xref: /linux/fs/bcachefs/fs.c (revision 3c2d73de49be528276474c1a53f78b38ee11c1fa)
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 #include "trace.h"
30 
31 #include <linux/aio.h>
32 #include <linux/backing-dev.h>
33 #include <linux/exportfs.h>
34 #include <linux/fiemap.h>
35 #include <linux/fs_context.h>
36 #include <linux/module.h>
37 #include <linux/pagemap.h>
38 #include <linux/posix_acl.h>
39 #include <linux/random.h>
40 #include <linux/seq_file.h>
41 #include <linux/statfs.h>
42 #include <linux/string.h>
43 #include <linux/xattr.h>
44 
45 static struct kmem_cache *bch2_inode_cache;
46 
47 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
48 				struct bch_inode_info *,
49 				struct bch_inode_unpacked *,
50 				struct bch_subvolume *);
51 
52 void bch2_inode_update_after_write(struct btree_trans *trans,
53 				   struct bch_inode_info *inode,
54 				   struct bch_inode_unpacked *bi,
55 				   unsigned fields)
56 {
57 	struct bch_fs *c = trans->c;
58 
59 	BUG_ON(bi->bi_inum != inode->v.i_ino);
60 
61 	bch2_assert_pos_locked(trans, BTREE_ID_inodes, POS(0, bi->bi_inum));
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 %llu:%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 bool subvol_inum_eq(subvol_inum a, subvol_inum b)
156 {
157 	return a.subvol == b.subvol && a.inum == b.inum;
158 }
159 
160 static int bch2_vfs_inode_cmp_fn(struct rhashtable_compare_arg *arg,
161 				 const void *obj)
162 {
163 	const struct bch_inode_info *inode = obj;
164 	const subvol_inum *v = arg->key;
165 
166 	return !subvol_inum_eq(inode->ei_inum, *v);
167 }
168 
169 static const struct rhashtable_params bch2_vfs_inodes_params = {
170 	.head_offset		= offsetof(struct bch_inode_info, hash),
171 	.key_offset		= offsetof(struct bch_inode_info, ei_inum),
172 	.key_len		= sizeof(subvol_inum),
173 	.obj_cmpfn		= bch2_vfs_inode_cmp_fn,
174 	.automatic_shrinking	= true,
175 };
176 
177 struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *c, subvol_inum inum)
178 {
179 	return rhashtable_lookup_fast(&c->vfs_inodes_table, &inum, bch2_vfs_inodes_params);
180 }
181 
182 static void __wait_on_freeing_inode(struct bch_fs *c,
183 				    struct bch_inode_info *inode,
184 				    subvol_inum inum)
185 {
186 	wait_queue_head_t *wq;
187 	DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
188 	wq = inode_bit_waitqueue(&wait, &inode->v, __I_NEW);
189 	prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
190 	spin_unlock(&inode->v.i_lock);
191 
192 	if (__bch2_inode_hash_find(c, inum) == inode)
193 		schedule_timeout(HZ * 10);
194 	finish_wait(wq, &wait.wq_entry);
195 }
196 
197 static struct bch_inode_info *bch2_inode_hash_find(struct bch_fs *c, struct btree_trans *trans,
198 						   subvol_inum inum)
199 {
200 	struct bch_inode_info *inode;
201 repeat:
202 	inode = __bch2_inode_hash_find(c, inum);
203 	if (inode) {
204 		spin_lock(&inode->v.i_lock);
205 		if (!test_bit(EI_INODE_HASHED, &inode->ei_flags)) {
206 			spin_unlock(&inode->v.i_lock);
207 			return NULL;
208 		}
209 		if ((inode->v.i_state & (I_FREEING|I_WILL_FREE))) {
210 			if (!trans) {
211 				__wait_on_freeing_inode(c, inode, inum);
212 			} else {
213 				bch2_trans_unlock(trans);
214 				__wait_on_freeing_inode(c, inode, inum);
215 				int ret = bch2_trans_relock(trans);
216 				if (ret)
217 					return ERR_PTR(ret);
218 			}
219 			goto repeat;
220 		}
221 		__iget(&inode->v);
222 		spin_unlock(&inode->v.i_lock);
223 	}
224 
225 	return inode;
226 }
227 
228 static void bch2_inode_hash_remove(struct bch_fs *c, struct bch_inode_info *inode)
229 {
230 	spin_lock(&inode->v.i_lock);
231 	bool remove = test_and_clear_bit(EI_INODE_HASHED, &inode->ei_flags);
232 	spin_unlock(&inode->v.i_lock);
233 
234 	if (remove) {
235 		int ret = rhashtable_remove_fast(&c->vfs_inodes_table,
236 					&inode->hash, bch2_vfs_inodes_params);
237 		BUG_ON(ret);
238 		inode->v.i_hash.pprev = NULL;
239 		/*
240 		 * This pairs with the bch2_inode_hash_find() ->
241 		 * __wait_on_freeing_inode() path
242 		 */
243 		inode_wake_up_bit(&inode->v, __I_NEW);
244 	}
245 }
246 
247 static struct bch_inode_info *bch2_inode_hash_insert(struct bch_fs *c,
248 						     struct btree_trans *trans,
249 						     struct bch_inode_info *inode)
250 {
251 	struct bch_inode_info *old = inode;
252 
253 	set_bit(EI_INODE_HASHED, &inode->ei_flags);
254 retry:
255 	if (unlikely(rhashtable_lookup_insert_fast(&c->vfs_inodes_table,
256 					&inode->hash,
257 					bch2_vfs_inodes_params))) {
258 		old = bch2_inode_hash_find(c, trans, inode->ei_inum);
259 		if (!old)
260 			goto retry;
261 
262 		clear_bit(EI_INODE_HASHED, &inode->ei_flags);
263 
264 		/*
265 		 * bcachefs doesn't use I_NEW; we have no use for it since we
266 		 * only insert fully created inodes in the inode hash table. But
267 		 * discard_new_inode() expects it to be set...
268 		 */
269 		inode->v.i_state |= I_NEW;
270 		/*
271 		 * We don't want bch2_evict_inode() to delete the inode on disk,
272 		 * we just raced and had another inode in cache. Normally new
273 		 * inodes don't have nlink == 0 - except tmpfiles do...
274 		 */
275 		set_nlink(&inode->v, 1);
276 		discard_new_inode(&inode->v);
277 		return old;
278 	} else {
279 		inode_fake_hash(&inode->v);
280 
281 		inode_sb_list_add(&inode->v);
282 
283 		mutex_lock(&c->vfs_inodes_lock);
284 		list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
285 		mutex_unlock(&c->vfs_inodes_lock);
286 		return inode;
287 	}
288 }
289 
290 #define memalloc_flags_do(_flags, _do)						\
291 ({										\
292 	unsigned _saved_flags = memalloc_flags_save(_flags);			\
293 	typeof(_do) _ret = _do;							\
294 	memalloc_noreclaim_restore(_saved_flags);				\
295 	_ret;									\
296 })
297 
298 static struct inode *bch2_alloc_inode(struct super_block *sb)
299 {
300 	BUG();
301 }
302 
303 static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
304 {
305 	struct bch_inode_info *inode = alloc_inode_sb(c->vfs_sb,
306 						bch2_inode_cache, GFP_NOFS);
307 	if (!inode)
308 		return NULL;
309 
310 	inode_init_once(&inode->v);
311 	mutex_init(&inode->ei_update_lock);
312 	two_state_lock_init(&inode->ei_pagecache_lock);
313 	INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
314 	inode->ei_flags = 0;
315 	mutex_init(&inode->ei_quota_lock);
316 	memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
317 
318 	if (unlikely(inode_init_always(c->vfs_sb, &inode->v))) {
319 		kmem_cache_free(bch2_inode_cache, inode);
320 		return NULL;
321 	}
322 
323 	return inode;
324 }
325 
326 /*
327  * Allocate a new inode, dropping/retaking btree locks if necessary:
328  */
329 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
330 {
331 	struct bch_inode_info *inode =
332 		memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
333 				  __bch2_new_inode(trans->c));
334 
335 	if (unlikely(!inode)) {
336 		int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c)) ? 0 : -ENOMEM);
337 		if (ret && inode) {
338 			__destroy_inode(&inode->v);
339 			kmem_cache_free(bch2_inode_cache, inode);
340 		}
341 		if (ret)
342 			return ERR_PTR(ret);
343 	}
344 
345 	return inode;
346 }
347 
348 static struct bch_inode_info *bch2_inode_hash_init_insert(struct btree_trans *trans,
349 							  subvol_inum inum,
350 							  struct bch_inode_unpacked *bi,
351 							  struct bch_subvolume *subvol)
352 {
353 	struct bch_inode_info *inode = bch2_new_inode(trans);
354 	if (IS_ERR(inode))
355 		return inode;
356 
357 	bch2_vfs_inode_init(trans, inum, inode, bi, subvol);
358 
359 	return bch2_inode_hash_insert(trans->c, trans, inode);
360 
361 }
362 
363 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
364 {
365 	struct bch_inode_info *inode = bch2_inode_hash_find(c, NULL, inum);
366 	if (inode)
367 		return &inode->v;
368 
369 	struct btree_trans *trans = bch2_trans_get(c);
370 
371 	struct bch_inode_unpacked inode_u;
372 	struct bch_subvolume subvol;
373 	int ret = lockrestart_do(trans,
374 		bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
375 		bch2_inode_find_by_inum_trans(trans, inum, &inode_u)) ?:
376 		PTR_ERR_OR_ZERO(inode = bch2_inode_hash_init_insert(trans, inum, &inode_u, &subvol));
377 	bch2_trans_put(trans);
378 
379 	return ret ? ERR_PTR(ret) : &inode->v;
380 }
381 
382 struct bch_inode_info *
383 __bch2_create(struct mnt_idmap *idmap,
384 	      struct bch_inode_info *dir, struct dentry *dentry,
385 	      umode_t mode, dev_t rdev, subvol_inum snapshot_src,
386 	      unsigned flags)
387 {
388 	struct bch_fs *c = dir->v.i_sb->s_fs_info;
389 	struct btree_trans *trans;
390 	struct bch_inode_unpacked dir_u;
391 	struct bch_inode_info *inode;
392 	struct bch_inode_unpacked inode_u;
393 	struct posix_acl *default_acl = NULL, *acl = NULL;
394 	subvol_inum inum;
395 	struct bch_subvolume subvol;
396 	u64 journal_seq = 0;
397 	kuid_t kuid;
398 	kgid_t kgid;
399 	int ret;
400 
401 	/*
402 	 * preallocate acls + vfs inode before btree transaction, so that
403 	 * nothing can fail after the transaction succeeds:
404 	 */
405 #ifdef CONFIG_BCACHEFS_POSIX_ACL
406 	ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
407 	if (ret)
408 		return ERR_PTR(ret);
409 #endif
410 	inode = __bch2_new_inode(c);
411 	if (unlikely(!inode)) {
412 		inode = ERR_PTR(-ENOMEM);
413 		goto err;
414 	}
415 
416 	bch2_inode_init_early(c, &inode_u);
417 
418 	if (!(flags & BCH_CREATE_TMPFILE))
419 		mutex_lock(&dir->ei_update_lock);
420 
421 	trans = bch2_trans_get(c);
422 retry:
423 	bch2_trans_begin(trans);
424 
425 	kuid = mapped_fsuid(idmap, i_user_ns(&dir->v));
426 	kgid = mapped_fsgid(idmap, i_user_ns(&dir->v));
427 	ret   = bch2_subvol_is_ro_trans(trans, dir->ei_inum.subvol) ?:
428 		bch2_create_trans(trans,
429 				  inode_inum(dir), &dir_u, &inode_u,
430 				  !(flags & BCH_CREATE_TMPFILE)
431 				  ? &dentry->d_name : NULL,
432 				  from_kuid(i_user_ns(&dir->v), kuid),
433 				  from_kgid(i_user_ns(&dir->v), kgid),
434 				  mode, rdev,
435 				  default_acl, acl, snapshot_src, flags) ?:
436 		bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
437 				KEY_TYPE_QUOTA_PREALLOC);
438 	if (unlikely(ret))
439 		goto err_before_quota;
440 
441 	inum.subvol = inode_u.bi_subvol ?: dir->ei_inum.subvol;
442 	inum.inum = inode_u.bi_inum;
443 
444 	ret   = bch2_subvolume_get(trans, inum.subvol, true,
445 				   BTREE_ITER_with_updates, &subvol) ?:
446 		bch2_trans_commit(trans, NULL, &journal_seq, 0);
447 	if (unlikely(ret)) {
448 		bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
449 				KEY_TYPE_QUOTA_WARN);
450 err_before_quota:
451 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
452 			goto retry;
453 		goto err_trans;
454 	}
455 
456 	if (!(flags & BCH_CREATE_TMPFILE)) {
457 		bch2_inode_update_after_write(trans, dir, &dir_u,
458 					      ATTR_MTIME|ATTR_CTIME);
459 		mutex_unlock(&dir->ei_update_lock);
460 	}
461 
462 	bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
463 
464 	set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
465 	set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
466 
467 	/*
468 	 * we must insert the new inode into the inode cache before calling
469 	 * bch2_trans_exit() and dropping locks, else we could race with another
470 	 * thread pulling the inode in and modifying it:
471 	 *
472 	 * also, calling bch2_inode_hash_insert() without passing in the
473 	 * transaction object is sketchy - if we could ever end up in
474 	 * __wait_on_freeing_inode(), we'd risk deadlock.
475 	 *
476 	 * But that shouldn't be possible, since we still have the inode locked
477 	 * that we just created, and we _really_ can't take a transaction
478 	 * restart here.
479 	 */
480 	inode = bch2_inode_hash_insert(c, NULL, inode);
481 	bch2_trans_put(trans);
482 err:
483 	posix_acl_release(default_acl);
484 	posix_acl_release(acl);
485 	return inode;
486 err_trans:
487 	if (!(flags & BCH_CREATE_TMPFILE))
488 		mutex_unlock(&dir->ei_update_lock);
489 
490 	bch2_trans_put(trans);
491 	make_bad_inode(&inode->v);
492 	iput(&inode->v);
493 	inode = ERR_PTR(ret);
494 	goto err;
495 }
496 
497 /* methods */
498 
499 static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
500 			subvol_inum dir, struct bch_hash_info *dir_hash_info,
501 			const struct qstr *name)
502 {
503 	struct bch_fs *c = trans->c;
504 	struct btree_iter dirent_iter = {};
505 	subvol_inum inum = {};
506 	struct printbuf buf = PRINTBUF;
507 
508 	struct bkey_s_c k = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
509 					     dir_hash_info, dir, name, 0);
510 	int ret = bkey_err(k);
511 	if (ret)
512 		return ERR_PTR(ret);
513 
514 	ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
515 	if (ret > 0)
516 		ret = -ENOENT;
517 	if (ret)
518 		goto err;
519 
520 	struct bch_inode_info *inode = bch2_inode_hash_find(c, trans, inum);
521 	if (inode)
522 		goto out;
523 
524 	struct bch_subvolume subvol;
525 	struct bch_inode_unpacked inode_u;
526 	ret =   bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
527 		bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
528 		PTR_ERR_OR_ZERO(inode = bch2_inode_hash_init_insert(trans, inum, &inode_u, &subvol));
529 
530 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
531 				c, "dirent to missing inode:\n  %s",
532 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
533 	if (ret)
534 		goto err;
535 
536 	/* regular files may have hardlinks: */
537 	if (bch2_fs_inconsistent_on(bch2_inode_should_have_bp(&inode_u) &&
538 				    !bkey_eq(k.k->p, POS(inode_u.bi_dir, inode_u.bi_dir_offset)),
539 				    c,
540 				    "dirent points to inode that does not point back:\n  %s",
541 				    (bch2_bkey_val_to_text(&buf, c, k),
542 				     prt_printf(&buf, "\n  "),
543 				     bch2_inode_unpacked_to_text(&buf, &inode_u),
544 				     buf.buf))) {
545 		ret = -ENOENT;
546 		goto err;
547 	}
548 out:
549 	bch2_trans_iter_exit(trans, &dirent_iter);
550 	printbuf_exit(&buf);
551 	return inode;
552 err:
553 	inode = ERR_PTR(ret);
554 	goto out;
555 }
556 
557 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
558 				  unsigned int flags)
559 {
560 	struct bch_fs *c = vdir->i_sb->s_fs_info;
561 	struct bch_inode_info *dir = to_bch_ei(vdir);
562 	struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
563 
564 	struct bch_inode_info *inode;
565 	bch2_trans_do(c, NULL, NULL, 0,
566 		PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
567 							  &hash, &dentry->d_name)));
568 	if (IS_ERR(inode))
569 		inode = NULL;
570 
571 	return d_splice_alias(&inode->v, dentry);
572 }
573 
574 static int bch2_mknod(struct mnt_idmap *idmap,
575 		      struct inode *vdir, struct dentry *dentry,
576 		      umode_t mode, dev_t rdev)
577 {
578 	struct bch_inode_info *inode =
579 		__bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
580 			      (subvol_inum) { 0 }, 0);
581 
582 	if (IS_ERR(inode))
583 		return bch2_err_class(PTR_ERR(inode));
584 
585 	d_instantiate(dentry, &inode->v);
586 	return 0;
587 }
588 
589 static int bch2_create(struct mnt_idmap *idmap,
590 		       struct inode *vdir, struct dentry *dentry,
591 		       umode_t mode, bool excl)
592 {
593 	return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
594 }
595 
596 static int __bch2_link(struct bch_fs *c,
597 		       struct bch_inode_info *inode,
598 		       struct bch_inode_info *dir,
599 		       struct dentry *dentry)
600 {
601 	struct bch_inode_unpacked dir_u, inode_u;
602 	int ret;
603 
604 	mutex_lock(&inode->ei_update_lock);
605 	struct btree_trans *trans = bch2_trans_get(c);
606 
607 	ret = commit_do(trans, NULL, NULL, 0,
608 			bch2_link_trans(trans,
609 					inode_inum(dir),   &dir_u,
610 					inode_inum(inode), &inode_u,
611 					&dentry->d_name));
612 
613 	if (likely(!ret)) {
614 		bch2_inode_update_after_write(trans, dir, &dir_u,
615 					      ATTR_MTIME|ATTR_CTIME);
616 		bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
617 	}
618 
619 	bch2_trans_put(trans);
620 	mutex_unlock(&inode->ei_update_lock);
621 	return ret;
622 }
623 
624 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
625 		     struct dentry *dentry)
626 {
627 	struct bch_fs *c = vdir->i_sb->s_fs_info;
628 	struct bch_inode_info *dir = to_bch_ei(vdir);
629 	struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
630 	int ret;
631 
632 	lockdep_assert_held(&inode->v.i_rwsem);
633 
634 	ret   = bch2_subvol_is_ro(c, dir->ei_inum.subvol) ?:
635 		bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
636 		__bch2_link(c, inode, dir, dentry);
637 	if (unlikely(ret))
638 		return bch2_err_class(ret);
639 
640 	ihold(&inode->v);
641 	d_instantiate(dentry, &inode->v);
642 	return 0;
643 }
644 
645 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
646 		  bool deleting_snapshot)
647 {
648 	struct bch_fs *c = vdir->i_sb->s_fs_info;
649 	struct bch_inode_info *dir = to_bch_ei(vdir);
650 	struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
651 	struct bch_inode_unpacked dir_u, inode_u;
652 	int ret;
653 
654 	bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
655 
656 	struct btree_trans *trans = bch2_trans_get(c);
657 
658 	ret = commit_do(trans, NULL, NULL,
659 			BCH_TRANS_COMMIT_no_enospc,
660 		bch2_unlink_trans(trans,
661 				  inode_inum(dir), &dir_u,
662 				  &inode_u, &dentry->d_name,
663 				  deleting_snapshot));
664 	if (unlikely(ret))
665 		goto err;
666 
667 	bch2_inode_update_after_write(trans, dir, &dir_u,
668 				      ATTR_MTIME|ATTR_CTIME);
669 	bch2_inode_update_after_write(trans, inode, &inode_u,
670 				      ATTR_MTIME);
671 
672 	if (inode_u.bi_subvol) {
673 		/*
674 		 * Subvolume deletion is asynchronous, but we still want to tell
675 		 * the VFS that it's been deleted here:
676 		 */
677 		set_nlink(&inode->v, 0);
678 	}
679 err:
680 	bch2_trans_put(trans);
681 	bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
682 
683 	return ret;
684 }
685 
686 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
687 {
688 	struct bch_inode_info *dir= to_bch_ei(vdir);
689 	struct bch_fs *c = dir->v.i_sb->s_fs_info;
690 
691 	int ret = bch2_subvol_is_ro(c, dir->ei_inum.subvol) ?:
692 		__bch2_unlink(vdir, dentry, false);
693 	return bch2_err_class(ret);
694 }
695 
696 static int bch2_symlink(struct mnt_idmap *idmap,
697 			struct inode *vdir, struct dentry *dentry,
698 			const char *symname)
699 {
700 	struct bch_fs *c = vdir->i_sb->s_fs_info;
701 	struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
702 	int ret;
703 
704 	inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
705 			      (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
706 	if (IS_ERR(inode))
707 		return bch2_err_class(PTR_ERR(inode));
708 
709 	inode_lock(&inode->v);
710 	ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
711 	inode_unlock(&inode->v);
712 
713 	if (unlikely(ret))
714 		goto err;
715 
716 	ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
717 	if (unlikely(ret))
718 		goto err;
719 
720 	ret = __bch2_link(c, inode, dir, dentry);
721 	if (unlikely(ret))
722 		goto err;
723 
724 	d_instantiate(dentry, &inode->v);
725 	return 0;
726 err:
727 	iput(&inode->v);
728 	return bch2_err_class(ret);
729 }
730 
731 static int bch2_mkdir(struct mnt_idmap *idmap,
732 		      struct inode *vdir, struct dentry *dentry, umode_t mode)
733 {
734 	return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
735 }
736 
737 static int bch2_rename2(struct mnt_idmap *idmap,
738 			struct inode *src_vdir, struct dentry *src_dentry,
739 			struct inode *dst_vdir, struct dentry *dst_dentry,
740 			unsigned flags)
741 {
742 	struct bch_fs *c = src_vdir->i_sb->s_fs_info;
743 	struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
744 	struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
745 	struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
746 	struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
747 	struct bch_inode_unpacked dst_dir_u, src_dir_u;
748 	struct bch_inode_unpacked src_inode_u, dst_inode_u, *whiteout_inode_u;
749 	struct btree_trans *trans;
750 	enum bch_rename_mode mode = flags & RENAME_EXCHANGE
751 		? BCH_RENAME_EXCHANGE
752 		: dst_dentry->d_inode
753 		? BCH_RENAME_OVERWRITE : BCH_RENAME;
754 	bool whiteout = !!(flags & RENAME_WHITEOUT);
755 	int ret;
756 
757 	if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE|RENAME_WHITEOUT))
758 		return -EINVAL;
759 
760 	if (mode == BCH_RENAME_OVERWRITE) {
761 		ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
762 						   0, LLONG_MAX);
763 		if (ret)
764 			return ret;
765 	}
766 
767 	bch2_lock_inodes(INODE_UPDATE_LOCK,
768 			 src_dir,
769 			 dst_dir,
770 			 src_inode,
771 			 dst_inode);
772 
773 	trans = bch2_trans_get(c);
774 
775 	ret   = bch2_subvol_is_ro_trans(trans, src_dir->ei_inum.subvol) ?:
776 		bch2_subvol_is_ro_trans(trans, dst_dir->ei_inum.subvol);
777 	if (ret)
778 		goto err;
779 
780 	if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
781 		ret = bch2_fs_quota_transfer(c, src_inode,
782 					     dst_dir->ei_qid,
783 					     1 << QTYP_PRJ,
784 					     KEY_TYPE_QUOTA_PREALLOC);
785 		if (ret)
786 			goto err;
787 	}
788 
789 	if (mode == BCH_RENAME_EXCHANGE &&
790 	    inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
791 		ret = bch2_fs_quota_transfer(c, dst_inode,
792 					     src_dir->ei_qid,
793 					     1 << QTYP_PRJ,
794 					     KEY_TYPE_QUOTA_PREALLOC);
795 		if (ret)
796 			goto err;
797 	}
798 retry:
799 	bch2_trans_begin(trans);
800 
801 	ret = bch2_rename_trans(trans,
802 				inode_inum(src_dir), &src_dir_u,
803 				inode_inum(dst_dir), &dst_dir_u,
804 				&src_inode_u,
805 				&dst_inode_u,
806 				&src_dentry->d_name,
807 				&dst_dentry->d_name,
808 				mode);
809 	if (unlikely(ret))
810 		goto err_tx_restart;
811 
812 	if (whiteout) {
813 		whiteout_inode_u = bch2_trans_kmalloc_nomemzero(trans, sizeof(*whiteout_inode_u));
814 		ret = PTR_ERR_OR_ZERO(whiteout_inode_u);
815 		if (unlikely(ret))
816 			goto err_tx_restart;
817 		bch2_inode_init_early(c, whiteout_inode_u);
818 
819 		ret = bch2_create_trans(trans,
820 					inode_inum(src_dir), &src_dir_u,
821 					whiteout_inode_u,
822 					&src_dentry->d_name,
823 					from_kuid(i_user_ns(&src_dir->v), current_fsuid()),
824 					from_kgid(i_user_ns(&src_dir->v), current_fsgid()),
825 					S_IFCHR|WHITEOUT_MODE, 0,
826 					NULL, NULL, (subvol_inum) { 0 }, 0) ?:
827 		      bch2_quota_acct(c, bch_qid(whiteout_inode_u), Q_INO, 1,
828 				      KEY_TYPE_QUOTA_PREALLOC);
829 		if (unlikely(ret))
830 			goto err_tx_restart;
831 	}
832 
833 	ret = bch2_trans_commit(trans, NULL, NULL, 0);
834 	if (unlikely(ret)) {
835 err_tx_restart:
836 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
837 			goto retry;
838 		goto err;
839 	}
840 
841 	BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
842 	BUG_ON(dst_inode &&
843 	       dst_inode->v.i_ino != dst_inode_u.bi_inum);
844 
845 	bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
846 				      ATTR_MTIME|ATTR_CTIME);
847 
848 	if (src_dir != dst_dir)
849 		bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
850 					      ATTR_MTIME|ATTR_CTIME);
851 
852 	bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
853 				      ATTR_CTIME);
854 
855 	if (dst_inode)
856 		bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
857 					      ATTR_CTIME);
858 err:
859 	bch2_trans_put(trans);
860 
861 	bch2_fs_quota_transfer(c, src_inode,
862 			       bch_qid(&src_inode->ei_inode),
863 			       1 << QTYP_PRJ,
864 			       KEY_TYPE_QUOTA_NOCHECK);
865 	if (dst_inode)
866 		bch2_fs_quota_transfer(c, dst_inode,
867 				       bch_qid(&dst_inode->ei_inode),
868 				       1 << QTYP_PRJ,
869 				       KEY_TYPE_QUOTA_NOCHECK);
870 
871 	bch2_unlock_inodes(INODE_UPDATE_LOCK,
872 			   src_dir,
873 			   dst_dir,
874 			   src_inode,
875 			   dst_inode);
876 
877 	return bch2_err_class(ret);
878 }
879 
880 static void bch2_setattr_copy(struct mnt_idmap *idmap,
881 			      struct bch_inode_info *inode,
882 			      struct bch_inode_unpacked *bi,
883 			      struct iattr *attr)
884 {
885 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
886 	unsigned int ia_valid = attr->ia_valid;
887 	kuid_t kuid;
888 	kgid_t kgid;
889 
890 	if (ia_valid & ATTR_UID) {
891 		kuid = from_vfsuid(idmap, i_user_ns(&inode->v), attr->ia_vfsuid);
892 		bi->bi_uid = from_kuid(i_user_ns(&inode->v), kuid);
893 	}
894 	if (ia_valid & ATTR_GID) {
895 		kgid = from_vfsgid(idmap, i_user_ns(&inode->v), attr->ia_vfsgid);
896 		bi->bi_gid = from_kgid(i_user_ns(&inode->v), kgid);
897 	}
898 
899 	if (ia_valid & ATTR_SIZE)
900 		bi->bi_size = attr->ia_size;
901 
902 	if (ia_valid & ATTR_ATIME)
903 		bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
904 	if (ia_valid & ATTR_MTIME)
905 		bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
906 	if (ia_valid & ATTR_CTIME)
907 		bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
908 
909 	if (ia_valid & ATTR_MODE) {
910 		umode_t mode = attr->ia_mode;
911 		kgid_t gid = ia_valid & ATTR_GID
912 			? kgid
913 			: inode->v.i_gid;
914 
915 		if (!in_group_or_capable(idmap, &inode->v,
916 			make_vfsgid(idmap, i_user_ns(&inode->v), gid)))
917 			mode &= ~S_ISGID;
918 		bi->bi_mode = mode;
919 	}
920 }
921 
922 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
923 			 struct bch_inode_info *inode,
924 			 struct iattr *attr)
925 {
926 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
927 	struct bch_qid qid;
928 	struct btree_trans *trans;
929 	struct btree_iter inode_iter = { NULL };
930 	struct bch_inode_unpacked inode_u;
931 	struct posix_acl *acl = NULL;
932 	kuid_t kuid;
933 	kgid_t kgid;
934 	int ret;
935 
936 	mutex_lock(&inode->ei_update_lock);
937 
938 	qid = inode->ei_qid;
939 
940 	if (attr->ia_valid & ATTR_UID) {
941 		kuid = from_vfsuid(idmap, i_user_ns(&inode->v), attr->ia_vfsuid);
942 		qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), kuid);
943 	}
944 
945 	if (attr->ia_valid & ATTR_GID) {
946 		kgid = from_vfsgid(idmap, i_user_ns(&inode->v), attr->ia_vfsgid);
947 		qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), kgid);
948 	}
949 
950 	ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
951 				     KEY_TYPE_QUOTA_PREALLOC);
952 	if (ret)
953 		goto err;
954 
955 	trans = bch2_trans_get(c);
956 retry:
957 	bch2_trans_begin(trans);
958 	kfree(acl);
959 	acl = NULL;
960 
961 	ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
962 			      BTREE_ITER_intent);
963 	if (ret)
964 		goto btree_err;
965 
966 	bch2_setattr_copy(idmap, inode, &inode_u, attr);
967 
968 	if (attr->ia_valid & ATTR_MODE) {
969 		ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
970 				     inode_u.bi_mode, &acl);
971 		if (ret)
972 			goto btree_err;
973 	}
974 
975 	ret =   bch2_inode_write(trans, &inode_iter, &inode_u) ?:
976 		bch2_trans_commit(trans, NULL, NULL,
977 				  BCH_TRANS_COMMIT_no_enospc);
978 btree_err:
979 	bch2_trans_iter_exit(trans, &inode_iter);
980 
981 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
982 		goto retry;
983 	if (unlikely(ret))
984 		goto err_trans;
985 
986 	bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
987 
988 	if (acl)
989 		set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
990 err_trans:
991 	bch2_trans_put(trans);
992 err:
993 	mutex_unlock(&inode->ei_update_lock);
994 
995 	return bch2_err_class(ret);
996 }
997 
998 static int bch2_getattr(struct mnt_idmap *idmap,
999 			const struct path *path, struct kstat *stat,
1000 			u32 request_mask, unsigned query_flags)
1001 {
1002 	struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
1003 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1004 	vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, &inode->v);
1005 	vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, &inode->v);
1006 
1007 	stat->dev	= inode->v.i_sb->s_dev;
1008 	stat->ino	= inode->v.i_ino;
1009 	stat->mode	= inode->v.i_mode;
1010 	stat->nlink	= inode->v.i_nlink;
1011 	stat->uid	= vfsuid_into_kuid(vfsuid);
1012 	stat->gid	= vfsgid_into_kgid(vfsgid);
1013 	stat->rdev	= inode->v.i_rdev;
1014 	stat->size	= i_size_read(&inode->v);
1015 	stat->atime	= inode_get_atime(&inode->v);
1016 	stat->mtime	= inode_get_mtime(&inode->v);
1017 	stat->ctime	= inode_get_ctime(&inode->v);
1018 	stat->blksize	= block_bytes(c);
1019 	stat->blocks	= inode->v.i_blocks;
1020 
1021 	stat->subvol	= inode->ei_inum.subvol;
1022 	stat->result_mask |= STATX_SUBVOL;
1023 
1024 	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->v.i_mode)) {
1025 		stat->result_mask |= STATX_DIOALIGN;
1026 		/*
1027 		 * this is incorrect; we should be tracking this in superblock,
1028 		 * and checking the alignment of open devices
1029 		 */
1030 		stat->dio_mem_align = SECTOR_SIZE;
1031 		stat->dio_offset_align = block_bytes(c);
1032 	}
1033 
1034 	if (request_mask & STATX_BTIME) {
1035 		stat->result_mask |= STATX_BTIME;
1036 		stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
1037 	}
1038 
1039 	if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
1040 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1041 	stat->attributes_mask	 |= STATX_ATTR_IMMUTABLE;
1042 
1043 	if (inode->ei_inode.bi_flags & BCH_INODE_append)
1044 		stat->attributes |= STATX_ATTR_APPEND;
1045 	stat->attributes_mask	 |= STATX_ATTR_APPEND;
1046 
1047 	if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
1048 		stat->attributes |= STATX_ATTR_NODUMP;
1049 	stat->attributes_mask	 |= STATX_ATTR_NODUMP;
1050 
1051 	return 0;
1052 }
1053 
1054 static int bch2_setattr(struct mnt_idmap *idmap,
1055 			struct dentry *dentry, struct iattr *iattr)
1056 {
1057 	struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
1058 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1059 	int ret;
1060 
1061 	lockdep_assert_held(&inode->v.i_rwsem);
1062 
1063 	ret   = bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
1064 		setattr_prepare(idmap, dentry, iattr);
1065 	if (ret)
1066 		return ret;
1067 
1068 	return iattr->ia_valid & ATTR_SIZE
1069 		? bchfs_truncate(idmap, inode, iattr)
1070 		: bch2_setattr_nonsize(idmap, inode, iattr);
1071 }
1072 
1073 static int bch2_tmpfile(struct mnt_idmap *idmap,
1074 			struct inode *vdir, struct file *file, umode_t mode)
1075 {
1076 	struct bch_inode_info *inode =
1077 		__bch2_create(idmap, to_bch_ei(vdir),
1078 			      file->f_path.dentry, mode, 0,
1079 			      (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
1080 
1081 	if (IS_ERR(inode))
1082 		return bch2_err_class(PTR_ERR(inode));
1083 
1084 	d_mark_tmpfile(file, &inode->v);
1085 	d_instantiate(file->f_path.dentry, &inode->v);
1086 	return finish_open_simple(file, 0);
1087 }
1088 
1089 static int bch2_fill_extent(struct bch_fs *c,
1090 			    struct fiemap_extent_info *info,
1091 			    struct bkey_s_c k, unsigned flags)
1092 {
1093 	if (bkey_extent_is_direct_data(k.k)) {
1094 		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1095 		const union bch_extent_entry *entry;
1096 		struct extent_ptr_decoded p;
1097 		int ret;
1098 
1099 		if (k.k->type == KEY_TYPE_reflink_v)
1100 			flags |= FIEMAP_EXTENT_SHARED;
1101 
1102 		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1103 			int flags2 = 0;
1104 			u64 offset = p.ptr.offset;
1105 
1106 			if (p.ptr.unwritten)
1107 				flags2 |= FIEMAP_EXTENT_UNWRITTEN;
1108 
1109 			if (p.crc.compression_type)
1110 				flags2 |= FIEMAP_EXTENT_ENCODED;
1111 			else
1112 				offset += p.crc.offset;
1113 
1114 			if ((offset & (block_sectors(c) - 1)) ||
1115 			    (k.k->size & (block_sectors(c) - 1)))
1116 				flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
1117 
1118 			ret = fiemap_fill_next_extent(info,
1119 						bkey_start_offset(k.k) << 9,
1120 						offset << 9,
1121 						k.k->size << 9, flags|flags2);
1122 			if (ret)
1123 				return ret;
1124 		}
1125 
1126 		return 0;
1127 	} else if (bkey_extent_is_inline_data(k.k)) {
1128 		return fiemap_fill_next_extent(info,
1129 					       bkey_start_offset(k.k) << 9,
1130 					       0, k.k->size << 9,
1131 					       flags|
1132 					       FIEMAP_EXTENT_DATA_INLINE);
1133 	} else if (k.k->type == KEY_TYPE_reservation) {
1134 		return fiemap_fill_next_extent(info,
1135 					       bkey_start_offset(k.k) << 9,
1136 					       0, k.k->size << 9,
1137 					       flags|
1138 					       FIEMAP_EXTENT_DELALLOC|
1139 					       FIEMAP_EXTENT_UNWRITTEN);
1140 	} else {
1141 		BUG();
1142 	}
1143 }
1144 
1145 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
1146 		       u64 start, u64 len)
1147 {
1148 	struct bch_fs *c = vinode->i_sb->s_fs_info;
1149 	struct bch_inode_info *ei = to_bch_ei(vinode);
1150 	struct btree_trans *trans;
1151 	struct btree_iter iter;
1152 	struct bkey_s_c k;
1153 	struct bkey_buf cur, prev;
1154 	unsigned offset_into_extent, sectors;
1155 	bool have_extent = false;
1156 	int ret = 0;
1157 
1158 	ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
1159 	if (ret)
1160 		return ret;
1161 
1162 	struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
1163 	if (start + len < start)
1164 		return -EINVAL;
1165 
1166 	start >>= 9;
1167 
1168 	bch2_bkey_buf_init(&cur);
1169 	bch2_bkey_buf_init(&prev);
1170 	trans = bch2_trans_get(c);
1171 
1172 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
1173 			     POS(ei->v.i_ino, start), 0);
1174 
1175 	while (true) {
1176 		enum btree_id data_btree = BTREE_ID_extents;
1177 
1178 		bch2_trans_begin(trans);
1179 
1180 		u32 snapshot;
1181 		ret = bch2_subvolume_get_snapshot(trans, ei->ei_inum.subvol, &snapshot);
1182 		if (ret)
1183 			goto err;
1184 
1185 		bch2_btree_iter_set_snapshot(&iter, snapshot);
1186 
1187 		k = bch2_btree_iter_peek_upto(&iter, end);
1188 		ret = bkey_err(k);
1189 		if (ret)
1190 			goto err;
1191 
1192 		if (!k.k)
1193 			break;
1194 
1195 		if (!bkey_extent_is_data(k.k) &&
1196 		    k.k->type != KEY_TYPE_reservation) {
1197 			bch2_btree_iter_advance(&iter);
1198 			continue;
1199 		}
1200 
1201 		offset_into_extent	= iter.pos.offset -
1202 			bkey_start_offset(k.k);
1203 		sectors			= k.k->size - offset_into_extent;
1204 
1205 		bch2_bkey_buf_reassemble(&cur, c, k);
1206 
1207 		ret = bch2_read_indirect_extent(trans, &data_btree,
1208 					&offset_into_extent, &cur);
1209 		if (ret)
1210 			break;
1211 
1212 		k = bkey_i_to_s_c(cur.k);
1213 		bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
1214 
1215 		sectors = min(sectors, k.k->size - offset_into_extent);
1216 
1217 		bch2_cut_front(POS(k.k->p.inode,
1218 				   bkey_start_offset(k.k) +
1219 				   offset_into_extent),
1220 			       cur.k);
1221 		bch2_key_resize(&cur.k->k, sectors);
1222 		cur.k->k.p = iter.pos;
1223 		cur.k->k.p.offset += cur.k->k.size;
1224 
1225 		if (have_extent) {
1226 			bch2_trans_unlock(trans);
1227 			ret = bch2_fill_extent(c, info,
1228 					bkey_i_to_s_c(prev.k), 0);
1229 			if (ret)
1230 				break;
1231 		}
1232 
1233 		bkey_copy(prev.k, cur.k);
1234 		have_extent = true;
1235 
1236 		bch2_btree_iter_set_pos(&iter,
1237 			POS(iter.pos.inode, iter.pos.offset + sectors));
1238 err:
1239 		if (ret &&
1240 		    !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1241 			break;
1242 	}
1243 	bch2_trans_iter_exit(trans, &iter);
1244 
1245 	if (!ret && have_extent) {
1246 		bch2_trans_unlock(trans);
1247 		ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
1248 				       FIEMAP_EXTENT_LAST);
1249 	}
1250 
1251 	bch2_trans_put(trans);
1252 	bch2_bkey_buf_exit(&cur, c);
1253 	bch2_bkey_buf_exit(&prev, c);
1254 	return ret < 0 ? ret : 0;
1255 }
1256 
1257 static const struct vm_operations_struct bch_vm_ops = {
1258 	.fault		= bch2_page_fault,
1259 	.map_pages	= filemap_map_pages,
1260 	.page_mkwrite   = bch2_page_mkwrite,
1261 };
1262 
1263 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1264 {
1265 	file_accessed(file);
1266 
1267 	vma->vm_ops = &bch_vm_ops;
1268 	return 0;
1269 }
1270 
1271 /* Directories: */
1272 
1273 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1274 {
1275 	return generic_file_llseek_size(file, offset, whence,
1276 					S64_MAX, S64_MAX);
1277 }
1278 
1279 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1280 {
1281 	struct bch_inode_info *inode = file_bch_inode(file);
1282 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1283 
1284 	if (!dir_emit_dots(file, ctx))
1285 		return 0;
1286 
1287 	int ret = bch2_readdir(c, inode_inum(inode), ctx);
1288 
1289 	bch_err_fn(c, ret);
1290 	return bch2_err_class(ret);
1291 }
1292 
1293 static int bch2_open(struct inode *vinode, struct file *file)
1294 {
1295 	if (file->f_flags & (O_WRONLY|O_RDWR)) {
1296 		struct bch_inode_info *inode = to_bch_ei(vinode);
1297 		struct bch_fs *c = inode->v.i_sb->s_fs_info;
1298 
1299 		int ret = bch2_subvol_is_ro(c, inode->ei_inum.subvol);
1300 		if (ret)
1301 			return ret;
1302 	}
1303 
1304 	file->f_mode |= FMODE_CAN_ODIRECT;
1305 
1306 	return generic_file_open(vinode, file);
1307 }
1308 
1309 static const struct file_operations bch_file_operations = {
1310 	.open		= bch2_open,
1311 	.llseek		= bch2_llseek,
1312 	.read_iter	= bch2_read_iter,
1313 	.write_iter	= bch2_write_iter,
1314 	.mmap		= bch2_mmap,
1315 	.get_unmapped_area = thp_get_unmapped_area,
1316 	.fsync		= bch2_fsync,
1317 	.splice_read	= filemap_splice_read,
1318 	.splice_write	= iter_file_splice_write,
1319 	.fallocate	= bch2_fallocate_dispatch,
1320 	.unlocked_ioctl = bch2_fs_file_ioctl,
1321 #ifdef CONFIG_COMPAT
1322 	.compat_ioctl	= bch2_compat_fs_ioctl,
1323 #endif
1324 	.remap_file_range = bch2_remap_file_range,
1325 };
1326 
1327 static const struct inode_operations bch_file_inode_operations = {
1328 	.getattr	= bch2_getattr,
1329 	.setattr	= bch2_setattr,
1330 	.fiemap		= bch2_fiemap,
1331 	.listxattr	= bch2_xattr_list,
1332 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1333 	.get_inode_acl	= bch2_get_acl,
1334 	.set_acl	= bch2_set_acl,
1335 #endif
1336 };
1337 
1338 static const struct inode_operations bch_dir_inode_operations = {
1339 	.lookup		= bch2_lookup,
1340 	.create		= bch2_create,
1341 	.link		= bch2_link,
1342 	.unlink		= bch2_unlink,
1343 	.symlink	= bch2_symlink,
1344 	.mkdir		= bch2_mkdir,
1345 	.rmdir		= bch2_unlink,
1346 	.mknod		= bch2_mknod,
1347 	.rename		= bch2_rename2,
1348 	.getattr	= bch2_getattr,
1349 	.setattr	= bch2_setattr,
1350 	.tmpfile	= bch2_tmpfile,
1351 	.listxattr	= bch2_xattr_list,
1352 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1353 	.get_inode_acl	= bch2_get_acl,
1354 	.set_acl	= bch2_set_acl,
1355 #endif
1356 };
1357 
1358 static const struct file_operations bch_dir_file_operations = {
1359 	.llseek		= bch2_dir_llseek,
1360 	.read		= generic_read_dir,
1361 	.iterate_shared	= bch2_vfs_readdir,
1362 	.fsync		= bch2_fsync,
1363 	.unlocked_ioctl = bch2_fs_file_ioctl,
1364 #ifdef CONFIG_COMPAT
1365 	.compat_ioctl	= bch2_compat_fs_ioctl,
1366 #endif
1367 };
1368 
1369 static const struct inode_operations bch_symlink_inode_operations = {
1370 	.get_link	= page_get_link,
1371 	.getattr	= bch2_getattr,
1372 	.setattr	= bch2_setattr,
1373 	.listxattr	= bch2_xattr_list,
1374 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1375 	.get_inode_acl	= bch2_get_acl,
1376 	.set_acl	= bch2_set_acl,
1377 #endif
1378 };
1379 
1380 static const struct inode_operations bch_special_inode_operations = {
1381 	.getattr	= bch2_getattr,
1382 	.setattr	= bch2_setattr,
1383 	.listxattr	= bch2_xattr_list,
1384 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1385 	.get_inode_acl	= bch2_get_acl,
1386 	.set_acl	= bch2_set_acl,
1387 #endif
1388 };
1389 
1390 static const struct address_space_operations bch_address_space_operations = {
1391 	.read_folio	= bch2_read_folio,
1392 	.writepages	= bch2_writepages,
1393 	.readahead	= bch2_readahead,
1394 	.dirty_folio	= filemap_dirty_folio,
1395 	.write_begin	= bch2_write_begin,
1396 	.write_end	= bch2_write_end,
1397 	.invalidate_folio = bch2_invalidate_folio,
1398 	.release_folio	= bch2_release_folio,
1399 #ifdef CONFIG_MIGRATION
1400 	.migrate_folio	= filemap_migrate_folio,
1401 #endif
1402 	.error_remove_folio = generic_error_remove_folio,
1403 };
1404 
1405 struct bcachefs_fid {
1406 	u64		inum;
1407 	u32		subvol;
1408 	u32		gen;
1409 } __packed;
1410 
1411 struct bcachefs_fid_with_parent {
1412 	struct bcachefs_fid	fid;
1413 	struct bcachefs_fid	dir;
1414 } __packed;
1415 
1416 static int bcachefs_fid_valid(int fh_len, int fh_type)
1417 {
1418 	switch (fh_type) {
1419 	case FILEID_BCACHEFS_WITHOUT_PARENT:
1420 		return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1421 	case FILEID_BCACHEFS_WITH_PARENT:
1422 		return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1423 	default:
1424 		return false;
1425 	}
1426 }
1427 
1428 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1429 {
1430 	return (struct bcachefs_fid) {
1431 		.inum	= inode->ei_inum.inum,
1432 		.subvol	= inode->ei_inum.subvol,
1433 		.gen	= inode->ei_inode.bi_generation,
1434 	};
1435 }
1436 
1437 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1438 			  struct inode *vdir)
1439 {
1440 	struct bch_inode_info *inode	= to_bch_ei(vinode);
1441 	struct bch_inode_info *dir	= to_bch_ei(vdir);
1442 	int min_len;
1443 
1444 	if (!S_ISDIR(inode->v.i_mode) && dir) {
1445 		struct bcachefs_fid_with_parent *fid = (void *) fh;
1446 
1447 		min_len = sizeof(*fid) / sizeof(u32);
1448 		if (*len < min_len) {
1449 			*len = min_len;
1450 			return FILEID_INVALID;
1451 		}
1452 
1453 		fid->fid = bch2_inode_to_fid(inode);
1454 		fid->dir = bch2_inode_to_fid(dir);
1455 
1456 		*len = min_len;
1457 		return FILEID_BCACHEFS_WITH_PARENT;
1458 	} else {
1459 		struct bcachefs_fid *fid = (void *) fh;
1460 
1461 		min_len = sizeof(*fid) / sizeof(u32);
1462 		if (*len < min_len) {
1463 			*len = min_len;
1464 			return FILEID_INVALID;
1465 		}
1466 		*fid = bch2_inode_to_fid(inode);
1467 
1468 		*len = min_len;
1469 		return FILEID_BCACHEFS_WITHOUT_PARENT;
1470 	}
1471 }
1472 
1473 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1474 					struct bcachefs_fid fid)
1475 {
1476 	struct bch_fs *c = sb->s_fs_info;
1477 	struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1478 				    .subvol = fid.subvol,
1479 				    .inum = fid.inum,
1480 	});
1481 	if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1482 		iput(vinode);
1483 		vinode = ERR_PTR(-ESTALE);
1484 	}
1485 	return vinode;
1486 }
1487 
1488 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1489 		int fh_len, int fh_type)
1490 {
1491 	struct bcachefs_fid *fid = (void *) _fid;
1492 
1493 	if (!bcachefs_fid_valid(fh_len, fh_type))
1494 		return NULL;
1495 
1496 	return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1497 }
1498 
1499 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1500 		int fh_len, int fh_type)
1501 {
1502 	struct bcachefs_fid_with_parent *fid = (void *) _fid;
1503 
1504 	if (!bcachefs_fid_valid(fh_len, fh_type) ||
1505 	    fh_type != FILEID_BCACHEFS_WITH_PARENT)
1506 		return NULL;
1507 
1508 	return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1509 }
1510 
1511 static struct dentry *bch2_get_parent(struct dentry *child)
1512 {
1513 	struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1514 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1515 	subvol_inum parent_inum = {
1516 		.subvol = inode->ei_inode.bi_parent_subvol ?:
1517 			inode->ei_inum.subvol,
1518 		.inum = inode->ei_inode.bi_dir,
1519 	};
1520 
1521 	return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1522 }
1523 
1524 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1525 {
1526 	struct bch_inode_info *inode	= to_bch_ei(child->d_inode);
1527 	struct bch_inode_info *dir	= to_bch_ei(parent->d_inode);
1528 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1529 	struct btree_trans *trans;
1530 	struct btree_iter iter1;
1531 	struct btree_iter iter2;
1532 	struct bkey_s_c k;
1533 	struct bkey_s_c_dirent d;
1534 	struct bch_inode_unpacked inode_u;
1535 	subvol_inum target;
1536 	u32 snapshot;
1537 	struct qstr dirent_name;
1538 	unsigned name_len = 0;
1539 	int ret;
1540 
1541 	if (!S_ISDIR(dir->v.i_mode))
1542 		return -EINVAL;
1543 
1544 	trans = bch2_trans_get(c);
1545 
1546 	bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1547 			     POS(dir->ei_inode.bi_inum, 0), 0);
1548 	bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1549 			     POS(dir->ei_inode.bi_inum, 0), 0);
1550 retry:
1551 	bch2_trans_begin(trans);
1552 
1553 	ret = bch2_subvolume_get_snapshot(trans, dir->ei_inum.subvol, &snapshot);
1554 	if (ret)
1555 		goto err;
1556 
1557 	bch2_btree_iter_set_snapshot(&iter1, snapshot);
1558 	bch2_btree_iter_set_snapshot(&iter2, snapshot);
1559 
1560 	ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1561 	if (ret)
1562 		goto err;
1563 
1564 	if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1565 		bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1566 
1567 		k = bch2_btree_iter_peek_slot(&iter1);
1568 		ret = bkey_err(k);
1569 		if (ret)
1570 			goto err;
1571 
1572 		if (k.k->type != KEY_TYPE_dirent) {
1573 			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1574 			goto err;
1575 		}
1576 
1577 		d = bkey_s_c_to_dirent(k);
1578 		ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1579 		if (ret > 0)
1580 			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1581 		if (ret)
1582 			goto err;
1583 
1584 		if (subvol_inum_eq(target, inode->ei_inum))
1585 			goto found;
1586 	} else {
1587 		/*
1588 		 * File with multiple hardlinks and our backref is to the wrong
1589 		 * directory - linear search:
1590 		 */
1591 		for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1592 			if (k.k->p.inode > dir->ei_inode.bi_inum)
1593 				break;
1594 
1595 			if (k.k->type != KEY_TYPE_dirent)
1596 				continue;
1597 
1598 			d = bkey_s_c_to_dirent(k);
1599 			ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1600 			if (ret < 0)
1601 				break;
1602 			if (ret)
1603 				continue;
1604 
1605 			if (subvol_inum_eq(target, inode->ei_inum))
1606 				goto found;
1607 		}
1608 	}
1609 
1610 	ret = -ENOENT;
1611 	goto err;
1612 found:
1613 	dirent_name = bch2_dirent_get_name(d);
1614 
1615 	name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1616 	memcpy(name, dirent_name.name, name_len);
1617 	name[name_len] = '\0';
1618 err:
1619 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1620 		goto retry;
1621 
1622 	bch2_trans_iter_exit(trans, &iter1);
1623 	bch2_trans_iter_exit(trans, &iter2);
1624 	bch2_trans_put(trans);
1625 
1626 	return ret;
1627 }
1628 
1629 static const struct export_operations bch_export_ops = {
1630 	.encode_fh	= bch2_encode_fh,
1631 	.fh_to_dentry	= bch2_fh_to_dentry,
1632 	.fh_to_parent	= bch2_fh_to_parent,
1633 	.get_parent	= bch2_get_parent,
1634 	.get_name	= bch2_get_name,
1635 };
1636 
1637 static void bch2_vfs_inode_init(struct btree_trans *trans,
1638 				subvol_inum inum,
1639 				struct bch_inode_info *inode,
1640 				struct bch_inode_unpacked *bi,
1641 				struct bch_subvolume *subvol)
1642 {
1643 	inode->v.i_ino		= inum.inum;
1644 	inode->ei_inum		= inum;
1645 	inode->ei_inode.bi_inum	= inum.inum;
1646 	bch2_inode_update_after_write(trans, inode, bi, ~0);
1647 
1648 	inode->v.i_blocks	= bi->bi_sectors;
1649 	inode->v.i_ino		= bi->bi_inum;
1650 	inode->v.i_rdev		= bi->bi_dev;
1651 	inode->v.i_generation	= bi->bi_generation;
1652 	inode->v.i_size		= bi->bi_size;
1653 
1654 	inode->ei_flags		= 0;
1655 	inode->ei_quota_reserved = 0;
1656 	inode->ei_qid		= bch_qid(bi);
1657 
1658 	if (BCH_SUBVOLUME_SNAP(subvol))
1659 		set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1660 
1661 	inode->v.i_mapping->a_ops = &bch_address_space_operations;
1662 
1663 	switch (inode->v.i_mode & S_IFMT) {
1664 	case S_IFREG:
1665 		inode->v.i_op	= &bch_file_inode_operations;
1666 		inode->v.i_fop	= &bch_file_operations;
1667 		break;
1668 	case S_IFDIR:
1669 		inode->v.i_op	= &bch_dir_inode_operations;
1670 		inode->v.i_fop	= &bch_dir_file_operations;
1671 		break;
1672 	case S_IFLNK:
1673 		inode_nohighmem(&inode->v);
1674 		inode->v.i_op	= &bch_symlink_inode_operations;
1675 		break;
1676 	default:
1677 		init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1678 		inode->v.i_op	= &bch_special_inode_operations;
1679 		break;
1680 	}
1681 
1682 	mapping_set_large_folios(inode->v.i_mapping);
1683 }
1684 
1685 static void bch2_free_inode(struct inode *vinode)
1686 {
1687 	kmem_cache_free(bch2_inode_cache, to_bch_ei(vinode));
1688 }
1689 
1690 static int inode_update_times_fn(struct btree_trans *trans,
1691 				 struct bch_inode_info *inode,
1692 				 struct bch_inode_unpacked *bi,
1693 				 void *p)
1694 {
1695 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1696 
1697 	bi->bi_atime	= timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1698 	bi->bi_mtime	= timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1699 	bi->bi_ctime	= timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1700 
1701 	return 0;
1702 }
1703 
1704 static int bch2_vfs_write_inode(struct inode *vinode,
1705 				struct writeback_control *wbc)
1706 {
1707 	struct bch_fs *c = vinode->i_sb->s_fs_info;
1708 	struct bch_inode_info *inode = to_bch_ei(vinode);
1709 	int ret;
1710 
1711 	mutex_lock(&inode->ei_update_lock);
1712 	ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1713 			       ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1714 	mutex_unlock(&inode->ei_update_lock);
1715 
1716 	return bch2_err_class(ret);
1717 }
1718 
1719 static void bch2_evict_inode(struct inode *vinode)
1720 {
1721 	struct bch_fs *c = vinode->i_sb->s_fs_info;
1722 	struct bch_inode_info *inode = to_bch_ei(vinode);
1723 	bool delete = !inode->v.i_nlink && !is_bad_inode(&inode->v);
1724 
1725 	/*
1726 	 * evict() has waited for outstanding writeback, we'll do no more IO
1727 	 * through this inode: it's safe to remove from VFS inode hashtable here
1728 	 *
1729 	 * Do that now so that other threads aren't blocked from pulling it back
1730 	 * in, there's no reason for them to be:
1731 	 */
1732 	if (!delete)
1733 		bch2_inode_hash_remove(c, inode);
1734 
1735 	truncate_inode_pages_final(&inode->v.i_data);
1736 
1737 	clear_inode(&inode->v);
1738 
1739 	BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1740 
1741 	if (delete) {
1742 		bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1743 				KEY_TYPE_QUOTA_WARN);
1744 		bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1745 				KEY_TYPE_QUOTA_WARN);
1746 		bch2_inode_rm(c, inode_inum(inode));
1747 
1748 		/*
1749 		 * If we are deleting, we need it present in the vfs hash table
1750 		 * so that fsck can check if unlinked inodes are still open:
1751 		 */
1752 		bch2_inode_hash_remove(c, inode);
1753 	}
1754 
1755 	mutex_lock(&c->vfs_inodes_lock);
1756 	list_del_init(&inode->ei_vfs_inode_list);
1757 	mutex_unlock(&c->vfs_inodes_lock);
1758 }
1759 
1760 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1761 {
1762 	struct bch_inode_info *inode;
1763 	DARRAY(struct bch_inode_info *) grabbed;
1764 	bool clean_pass = false, this_pass_clean;
1765 
1766 	/*
1767 	 * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1768 	 * be pruned with d_mark_dontcache().
1769 	 *
1770 	 * Once we've had a clean pass where we didn't find any inodes without
1771 	 * I_DONTCACHE, we wait for them to be freed:
1772 	 */
1773 
1774 	darray_init(&grabbed);
1775 	darray_make_room(&grabbed, 1024);
1776 again:
1777 	cond_resched();
1778 	this_pass_clean = true;
1779 
1780 	mutex_lock(&c->vfs_inodes_lock);
1781 	list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1782 		if (!snapshot_list_has_id(s, inode->ei_inum.subvol))
1783 			continue;
1784 
1785 		if (!(inode->v.i_state & I_DONTCACHE) &&
1786 		    !(inode->v.i_state & I_FREEING) &&
1787 		    igrab(&inode->v)) {
1788 			this_pass_clean = false;
1789 
1790 			if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1791 				iput(&inode->v);
1792 				break;
1793 			}
1794 		} else if (clean_pass && this_pass_clean) {
1795 			struct wait_bit_queue_entry wqe;
1796 			struct wait_queue_head *wq_head;
1797 
1798 			wq_head = inode_bit_waitqueue(&wqe, &inode->v, __I_NEW);
1799 			prepare_to_wait_event(wq_head, &wqe.wq_entry,
1800 					      TASK_UNINTERRUPTIBLE);
1801 			mutex_unlock(&c->vfs_inodes_lock);
1802 
1803 			schedule();
1804 			finish_wait(wq_head, &wqe.wq_entry);
1805 			goto again;
1806 		}
1807 	}
1808 	mutex_unlock(&c->vfs_inodes_lock);
1809 
1810 	darray_for_each(grabbed, i) {
1811 		inode = *i;
1812 		d_mark_dontcache(&inode->v);
1813 		d_prune_aliases(&inode->v);
1814 		iput(&inode->v);
1815 	}
1816 	grabbed.nr = 0;
1817 
1818 	if (!clean_pass || !this_pass_clean) {
1819 		clean_pass = this_pass_clean;
1820 		goto again;
1821 	}
1822 
1823 	darray_exit(&grabbed);
1824 }
1825 
1826 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1827 {
1828 	struct super_block *sb = dentry->d_sb;
1829 	struct bch_fs *c = sb->s_fs_info;
1830 	struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1831 	unsigned shift = sb->s_blocksize_bits - 9;
1832 	/*
1833 	 * this assumes inodes take up 64 bytes, which is a decent average
1834 	 * number:
1835 	 */
1836 	u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1837 
1838 	buf->f_type	= BCACHEFS_STATFS_MAGIC;
1839 	buf->f_bsize	= sb->s_blocksize;
1840 	buf->f_blocks	= usage.capacity >> shift;
1841 	buf->f_bfree	= usage.free >> shift;
1842 	buf->f_bavail	= avail_factor(usage.free) >> shift;
1843 
1844 	buf->f_files	= usage.nr_inodes + avail_inodes;
1845 	buf->f_ffree	= avail_inodes;
1846 
1847 	buf->f_fsid	= uuid_to_fsid(c->sb.user_uuid.b);
1848 	buf->f_namelen	= BCH_NAME_MAX;
1849 
1850 	return 0;
1851 }
1852 
1853 static int bch2_sync_fs(struct super_block *sb, int wait)
1854 {
1855 	struct bch_fs *c = sb->s_fs_info;
1856 	int ret;
1857 
1858 	trace_bch2_sync_fs(sb, wait);
1859 
1860 	if (c->opts.journal_flush_disabled)
1861 		return 0;
1862 
1863 	if (!wait) {
1864 		bch2_journal_flush_async(&c->journal, NULL);
1865 		return 0;
1866 	}
1867 
1868 	ret = bch2_journal_flush(&c->journal);
1869 	return bch2_err_class(ret);
1870 }
1871 
1872 static struct bch_fs *bch2_path_to_fs(const char *path)
1873 {
1874 	struct bch_fs *c;
1875 	dev_t dev;
1876 	int ret;
1877 
1878 	ret = lookup_bdev(path, &dev);
1879 	if (ret)
1880 		return ERR_PTR(ret);
1881 
1882 	c = bch2_dev_to_fs(dev);
1883 	if (c)
1884 		closure_put(&c->cl);
1885 	return c ?: ERR_PTR(-ENOENT);
1886 }
1887 
1888 static int bch2_remount(struct super_block *sb, int *flags,
1889 			struct bch_opts opts)
1890 {
1891 	struct bch_fs *c = sb->s_fs_info;
1892 	int ret = 0;
1893 
1894 	opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1895 
1896 	if (opts.read_only != c->opts.read_only) {
1897 		down_write(&c->state_lock);
1898 
1899 		if (opts.read_only) {
1900 			bch2_fs_read_only(c);
1901 
1902 			sb->s_flags |= SB_RDONLY;
1903 		} else {
1904 			ret = bch2_fs_read_write(c);
1905 			if (ret) {
1906 				bch_err(c, "error going rw: %i", ret);
1907 				up_write(&c->state_lock);
1908 				ret = -EINVAL;
1909 				goto err;
1910 			}
1911 
1912 			sb->s_flags &= ~SB_RDONLY;
1913 		}
1914 
1915 		c->opts.read_only = opts.read_only;
1916 
1917 		up_write(&c->state_lock);
1918 	}
1919 
1920 	if (opt_defined(opts, errors))
1921 		c->opts.errors = opts.errors;
1922 err:
1923 	return bch2_err_class(ret);
1924 }
1925 
1926 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1927 {
1928 	struct bch_fs *c = root->d_sb->s_fs_info;
1929 	bool first = true;
1930 
1931 	for_each_online_member(c, ca) {
1932 		if (!first)
1933 			seq_putc(seq, ':');
1934 		first = false;
1935 		seq_puts(seq, ca->disk_sb.sb_name);
1936 	}
1937 
1938 	return 0;
1939 }
1940 
1941 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1942 {
1943 	struct bch_fs *c = root->d_sb->s_fs_info;
1944 	struct printbuf buf = PRINTBUF;
1945 
1946 	bch2_opts_to_text(&buf, c->opts, c, c->disk_sb.sb,
1947 			  OPT_MOUNT, OPT_HIDDEN, OPT_SHOW_MOUNT_STYLE);
1948 	printbuf_nul_terminate(&buf);
1949 	seq_puts(seq, buf.buf);
1950 
1951 	int ret = buf.allocation_failure ? -ENOMEM : 0;
1952 	printbuf_exit(&buf);
1953 	return ret;
1954 }
1955 
1956 static void bch2_put_super(struct super_block *sb)
1957 {
1958 	struct bch_fs *c = sb->s_fs_info;
1959 
1960 	__bch2_fs_stop(c);
1961 }
1962 
1963 /*
1964  * bcachefs doesn't currently integrate intwrite freeze protection but the
1965  * internal write references serve the same purpose. Therefore reuse the
1966  * read-only transition code to perform the quiesce. The caveat is that we don't
1967  * currently have the ability to block tasks that want a write reference while
1968  * the superblock is frozen. This is fine for now, but we should either add
1969  * blocking support or find a way to integrate sb_start_intwrite() and friends.
1970  */
1971 static int bch2_freeze(struct super_block *sb)
1972 {
1973 	struct bch_fs *c = sb->s_fs_info;
1974 
1975 	down_write(&c->state_lock);
1976 	bch2_fs_read_only(c);
1977 	up_write(&c->state_lock);
1978 	return 0;
1979 }
1980 
1981 static int bch2_unfreeze(struct super_block *sb)
1982 {
1983 	struct bch_fs *c = sb->s_fs_info;
1984 	int ret;
1985 
1986 	if (test_bit(BCH_FS_emergency_ro, &c->flags))
1987 		return 0;
1988 
1989 	down_write(&c->state_lock);
1990 	ret = bch2_fs_read_write(c);
1991 	up_write(&c->state_lock);
1992 	return ret;
1993 }
1994 
1995 static const struct super_operations bch_super_operations = {
1996 	.alloc_inode	= bch2_alloc_inode,
1997 	.free_inode	= bch2_free_inode,
1998 	.write_inode	= bch2_vfs_write_inode,
1999 	.evict_inode	= bch2_evict_inode,
2000 	.sync_fs	= bch2_sync_fs,
2001 	.statfs		= bch2_statfs,
2002 	.show_devname	= bch2_show_devname,
2003 	.show_options	= bch2_show_options,
2004 	.put_super	= bch2_put_super,
2005 	.freeze_fs	= bch2_freeze,
2006 	.unfreeze_fs	= bch2_unfreeze,
2007 };
2008 
2009 static int bch2_set_super(struct super_block *s, void *data)
2010 {
2011 	s->s_fs_info = data;
2012 	return 0;
2013 }
2014 
2015 static int bch2_noset_super(struct super_block *s, void *data)
2016 {
2017 	return -EBUSY;
2018 }
2019 
2020 typedef DARRAY(struct bch_fs *) darray_fs;
2021 
2022 static int bch2_test_super(struct super_block *s, void *data)
2023 {
2024 	struct bch_fs *c = s->s_fs_info;
2025 	darray_fs *d = data;
2026 
2027 	if (!c)
2028 		return false;
2029 
2030 	darray_for_each(*d, i)
2031 		if (c != *i)
2032 			return false;
2033 	return true;
2034 }
2035 
2036 static int bch2_fs_get_tree(struct fs_context *fc)
2037 {
2038 	struct bch_fs *c;
2039 	struct super_block *sb;
2040 	struct inode *vinode;
2041 	struct bch2_opts_parse *opts_parse = fc->fs_private;
2042 	struct bch_opts opts = opts_parse->opts;
2043 	darray_str devs;
2044 	darray_fs devs_to_fs = {};
2045 	int ret;
2046 
2047 	opt_set(opts, read_only, (fc->sb_flags & SB_RDONLY) != 0);
2048 	opt_set(opts, nostart, true);
2049 
2050 	if (!fc->source || strlen(fc->source) == 0)
2051 		return -EINVAL;
2052 
2053 	ret = bch2_split_devs(fc->source, &devs);
2054 	if (ret)
2055 		return ret;
2056 
2057 	darray_for_each(devs, i) {
2058 		ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
2059 		if (ret)
2060 			goto err;
2061 	}
2062 
2063 	sb = sget(fc->fs_type, bch2_test_super, bch2_noset_super, fc->sb_flags|SB_NOSEC, &devs_to_fs);
2064 	if (!IS_ERR(sb))
2065 		goto got_sb;
2066 
2067 	c = bch2_fs_open(devs.data, devs.nr, opts);
2068 	ret = PTR_ERR_OR_ZERO(c);
2069 	if (ret)
2070 		goto err;
2071 
2072 	/* Some options can't be parsed until after the fs is started: */
2073 	opts = bch2_opts_empty();
2074 	ret = bch2_parse_mount_opts(c, &opts, NULL, opts_parse->parse_later.buf);
2075 	if (ret)
2076 		goto err_stop_fs;
2077 
2078 	bch2_opts_apply(&c->opts, opts);
2079 
2080 	ret = bch2_fs_start(c);
2081 	if (ret)
2082 		goto err_stop_fs;
2083 
2084 	sb = sget(fc->fs_type, NULL, bch2_set_super, fc->sb_flags|SB_NOSEC, c);
2085 	ret = PTR_ERR_OR_ZERO(sb);
2086 	if (ret)
2087 		goto err_stop_fs;
2088 got_sb:
2089 	c = sb->s_fs_info;
2090 
2091 	if (sb->s_root) {
2092 		if ((fc->sb_flags ^ sb->s_flags) & SB_RDONLY) {
2093 			ret = -EBUSY;
2094 			goto err_put_super;
2095 		}
2096 		goto out;
2097 	}
2098 
2099 	sb->s_blocksize		= block_bytes(c);
2100 	sb->s_blocksize_bits	= ilog2(block_bytes(c));
2101 	sb->s_maxbytes		= MAX_LFS_FILESIZE;
2102 	sb->s_op		= &bch_super_operations;
2103 	sb->s_export_op		= &bch_export_ops;
2104 #ifdef CONFIG_BCACHEFS_QUOTA
2105 	sb->s_qcop		= &bch2_quotactl_operations;
2106 	sb->s_quota_types	= QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
2107 #endif
2108 	sb->s_xattr		= bch2_xattr_handlers;
2109 	sb->s_magic		= BCACHEFS_STATFS_MAGIC;
2110 	sb->s_time_gran		= c->sb.nsec_per_time_unit;
2111 	sb->s_time_min		= div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
2112 	sb->s_time_max		= div_s64(S64_MAX, c->sb.time_units_per_sec);
2113 	sb->s_uuid		= c->sb.user_uuid;
2114 	sb->s_shrink->seeks	= 0;
2115 	c->vfs_sb		= sb;
2116 	strscpy(sb->s_id, c->name, sizeof(sb->s_id));
2117 
2118 	ret = super_setup_bdi(sb);
2119 	if (ret)
2120 		goto err_put_super;
2121 
2122 	sb->s_bdi->ra_pages		= VM_READAHEAD_PAGES;
2123 
2124 	for_each_online_member(c, ca) {
2125 		struct block_device *bdev = ca->disk_sb.bdev;
2126 
2127 		/* XXX: create an anonymous device for multi device filesystems */
2128 		sb->s_bdev	= bdev;
2129 		sb->s_dev	= bdev->bd_dev;
2130 		percpu_ref_put(&ca->io_ref);
2131 		break;
2132 	}
2133 
2134 	c->dev = sb->s_dev;
2135 
2136 #ifdef CONFIG_BCACHEFS_POSIX_ACL
2137 	if (c->opts.acl)
2138 		sb->s_flags	|= SB_POSIXACL;
2139 #endif
2140 
2141 	sb->s_shrink->seeks = 0;
2142 
2143 	vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
2144 	ret = PTR_ERR_OR_ZERO(vinode);
2145 	bch_err_msg(c, ret, "mounting: error getting root inode");
2146 	if (ret)
2147 		goto err_put_super;
2148 
2149 	sb->s_root = d_make_root(vinode);
2150 	if (!sb->s_root) {
2151 		bch_err(c, "error mounting: error allocating root dentry");
2152 		ret = -ENOMEM;
2153 		goto err_put_super;
2154 	}
2155 
2156 	sb->s_flags |= SB_ACTIVE;
2157 out:
2158 	fc->root = dget(sb->s_root);
2159 err:
2160 	darray_exit(&devs_to_fs);
2161 	bch2_darray_str_exit(&devs);
2162 	if (ret)
2163 		pr_err("error: %s", bch2_err_str(ret));
2164 	/*
2165 	 * On an inconsistency error in recovery we might see an -EROFS derived
2166 	 * errorcode (from the journal), but we don't want to return that to
2167 	 * userspace as that causes util-linux to retry the mount RO - which is
2168 	 * confusing:
2169 	 */
2170 	if (bch2_err_matches(ret, EROFS) && ret != -EROFS)
2171 		ret = -EIO;
2172 	return bch2_err_class(ret);
2173 
2174 err_stop_fs:
2175 	bch2_fs_stop(c);
2176 	goto err;
2177 
2178 err_put_super:
2179 	__bch2_fs_stop(c);
2180 	deactivate_locked_super(sb);
2181 	goto err;
2182 }
2183 
2184 static void bch2_kill_sb(struct super_block *sb)
2185 {
2186 	struct bch_fs *c = sb->s_fs_info;
2187 
2188 	generic_shutdown_super(sb);
2189 	bch2_fs_free(c);
2190 }
2191 
2192 static void bch2_fs_context_free(struct fs_context *fc)
2193 {
2194 	struct bch2_opts_parse *opts = fc->fs_private;
2195 
2196 	if (opts) {
2197 		printbuf_exit(&opts->parse_later);
2198 		kfree(opts);
2199 	}
2200 }
2201 
2202 static int bch2_fs_parse_param(struct fs_context *fc,
2203 			       struct fs_parameter *param)
2204 {
2205 	/*
2206 	 * the "source" param, i.e., the name of the device(s) to mount,
2207 	 * is handled by the VFS layer.
2208 	 */
2209 	if (!strcmp(param->key, "source"))
2210 		return -ENOPARAM;
2211 
2212 	struct bch2_opts_parse *opts = fc->fs_private;
2213 	struct bch_fs *c = NULL;
2214 
2215 	/* for reconfigure, we already have a struct bch_fs */
2216 	if (fc->root)
2217 		c = fc->root->d_sb->s_fs_info;
2218 
2219 	int ret = bch2_parse_one_mount_opt(c, &opts->opts,
2220 					   &opts->parse_later, param->key,
2221 					   param->string);
2222 
2223 	return bch2_err_class(ret);
2224 }
2225 
2226 static int bch2_fs_reconfigure(struct fs_context *fc)
2227 {
2228 	struct super_block *sb = fc->root->d_sb;
2229 	struct bch2_opts_parse *opts = fc->fs_private;
2230 
2231 	return bch2_remount(sb, &fc->sb_flags, opts->opts);
2232 }
2233 
2234 static const struct fs_context_operations bch2_context_ops = {
2235 	.free        = bch2_fs_context_free,
2236 	.parse_param = bch2_fs_parse_param,
2237 	.get_tree    = bch2_fs_get_tree,
2238 	.reconfigure = bch2_fs_reconfigure,
2239 };
2240 
2241 static int bch2_init_fs_context(struct fs_context *fc)
2242 {
2243 	struct bch2_opts_parse *opts = kzalloc(sizeof(*opts), GFP_KERNEL);
2244 
2245 	if (!opts)
2246 		return -ENOMEM;
2247 
2248 	opts->parse_later = PRINTBUF;
2249 
2250 	fc->ops = &bch2_context_ops;
2251 	fc->fs_private = opts;
2252 
2253 	return 0;
2254 }
2255 
2256 void bch2_fs_vfs_exit(struct bch_fs *c)
2257 {
2258 	if (c->vfs_inodes_table.tbl)
2259 		rhashtable_destroy(&c->vfs_inodes_table);
2260 }
2261 
2262 int bch2_fs_vfs_init(struct bch_fs *c)
2263 {
2264 	return rhashtable_init(&c->vfs_inodes_table, &bch2_vfs_inodes_params);
2265 }
2266 
2267 static struct file_system_type bcache_fs_type = {
2268 	.owner			= THIS_MODULE,
2269 	.name			= "bcachefs",
2270 	.init_fs_context	= bch2_init_fs_context,
2271 	.kill_sb		= bch2_kill_sb,
2272 	.fs_flags		= FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
2273 };
2274 
2275 MODULE_ALIAS_FS("bcachefs");
2276 
2277 void bch2_vfs_exit(void)
2278 {
2279 	unregister_filesystem(&bcache_fs_type);
2280 	kmem_cache_destroy(bch2_inode_cache);
2281 }
2282 
2283 int __init bch2_vfs_init(void)
2284 {
2285 	int ret = -ENOMEM;
2286 
2287 	bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT |
2288 				      SLAB_ACCOUNT);
2289 	if (!bch2_inode_cache)
2290 		goto err;
2291 
2292 	ret = register_filesystem(&bcache_fs_type);
2293 	if (ret)
2294 		goto err;
2295 
2296 	return 0;
2297 err:
2298 	bch2_vfs_exit();
2299 	return ret;
2300 }
2301 
2302 #endif /* NO_BCACHEFS_FS */
2303