76 
77 	bch2_inode_flags_to_vfs(inode);
78 }
79 
80 int __must_check bch2_write_inode(struct bch_fs *c,
81 				  struct bch_inode_info *inode,
82 				  inode_set_fn set,
83 				  void *p, unsigned fields)
84 {
85 	struct btree_trans *trans = bch2_trans_get(c);
86 	struct btree_iter iter = { NULL };
87 	struct bch_inode_unpacked inode_u;
88 	int ret;
89 retry:
90 	bch2_trans_begin(trans);
91 
92 	ret   = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
93 				BTREE_ITER_intent) ?:
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);
97 
98 	/*
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
101 	 */
102 	if (!ret)
103 		bch2_inode_update_after_write(trans, inode, &inode_u, fields);
104 
105 	bch2_trans_iter_exit(trans, &iter);
106 
107 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
108 		goto retry;
109 
110 	bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
111 			     "%s: inode %u:%llu not found when updating",
112 			     bch2_err_str(ret),
113 			     inode_inum(inode).subvol,
114 			     inode_inum(inode).inum);
115 
116 	bch2_trans_put(trans);
117 	return ret < 0 ? ret : 0;
118 }
119 
120 int bch2_fs_quota_transfer(struct bch_fs *c,
121 			   struct bch_inode_info *inode,
122 			   struct bch_qid new_qid,
123 			   unsigned qtypes,
124 			   enum quota_acct_mode mode)
125 {
126 	unsigned i;
127 	int ret;
128 
129 	qtypes &= enabled_qtypes(c);
130 
131 	for (i = 0; i < QTYP_NR; i++)
132 		if (new_qid.q[i] == inode->ei_qid.q[i])
133 			qtypes &= ~(1U << i);
134 
135 	if (!qtypes)
136 		return 0;
137 
138 	mutex_lock(&inode->ei_quota_lock);
139 
140 	ret = bch2_quota_transfer(c, qtypes, new_qid,
141 				  inode->ei_qid,
142 				  inode->v.i_blocks +
143 				  inode->ei_quota_reserved,
144 				  mode);
145 	if (!ret)
146 		for (i = 0; i < QTYP_NR; i++)
147 			if (qtypes & (1 << i))
148 				inode->ei_qid.q[i] = new_qid.q[i];
149 
150 	mutex_unlock(&inode->ei_quota_lock);
151 
152 	return ret;
153 }
154 
155 static int bch2_iget5_test(struct inode *vinode, void *p)
156 {
157 	struct bch_inode_info *inode = to_bch_ei(vinode);
158 	subvol_inum *inum = p;
159 
160 	return inode->ei_subvol == inum->subvol &&
161 		inode->ei_inode.bi_inum == inum->inum;
162 }
163 
164 static int bch2_iget5_set(struct inode *vinode, void *p)
165 {
166 	struct bch_inode_info *inode = to_bch_ei(vinode);
167 	subvol_inum *inum = p;
168 
169 	inode->v.i_ino		= inum->inum;
170 	inode->ei_subvol	= inum->subvol;
171 	inode->ei_inode.bi_inum	= inum->inum;
172 	return 0;
173 }
174 
175 static unsigned bch2_inode_hash(subvol_inum inum)
176 {
177 	return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
178 }
179 
180 static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_inode_info *inode)
181 {
182 	subvol_inum inum = inode_inum(inode);
183 	struct bch_inode_info *old = to_bch_ei(inode_insert5(&inode->v,
184 				      bch2_inode_hash(inum),
185 				      bch2_iget5_test,
186 				      bch2_iget5_set,
187 				      &inum));
188 	BUG_ON(!old);
189 
190 	if (unlikely(old != inode)) {
191 		__destroy_inode(&inode->v);
192 		kmem_cache_free(bch2_inode_cache, inode);
193 		inode = old;
194 	} else {
195 		mutex_lock(&c->vfs_inodes_lock);
196 		list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
197 		mutex_unlock(&c->vfs_inodes_lock);
198 		/*
199 		 * we really don't want insert_inode_locked2() to be setting
200 		 * I_NEW...
201 		 */
202 		unlock_new_inode(&inode->v);
203 	}
204 
205 	return inode;
206 }
207 
208 #define memalloc_flags_do(_flags, _do)						\
209 ({										\
210 	unsigned _saved_flags = memalloc_flags_save(_flags);			\
211 	typeof(_do) _ret = _do;							\
212 	memalloc_noreclaim_restore(_saved_flags);				\
213 	_ret;									\
214 })
215 
216 static struct inode *bch2_alloc_inode(struct super_block *sb)
217 {
218 	BUG();
219 }
220 
221 static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
222 {
223 	struct bch_inode_info *inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
224 	if (!inode)
225 		return NULL;
226 
227 	inode_init_once(&inode->v);
228 	mutex_init(&inode->ei_update_lock);
229 	two_state_lock_init(&inode->ei_pagecache_lock);
230 	INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
231 	mutex_init(&inode->ei_quota_lock);
232 	inode->v.i_state = 0;
233 
234 	if (unlikely(inode_init_always(c->vfs_sb, &inode->v))) {
235 		kmem_cache_free(bch2_inode_cache, inode);
236 		return NULL;
237 	}
238 
239 	return inode;
240 }
241 
242 /*
243  * Allocate a new inode, dropping/retaking btree locks if necessary:
244  */
245 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
246 {
247 	struct bch_inode_info *inode =
248 		memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
249 				  __bch2_new_inode(trans->c));
250 
251 	if (unlikely(!inode)) {
252 		int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c)) ? 0 : -ENOMEM);
253 		if (ret && inode) {
254 			__destroy_inode(&inode->v);
255 			kmem_cache_free(bch2_inode_cache, inode);
256 		}
257 		if (ret)
258 			return ERR_PTR(ret);
259 	}
260 
261 	return inode;
262 }
263 
264 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
265 {
266 	struct bch_inode_info *inode =
267 		to_bch_ei(ilookup5_nowait(c->vfs_sb,
268 					  bch2_inode_hash(inum),
269 					  bch2_iget5_test,
270 					  &inum));
271 	if (inode)
272 		return &inode->v;
273 
274 	struct btree_trans *trans = bch2_trans_get(c);
275 
276 	struct bch_inode_unpacked inode_u;
277 	struct bch_subvolume subvol;
278 	int ret = lockrestart_do(trans,
279 		bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
280 		bch2_inode_find_by_inum_trans(trans, inum, &inode_u)) ?:
281 		PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
282 	if (!ret) {
283 		bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
284 		inode = bch2_inode_insert(c, inode);
285 	}
286 	bch2_trans_put(trans);
287 
288 	return ret ? ERR_PTR(ret) : &inode->v;
289 }
290 
291 struct bch_inode_info *
292 __bch2_create(struct mnt_idmap *idmap,
293 	      struct bch_inode_info *dir, struct dentry *dentry,
294 	      umode_t mode, dev_t rdev, subvol_inum snapshot_src,
295 	      unsigned flags)
296 {
297 	struct bch_fs *c = dir->v.i_sb->s_fs_info;
298 	struct btree_trans *trans;
299 	struct bch_inode_unpacked dir_u;
300 	struct bch_inode_info *inode;
301 	struct bch_inode_unpacked inode_u;
302 	struct posix_acl *default_acl = NULL, *acl = NULL;
303 	subvol_inum inum;
304 	struct bch_subvolume subvol;
305 	u64 journal_seq = 0;
306 	int ret;
307 
308 	/*
309 	 * preallocate acls + vfs inode before btree transaction, so that
310 	 * nothing can fail after the transaction succeeds:
311 	 */
312 #ifdef CONFIG_BCACHEFS_POSIX_ACL
313 	ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
314 	if (ret)
315 		return ERR_PTR(ret);
316 #endif
317 	inode = __bch2_new_inode(c);
318 	if (unlikely(!inode)) {
319 		inode = ERR_PTR(-ENOMEM);
320 		goto err;
321 	}
322 
323 	bch2_inode_init_early(c, &inode_u);
324 
325 	if (!(flags & BCH_CREATE_TMPFILE))
326 		mutex_lock(&dir->ei_update_lock);
327 
328 	trans = bch2_trans_get(c);
329 retry:
330 	bch2_trans_begin(trans);
331 
332 	ret   = bch2_subvol_is_ro_trans(trans, dir->ei_subvol) ?:
333 		bch2_create_trans(trans,
334 				  inode_inum(dir), &dir_u, &inode_u,
335 				  !(flags & BCH_CREATE_TMPFILE)
336 				  ? &dentry->d_name : NULL,
337 				  from_kuid(i_user_ns(&dir->v), current_fsuid()),
338 				  from_kgid(i_user_ns(&dir->v), current_fsgid()),
339 				  mode, rdev,
340 				  default_acl, acl, snapshot_src, flags) ?:
341 		bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
342 				KEY_TYPE_QUOTA_PREALLOC);
343 	if (unlikely(ret))
344 		goto err_before_quota;
345 
346 	inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
347 	inum.inum = inode_u.bi_inum;
348 
349 	ret   = bch2_subvolume_get(trans, inum.subvol, true,
350 				   BTREE_ITER_with_updates, &subvol) ?:
351 		bch2_trans_commit(trans, NULL, &journal_seq, 0);
352 	if (unlikely(ret)) {
353 		bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
354 				KEY_TYPE_QUOTA_WARN);
355 err_before_quota:
356 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
357 			goto retry;
358 		goto err_trans;
359 	}
360 
361 	if (!(flags & BCH_CREATE_TMPFILE)) {
362 		bch2_inode_update_after_write(trans, dir, &dir_u,
363 					      ATTR_MTIME|ATTR_CTIME);
364 		mutex_unlock(&dir->ei_update_lock);
365 	}
366 
367 	bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
368 
369 	set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
370 	set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
371 
372 	/*
373 	 * we must insert the new inode into the inode cache before calling
374 	 * bch2_trans_exit() and dropping locks, else we could race with another
375 	 * thread pulling the inode in and modifying it:
376 	 */
377 	inode = bch2_inode_insert(c, inode);
378 	bch2_trans_put(trans);
379 err:
380 	posix_acl_release(default_acl);
381 	posix_acl_release(acl);
382 	return inode;
383 err_trans:
384 	if (!(flags & BCH_CREATE_TMPFILE))
385 		mutex_unlock(&dir->ei_update_lock);
386 
387 	bch2_trans_put(trans);
388 	make_bad_inode(&inode->v);
389 	iput(&inode->v);
390 	inode = ERR_PTR(ret);
391 	goto err;
392 }
393 
394 /* methods */
395 
396 static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
397 			subvol_inum dir, struct bch_hash_info *dir_hash_info,
398 			const struct qstr *name)
399 {
400 	struct bch_fs *c = trans->c;
401 	struct btree_iter dirent_iter = {};
402 	subvol_inum inum = {};
403 	struct printbuf buf = PRINTBUF;
404 
405 	struct bkey_s_c k = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
406 					     dir_hash_info, dir, name, 0);
407 	int ret = bkey_err(k);
408 	if (ret)
409 		return ERR_PTR(ret);
410 
411 	ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
412 	if (ret > 0)
413 		ret = -ENOENT;
414 	if (ret)
415 		goto err;
416 
417 	struct bch_inode_info *inode =
418 		to_bch_ei(ilookup5_nowait(c->vfs_sb,
419 					  bch2_inode_hash(inum),
420 					  bch2_iget5_test,
421 					  &inum));
422 	if (inode)
423 		goto out;
424 
425 	struct bch_subvolume subvol;
426 	struct bch_inode_unpacked inode_u;
427 	ret =   bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
428 		bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
429 		PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
430 
431 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
432 				c, "dirent to missing inode:\n  %s",
433 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
434 	if (ret)
435 		goto err;
436 
437 	/* regular files may have hardlinks: */
438 	if (bch2_fs_inconsistent_on(bch2_inode_should_have_bp(&inode_u) &&
439 				    !bkey_eq(k.k->p, POS(inode_u.bi_dir, inode_u.bi_dir_offset)),
440 				    c,
441 				    "dirent points to inode that does not point back:\n  %s",
442 				    (bch2_bkey_val_to_text(&buf, c, k),
443 				     prt_printf(&buf, "\n  "),
444 				     bch2_inode_unpacked_to_text(&buf, &inode_u),
445 				     buf.buf))) {
446 		ret = -ENOENT;
447 		goto err;
448 	}
449 
450 	bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
451 	inode = bch2_inode_insert(c, inode);
452 out:
453 	bch2_trans_iter_exit(trans, &dirent_iter);
454 	printbuf_exit(&buf);
455 	return inode;
456 err:
457 	inode = ERR_PTR(ret);
458 	goto out;
459 }
460 
461 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
462 				  unsigned int flags)
463 {
464 	struct bch_fs *c = vdir->i_sb->s_fs_info;
465 	struct bch_inode_info *dir = to_bch_ei(vdir);
466 	struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
467 
468 	struct bch_inode_info *inode;
469 	bch2_trans_do(c, NULL, NULL, 0,
470 		PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
471 							  &hash, &dentry->d_name)));
472 	if (IS_ERR(inode))
473 		inode = NULL;
474 
475 	return d_splice_alias(&inode->v, dentry);
476 }
477 
478 static int bch2_mknod(struct mnt_idmap *idmap,
479 		      struct inode *vdir, struct dentry *dentry,
480 		      umode_t mode, dev_t rdev)
481 {
482 	struct bch_inode_info *inode =
483 		__bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
484 			      (subvol_inum) { 0 }, 0);
485 
486 	if (IS_ERR(inode))
487 		return bch2_err_class(PTR_ERR(inode));
488 
489 	d_instantiate(dentry, &inode->v);
490 	return 0;
491 }
492 
493 static int bch2_create(struct mnt_idmap *idmap,
494 		       struct inode *vdir, struct dentry *dentry,
495 		       umode_t mode, bool excl)
496 {
497 	return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
498 }
499 
500 static int __bch2_link(struct bch_fs *c,
501 		       struct bch_inode_info *inode,
502 		       struct bch_inode_info *dir,
503 		       struct dentry *dentry)
504 {
505 	struct btree_trans *trans = bch2_trans_get(c);
506 	struct bch_inode_unpacked dir_u, inode_u;
507 	int ret;
508 
509 	mutex_lock(&inode->ei_update_lock);
510 
511 	ret = commit_do(trans, NULL, NULL, 0,
512 			bch2_link_trans(trans,
513 					inode_inum(dir),   &dir_u,
514 					inode_inum(inode), &inode_u,
515 					&dentry->d_name));
516 
517 	if (likely(!ret)) {
518 		bch2_inode_update_after_write(trans, dir, &dir_u,
519 					      ATTR_MTIME|ATTR_CTIME);
520 		bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
521 	}
522 
523 	bch2_trans_put(trans);
524 	mutex_unlock(&inode->ei_update_lock);
525 	return ret;
526 }
527 
528 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
529 		     struct dentry *dentry)
530 {
531 	struct bch_fs *c = vdir->i_sb->s_fs_info;
532 	struct bch_inode_info *dir = to_bch_ei(vdir);
533 	struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
534 	int ret;
535 
536 	lockdep_assert_held(&inode->v.i_rwsem);
537 
538 	ret   = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
539 		bch2_subvol_is_ro(c, inode->ei_subvol) ?:
540 		__bch2_link(c, inode, dir, dentry);
541 	if (unlikely(ret))
542 		return bch2_err_class(ret);
543 
544 	ihold(&inode->v);
545 	d_instantiate(dentry, &inode->v);
546 	return 0;
547 }
548 
549 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
550 		  bool deleting_snapshot)
551 {
552 	struct bch_fs *c = vdir->i_sb->s_fs_info;
553 	struct bch_inode_info *dir = to_bch_ei(vdir);
554 	struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
555 	struct bch_inode_unpacked dir_u, inode_u;
556 	struct btree_trans *trans = bch2_trans_get(c);
557 	int ret;
558 
559 	bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
560 
561 	ret = commit_do(trans, NULL, NULL,
562 			BCH_TRANS_COMMIT_no_enospc,
563 		bch2_unlink_trans(trans,
564 				  inode_inum(dir), &dir_u,
565 				  &inode_u, &dentry->d_name,
566 				  deleting_snapshot));
567 	if (unlikely(ret))
568 		goto err;
569 
570 	bch2_inode_update_after_write(trans, dir, &dir_u,
571 				      ATTR_MTIME|ATTR_CTIME);
572 	bch2_inode_update_after_write(trans, inode, &inode_u,
573 				      ATTR_MTIME);
574 
575 	if (inode_u.bi_subvol) {
576 		/*
577 		 * Subvolume deletion is asynchronous, but we still want to tell
578 		 * the VFS that it's been deleted here:
579 		 */
580 		set_nlink(&inode->v, 0);
581 	}
582 err:
583 	bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
584 	bch2_trans_put(trans);
585 
586 	return ret;
587 }
588 
589 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
590 {
591 	struct bch_inode_info *dir= to_bch_ei(vdir);
592 	struct bch_fs *c = dir->v.i_sb->s_fs_info;
593 
594 	int ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
595 		__bch2_unlink(vdir, dentry, false);
596 	return bch2_err_class(ret);
597 }
598 
599 static int bch2_symlink(struct mnt_idmap *idmap,
600 			struct inode *vdir, struct dentry *dentry,
601 			const char *symname)
602 {
603 	struct bch_fs *c = vdir->i_sb->s_fs_info;
604 	struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
605 	int ret;
606 
607 	inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
608 			      (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
609 	if (IS_ERR(inode))
610 		return bch2_err_class(PTR_ERR(inode));
611 
612 	inode_lock(&inode->v);
613 	ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
614 	inode_unlock(&inode->v);
615 
616 	if (unlikely(ret))
617 		goto err;
618 
619 	ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
620 	if (unlikely(ret))
621 		goto err;
622 
623 	ret = __bch2_link(c, inode, dir, dentry);
624 	if (unlikely(ret))
625 		goto err;
626 
627 	d_instantiate(dentry, &inode->v);
628 	return 0;
629 err:
630 	iput(&inode->v);
631 	return bch2_err_class(ret);
632 }
633 
634 static int bch2_mkdir(struct mnt_idmap *idmap,
635 		      struct inode *vdir, struct dentry *dentry, umode_t mode)
636 {
637 	return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
638 }
639 
640 static int bch2_rename2(struct mnt_idmap *idmap,
641 			struct inode *src_vdir, struct dentry *src_dentry,
642 			struct inode *dst_vdir, struct dentry *dst_dentry,
643 			unsigned flags)
644 {
645 	struct bch_fs *c = src_vdir->i_sb->s_fs_info;
646 	struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
647 	struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
648 	struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
649 	struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
650 	struct bch_inode_unpacked dst_dir_u, src_dir_u;
651 	struct bch_inode_unpacked src_inode_u, dst_inode_u;
652 	struct btree_trans *trans;
653 	enum bch_rename_mode mode = flags & RENAME_EXCHANGE
654 		? BCH_RENAME_EXCHANGE
655 		: dst_dentry->d_inode
656 		? BCH_RENAME_OVERWRITE : BCH_RENAME;
657 	int ret;
658 
659 	if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
660 		return -EINVAL;
661 
662 	if (mode == BCH_RENAME_OVERWRITE) {
663 		ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
664 						   0, LLONG_MAX);
665 		if (ret)
666 			return ret;
667 	}
668 
669 	trans = bch2_trans_get(c);
670 
671 	bch2_lock_inodes(INODE_UPDATE_LOCK,
672 			 src_dir,
673 			 dst_dir,
674 			 src_inode,
675 			 dst_inode);
676 
677 	ret   = bch2_subvol_is_ro_trans(trans, src_dir->ei_subvol) ?:
678 		bch2_subvol_is_ro_trans(trans, dst_dir->ei_subvol);
679 	if (ret)
680 		goto err;
681 
682 	if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
683 		ret = bch2_fs_quota_transfer(c, src_inode,
684 					     dst_dir->ei_qid,
685 					     1 << QTYP_PRJ,
686 					     KEY_TYPE_QUOTA_PREALLOC);
687 		if (ret)
688 			goto err;
689 	}
690 
691 	if (mode == BCH_RENAME_EXCHANGE &&
692 	    inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
693 		ret = bch2_fs_quota_transfer(c, dst_inode,
694 					     src_dir->ei_qid,
695 					     1 << QTYP_PRJ,
696 					     KEY_TYPE_QUOTA_PREALLOC);
697 		if (ret)
698 			goto err;
699 	}
700 
701 	ret = commit_do(trans, NULL, NULL, 0,
702 			bch2_rename_trans(trans,
703 					  inode_inum(src_dir), &src_dir_u,
704 					  inode_inum(dst_dir), &dst_dir_u,
705 					  &src_inode_u,
706 					  &dst_inode_u,
707 					  &src_dentry->d_name,
708 					  &dst_dentry->d_name,
709 					  mode));
710 	if (unlikely(ret))
711 		goto err;
712 
713 	BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
714 	BUG_ON(dst_inode &&
715 	       dst_inode->v.i_ino != dst_inode_u.bi_inum);
716 
717 	bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
718 				      ATTR_MTIME|ATTR_CTIME);
719 
720 	if (src_dir != dst_dir)
721 		bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
722 					      ATTR_MTIME|ATTR_CTIME);
723 
724 	bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
725 				      ATTR_CTIME);
726 
727 	if (dst_inode)
728 		bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
729 					      ATTR_CTIME);
730 err:
731 	bch2_trans_put(trans);
732 
733 	bch2_fs_quota_transfer(c, src_inode,
734 			       bch_qid(&src_inode->ei_inode),
735 			       1 << QTYP_PRJ,
736 			       KEY_TYPE_QUOTA_NOCHECK);
737 	if (dst_inode)
738 		bch2_fs_quota_transfer(c, dst_inode,
739 				       bch_qid(&dst_inode->ei_inode),
740 				       1 << QTYP_PRJ,
741 				       KEY_TYPE_QUOTA_NOCHECK);
742 
743 	bch2_unlock_inodes(INODE_UPDATE_LOCK,
744 			   src_dir,
745 			   dst_dir,
746 			   src_inode,
747 			   dst_inode);
748 
749 	return bch2_err_class(ret);
750 }
751 
752 static void bch2_setattr_copy(struct mnt_idmap *idmap,
753 			      struct bch_inode_info *inode,
754 			      struct bch_inode_unpacked *bi,
755 			      struct iattr *attr)
756 {
757 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
758 	unsigned int ia_valid = attr->ia_valid;
759 
760 	if (ia_valid & ATTR_UID)
761 		bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
762 	if (ia_valid & ATTR_GID)
763 		bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
764 
765 	if (ia_valid & ATTR_SIZE)
766 		bi->bi_size = attr->ia_size;
767 
768 	if (ia_valid & ATTR_ATIME)
769 		bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
770 	if (ia_valid & ATTR_MTIME)
771 		bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
772 	if (ia_valid & ATTR_CTIME)
773 		bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
774 
775 	if (ia_valid & ATTR_MODE) {
776 		umode_t mode = attr->ia_mode;
777 		kgid_t gid = ia_valid & ATTR_GID
778 			? attr->ia_gid
779 			: inode->v.i_gid;
780 
781 		if (!in_group_p(gid) &&
782 		    !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
783 			mode &= ~S_ISGID;
784 		bi->bi_mode = mode;
785 	}
786 }
787 
788 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
789 			 struct bch_inode_info *inode,
790 			 struct iattr *attr)
791 {
792 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
793 	struct bch_qid qid;
794 	struct btree_trans *trans;
795 	struct btree_iter inode_iter = { NULL };
796 	struct bch_inode_unpacked inode_u;
797 	struct posix_acl *acl = NULL;
798 	int ret;
799 
800 	mutex_lock(&inode->ei_update_lock);
801 
802 	qid = inode->ei_qid;
803 
804 	if (attr->ia_valid & ATTR_UID)
805 		qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
806 
807 	if (attr->ia_valid & ATTR_GID)
808 		qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
809 
810 	ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
811 				     KEY_TYPE_QUOTA_PREALLOC);
812 	if (ret)
813 		goto err;
814 
815 	trans = bch2_trans_get(c);
816 retry:
817 	bch2_trans_begin(trans);
818 	kfree(acl);
819 	acl = NULL;
820 
821 	ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
822 			      BTREE_ITER_intent);
823 	if (ret)
824 		goto btree_err;
825 
826 	bch2_setattr_copy(idmap, inode, &inode_u, attr);
827 
828 	if (attr->ia_valid & ATTR_MODE) {
829 		ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
830 				     inode_u.bi_mode, &acl);
831 		if (ret)
832 			goto btree_err;
833 	}
834 
835 	ret =   bch2_inode_write(trans, &inode_iter, &inode_u) ?:
836 		bch2_trans_commit(trans, NULL, NULL,
837 				  BCH_TRANS_COMMIT_no_enospc);
838 btree_err:
839 	bch2_trans_iter_exit(trans, &inode_iter);
840 
841 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
842 		goto retry;
843 	if (unlikely(ret))
844 		goto err_trans;
845 
846 	bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
847 
848 	if (acl)
849 		set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
850 err_trans:
851 	bch2_trans_put(trans);
852 err:
853 	mutex_unlock(&inode->ei_update_lock);
854 
855 	return bch2_err_class(ret);
856 }
857 
858 static int bch2_getattr(struct mnt_idmap *idmap,
859 			const struct path *path, struct kstat *stat,
860 			u32 request_mask, unsigned query_flags)
861 {
862 	struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
863 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
864 
865 	stat->dev	= inode->v.i_sb->s_dev;
866 	stat->ino	= inode->v.i_ino;
867 	stat->mode	= inode->v.i_mode;
868 	stat->nlink	= inode->v.i_nlink;
869 	stat->uid	= inode->v.i_uid;
870 	stat->gid	= inode->v.i_gid;
871 	stat->rdev	= inode->v.i_rdev;
872 	stat->size	= i_size_read(&inode->v);
873 	stat->atime	= inode_get_atime(&inode->v);
874 	stat->mtime	= inode_get_mtime(&inode->v);
875 	stat->ctime	= inode_get_ctime(&inode->v);
876 	stat->blksize	= block_bytes(c);
877 	stat->blocks	= inode->v.i_blocks;
878 
879 	stat->subvol	= inode->ei_subvol;
880 	stat->result_mask |= STATX_SUBVOL;
881 
882 	if (request_mask & STATX_BTIME) {
883 		stat->result_mask |= STATX_BTIME;
884 		stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
885 	}
886 
887 	if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
888 		stat->attributes |= STATX_ATTR_IMMUTABLE;
889 	stat->attributes_mask	 |= STATX_ATTR_IMMUTABLE;
890 
891 	if (inode->ei_inode.bi_flags & BCH_INODE_append)
892 		stat->attributes |= STATX_ATTR_APPEND;
893 	stat->attributes_mask	 |= STATX_ATTR_APPEND;
894 
895 	if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
896 		stat->attributes |= STATX_ATTR_NODUMP;
897 	stat->attributes_mask	 |= STATX_ATTR_NODUMP;
898 
899 	return 0;
900 }
901 
902 static int bch2_setattr(struct mnt_idmap *idmap,
903 			struct dentry *dentry, struct iattr *iattr)
904 {
905 	struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
906 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
907 	int ret;
908 
909 	lockdep_assert_held(&inode->v.i_rwsem);
910 
911 	ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
912 		setattr_prepare(idmap, dentry, iattr);
913 	if (ret)
914 		return ret;
915 
916 	return iattr->ia_valid & ATTR_SIZE
917 		? bchfs_truncate(idmap, inode, iattr)
918 		: bch2_setattr_nonsize(idmap, inode, iattr);
919 }
920 
921 static int bch2_tmpfile(struct mnt_idmap *idmap,
922 			struct inode *vdir, struct file *file, umode_t mode)
923 {
924 	struct bch_inode_info *inode =
925 		__bch2_create(idmap, to_bch_ei(vdir),
926 			      file->f_path.dentry, mode, 0,
927 			      (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
928 
929 	if (IS_ERR(inode))
930 		return bch2_err_class(PTR_ERR(inode));
931 
932 	d_mark_tmpfile(file, &inode->v);
933 	d_instantiate(file->f_path.dentry, &inode->v);
934 	return finish_open_simple(file, 0);
935 }
936 
937 static int bch2_fill_extent(struct bch_fs *c,
938 			    struct fiemap_extent_info *info,
939 			    struct bkey_s_c k, unsigned flags)
940 {
941 	if (bkey_extent_is_direct_data(k.k)) {
942 		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
943 		const union bch_extent_entry *entry;
944 		struct extent_ptr_decoded p;
945 		int ret;
946 
947 		if (k.k->type == KEY_TYPE_reflink_v)
948 			flags |= FIEMAP_EXTENT_SHARED;
949 
950 		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
951 			int flags2 = 0;
952 			u64 offset = p.ptr.offset;
953 
954 			if (p.ptr.unwritten)
955 				flags2 |= FIEMAP_EXTENT_UNWRITTEN;
956 
957 			if (p.crc.compression_type)
958 				flags2 |= FIEMAP_EXTENT_ENCODED;
959 			else
960 				offset += p.crc.offset;
961 
962 			if ((offset & (block_sectors(c) - 1)) ||
963 			    (k.k->size & (block_sectors(c) - 1)))
964 				flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
965 
966 			ret = fiemap_fill_next_extent(info,
967 						bkey_start_offset(k.k) << 9,
968 						offset << 9,
969 						k.k->size << 9, flags|flags2);
970 			if (ret)
971 				return ret;
972 		}
973 
974 		return 0;
975 	} else if (bkey_extent_is_inline_data(k.k)) {
976 		return fiemap_fill_next_extent(info,
977 					       bkey_start_offset(k.k) << 9,
978 					       0, k.k->size << 9,
979 					       flags|
980 					       FIEMAP_EXTENT_DATA_INLINE);
981 	} else if (k.k->type == KEY_TYPE_reservation) {
982 		return fiemap_fill_next_extent(info,
983 					       bkey_start_offset(k.k) << 9,
984 					       0, k.k->size << 9,
985 					       flags|
986 					       FIEMAP_EXTENT_DELALLOC|
987 					       FIEMAP_EXTENT_UNWRITTEN);
988 	} else {
989 		BUG();
990 	}
991 }
992 
993 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
994 		       u64 start, u64 len)
995 {
996 	struct bch_fs *c = vinode->i_sb->s_fs_info;
997 	struct bch_inode_info *ei = to_bch_ei(vinode);
998 	struct btree_trans *trans;
999 	struct btree_iter iter;
1000 	struct bkey_s_c k;
1001 	struct bkey_buf cur, prev;
1002 	unsigned offset_into_extent, sectors;
1003 	bool have_extent = false;
1004 	u32 snapshot;
1005 	int ret = 0;
1006 
1007 	ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
1008 	if (ret)
1009 		return ret;
1010 
1011 	struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
1012 	if (start + len < start)
1013 		return -EINVAL;
1014 
1015 	start >>= 9;
1016 
1017 	bch2_bkey_buf_init(&cur);
1018 	bch2_bkey_buf_init(&prev);
1019 	trans = bch2_trans_get(c);
1020 retry:
1021 	bch2_trans_begin(trans);
1022 
1023 	ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
1024 	if (ret)
1025 		goto err;
1026 
1027 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
1028 			     SPOS(ei->v.i_ino, start, snapshot), 0);
1029 
1030 	while (!(ret = btree_trans_too_many_iters(trans)) &&
1031 	       (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
1032 	       !(ret = bkey_err(k))) {
1033 		enum btree_id data_btree = BTREE_ID_extents;
1034 
1035 		if (!bkey_extent_is_data(k.k) &&
1036 		    k.k->type != KEY_TYPE_reservation) {
1037 			bch2_btree_iter_advance(&iter);
1038 			continue;
1039 		}
1040 
1041 		offset_into_extent	= iter.pos.offset -
1042 			bkey_start_offset(k.k);
1043 		sectors			= k.k->size - offset_into_extent;
1044 
1045 		bch2_bkey_buf_reassemble(&cur, c, k);
1046 
1047 		ret = bch2_read_indirect_extent(trans, &data_btree,
1048 					&offset_into_extent, &cur);
1049 		if (ret)
1050 			break;
1051 
1052 		k = bkey_i_to_s_c(cur.k);
1053 		bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
1054 
1055 		sectors = min(sectors, k.k->size - offset_into_extent);
1056 
1057 		bch2_cut_front(POS(k.k->p.inode,
1058 				   bkey_start_offset(k.k) +
1059 				   offset_into_extent),
1060 			       cur.k);
1061 		bch2_key_resize(&cur.k->k, sectors);
1062 		cur.k->k.p = iter.pos;
1063 		cur.k->k.p.offset += cur.k->k.size;
1064 
1065 		if (have_extent) {
1066 			bch2_trans_unlock(trans);
1067 			ret = bch2_fill_extent(c, info,
1068 					bkey_i_to_s_c(prev.k), 0);
1069 			if (ret)
1070 				break;
1071 		}
1072 
1073 		bkey_copy(prev.k, cur.k);
1074 		have_extent = true;
1075 
1076 		bch2_btree_iter_set_pos(&iter,
1077 			POS(iter.pos.inode, iter.pos.offset + sectors));
1078 
1079 		ret = bch2_trans_relock(trans);
1080 		if (ret)
1081 			break;
1082 	}
1083 	start = iter.pos.offset;
1084 	bch2_trans_iter_exit(trans, &iter);
1085 err:
1086 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1087 		goto retry;
1088 
1089 	if (!ret && have_extent) {
1090 		bch2_trans_unlock(trans);
1091 		ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
1092 				       FIEMAP_EXTENT_LAST);
1093 	}
1094 
1095 	bch2_trans_put(trans);
1096 	bch2_bkey_buf_exit(&cur, c);
1097 	bch2_bkey_buf_exit(&prev, c);
1098 	return ret < 0 ? ret : 0;
1099 }
1100 
1101 static const struct vm_operations_struct bch_vm_ops = {
1102 	.fault		= bch2_page_fault,
1103 	.map_pages	= filemap_map_pages,
1104 	.page_mkwrite   = bch2_page_mkwrite,
1105 };
1106 
1107 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1108 {
1109 	file_accessed(file);
1110 
1111 	vma->vm_ops = &bch_vm_ops;
1112 	return 0;
1113 }
1114 
1115 /* Directories: */
1116 
1117 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1118 {
1119 	return generic_file_llseek_size(file, offset, whence,
1120 					S64_MAX, S64_MAX);
1121 }
1122 
1123 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1124 {
1125 	struct bch_inode_info *inode = file_bch_inode(file);
1126 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1127 
1128 	if (!dir_emit_dots(file, ctx))
1129 		return 0;
1130 
1131 	int ret = bch2_readdir(c, inode_inum(inode), ctx);
1132 
1133 	bch_err_fn(c, ret);
1134 	return bch2_err_class(ret);
1135 }
1136 
1137 static int bch2_open(struct inode *vinode, struct file *file)
1138 {
1139 	if (file->f_flags & (O_WRONLY|O_RDWR)) {
1140 		struct bch_inode_info *inode = to_bch_ei(vinode);
1141 		struct bch_fs *c = inode->v.i_sb->s_fs_info;
1142 
1143 		int ret = bch2_subvol_is_ro(c, inode->ei_subvol);
1144 		if (ret)
1145 			return ret;
1146 	}
1147 
1148 	return generic_file_open(vinode, file);
1149 }
1150 
1151 static const struct file_operations bch_file_operations = {
1152 	.open		= bch2_open,
1153 	.llseek		= bch2_llseek,
1154 	.read_iter	= bch2_read_iter,
1155 	.write_iter	= bch2_write_iter,
1156 	.mmap		= bch2_mmap,
1157 	.fsync		= bch2_fsync,
1158 	.splice_read	= filemap_splice_read,
1159 	.splice_write	= iter_file_splice_write,
1160 	.fallocate	= bch2_fallocate_dispatch,
1161 	.unlocked_ioctl = bch2_fs_file_ioctl,
1162 #ifdef CONFIG_COMPAT
1163 	.compat_ioctl	= bch2_compat_fs_ioctl,
1164 #endif
1165 	.remap_file_range = bch2_remap_file_range,
1166 };
1167 
1168 static const struct inode_operations bch_file_inode_operations = {
1169 	.getattr	= bch2_getattr,
1170 	.setattr	= bch2_setattr,
1171 	.fiemap		= bch2_fiemap,
1172 	.listxattr	= bch2_xattr_list,
1173 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1174 	.get_acl	= bch2_get_acl,
1175 	.set_acl	= bch2_set_acl,
1176 #endif
1177 };
1178 
1179 static const struct inode_operations bch_dir_inode_operations = {
1180 	.lookup		= bch2_lookup,
1181 	.create		= bch2_create,
1182 	.link		= bch2_link,
1183 	.unlink		= bch2_unlink,
1184 	.symlink	= bch2_symlink,
1185 	.mkdir		= bch2_mkdir,
1186 	.rmdir		= bch2_unlink,
1187 	.mknod		= bch2_mknod,
1188 	.rename		= bch2_rename2,
1189 	.getattr	= bch2_getattr,
1190 	.setattr	= bch2_setattr,
1191 	.tmpfile	= bch2_tmpfile,
1192 	.listxattr	= bch2_xattr_list,
1193 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1194 	.get_acl	= bch2_get_acl,
1195 	.set_acl	= bch2_set_acl,
1196 #endif
1197 };
1198 
1199 static const struct file_operations bch_dir_file_operations = {
1200 	.llseek		= bch2_dir_llseek,
1201 	.read		= generic_read_dir,
1202 	.iterate_shared	= bch2_vfs_readdir,
1203 	.fsync		= bch2_fsync,
1204 	.unlocked_ioctl = bch2_fs_file_ioctl,
1205 #ifdef CONFIG_COMPAT
1206 	.compat_ioctl	= bch2_compat_fs_ioctl,
1207 #endif
1208 };
1209 
1210 static const struct inode_operations bch_symlink_inode_operations = {
1211 	.get_link	= page_get_link,
1212 	.getattr	= bch2_getattr,
1213 	.setattr	= bch2_setattr,
1214 	.listxattr	= bch2_xattr_list,
1215 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1216 	.get_acl	= bch2_get_acl,
1217 	.set_acl	= bch2_set_acl,
1218 #endif
1219 };
1220 
1221 static const struct inode_operations bch_special_inode_operations = {
1222 	.getattr	= bch2_getattr,
1223 	.setattr	= bch2_setattr,
1224 	.listxattr	= bch2_xattr_list,
1225 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1226 	.get_acl	= bch2_get_acl,
1227 	.set_acl	= bch2_set_acl,
1228 #endif
1229 };
1230 
1231 static const struct address_space_operations bch_address_space_operations = {
1232 	.read_folio	= bch2_read_folio,
1233 	.writepages	= bch2_writepages,
1234 	.readahead	= bch2_readahead,
1235 	.dirty_folio	= filemap_dirty_folio,
1236 	.write_begin	= bch2_write_begin,
1237 	.write_end	= bch2_write_end,
1238 	.invalidate_folio = bch2_invalidate_folio,
1239 	.release_folio	= bch2_release_folio,
1240 	.direct_IO	= noop_direct_IO,
1241 #ifdef CONFIG_MIGRATION
1242 	.migrate_folio	= filemap_migrate_folio,
1243 #endif
1244 	.error_remove_folio = generic_error_remove_folio,
1245 };
1246 
1247 struct bcachefs_fid {
1248 	u64		inum;
1249 	u32		subvol;
1250 	u32		gen;
1251 } __packed;
1252 
1253 struct bcachefs_fid_with_parent {
1254 	struct bcachefs_fid	fid;
1255 	struct bcachefs_fid	dir;
1256 } __packed;
1257 
1258 static int bcachefs_fid_valid(int fh_len, int fh_type)
1259 {
1260 	switch (fh_type) {
1261 	case FILEID_BCACHEFS_WITHOUT_PARENT:
1262 		return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1263 	case FILEID_BCACHEFS_WITH_PARENT:
1264 		return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1265 	default:
1266 		return false;
1267 	}
1268 }
1269 
1270 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1271 {
1272 	return (struct bcachefs_fid) {
1273 		.inum	= inode->ei_inode.bi_inum,
1274 		.subvol	= inode->ei_subvol,
1275 		.gen	= inode->ei_inode.bi_generation,
1276 	};
1277 }
1278 
1279 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1280 			  struct inode *vdir)
1281 {
1282 	struct bch_inode_info *inode	= to_bch_ei(vinode);
1283 	struct bch_inode_info *dir	= to_bch_ei(vdir);
1284 	int min_len;
1285 
1286 	if (!S_ISDIR(inode->v.i_mode) && dir) {
1287 		struct bcachefs_fid_with_parent *fid = (void *) fh;
1288 
1289 		min_len = sizeof(*fid) / sizeof(u32);
1290 		if (*len < min_len) {
1291 			*len = min_len;
1292 			return FILEID_INVALID;
1293 		}
1294 
1295 		fid->fid = bch2_inode_to_fid(inode);
1296 		fid->dir = bch2_inode_to_fid(dir);
1297 
1298 		*len = min_len;
1299 		return FILEID_BCACHEFS_WITH_PARENT;
1300 	} else {
1301 		struct bcachefs_fid *fid = (void *) fh;
1302 
1303 		min_len = sizeof(*fid) / sizeof(u32);
1304 		if (*len < min_len) {
1305 			*len = min_len;
1306 			return FILEID_INVALID;
1307 		}
1308 		*fid = bch2_inode_to_fid(inode);
1309 
1310 		*len = min_len;
1311 		return FILEID_BCACHEFS_WITHOUT_PARENT;
1312 	}
1313 }
1314 
1315 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1316 					struct bcachefs_fid fid)
1317 {
1318 	struct bch_fs *c = sb->s_fs_info;
1319 	struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1320 				    .subvol = fid.subvol,
1321 				    .inum = fid.inum,
1322 	});
1323 	if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1324 		iput(vinode);
1325 		vinode = ERR_PTR(-ESTALE);
1326 	}
1327 	return vinode;
1328 }
1329 
1330 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1331 		int fh_len, int fh_type)
1332 {
1333 	struct bcachefs_fid *fid = (void *) _fid;
1334 
1335 	if (!bcachefs_fid_valid(fh_len, fh_type))
1336 		return NULL;
1337 
1338 	return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1339 }
1340 
1341 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1342 		int fh_len, int fh_type)
1343 {
1344 	struct bcachefs_fid_with_parent *fid = (void *) _fid;
1345 
1346 	if (!bcachefs_fid_valid(fh_len, fh_type) ||
1347 	    fh_type != FILEID_BCACHEFS_WITH_PARENT)
1348 		return NULL;
1349 
1350 	return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1351 }
1352 
1353 static struct dentry *bch2_get_parent(struct dentry *child)
1354 {
1355 	struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1356 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1357 	subvol_inum parent_inum = {
1358 		.subvol = inode->ei_inode.bi_parent_subvol ?:
1359 			inode->ei_subvol,
1360 		.inum = inode->ei_inode.bi_dir,
1361 	};
1362 
1363 	return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1364 }
1365 
1366 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1367 {
1368 	struct bch_inode_info *inode	= to_bch_ei(child->d_inode);
1369 	struct bch_inode_info *dir	= to_bch_ei(parent->d_inode);
1370 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1371 	struct btree_trans *trans;
1372 	struct btree_iter iter1;
1373 	struct btree_iter iter2;
1374 	struct bkey_s_c k;
1375 	struct bkey_s_c_dirent d;
1376 	struct bch_inode_unpacked inode_u;
1377 	subvol_inum target;
1378 	u32 snapshot;
1379 	struct qstr dirent_name;
1380 	unsigned name_len = 0;
1381 	int ret;
1382 
1383 	if (!S_ISDIR(dir->v.i_mode))
1384 		return -EINVAL;
1385 
1386 	trans = bch2_trans_get(c);
1387 
1388 	bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1389 			     POS(dir->ei_inode.bi_inum, 0), 0);
1390 	bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1391 			     POS(dir->ei_inode.bi_inum, 0), 0);
1392 retry:
1393 	bch2_trans_begin(trans);
1394 
1395 	ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1396 	if (ret)
1397 		goto err;
1398 
1399 	bch2_btree_iter_set_snapshot(&iter1, snapshot);
1400 	bch2_btree_iter_set_snapshot(&iter2, snapshot);
1401 
1402 	ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1403 	if (ret)
1404 		goto err;
1405 
1406 	if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1407 		bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1408 
1409 		k = bch2_btree_iter_peek_slot(&iter1);
1410 		ret = bkey_err(k);
1411 		if (ret)
1412 			goto err;
1413 
1414 		if (k.k->type != KEY_TYPE_dirent) {
1415 			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1416 			goto err;
1417 		}
1418 
1419 		d = bkey_s_c_to_dirent(k);
1420 		ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1421 		if (ret > 0)
1422 			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1423 		if (ret)
1424 			goto err;
1425 
1426 		if (target.subvol	== inode->ei_subvol &&
1427 		    target.inum		== inode->ei_inode.bi_inum)
1428 			goto found;
1429 	} else {
1430 		/*
1431 		 * File with multiple hardlinks and our backref is to the wrong
1432 		 * directory - linear search:
1433 		 */
1434 		for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1435 			if (k.k->p.inode > dir->ei_inode.bi_inum)
1436 				break;
1437 
1438 			if (k.k->type != KEY_TYPE_dirent)
1439 				continue;
1440 
1441 			d = bkey_s_c_to_dirent(k);
1442 			ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1443 			if (ret < 0)
1444 				break;
1445 			if (ret)
1446 				continue;
1447 
1448 			if (target.subvol	== inode->ei_subvol &&
1449 			    target.inum		== inode->ei_inode.bi_inum)
1450 				goto found;
1451 		}
1452 	}
1453 
1454 	ret = -ENOENT;
1455 	goto err;
1456 found:
1457 	dirent_name = bch2_dirent_get_name(d);
1458 
1459 	name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1460 	memcpy(name, dirent_name.name, name_len);
1461 	name[name_len] = '\0';
1462 err:
1463 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1464 		goto retry;
1465 
1466 	bch2_trans_iter_exit(trans, &iter1);
1467 	bch2_trans_iter_exit(trans, &iter2);
1468 	bch2_trans_put(trans);
1469 
1470 	return ret;
1471 }
1472 
1473 static const struct export_operations bch_export_ops = {
1474 	.encode_fh	= bch2_encode_fh,
1475 	.fh_to_dentry	= bch2_fh_to_dentry,
1476 	.fh_to_parent	= bch2_fh_to_parent,
1477 	.get_parent	= bch2_get_parent,
1478 	.get_name	= bch2_get_name,
1479 };
1480 
1481 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1482 				struct bch_inode_info *inode,
1483 				struct bch_inode_unpacked *bi,
1484 				struct bch_subvolume *subvol)
1485 {
1486 	bch2_iget5_set(&inode->v, &inum);
1487 	bch2_inode_update_after_write(trans, inode, bi, ~0);
1488 
1489 	if (BCH_SUBVOLUME_SNAP(subvol))
1490 		set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1491 	else
1492 		clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1493 
1494 	inode->v.i_blocks	= bi->bi_sectors;
1495 	inode->v.i_ino		= bi->bi_inum;
1496 	inode->v.i_rdev		= bi->bi_dev;
1497 	inode->v.i_generation	= bi->bi_generation;
1498 	inode->v.i_size		= bi->bi_size;
1499 
1500 	inode->ei_flags		= 0;
1501 	inode->ei_quota_reserved = 0;
1502 	inode->ei_qid		= bch_qid(bi);
1503 	inode->ei_subvol	= inum.subvol;
1504 
1505 	inode->v.i_mapping->a_ops = &bch_address_space_operations;
1506 
1507 	switch (inode->v.i_mode & S_IFMT) {
1508 	case S_IFREG:
1509 		inode->v.i_op	= &bch_file_inode_operations;
1510 		inode->v.i_fop	= &bch_file_operations;
1511 		break;
1512 	case S_IFDIR:
1513 		inode->v.i_op	= &bch_dir_inode_operations;
1514 		inode->v.i_fop	= &bch_dir_file_operations;
1515 		break;
1516 	case S_IFLNK:
1517 		inode_nohighmem(&inode->v);
1518 		inode->v.i_op	= &bch_symlink_inode_operations;
1519 		break;
1520 	default:
1521 		init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1522 		inode->v.i_op	= &bch_special_inode_operations;
1523 		break;
1524 	}
1525 
1526 	mapping_set_large_folios(inode->v.i_mapping);
1527 }
1528 
1529 static void bch2_free_inode(struct inode *vinode)
1530 {
1531 	kmem_cache_free(bch2_inode_cache, to_bch_ei(vinode));
1532 }
1533 
1534 static int inode_update_times_fn(struct btree_trans *trans,
1535 				 struct bch_inode_info *inode,
1536 				 struct bch_inode_unpacked *bi,
1537 				 void *p)
1538 {
1539 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1540 
1541 	bi->bi_atime	= timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1542 	bi->bi_mtime	= timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1543 	bi->bi_ctime	= timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1544 
1545 	return 0;
1546 }
1547 
1548 static int bch2_vfs_write_inode(struct inode *vinode,
1549 				struct writeback_control *wbc)
1550 {
1551 	struct bch_fs *c = vinode->i_sb->s_fs_info;
1552 	struct bch_inode_info *inode = to_bch_ei(vinode);
1553 	int ret;
1554 
1555 	mutex_lock(&inode->ei_update_lock);
1556 	ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1557 			       ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1558 	mutex_unlock(&inode->ei_update_lock);
1559 
1560 	return bch2_err_class(ret);
1561 }
1562 
1563 static void bch2_evict_inode(struct inode *vinode)
1564 {
1565 	struct bch_fs *c = vinode->i_sb->s_fs_info;
1566 	struct bch_inode_info *inode = to_bch_ei(vinode);
1567 
1568 	truncate_inode_pages_final(&inode->v.i_data);
1569 
1570 	clear_inode(&inode->v);
1571 
1572 	BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1573 
1574 	if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1575 		bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1576 				KEY_TYPE_QUOTA_WARN);
1577 		bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1578 				KEY_TYPE_QUOTA_WARN);
1579 		bch2_inode_rm(c, inode_inum(inode));
1580 	}
1581 
1582 	mutex_lock(&c->vfs_inodes_lock);
1583 	list_del_init(&inode->ei_vfs_inode_list);
1584 	mutex_unlock(&c->vfs_inodes_lock);
1585 }
1586 
1587 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1588 {
1589 	struct bch_inode_info *inode;
1590 	DARRAY(struct bch_inode_info *) grabbed;
1591 	bool clean_pass = false, this_pass_clean;
1592 
1593 	/*
1594 	 * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1595 	 * be pruned with d_mark_dontcache().
1596 	 *
1597 	 * Once we've had a clean pass where we didn't find any inodes without
1598 	 * I_DONTCACHE, we wait for them to be freed:
1599 	 */
1600 
1601 	darray_init(&grabbed);
1602 	darray_make_room(&grabbed, 1024);
1603 again:
1604 	cond_resched();
1605 	this_pass_clean = true;
1606 
1607 	mutex_lock(&c->vfs_inodes_lock);
1608 	list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1609 		if (!snapshot_list_has_id(s, inode->ei_subvol))
1610 			continue;
1611 
1612 		if (!(inode->v.i_state & I_DONTCACHE) &&
1613 		    !(inode->v.i_state & I_FREEING) &&
1614 		    igrab(&inode->v)) {
1615 			this_pass_clean = false;
1616 
1617 			if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1618 				iput(&inode->v);
1619 				break;
1620 			}
1621 		} else if (clean_pass && this_pass_clean) {
1622 			wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1623 			DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1624 
1625 			prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1626 			mutex_unlock(&c->vfs_inodes_lock);
1627 
1628 			schedule();
1629 			finish_wait(wq, &wait.wq_entry);
1630 			goto again;
1631 		}
1632 	}
1633 	mutex_unlock(&c->vfs_inodes_lock);
1634 
1635 	darray_for_each(grabbed, i) {
1636 		inode = *i;
1637 		d_mark_dontcache(&inode->v);
1638 		d_prune_aliases(&inode->v);
1639 		iput(&inode->v);
1640 	}
1641 	grabbed.nr = 0;
1642 
1643 	if (!clean_pass || !this_pass_clean) {
1644 		clean_pass = this_pass_clean;
1645 		goto again;
1646 	}
1647 
1648 	darray_exit(&grabbed);
1649 }
1650 
1651 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1652 {
1653 	struct super_block *sb = dentry->d_sb;
1654 	struct bch_fs *c = sb->s_fs_info;
1655 	struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1656 	unsigned shift = sb->s_blocksize_bits - 9;
1657 	/*
1658 	 * this assumes inodes take up 64 bytes, which is a decent average
1659 	 * number:
1660 	 */
1661 	u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1662 
1663 	buf->f_type	= BCACHEFS_STATFS_MAGIC;
1664 	buf->f_bsize	= sb->s_blocksize;
1665 	buf->f_blocks	= usage.capacity >> shift;
1666 	buf->f_bfree	= usage.free >> shift;
1667 	buf->f_bavail	= avail_factor(usage.free) >> shift;
1668 
1669 	buf->f_files	= usage.nr_inodes + avail_inodes;
1670 	buf->f_ffree	= avail_inodes;
1671 
1672 	buf->f_fsid	= uuid_to_fsid(c->sb.user_uuid.b);
1673 	buf->f_namelen	= BCH_NAME_MAX;
1674 
1675 	return 0;
1676 }
1677 
1678 static int bch2_sync_fs(struct super_block *sb, int wait)
1679 {
1680 	struct bch_fs *c = sb->s_fs_info;
1681 	int ret;
1682 
1683 	if (c->opts.journal_flush_disabled)
1684 		return 0;
1685 
1686 	if (!wait) {
1687 		bch2_journal_flush_async(&c->journal, NULL);
1688 		return 0;
1689 	}
1690 
1691 	ret = bch2_journal_flush(&c->journal);
1692 	return bch2_err_class(ret);
1693 }
1694 
1695 static struct bch_fs *bch2_path_to_fs(const char *path)
1696 {
1697 	struct bch_fs *c;
1698 	dev_t dev;
1699 	int ret;
1700 
1701 	ret = lookup_bdev(path, &dev);
1702 	if (ret)
1703 		return ERR_PTR(ret);
1704 
1705 	c = bch2_dev_to_fs(dev);
1706 	if (c)
1707 		closure_put(&c->cl);
1708 	return c ?: ERR_PTR(-ENOENT);
1709 }
1710 
1711 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1712 {
1713 	struct bch_fs *c = sb->s_fs_info;
1714 	struct bch_opts opts = bch2_opts_empty();
1715 	int ret;
1716 
1717 	ret = bch2_parse_mount_opts(c, &opts, data);
1718 	if (ret)
1719 		goto err;
1720 
1721 	opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1722 
1723 	if (opts.read_only != c->opts.read_only) {
1724 		down_write(&c->state_lock);
1725 
1726 		if (opts.read_only) {
1727 			bch2_fs_read_only(c);
1728 
1729 			sb->s_flags |= SB_RDONLY;
1730 		} else {
1731 			ret = bch2_fs_read_write(c);
1732 			if (ret) {
1733 				bch_err(c, "error going rw: %i", ret);
1734 				up_write(&c->state_lock);
1735 				ret = -EINVAL;
1736 				goto err;
1737 			}
1738 
1739 			sb->s_flags &= ~SB_RDONLY;
1740 		}
1741 
1742 		c->opts.read_only = opts.read_only;
1743 
1744 		up_write(&c->state_lock);
1745 	}
1746 
1747 	if (opt_defined(opts, errors))
1748 		c->opts.errors = opts.errors;
1749 err:
1750 	return bch2_err_class(ret);
1751 }
1752 
1753 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1754 {
1755 	struct bch_fs *c = root->d_sb->s_fs_info;
1756 	bool first = true;
1757 
1758 	for_each_online_member(c, ca) {
1759 		if (!first)
1760 			seq_putc(seq, ':');
1761 		first = false;
1762 		seq_puts(seq, ca->disk_sb.sb_name);
1763 	}
1764 
1765 	return 0;
1766 }
1767 
1768 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1769 {
1770 	struct bch_fs *c = root->d_sb->s_fs_info;
1771 	enum bch_opt_id i;
1772 	struct printbuf buf = PRINTBUF;
1773 	int ret = 0;
1774 
1775 	for (i = 0; i < bch2_opts_nr; i++) {
1776 		const struct bch_option *opt = &bch2_opt_table[i];
1777 		u64 v = bch2_opt_get_by_id(&c->opts, i);
1778 
1779 		if (!(opt->flags & OPT_MOUNT))
1780 			continue;
1781 
1782 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1783 			continue;
1784 
1785 		printbuf_reset(&buf);
1786 		bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1787 				 OPT_SHOW_MOUNT_STYLE);
1788 		seq_putc(seq, ',');
1789 		seq_puts(seq, buf.buf);
1790 	}
1791 
1792 	if (buf.allocation_failure)
1793 		ret = -ENOMEM;
1794 	printbuf_exit(&buf);
1795 	return ret;
1796 }
1797 
1798 static void bch2_put_super(struct super_block *sb)
1799 {
1800 	struct bch_fs *c = sb->s_fs_info;
1801 
1802 	__bch2_fs_stop(c);
1803 }
1804 
1805 /*
1806  * bcachefs doesn't currently integrate intwrite freeze protection but the
1807  * internal write references serve the same purpose. Therefore reuse the
1808  * read-only transition code to perform the quiesce. The caveat is that we don't
1809  * currently have the ability to block tasks that want a write reference while
1810  * the superblock is frozen. This is fine for now, but we should either add
1811  * blocking support or find a way to integrate sb_start_intwrite() and friends.
1812  */
1813 static int bch2_freeze(struct super_block *sb)
1814 {
1815 	struct bch_fs *c = sb->s_fs_info;
1816 
1817 	down_write(&c->state_lock);
1818 	bch2_fs_read_only(c);
1819 	up_write(&c->state_lock);
1820 	return 0;
1821 }
1822 
1823 static int bch2_unfreeze(struct super_block *sb)
1824 {
1825 	struct bch_fs *c = sb->s_fs_info;
1826 	int ret;
1827 
1828 	if (test_bit(BCH_FS_emergency_ro, &c->flags))
1829 		return 0;
1830 
1831 	down_write(&c->state_lock);
1832 	ret = bch2_fs_read_write(c);
1833 	up_write(&c->state_lock);
1834 	return ret;
1835 }
1836 
1837 static const struct super_operations bch_super_operations = {
1838 	.alloc_inode	= bch2_alloc_inode,
1839 	.free_inode	= bch2_free_inode,
1840 	.write_inode	= bch2_vfs_write_inode,
1841 	.evict_inode	= bch2_evict_inode,
1842 	.sync_fs	= bch2_sync_fs,
1843 	.statfs		= bch2_statfs,
1844 	.show_devname	= bch2_show_devname,
1845 	.show_options	= bch2_show_options,
1846 	.remount_fs	= bch2_remount,
1847 	.put_super	= bch2_put_super,
1848 	.freeze_fs	= bch2_freeze,
1849 	.unfreeze_fs	= bch2_unfreeze,
1850 };
1851 
1852 static int bch2_set_super(struct super_block *s, void *data)
1853 {
1854 	s->s_fs_info = data;
1855 	return 0;
1856 }
1857 
1858 static int bch2_noset_super(struct super_block *s, void *data)
1859 {
1860 	return -EBUSY;
1861 }
1862 
1863 typedef DARRAY(struct bch_fs *) darray_fs;
1864 
1865 static int bch2_test_super(struct super_block *s, void *data)
1866 {
1867 	struct bch_fs *c = s->s_fs_info;
1868 	darray_fs *d = data;
1869 
1870 	if (!c)
1871 		return false;
1872 
1873 	darray_for_each(*d, i)
1874 		if (c != *i)
1875 			return false;
1876 	return true;
1877 }
1878 
1879 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1880 				 int flags, const char *dev_name, void *data)
1881 {
1882 	struct bch_fs *c;
1883 	struct super_block *sb;
1884 	struct inode *vinode;
1885 	struct bch_opts opts = bch2_opts_empty();
1886 	int ret;
1887 
1888 	opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1889 
1890 	ret = bch2_parse_mount_opts(NULL, &opts, data);
1891 	if (ret) {
1892 		ret = bch2_err_class(ret);
1893 		return ERR_PTR(ret);
1894 	}
1895 
1896 	if (!dev_name || strlen(dev_name) == 0)
1897 		return ERR_PTR(-EINVAL);
1898 
1899 	darray_str devs;
1900 	ret = bch2_split_devs(dev_name, &devs);
1901 	if (ret)
1902 		return ERR_PTR(ret);
1903 
1904 	darray_fs devs_to_fs = {};
1905 	darray_for_each(devs, i) {
1906 		ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
1907 		if (ret) {
1908 			sb = ERR_PTR(ret);
1909 			goto got_sb;
1910 		}
1911 	}
1912 
1913 	sb = sget(fs_type, bch2_test_super, bch2_noset_super, flags|SB_NOSEC, &devs_to_fs);
1914 	if (!IS_ERR(sb))
1915 		goto got_sb;
1916 
1917 	c = bch2_fs_open(devs.data, devs.nr, opts);
1918 	if (IS_ERR(c)) {
1919 		sb = ERR_CAST(c);
1920 		goto got_sb;
1921 	}
1922 
1923 	/* Some options can't be parsed until after the fs is started: */
1924 	ret = bch2_parse_mount_opts(c, &opts, data);
1925 	if (ret) {
1926 		bch2_fs_stop(c);
1927 		sb = ERR_PTR(ret);
1928 		goto got_sb;
1929 	}
1930 
1931 	bch2_opts_apply(&c->opts, opts);
1932 
1933 	sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1934 	if (IS_ERR(sb))
1935 		bch2_fs_stop(c);
1936 got_sb:
1937 	darray_exit(&devs_to_fs);
1938 	bch2_darray_str_exit(&devs);
1939 
1940 	if (IS_ERR(sb)) {
1941 		ret = PTR_ERR(sb);
1942 		ret = bch2_err_class(ret);
1943 		return ERR_PTR(ret);
1944 	}
1945 
1946 	c = sb->s_fs_info;
1947 
1948 	if (sb->s_root) {
1949 		if ((flags ^ sb->s_flags) & SB_RDONLY) {
1950 			ret = -EBUSY;
1951 			goto err_put_super;
1952 		}
1953 		goto out;
1954 	}
1955 
1956 	sb->s_blocksize		= block_bytes(c);
1957 	sb->s_blocksize_bits	= ilog2(block_bytes(c));
1958 	sb->s_maxbytes		= MAX_LFS_FILESIZE;
1959 	sb->s_op		= &bch_super_operations;
1960 	sb->s_export_op		= &bch_export_ops;
1961 #ifdef CONFIG_BCACHEFS_QUOTA
1962 	sb->s_qcop		= &bch2_quotactl_operations;
1963 	sb->s_quota_types	= QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1964 #endif
1965 	sb->s_xattr		= bch2_xattr_handlers;
1966 	sb->s_magic		= BCACHEFS_STATFS_MAGIC;
1967 	sb->s_time_gran		= c->sb.nsec_per_time_unit;
1968 	sb->s_time_min		= div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1969 	sb->s_time_max		= div_s64(S64_MAX, c->sb.time_units_per_sec);
1970 	sb->s_uuid		= c->sb.user_uuid;
1971 	c->vfs_sb		= sb;
1972 	strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1973 
1974 	ret = super_setup_bdi(sb);
1975 	if (ret)
1976 		goto err_put_super;
1977 
1978 	sb->s_bdi->ra_pages		= VM_READAHEAD_PAGES;
1979 
1980 	for_each_online_member(c, ca) {
1981 		struct block_device *bdev = ca->disk_sb.bdev;
1982 
1983 		/* XXX: create an anonymous device for multi device filesystems */
1984 		sb->s_bdev	= bdev;
1985 		sb->s_dev	= bdev->bd_dev;
1986 		percpu_ref_put(&ca->io_ref);
1987 		break;
1988 	}
1989 
1990 	c->dev = sb->s_dev;
1991 
1992 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1993 	if (c->opts.acl)
1994 		sb->s_flags	|= SB_POSIXACL;
1995 #endif
1996 
1997 	sb->s_shrink->seeks = 0;
1998 
1999 	vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
2000 	ret = PTR_ERR_OR_ZERO(vinode);
2001 	bch_err_msg(c, ret, "mounting: error getting root inode");
2002 	if (ret)
2003 		goto err_put_super;
2004 
2005 	sb->s_root = d_make_root(vinode);
2006 	if (!sb->s_root) {
2007 		bch_err(c, "error mounting: error allocating root dentry");
2008 		ret = -ENOMEM;
2009 		goto err_put_super;
2010 	}
2011 
2012 	sb->s_flags |= SB_ACTIVE;
2013 out:
2014 	return dget(sb->s_root);
2015 
2016 err_put_super:
2017 	__bch2_fs_stop(c);
2018 	deactivate_locked_super(sb);
2019 	return ERR_PTR(bch2_err_class(ret));
2020 }
2021 
2022 static void bch2_kill_sb(struct super_block *sb)
2023 {
2024 	struct bch_fs *c = sb->s_fs_info;
2025 
2026 	generic_shutdown_super(sb);
2027 	bch2_fs_free(c);
2028 }
2029 
2030 static struct file_system_type bcache_fs_type = {
2031 	.owner		= THIS_MODULE,
2032 	.name		= "bcachefs",
2033 	.mount		= bch2_mount,
2034 	.kill_sb	= bch2_kill_sb,
2035 	.fs_flags	= FS_REQUIRES_DEV,
2036 };
2037 
2038 MODULE_ALIAS_FS("bcachefs");
2039 
2040 void bch2_vfs_exit(void)
2041 {
2042 	unregister_filesystem(&bcache_fs_type);
2043 	kmem_cache_destroy(bch2_inode_cache);
2044 }
2045 
2046 int __init bch2_vfs_init(void)
2047 {
2048 	int ret = -ENOMEM;
2049 
2050 	bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
2051 	if (!bch2_inode_cache)
2052 		goto err;
2053 
2054 	ret = register_filesystem(&bcache_fs_type);
2055 	if (ret)
2056 		goto err;
2057 
2058 	return 0;
2059 err:
2060 	bch2_vfs_exit();
2061 	return ret;
2062 }
2063 
2064 #endif /* NO_BCACHEFS_FS */
2065