]> git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/fs.c
Update bcachefs sources to ae6f512de8 bcachefs: Fix out of bounds read in fs usage...
[bcachefs-tools-debian] / libbcachefs / fs.c
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 "extents.h"
12 #include "fs.h"
13 #include "fs-common.h"
14 #include "fs-io.h"
15 #include "fs-ioctl.h"
16 #include "fsck.h"
17 #include "inode.h"
18 #include "io.h"
19 #include "journal.h"
20 #include "keylist.h"
21 #include "quota.h"
22 #include "super.h"
23 #include "xattr.h"
24
25 #include <linux/aio.h>
26 #include <linux/backing-dev.h>
27 #include <linux/exportfs.h>
28 #include <linux/fiemap.h>
29 #include <linux/module.h>
30 #include <linux/posix_acl.h>
31 #include <linux/random.h>
32 #include <linux/statfs.h>
33 #include <linux/string.h>
34 #include <linux/xattr.h>
35
36 static struct kmem_cache *bch2_inode_cache;
37
38 static void bch2_vfs_inode_init(struct bch_fs *,
39                                 struct bch_inode_info *,
40                                 struct bch_inode_unpacked *);
41
42 static void journal_seq_copy(struct bch_fs *c,
43                              struct bch_inode_info *dst,
44                              u64 journal_seq)
45 {
46         /*
47          * atomic64_cmpxchg has a fallback for archs that don't support it,
48          * cmpxchg does not:
49          */
50         atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
51         u64 old, v = READ_ONCE(dst->ei_journal_seq);
52
53         do {
54                 old = v;
55
56                 if (old >= journal_seq)
57                         break;
58         } while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
59
60         bch2_journal_set_has_inum(&c->journal, dst->v.i_ino, journal_seq);
61 }
62
63 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
64 {
65         BUG_ON(atomic_long_read(&lock->v) == 0);
66
67         if (atomic_long_sub_return_release(i, &lock->v) == 0)
68                 wake_up_all(&lock->wait);
69 }
70
71 static bool __pagecache_lock_tryget(struct pagecache_lock *lock, long i)
72 {
73         long v = atomic_long_read(&lock->v), old;
74
75         do {
76                 old = v;
77
78                 if (i > 0 ? v < 0 : v > 0)
79                         return false;
80         } while ((v = atomic_long_cmpxchg_acquire(&lock->v,
81                                         old, old + i)) != old);
82         return true;
83 }
84
85 static void __pagecache_lock_get(struct pagecache_lock *lock, long i)
86 {
87         wait_event(lock->wait, __pagecache_lock_tryget(lock, i));
88 }
89
90 void bch2_pagecache_add_put(struct pagecache_lock *lock)
91 {
92         __pagecache_lock_put(lock, 1);
93 }
94
95 bool bch2_pagecache_add_tryget(struct pagecache_lock *lock)
96 {
97         return __pagecache_lock_tryget(lock, 1);
98 }
99
100 void bch2_pagecache_add_get(struct pagecache_lock *lock)
101 {
102         __pagecache_lock_get(lock, 1);
103 }
104
105 void bch2_pagecache_block_put(struct pagecache_lock *lock)
106 {
107         __pagecache_lock_put(lock, -1);
108 }
109
110 void bch2_pagecache_block_get(struct pagecache_lock *lock)
111 {
112         __pagecache_lock_get(lock, -1);
113 }
114
115 void bch2_inode_update_after_write(struct bch_fs *c,
116                                    struct bch_inode_info *inode,
117                                    struct bch_inode_unpacked *bi,
118                                    unsigned fields)
119 {
120         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
121         i_uid_write(&inode->v, bi->bi_uid);
122         i_gid_write(&inode->v, bi->bi_gid);
123         inode->v.i_mode = bi->bi_mode;
124
125         if (fields & ATTR_ATIME)
126                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
127         if (fields & ATTR_MTIME)
128                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
129         if (fields & ATTR_CTIME)
130                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
131
132         inode->ei_inode         = *bi;
133
134         bch2_inode_flags_to_vfs(inode);
135 }
136
137 int __must_check bch2_write_inode(struct bch_fs *c,
138                                   struct bch_inode_info *inode,
139                                   inode_set_fn set,
140                                   void *p, unsigned fields)
141 {
142         struct btree_trans trans;
143         struct btree_iter *iter;
144         struct bch_inode_unpacked inode_u;
145         int ret;
146
147         bch2_trans_init(&trans, c, 0, 256);
148 retry:
149         bch2_trans_begin(&trans);
150
151         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
152                                BTREE_ITER_INTENT);
153         ret   = PTR_ERR_OR_ZERO(iter) ?:
154                 (set ? set(inode, &inode_u, p) : 0) ?:
155                 bch2_inode_write(&trans, iter, &inode_u) ?:
156                 bch2_trans_commit(&trans, NULL,
157                                   &inode->ei_journal_seq,
158                                   BTREE_INSERT_NOUNLOCK|
159                                   BTREE_INSERT_NOFAIL);
160
161         /*
162          * the btree node lock protects inode->ei_inode, not ei_update_lock;
163          * this is important for inode updates via bchfs_write_index_update
164          */
165         if (!ret)
166                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
167
168         bch2_trans_iter_put(&trans, iter);
169
170         if (ret == -EINTR)
171                 goto retry;
172
173         bch2_trans_exit(&trans);
174         return ret < 0 ? ret : 0;
175 }
176
177 int bch2_fs_quota_transfer(struct bch_fs *c,
178                            struct bch_inode_info *inode,
179                            struct bch_qid new_qid,
180                            unsigned qtypes,
181                            enum quota_acct_mode mode)
182 {
183         unsigned i;
184         int ret;
185
186         qtypes &= enabled_qtypes(c);
187
188         for (i = 0; i < QTYP_NR; i++)
189                 if (new_qid.q[i] == inode->ei_qid.q[i])
190                         qtypes &= ~(1U << i);
191
192         if (!qtypes)
193                 return 0;
194
195         mutex_lock(&inode->ei_quota_lock);
196
197         ret = bch2_quota_transfer(c, qtypes, new_qid,
198                                   inode->ei_qid,
199                                   inode->v.i_blocks +
200                                   inode->ei_quota_reserved,
201                                   mode);
202         if (!ret)
203                 for (i = 0; i < QTYP_NR; i++)
204                         if (qtypes & (1 << i))
205                                 inode->ei_qid.q[i] = new_qid.q[i];
206
207         mutex_unlock(&inode->ei_quota_lock);
208
209         return ret;
210 }
211
212 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
213 {
214         struct bch_inode_unpacked inode_u;
215         struct bch_inode_info *inode;
216         int ret;
217
218         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
219         if (unlikely(!inode))
220                 return ERR_PTR(-ENOMEM);
221         if (!(inode->v.i_state & I_NEW))
222                 return &inode->v;
223
224         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
225         if (ret) {
226                 iget_failed(&inode->v);
227                 return ERR_PTR(ret);
228         }
229
230         bch2_vfs_inode_init(c, inode, &inode_u);
231
232         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
233
234         unlock_new_inode(&inode->v);
235
236         return &inode->v;
237 }
238
239 static int inum_test(struct inode *inode, void *p)
240 {
241         unsigned long *ino = p;
242
243         return *ino == inode->i_ino;
244 }
245
246 static struct bch_inode_info *
247 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
248               umode_t mode, dev_t rdev, bool tmpfile)
249 {
250         struct bch_fs *c = dir->v.i_sb->s_fs_info;
251         struct user_namespace *ns = dir->v.i_sb->s_user_ns;
252         struct btree_trans trans;
253         struct bch_inode_unpacked dir_u;
254         struct bch_inode_info *inode, *old;
255         struct bch_inode_unpacked inode_u;
256         struct posix_acl *default_acl = NULL, *acl = NULL;
257         u64 journal_seq = 0;
258         int ret;
259
260         /*
261          * preallocate acls + vfs inode before btree transaction, so that
262          * nothing can fail after the transaction succeeds:
263          */
264 #ifdef CONFIG_BCACHEFS_POSIX_ACL
265         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
266         if (ret)
267                 return ERR_PTR(ret);
268 #endif
269         inode = to_bch_ei(new_inode(c->vfs_sb));
270         if (unlikely(!inode)) {
271                 inode = ERR_PTR(-ENOMEM);
272                 goto err;
273         }
274
275         bch2_inode_init_early(c, &inode_u);
276
277         if (!tmpfile)
278                 mutex_lock(&dir->ei_update_lock);
279
280         bch2_trans_init(&trans, c, 8,
281                         2048 + (!tmpfile ? dentry->d_name.len : 0));
282 retry:
283         bch2_trans_begin(&trans);
284
285         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
286                                   !tmpfile ? &dentry->d_name : NULL,
287                                   from_kuid(ns, current_fsuid()),
288                                   from_kgid(ns, current_fsgid()),
289                                   mode, rdev,
290                                   default_acl, acl) ?:
291                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
292                                 KEY_TYPE_QUOTA_PREALLOC);
293         if (unlikely(ret))
294                 goto err_before_quota;
295
296         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
297                                   BTREE_INSERT_NOUNLOCK);
298         if (unlikely(ret)) {
299                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
300                                 KEY_TYPE_QUOTA_WARN);
301 err_before_quota:
302                 if (ret == -EINTR)
303                         goto retry;
304                 goto err_trans;
305         }
306
307         if (!tmpfile) {
308                 bch2_inode_update_after_write(c, dir, &dir_u,
309                                               ATTR_MTIME|ATTR_CTIME);
310                 journal_seq_copy(c, dir, journal_seq);
311                 mutex_unlock(&dir->ei_update_lock);
312         }
313
314         bch2_vfs_inode_init(c, inode, &inode_u);
315         journal_seq_copy(c, inode, journal_seq);
316
317         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
318         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
319
320         /*
321          * we must insert the new inode into the inode cache before calling
322          * bch2_trans_exit() and dropping locks, else we could race with another
323          * thread pulling the inode in and modifying it:
324          */
325
326         inode->v.i_state |= I_CREATING;
327         old = to_bch_ei(inode_insert5(&inode->v, inode->v.i_ino,
328                                       inum_test, NULL, &inode->v.i_ino));
329         BUG_ON(!old);
330
331         if (unlikely(old != inode)) {
332                 /*
333                  * We raced, another process pulled the new inode into cache
334                  * before us:
335                  */
336                 journal_seq_copy(c, old, journal_seq);
337                 make_bad_inode(&inode->v);
338                 iput(&inode->v);
339
340                 inode = old;
341         } else {
342                 /*
343                  * we really don't want insert_inode_locked2() to be setting
344                  * I_NEW...
345                  */
346                 unlock_new_inode(&inode->v);
347         }
348
349         bch2_trans_exit(&trans);
350 err:
351         posix_acl_release(default_acl);
352         posix_acl_release(acl);
353         return inode;
354 err_trans:
355         if (!tmpfile)
356                 mutex_unlock(&dir->ei_update_lock);
357
358         bch2_trans_exit(&trans);
359         make_bad_inode(&inode->v);
360         iput(&inode->v);
361         inode = ERR_PTR(ret);
362         goto err;
363 }
364
365 /* methods */
366
367 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
368                                   unsigned int flags)
369 {
370         struct bch_fs *c = vdir->i_sb->s_fs_info;
371         struct bch_inode_info *dir = to_bch_ei(vdir);
372         struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
373         struct inode *vinode = NULL;
374         u64 inum;
375
376         inum = bch2_dirent_lookup(c, dir->v.i_ino, &hash,
377                                   &dentry->d_name);
378
379         if (inum)
380                 vinode = bch2_vfs_inode_get(c, inum);
381
382         return d_splice_alias(vinode, dentry);
383 }
384
385 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
386                       umode_t mode, dev_t rdev)
387 {
388         struct bch_inode_info *inode =
389                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
390
391         if (IS_ERR(inode))
392                 return PTR_ERR(inode);
393
394         d_instantiate(dentry, &inode->v);
395         return 0;
396 }
397
398 static int bch2_create(struct inode *vdir, struct dentry *dentry,
399                        umode_t mode, bool excl)
400 {
401         return bch2_mknod(vdir, dentry, mode|S_IFREG, 0);
402 }
403
404 static int __bch2_link(struct bch_fs *c,
405                        struct bch_inode_info *inode,
406                        struct bch_inode_info *dir,
407                        struct dentry *dentry)
408 {
409         struct btree_trans trans;
410         struct bch_inode_unpacked dir_u, inode_u;
411         int ret;
412
413         mutex_lock(&inode->ei_update_lock);
414         bch2_trans_init(&trans, c, 4, 1024);
415
416         ret = __bch2_trans_do(&trans, NULL, &inode->ei_journal_seq,
417                               BTREE_INSERT_NOUNLOCK,
418                         bch2_link_trans(&trans,
419                                         dir->v.i_ino,
420                                         inode->v.i_ino, &dir_u, &inode_u,
421                                         &dentry->d_name));
422
423         if (likely(!ret)) {
424                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
425
426                 journal_seq_copy(c, inode, dir->ei_journal_seq);
427                 bch2_inode_update_after_write(c, dir, &dir_u,
428                                               ATTR_MTIME|ATTR_CTIME);
429                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
430         }
431
432         bch2_trans_exit(&trans);
433         mutex_unlock(&inode->ei_update_lock);
434         return ret;
435 }
436
437 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
438                      struct dentry *dentry)
439 {
440         struct bch_fs *c = vdir->i_sb->s_fs_info;
441         struct bch_inode_info *dir = to_bch_ei(vdir);
442         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
443         int ret;
444
445         lockdep_assert_held(&inode->v.i_rwsem);
446
447         ret = __bch2_link(c, inode, dir, dentry);
448         if (unlikely(ret))
449                 return ret;
450
451         ihold(&inode->v);
452         d_instantiate(dentry, &inode->v);
453         return 0;
454 }
455
456 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
457 {
458         struct bch_fs *c = vdir->i_sb->s_fs_info;
459         struct bch_inode_info *dir = to_bch_ei(vdir);
460         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
461         struct bch_inode_unpacked dir_u, inode_u;
462         struct btree_trans trans;
463         int ret;
464
465         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
466         bch2_trans_init(&trans, c, 4, 1024);
467
468         ret = __bch2_trans_do(&trans, NULL, &dir->ei_journal_seq,
469                               BTREE_INSERT_NOUNLOCK|
470                               BTREE_INSERT_NOFAIL,
471                         bch2_unlink_trans(&trans,
472                                           dir->v.i_ino, &dir_u,
473                                           &inode_u, &dentry->d_name));
474
475         if (likely(!ret)) {
476                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
477
478                 journal_seq_copy(c, inode, dir->ei_journal_seq);
479                 bch2_inode_update_after_write(c, dir, &dir_u,
480                                               ATTR_MTIME|ATTR_CTIME);
481                 bch2_inode_update_after_write(c, inode, &inode_u,
482                                               ATTR_MTIME);
483         }
484
485         bch2_trans_exit(&trans);
486         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
487
488         return ret;
489 }
490
491 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
492                         const char *symname)
493 {
494         struct bch_fs *c = vdir->i_sb->s_fs_info;
495         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
496         int ret;
497
498         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
499         if (unlikely(IS_ERR(inode)))
500                 return PTR_ERR(inode);
501
502         inode_lock(&inode->v);
503         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
504         inode_unlock(&inode->v);
505
506         if (unlikely(ret))
507                 goto err;
508
509         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
510         if (unlikely(ret))
511                 goto err;
512
513         journal_seq_copy(c, dir, inode->ei_journal_seq);
514
515         ret = __bch2_link(c, inode, dir, dentry);
516         if (unlikely(ret))
517                 goto err;
518
519         d_instantiate(dentry, &inode->v);
520         return 0;
521 err:
522         iput(&inode->v);
523         return ret;
524 }
525
526 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
527 {
528         return bch2_mknod(vdir, dentry, mode|S_IFDIR, 0);
529 }
530
531 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
532                         struct inode *dst_vdir, struct dentry *dst_dentry,
533                         unsigned flags)
534 {
535         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
536         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
537         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
538         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
539         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
540         struct bch_inode_unpacked dst_dir_u, src_dir_u;
541         struct bch_inode_unpacked src_inode_u, dst_inode_u;
542         struct btree_trans trans;
543         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
544                 ? BCH_RENAME_EXCHANGE
545                 : dst_dentry->d_inode
546                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
547         u64 journal_seq = 0;
548         int ret;
549
550         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
551                 return -EINVAL;
552
553         if (mode == BCH_RENAME_OVERWRITE) {
554                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
555                                                    0, LLONG_MAX);
556                 if (ret)
557                         return ret;
558         }
559
560         bch2_trans_init(&trans, c, 8, 2048);
561
562         bch2_lock_inodes(INODE_UPDATE_LOCK,
563                          src_dir,
564                          dst_dir,
565                          src_inode,
566                          dst_inode);
567
568         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
569                 ret = bch2_fs_quota_transfer(c, src_inode,
570                                              dst_dir->ei_qid,
571                                              1 << QTYP_PRJ,
572                                              KEY_TYPE_QUOTA_PREALLOC);
573                 if (ret)
574                         goto err;
575         }
576
577         if (mode == BCH_RENAME_EXCHANGE &&
578             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
579                 ret = bch2_fs_quota_transfer(c, dst_inode,
580                                              src_dir->ei_qid,
581                                              1 << QTYP_PRJ,
582                                              KEY_TYPE_QUOTA_PREALLOC);
583                 if (ret)
584                         goto err;
585         }
586
587         ret = __bch2_trans_do(&trans, NULL, &journal_seq,
588                               BTREE_INSERT_NOUNLOCK,
589                         bch2_rename_trans(&trans,
590                                           src_dir->v.i_ino, &src_dir_u,
591                                           dst_dir->v.i_ino, &dst_dir_u,
592                                           &src_inode_u,
593                                           &dst_inode_u,
594                                           &src_dentry->d_name,
595                                           &dst_dentry->d_name,
596                                           mode));
597         if (unlikely(ret))
598                 goto err;
599
600         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
601         BUG_ON(dst_inode &&
602                dst_inode->v.i_ino != dst_inode_u.bi_inum);
603
604         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
605                                       ATTR_MTIME|ATTR_CTIME);
606         journal_seq_copy(c, src_dir, journal_seq);
607
608         if (src_dir != dst_dir) {
609                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
610                                               ATTR_MTIME|ATTR_CTIME);
611                 journal_seq_copy(c, dst_dir, journal_seq);
612         }
613
614         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
615                                       ATTR_CTIME);
616         journal_seq_copy(c, src_inode, journal_seq);
617
618         if (dst_inode) {
619                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
620                                               ATTR_CTIME);
621                 journal_seq_copy(c, dst_inode, journal_seq);
622         }
623 err:
624         bch2_trans_exit(&trans);
625
626         bch2_fs_quota_transfer(c, src_inode,
627                                bch_qid(&src_inode->ei_inode),
628                                1 << QTYP_PRJ,
629                                KEY_TYPE_QUOTA_NOCHECK);
630         if (dst_inode)
631                 bch2_fs_quota_transfer(c, dst_inode,
632                                        bch_qid(&dst_inode->ei_inode),
633                                        1 << QTYP_PRJ,
634                                        KEY_TYPE_QUOTA_NOCHECK);
635
636         bch2_unlock_inodes(INODE_UPDATE_LOCK,
637                            src_dir,
638                            dst_dir,
639                            src_inode,
640                            dst_inode);
641
642         return ret;
643 }
644
645 void bch2_setattr_copy(struct bch_inode_info *inode,
646                        struct bch_inode_unpacked *bi,
647                        struct iattr *attr)
648 {
649         struct bch_fs *c = inode->v.i_sb->s_fs_info;
650         unsigned int ia_valid = attr->ia_valid;
651
652         if (ia_valid & ATTR_UID)
653                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
654         if (ia_valid & ATTR_GID)
655                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
656
657         if (ia_valid & ATTR_ATIME)
658                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
659         if (ia_valid & ATTR_MTIME)
660                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
661         if (ia_valid & ATTR_CTIME)
662                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
663
664         if (ia_valid & ATTR_MODE) {
665                 umode_t mode = attr->ia_mode;
666                 kgid_t gid = ia_valid & ATTR_GID
667                         ? attr->ia_gid
668                         : inode->v.i_gid;
669
670                 if (!in_group_p(gid) &&
671                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
672                         mode &= ~S_ISGID;
673                 bi->bi_mode = mode;
674         }
675 }
676
677 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
678                                 struct iattr *attr)
679 {
680         struct bch_fs *c = inode->v.i_sb->s_fs_info;
681         struct bch_qid qid;
682         struct btree_trans trans;
683         struct btree_iter *inode_iter;
684         struct bch_inode_unpacked inode_u;
685         struct posix_acl *acl = NULL;
686         int ret;
687
688         mutex_lock(&inode->ei_update_lock);
689
690         qid = inode->ei_qid;
691
692         if (attr->ia_valid & ATTR_UID)
693                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
694
695         if (attr->ia_valid & ATTR_GID)
696                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
697
698         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
699                                      KEY_TYPE_QUOTA_PREALLOC);
700         if (ret)
701                 goto err;
702
703         bch2_trans_init(&trans, c, 0, 0);
704 retry:
705         bch2_trans_begin(&trans);
706         kfree(acl);
707         acl = NULL;
708
709         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
710                                      BTREE_ITER_INTENT);
711         ret = PTR_ERR_OR_ZERO(inode_iter);
712         if (ret)
713                 goto btree_err;
714
715         bch2_setattr_copy(inode, &inode_u, attr);
716
717         if (attr->ia_valid & ATTR_MODE) {
718                 ret = bch2_acl_chmod(&trans, &inode_u, inode_u.bi_mode, &acl);
719                 if (ret)
720                         goto btree_err;
721         }
722
723         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
724                 bch2_trans_commit(&trans, NULL,
725                                   &inode->ei_journal_seq,
726                                   BTREE_INSERT_NOUNLOCK|
727                                   BTREE_INSERT_NOFAIL);
728 btree_err:
729         bch2_trans_iter_put(&trans, inode_iter);
730
731         if (ret == -EINTR)
732                 goto retry;
733         if (unlikely(ret))
734                 goto err_trans;
735
736         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
737
738         if (acl)
739                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
740 err_trans:
741         bch2_trans_exit(&trans);
742 err:
743         mutex_unlock(&inode->ei_update_lock);
744
745         return ret;
746 }
747
748 static int bch2_getattr(const struct path *path, struct kstat *stat,
749                         u32 request_mask, unsigned query_flags)
750 {
751         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
752         struct bch_fs *c = inode->v.i_sb->s_fs_info;
753
754         stat->dev       = inode->v.i_sb->s_dev;
755         stat->ino       = inode->v.i_ino;
756         stat->mode      = inode->v.i_mode;
757         stat->nlink     = inode->v.i_nlink;
758         stat->uid       = inode->v.i_uid;
759         stat->gid       = inode->v.i_gid;
760         stat->rdev      = inode->v.i_rdev;
761         stat->size      = i_size_read(&inode->v);
762         stat->atime     = inode->v.i_atime;
763         stat->mtime     = inode->v.i_mtime;
764         stat->ctime     = inode->v.i_ctime;
765         stat->blksize   = block_bytes(c);
766         stat->blocks    = inode->v.i_blocks;
767
768         if (request_mask & STATX_BTIME) {
769                 stat->result_mask |= STATX_BTIME;
770                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
771         }
772
773         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
774                 stat->attributes |= STATX_ATTR_IMMUTABLE;
775         stat->attributes_mask    |= STATX_ATTR_IMMUTABLE;
776
777         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
778                 stat->attributes |= STATX_ATTR_APPEND;
779         stat->attributes_mask    |= STATX_ATTR_APPEND;
780
781         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
782                 stat->attributes |= STATX_ATTR_NODUMP;
783         stat->attributes_mask    |= STATX_ATTR_NODUMP;
784
785         return 0;
786 }
787
788 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
789 {
790         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
791         int ret;
792
793         lockdep_assert_held(&inode->v.i_rwsem);
794
795         ret = setattr_prepare(dentry, iattr);
796         if (ret)
797                 return ret;
798
799         return iattr->ia_valid & ATTR_SIZE
800                 ? bch2_truncate(inode, iattr)
801                 : bch2_setattr_nonsize(inode, iattr);
802 }
803
804 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
805 {
806         struct bch_inode_info *inode =
807                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
808
809         if (IS_ERR(inode))
810                 return PTR_ERR(inode);
811
812         d_mark_tmpfile(dentry, &inode->v);
813         d_instantiate(dentry, &inode->v);
814         return 0;
815 }
816
817 static int bch2_fill_extent(struct bch_fs *c,
818                             struct fiemap_extent_info *info,
819                             struct bkey_s_c k, unsigned flags)
820 {
821         if (bkey_extent_is_direct_data(k.k)) {
822                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
823                 const union bch_extent_entry *entry;
824                 struct extent_ptr_decoded p;
825                 int ret;
826
827                 if (k.k->type == KEY_TYPE_reflink_v)
828                         flags |= FIEMAP_EXTENT_SHARED;
829
830                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
831                         int flags2 = 0;
832                         u64 offset = p.ptr.offset;
833
834                         if (p.crc.compression_type)
835                                 flags2 |= FIEMAP_EXTENT_ENCODED;
836                         else
837                                 offset += p.crc.offset;
838
839                         if ((offset & (c->opts.block_size - 1)) ||
840                             (k.k->size & (c->opts.block_size - 1)))
841                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
842
843                         ret = fiemap_fill_next_extent(info,
844                                                 bkey_start_offset(k.k) << 9,
845                                                 offset << 9,
846                                                 k.k->size << 9, flags|flags2);
847                         if (ret)
848                                 return ret;
849                 }
850
851                 return 0;
852         } else if (bkey_extent_is_inline_data(k.k)) {
853                 return fiemap_fill_next_extent(info,
854                                                bkey_start_offset(k.k) << 9,
855                                                0, k.k->size << 9,
856                                                flags|
857                                                FIEMAP_EXTENT_DATA_INLINE);
858         } else if (k.k->type == KEY_TYPE_reservation) {
859                 return fiemap_fill_next_extent(info,
860                                                bkey_start_offset(k.k) << 9,
861                                                0, k.k->size << 9,
862                                                flags|
863                                                FIEMAP_EXTENT_DELALLOC|
864                                                FIEMAP_EXTENT_UNWRITTEN);
865         } else {
866                 BUG();
867         }
868 }
869
870 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
871                        u64 start, u64 len)
872 {
873         struct bch_fs *c = vinode->i_sb->s_fs_info;
874         struct bch_inode_info *ei = to_bch_ei(vinode);
875         struct btree_trans trans;
876         struct btree_iter *iter;
877         struct bkey_s_c k;
878         struct bkey_buf cur, prev;
879         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
880         unsigned offset_into_extent, sectors;
881         bool have_extent = false;
882         int ret = 0;
883
884         ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
885         if (ret)
886                 return ret;
887
888         if (start + len < start)
889                 return -EINVAL;
890
891         bch2_bkey_buf_init(&cur);
892         bch2_bkey_buf_init(&prev);
893         bch2_trans_init(&trans, c, 0, 0);
894
895         iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
896                                    POS(ei->v.i_ino, start >> 9), 0);
897 retry:
898         while ((k = bch2_btree_iter_peek(iter)).k &&
899                !(ret = bkey_err(k)) &&
900                bkey_cmp(iter->pos, end) < 0) {
901                 enum btree_id data_btree = BTREE_ID_extents;
902
903                 if (!bkey_extent_is_data(k.k) &&
904                     k.k->type != KEY_TYPE_reservation) {
905                         bch2_btree_iter_advance(iter);
906                         continue;
907                 }
908
909                 offset_into_extent      = iter->pos.offset -
910                         bkey_start_offset(k.k);
911                 sectors                 = k.k->size - offset_into_extent;
912
913                 bch2_bkey_buf_reassemble(&cur, c, k);
914
915                 ret = bch2_read_indirect_extent(&trans, &data_btree,
916                                         &offset_into_extent, &cur);
917                 if (ret)
918                         break;
919
920                 k = bkey_i_to_s_c(cur.k);
921                 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
922
923                 sectors = min(sectors, k.k->size - offset_into_extent);
924
925                 bch2_cut_front(POS(k.k->p.inode,
926                                    bkey_start_offset(k.k) +
927                                    offset_into_extent),
928                                cur.k);
929                 bch2_key_resize(&cur.k->k, sectors);
930                 cur.k->k.p = iter->pos;
931                 cur.k->k.p.offset += cur.k->k.size;
932
933                 if (have_extent) {
934                         ret = bch2_fill_extent(c, info,
935                                         bkey_i_to_s_c(prev.k), 0);
936                         if (ret)
937                                 break;
938                 }
939
940                 bkey_copy(prev.k, cur.k);
941                 have_extent = true;
942
943                 bch2_btree_iter_set_pos(iter,
944                         POS(iter->pos.inode, iter->pos.offset + sectors));
945         }
946
947         if (ret == -EINTR)
948                 goto retry;
949
950         if (!ret && have_extent)
951                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
952                                        FIEMAP_EXTENT_LAST);
953
954         bch2_trans_iter_put(&trans, iter);
955         ret = bch2_trans_exit(&trans) ?: ret;
956         bch2_bkey_buf_exit(&cur, c);
957         bch2_bkey_buf_exit(&prev, c);
958         return ret < 0 ? ret : 0;
959 }
960
961 static const struct vm_operations_struct bch_vm_ops = {
962         .fault          = bch2_page_fault,
963         .map_pages      = filemap_map_pages,
964         .page_mkwrite   = bch2_page_mkwrite,
965 };
966
967 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
968 {
969         file_accessed(file);
970
971         vma->vm_ops = &bch_vm_ops;
972         return 0;
973 }
974
975 /* Directories: */
976
977 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
978 {
979         return generic_file_llseek_size(file, offset, whence,
980                                         S64_MAX, S64_MAX);
981 }
982
983 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
984 {
985         struct bch_inode_info *inode = file_bch_inode(file);
986         struct bch_fs *c = inode->v.i_sb->s_fs_info;
987
988         if (!dir_emit_dots(file, ctx))
989                 return 0;
990
991         return bch2_readdir(c, inode->v.i_ino, ctx);
992 }
993
994 static const struct file_operations bch_file_operations = {
995         .llseek         = bch2_llseek,
996         .read_iter      = bch2_read_iter,
997         .write_iter     = bch2_write_iter,
998         .mmap           = bch2_mmap,
999         .open           = generic_file_open,
1000         .fsync          = bch2_fsync,
1001         .splice_read    = generic_file_splice_read,
1002         .splice_write   = iter_file_splice_write,
1003         .fallocate      = bch2_fallocate_dispatch,
1004         .unlocked_ioctl = bch2_fs_file_ioctl,
1005 #ifdef CONFIG_COMPAT
1006         .compat_ioctl   = bch2_compat_fs_ioctl,
1007 #endif
1008         .remap_file_range = bch2_remap_file_range,
1009 };
1010
1011 static const struct inode_operations bch_file_inode_operations = {
1012         .getattr        = bch2_getattr,
1013         .setattr        = bch2_setattr,
1014         .fiemap         = bch2_fiemap,
1015         .listxattr      = bch2_xattr_list,
1016 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1017         .get_acl        = bch2_get_acl,
1018         .set_acl        = bch2_set_acl,
1019 #endif
1020 };
1021
1022 static const struct inode_operations bch_dir_inode_operations = {
1023         .lookup         = bch2_lookup,
1024         .create         = bch2_create,
1025         .link           = bch2_link,
1026         .unlink         = bch2_unlink,
1027         .symlink        = bch2_symlink,
1028         .mkdir          = bch2_mkdir,
1029         .rmdir          = bch2_unlink,
1030         .mknod          = bch2_mknod,
1031         .rename         = bch2_rename2,
1032         .getattr        = bch2_getattr,
1033         .setattr        = bch2_setattr,
1034         .tmpfile        = bch2_tmpfile,
1035         .listxattr      = bch2_xattr_list,
1036 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1037         .get_acl        = bch2_get_acl,
1038         .set_acl        = bch2_set_acl,
1039 #endif
1040 };
1041
1042 static const struct file_operations bch_dir_file_operations = {
1043         .llseek         = bch2_dir_llseek,
1044         .read           = generic_read_dir,
1045         .iterate_shared = bch2_vfs_readdir,
1046         .fsync          = bch2_fsync,
1047         .unlocked_ioctl = bch2_fs_file_ioctl,
1048 #ifdef CONFIG_COMPAT
1049         .compat_ioctl   = bch2_compat_fs_ioctl,
1050 #endif
1051 };
1052
1053 static const struct inode_operations bch_symlink_inode_operations = {
1054         .get_link       = page_get_link,
1055         .getattr        = bch2_getattr,
1056         .setattr        = bch2_setattr,
1057         .listxattr      = bch2_xattr_list,
1058 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1059         .get_acl        = bch2_get_acl,
1060         .set_acl        = bch2_set_acl,
1061 #endif
1062 };
1063
1064 static const struct inode_operations bch_special_inode_operations = {
1065         .getattr        = bch2_getattr,
1066         .setattr        = bch2_setattr,
1067         .listxattr      = bch2_xattr_list,
1068 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1069         .get_acl        = bch2_get_acl,
1070         .set_acl        = bch2_set_acl,
1071 #endif
1072 };
1073
1074 static const struct address_space_operations bch_address_space_operations = {
1075         .writepage      = bch2_writepage,
1076         .readpage       = bch2_readpage,
1077         .writepages     = bch2_writepages,
1078         .readahead      = bch2_readahead,
1079         .set_page_dirty = __set_page_dirty_nobuffers,
1080         .write_begin    = bch2_write_begin,
1081         .write_end      = bch2_write_end,
1082         .invalidatepage = bch2_invalidatepage,
1083         .releasepage    = bch2_releasepage,
1084         .direct_IO      = noop_direct_IO,
1085 #ifdef CONFIG_MIGRATION
1086         .migratepage    = bch2_migrate_page,
1087 #endif
1088         .error_remove_page = generic_error_remove_page,
1089 };
1090
1091 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1092                 u64 ino, u32 generation)
1093 {
1094         struct bch_fs *c = sb->s_fs_info;
1095         struct inode *vinode;
1096
1097         if (ino < BCACHEFS_ROOT_INO)
1098                 return ERR_PTR(-ESTALE);
1099
1100         vinode = bch2_vfs_inode_get(c, ino);
1101         if (IS_ERR(vinode))
1102                 return ERR_CAST(vinode);
1103         if (generation && vinode->i_generation != generation) {
1104                 /* we didn't find the right inode.. */
1105                 iput(vinode);
1106                 return ERR_PTR(-ESTALE);
1107         }
1108         return vinode;
1109 }
1110
1111 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1112                 int fh_len, int fh_type)
1113 {
1114         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1115                                     bch2_nfs_get_inode);
1116 }
1117
1118 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1119                 int fh_len, int fh_type)
1120 {
1121         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1122                                     bch2_nfs_get_inode);
1123 }
1124
1125 static const struct export_operations bch_export_ops = {
1126         .fh_to_dentry   = bch2_fh_to_dentry,
1127         .fh_to_parent   = bch2_fh_to_parent,
1128         //.get_parent   = bch2_get_parent,
1129 };
1130
1131 static void bch2_vfs_inode_init(struct bch_fs *c,
1132                                 struct bch_inode_info *inode,
1133                                 struct bch_inode_unpacked *bi)
1134 {
1135         bch2_inode_update_after_write(c, inode, bi, ~0);
1136
1137         inode->v.i_blocks       = bi->bi_sectors;
1138         inode->v.i_ino          = bi->bi_inum;
1139         inode->v.i_rdev         = bi->bi_dev;
1140         inode->v.i_generation   = bi->bi_generation;
1141         inode->v.i_size         = bi->bi_size;
1142
1143         inode->ei_flags         = 0;
1144         inode->ei_journal_seq   = 0;
1145         inode->ei_quota_reserved = 0;
1146         inode->ei_qid           = bch_qid(bi);
1147
1148         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1149
1150         switch (inode->v.i_mode & S_IFMT) {
1151         case S_IFREG:
1152                 inode->v.i_op   = &bch_file_inode_operations;
1153                 inode->v.i_fop  = &bch_file_operations;
1154                 break;
1155         case S_IFDIR:
1156                 inode->v.i_op   = &bch_dir_inode_operations;
1157                 inode->v.i_fop  = &bch_dir_file_operations;
1158                 break;
1159         case S_IFLNK:
1160                 inode_nohighmem(&inode->v);
1161                 inode->v.i_op   = &bch_symlink_inode_operations;
1162                 break;
1163         default:
1164                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1165                 inode->v.i_op   = &bch_special_inode_operations;
1166                 break;
1167         }
1168 }
1169
1170 static struct inode *bch2_alloc_inode(struct super_block *sb)
1171 {
1172         struct bch_inode_info *inode;
1173
1174         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1175         if (!inode)
1176                 return NULL;
1177
1178         inode_init_once(&inode->v);
1179         mutex_init(&inode->ei_update_lock);
1180         pagecache_lock_init(&inode->ei_pagecache_lock);
1181         mutex_init(&inode->ei_quota_lock);
1182         inode->ei_journal_seq = 0;
1183
1184         return &inode->v;
1185 }
1186
1187 static void bch2_i_callback(struct rcu_head *head)
1188 {
1189         struct inode *vinode = container_of(head, struct inode, i_rcu);
1190         struct bch_inode_info *inode = to_bch_ei(vinode);
1191
1192         kmem_cache_free(bch2_inode_cache, inode);
1193 }
1194
1195 static void bch2_destroy_inode(struct inode *vinode)
1196 {
1197         call_rcu(&vinode->i_rcu, bch2_i_callback);
1198 }
1199
1200 static int inode_update_times_fn(struct bch_inode_info *inode,
1201                                  struct bch_inode_unpacked *bi,
1202                                  void *p)
1203 {
1204         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1205
1206         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1207         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1208         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1209
1210         return 0;
1211 }
1212
1213 static int bch2_vfs_write_inode(struct inode *vinode,
1214                                 struct writeback_control *wbc)
1215 {
1216         struct bch_fs *c = vinode->i_sb->s_fs_info;
1217         struct bch_inode_info *inode = to_bch_ei(vinode);
1218         int ret;
1219
1220         mutex_lock(&inode->ei_update_lock);
1221         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1222                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1223         mutex_unlock(&inode->ei_update_lock);
1224
1225         return ret;
1226 }
1227
1228 static void bch2_evict_inode(struct inode *vinode)
1229 {
1230         struct bch_fs *c = vinode->i_sb->s_fs_info;
1231         struct bch_inode_info *inode = to_bch_ei(vinode);
1232
1233         truncate_inode_pages_final(&inode->v.i_data);
1234
1235         clear_inode(&inode->v);
1236
1237         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1238
1239         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1240                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1241                                 KEY_TYPE_QUOTA_WARN);
1242                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1243                                 KEY_TYPE_QUOTA_WARN);
1244                 bch2_inode_rm(c, inode->v.i_ino, true);
1245         }
1246 }
1247
1248 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1249 {
1250         struct super_block *sb = dentry->d_sb;
1251         struct bch_fs *c = sb->s_fs_info;
1252         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1253         unsigned shift = sb->s_blocksize_bits - 9;
1254         /*
1255          * this assumes inodes take up 64 bytes, which is a decent average
1256          * number:
1257          */
1258         u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1259         u64 fsid;
1260
1261         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1262         buf->f_bsize    = sb->s_blocksize;
1263         buf->f_blocks   = usage.capacity >> shift;
1264         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1265         buf->f_bavail   = buf->f_bfree;
1266
1267         buf->f_files    = usage.nr_inodes + avail_inodes;
1268         buf->f_ffree    = avail_inodes;
1269
1270         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1271                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1272         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1273         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1274         buf->f_namelen  = BCH_NAME_MAX;
1275
1276         return 0;
1277 }
1278
1279 static int bch2_sync_fs(struct super_block *sb, int wait)
1280 {
1281         struct bch_fs *c = sb->s_fs_info;
1282
1283         if (c->opts.journal_flush_disabled)
1284                 return 0;
1285
1286         if (!wait) {
1287                 bch2_journal_flush_async(&c->journal, NULL);
1288                 return 0;
1289         }
1290
1291         return bch2_journal_flush(&c->journal);
1292 }
1293
1294 static struct bch_fs *bch2_path_to_fs(const char *path)
1295 {
1296         struct bch_fs *c;
1297         dev_t dev;
1298         int ret;
1299
1300         ret = lookup_bdev(path, &dev);
1301         if (ret)
1302                 return ERR_PTR(ret);
1303
1304         c = bch2_dev_to_fs(dev);
1305         if (c)
1306                 closure_put(&c->cl);
1307         return c ?: ERR_PTR(-ENOENT);
1308 }
1309
1310 static char **split_devs(const char *_dev_name, unsigned *nr)
1311 {
1312         char *dev_name = NULL, **devs = NULL, *s;
1313         size_t i, nr_devs = 0;
1314
1315         if (strlen(_dev_name) == 0)
1316                 return NULL;
1317
1318         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1319         if (!dev_name)
1320                 return NULL;
1321
1322         for (s = dev_name; s; s = strchr(s + 1, ':'))
1323                 nr_devs++;
1324
1325         devs = kcalloc(nr_devs + 1, sizeof(const char *), GFP_KERNEL);
1326         if (!devs) {
1327                 kfree(dev_name);
1328                 return NULL;
1329         }
1330
1331         for (i = 0, s = dev_name;
1332              s;
1333              (s = strchr(s, ':')) && (*s++ = '\0'))
1334                 devs[i++] = s;
1335
1336         *nr = nr_devs;
1337         return devs;
1338 }
1339
1340 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1341 {
1342         struct bch_fs *c = sb->s_fs_info;
1343         struct bch_opts opts = bch2_opts_empty();
1344         int ret;
1345
1346         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1347
1348         ret = bch2_parse_mount_opts(c, &opts, data);
1349         if (ret)
1350                 return ret;
1351
1352         if (opts.read_only != c->opts.read_only) {
1353                 down_write(&c->state_lock);
1354
1355                 if (opts.read_only) {
1356                         bch2_fs_read_only(c);
1357
1358                         sb->s_flags |= SB_RDONLY;
1359                 } else {
1360                         ret = bch2_fs_read_write(c);
1361                         if (ret) {
1362                                 bch_err(c, "error going rw: %i", ret);
1363                                 up_write(&c->state_lock);
1364                                 return -EINVAL;
1365                         }
1366
1367                         sb->s_flags &= ~SB_RDONLY;
1368                 }
1369
1370                 c->opts.read_only = opts.read_only;
1371
1372                 up_write(&c->state_lock);
1373         }
1374
1375         if (opts.errors >= 0)
1376                 c->opts.errors = opts.errors;
1377
1378         return ret;
1379 }
1380
1381 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1382 {
1383         struct bch_fs *c = root->d_sb->s_fs_info;
1384         struct bch_dev *ca;
1385         unsigned i;
1386         bool first = true;
1387
1388         for_each_online_member(ca, c, i) {
1389                 if (!first)
1390                         seq_putc(seq, ':');
1391                 first = false;
1392                 seq_puts(seq, "/dev/");
1393                 seq_puts(seq, ca->name);
1394         }
1395
1396         return 0;
1397 }
1398
1399 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1400 {
1401         struct bch_fs *c = root->d_sb->s_fs_info;
1402         enum bch_opt_id i;
1403         char buf[512];
1404
1405         for (i = 0; i < bch2_opts_nr; i++) {
1406                 const struct bch_option *opt = &bch2_opt_table[i];
1407                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1408
1409                 if (!(opt->mode & OPT_MOUNT))
1410                         continue;
1411
1412                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1413                         continue;
1414
1415                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1416                                  OPT_SHOW_MOUNT_STYLE);
1417                 seq_putc(seq, ',');
1418                 seq_puts(seq, buf);
1419         }
1420
1421         return 0;
1422 }
1423
1424 static void bch2_put_super(struct super_block *sb)
1425 {
1426         struct bch_fs *c = sb->s_fs_info;
1427
1428         __bch2_fs_stop(c);
1429 }
1430
1431 static const struct super_operations bch_super_operations = {
1432         .alloc_inode    = bch2_alloc_inode,
1433         .destroy_inode  = bch2_destroy_inode,
1434         .write_inode    = bch2_vfs_write_inode,
1435         .evict_inode    = bch2_evict_inode,
1436         .sync_fs        = bch2_sync_fs,
1437         .statfs         = bch2_statfs,
1438         .show_devname   = bch2_show_devname,
1439         .show_options   = bch2_show_options,
1440         .remount_fs     = bch2_remount,
1441         .put_super      = bch2_put_super,
1442 #if 0
1443         .freeze_fs      = bch2_freeze,
1444         .unfreeze_fs    = bch2_unfreeze,
1445 #endif
1446 };
1447
1448 static int bch2_set_super(struct super_block *s, void *data)
1449 {
1450         s->s_fs_info = data;
1451         return 0;
1452 }
1453
1454 static int bch2_noset_super(struct super_block *s, void *data)
1455 {
1456         return -EBUSY;
1457 }
1458
1459 static int bch2_test_super(struct super_block *s, void *data)
1460 {
1461         struct bch_fs *c = s->s_fs_info;
1462         struct bch_fs **devs = data;
1463         unsigned i;
1464
1465         if (!c)
1466                 return false;
1467
1468         for (i = 0; devs[i]; i++)
1469                 if (c != devs[i])
1470                         return false;
1471         return true;
1472 }
1473
1474 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1475                                  int flags, const char *dev_name, void *data)
1476 {
1477         struct bch_fs *c;
1478         struct bch_dev *ca;
1479         struct super_block *sb;
1480         struct inode *vinode;
1481         struct bch_opts opts = bch2_opts_empty();
1482         char **devs;
1483         struct bch_fs **devs_to_fs = NULL;
1484         unsigned i, nr_devs;
1485         int ret;
1486
1487         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1488
1489         ret = bch2_parse_mount_opts(NULL, &opts, data);
1490         if (ret)
1491                 return ERR_PTR(ret);
1492
1493         devs = split_devs(dev_name, &nr_devs);
1494         if (!devs)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         devs_to_fs = kcalloc(nr_devs + 1, sizeof(void *), GFP_KERNEL);
1498         if (!devs_to_fs) {
1499                 sb = ERR_PTR(-ENOMEM);
1500                 goto got_sb;
1501         }
1502
1503         for (i = 0; i < nr_devs; i++)
1504                 devs_to_fs[i] = bch2_path_to_fs(devs[i]);
1505
1506         sb = sget(fs_type, bch2_test_super, bch2_noset_super,
1507                   flags|SB_NOSEC, devs_to_fs);
1508         if (!IS_ERR(sb))
1509                 goto got_sb;
1510
1511         c = bch2_fs_open(devs, nr_devs, opts);
1512         if (IS_ERR(c)) {
1513                 sb = ERR_CAST(c);
1514                 goto got_sb;
1515         }
1516
1517         /* Some options can't be parsed until after the fs is started: */
1518         ret = bch2_parse_mount_opts(c, &opts, data);
1519         if (ret) {
1520                 bch2_fs_stop(c);
1521                 sb = ERR_PTR(ret);
1522                 goto got_sb;
1523         }
1524
1525         bch2_opts_apply(&c->opts, opts);
1526
1527         sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1528         if (IS_ERR(sb))
1529                 bch2_fs_stop(c);
1530 got_sb:
1531         kfree(devs_to_fs);
1532         kfree(devs[0]);
1533         kfree(devs);
1534
1535         if (IS_ERR(sb))
1536                 return ERR_CAST(sb);
1537
1538         c = sb->s_fs_info;
1539
1540         if (sb->s_root) {
1541                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1542                         ret = -EBUSY;
1543                         goto err_put_super;
1544                 }
1545                 goto out;
1546         }
1547
1548         sb->s_blocksize         = block_bytes(c);
1549         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1550         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1551         sb->s_op                = &bch_super_operations;
1552         sb->s_export_op         = &bch_export_ops;
1553 #ifdef CONFIG_BCACHEFS_QUOTA
1554         sb->s_qcop              = &bch2_quotactl_operations;
1555         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1556 #endif
1557         sb->s_xattr             = bch2_xattr_handlers;
1558         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1559         sb->s_time_gran         = c->sb.nsec_per_time_unit;
1560         sb->s_time_min          = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1561         sb->s_time_max          = div_s64(S64_MAX, c->sb.time_units_per_sec);
1562         c->vfs_sb               = sb;
1563         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1564
1565         ret = super_setup_bdi(sb);
1566         if (ret)
1567                 goto err_put_super;
1568
1569         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1570
1571         for_each_online_member(ca, c, i) {
1572                 struct block_device *bdev = ca->disk_sb.bdev;
1573
1574                 /* XXX: create an anonymous device for multi device filesystems */
1575                 sb->s_bdev      = bdev;
1576                 sb->s_dev       = bdev->bd_dev;
1577                 percpu_ref_put(&ca->io_ref);
1578                 break;
1579         }
1580
1581 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1582         if (c->opts.acl)
1583                 sb->s_flags     |= SB_POSIXACL;
1584 #endif
1585
1586         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1587         if (IS_ERR(vinode)) {
1588                 bch_err(c, "error mounting: error getting root inode %i",
1589                         (int) PTR_ERR(vinode));
1590                 ret = PTR_ERR(vinode);
1591                 goto err_put_super;
1592         }
1593
1594         sb->s_root = d_make_root(vinode);
1595         if (!sb->s_root) {
1596                 bch_err(c, "error mounting: error allocating root dentry");
1597                 ret = -ENOMEM;
1598                 goto err_put_super;
1599         }
1600
1601         sb->s_flags |= SB_ACTIVE;
1602 out:
1603         return dget(sb->s_root);
1604
1605 err_put_super:
1606         deactivate_locked_super(sb);
1607         return ERR_PTR(ret);
1608 }
1609
1610 static void bch2_kill_sb(struct super_block *sb)
1611 {
1612         struct bch_fs *c = sb->s_fs_info;
1613
1614         generic_shutdown_super(sb);
1615         bch2_fs_free(c);
1616 }
1617
1618 static struct file_system_type bcache_fs_type = {
1619         .owner          = THIS_MODULE,
1620         .name           = "bcachefs",
1621         .mount          = bch2_mount,
1622         .kill_sb        = bch2_kill_sb,
1623         .fs_flags       = FS_REQUIRES_DEV,
1624 };
1625
1626 MODULE_ALIAS_FS("bcachefs");
1627
1628 void bch2_vfs_exit(void)
1629 {
1630         unregister_filesystem(&bcache_fs_type);
1631         if (bch2_inode_cache)
1632                 kmem_cache_destroy(bch2_inode_cache);
1633 }
1634
1635 int __init bch2_vfs_init(void)
1636 {
1637         int ret = -ENOMEM;
1638
1639         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1640         if (!bch2_inode_cache)
1641                 goto err;
1642
1643         ret = register_filesystem(&bcache_fs_type);
1644         if (ret)
1645                 goto err;
1646
1647         return 0;
1648 err:
1649         bch2_vfs_exit();
1650         return ret;
1651 }
1652
1653 #endif /* NO_BCACHEFS_FS */