1 // SPDX-License-Identifier: GPL-2.0
2 #ifndef NO_BCACHEFS_FS
3
4 #include "bcachefs.h"
5 #include "acl.h"
6 #include "bkey_buf.h"
7 #include "btree_update.h"
8 #include "buckets.h"
9 #include "chardev.h"
10 #include "dirent.h"
11 #include "errcode.h"
12 #include "extents.h"
13 #include "fs.h"
14 #include "fs-common.h"
15 #include "fs-io.h"
16 #include "fs-ioctl.h"
17 #include "fs-io-buffered.h"
18 #include "fs-io-direct.h"
19 #include "fs-io-pagecache.h"
20 #include "fsck.h"
21 #include "inode.h"
22 #include "io_read.h"
23 #include "journal.h"
24 #include "keylist.h"
25 #include "quota.h"
26 #include "snapshot.h"
27 #include "super.h"
28 #include "xattr.h"
29
30 #include <linux/aio.h>
31 #include <linux/backing-dev.h>
32 #include <linux/exportfs.h>
33 #include <linux/fiemap.h>
34 #include <linux/module.h>
35 #include <linux/pagemap.h>
36 #include <linux/posix_acl.h>
37 #include <linux/random.h>
38 #include <linux/seq_file.h>
39 #include <linux/statfs.h>
40 #include <linux/string.h>
41 #include <linux/xattr.h>
42
43 static struct kmem_cache *bch2_inode_cache;
44
45 static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
46 struct bch_inode_info *,
47 struct bch_inode_unpacked *,
48 struct bch_subvolume *);
49
bch2_inode_update_after_write(struct btree_trans * trans,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,unsigned fields)50 void bch2_inode_update_after_write(struct btree_trans *trans,
51 struct bch_inode_info *inode,
52 struct bch_inode_unpacked *bi,
53 unsigned fields)
54 {
55 struct bch_fs *c = trans->c;
56
57 BUG_ON(bi->bi_inum != inode->v.i_ino);
58
59 bch2_assert_pos_locked(trans, BTREE_ID_inodes,
60 POS(0, bi->bi_inum),
61 c->opts.inodes_use_key_cache);
62
63 set_nlink(&inode->v, bch2_inode_nlink_get(bi));
64 i_uid_write(&inode->v, bi->bi_uid);
65 i_gid_write(&inode->v, bi->bi_gid);
66 inode->v.i_mode = bi->bi_mode;
67
68 if (fields & ATTR_ATIME)
69 inode_set_atime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_atime));
70 if (fields & ATTR_MTIME)
71 inode_set_mtime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_mtime));
72 if (fields & ATTR_CTIME)
73 inode_set_ctime_to_ts(&inode->v, bch2_time_to_timespec(c, bi->bi_ctime));
74
75 inode->ei_inode = *bi;
76
77 bch2_inode_flags_to_vfs(inode);
78 }
79
bch2_write_inode(struct bch_fs * c,struct bch_inode_info * inode,inode_set_fn set,void * p,unsigned fields)80 int __must_check bch2_write_inode(struct bch_fs *c,
81 struct bch_inode_info *inode,
82 inode_set_fn set,
83 void *p, unsigned fields)
84 {
85 struct btree_trans *trans = bch2_trans_get(c);
86 struct btree_iter iter = { NULL };
87 struct bch_inode_unpacked inode_u;
88 int ret;
89 retry:
90 bch2_trans_begin(trans);
91
92 ret = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
93 BTREE_ITER_intent) ?:
94 (set ? set(trans, inode, &inode_u, p) : 0) ?:
95 bch2_inode_write(trans, &iter, &inode_u) ?:
96 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
97
98 /*
99 * the btree node lock protects inode->ei_inode, not ei_update_lock;
100 * this is important for inode updates via bchfs_write_index_update
101 */
102 if (!ret)
103 bch2_inode_update_after_write(trans, inode, &inode_u, fields);
104
105 bch2_trans_iter_exit(trans, &iter);
106
107 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
108 goto retry;
109
110 bch2_fs_fatal_err_on(bch2_err_matches(ret, ENOENT), c,
111 "%s: inode %u:%llu not found when updating",
112 bch2_err_str(ret),
113 inode_inum(inode).subvol,
114 inode_inum(inode).inum);
115
116 bch2_trans_put(trans);
117 return ret < 0 ? ret : 0;
118 }
119
bch2_fs_quota_transfer(struct bch_fs * c,struct bch_inode_info * inode,struct bch_qid new_qid,unsigned qtypes,enum quota_acct_mode mode)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
bch2_iget5_test(struct inode * vinode,void * p)155 static int bch2_iget5_test(struct inode *vinode, void *p)
156 {
157 struct bch_inode_info *inode = to_bch_ei(vinode);
158 subvol_inum *inum = p;
159
160 return inode->ei_subvol == inum->subvol &&
161 inode->ei_inode.bi_inum == inum->inum;
162 }
163
bch2_iget5_set(struct inode * vinode,void * p)164 static int bch2_iget5_set(struct inode *vinode, void *p)
165 {
166 struct bch_inode_info *inode = to_bch_ei(vinode);
167 subvol_inum *inum = p;
168
169 inode->v.i_ino = inum->inum;
170 inode->ei_subvol = inum->subvol;
171 inode->ei_inode.bi_inum = inum->inum;
172 return 0;
173 }
174
bch2_inode_hash(subvol_inum inum)175 static unsigned bch2_inode_hash(subvol_inum inum)
176 {
177 return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
178 }
179
bch2_inode_insert(struct bch_fs * c,struct bch_inode_info * inode)180 static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_inode_info *inode)
181 {
182 subvol_inum inum = inode_inum(inode);
183 struct bch_inode_info *old = to_bch_ei(inode_insert5(&inode->v,
184 bch2_inode_hash(inum),
185 bch2_iget5_test,
186 bch2_iget5_set,
187 &inum));
188 BUG_ON(!old);
189
190 if (unlikely(old != inode)) {
191 discard_new_inode(&inode->v);
192 inode = old;
193 } else {
194 mutex_lock(&c->vfs_inodes_lock);
195 list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
196 mutex_unlock(&c->vfs_inodes_lock);
197 /*
198 * we really don't want insert_inode_locked2() to be setting
199 * I_NEW...
200 */
201 unlock_new_inode(&inode->v);
202 }
203
204 return inode;
205 }
206
207 #define memalloc_flags_do(_flags, _do) \
208 ({ \
209 unsigned _saved_flags = memalloc_flags_save(_flags); \
210 typeof(_do) _ret = _do; \
211 memalloc_noreclaim_restore(_saved_flags); \
212 _ret; \
213 })
214
bch2_alloc_inode(struct super_block * sb)215 static struct inode *bch2_alloc_inode(struct super_block *sb)
216 {
217 BUG();
218 }
219
__bch2_new_inode(struct bch_fs * c)220 static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
221 {
222 struct bch_inode_info *inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
223 if (!inode)
224 return NULL;
225
226 inode_init_once(&inode->v);
227 mutex_init(&inode->ei_update_lock);
228 two_state_lock_init(&inode->ei_pagecache_lock);
229 INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
230 inode->ei_flags = 0;
231 mutex_init(&inode->ei_quota_lock);
232 memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
233 inode->v.i_state = 0;
234
235 if (unlikely(inode_init_always(c->vfs_sb, &inode->v))) {
236 kmem_cache_free(bch2_inode_cache, inode);
237 return NULL;
238 }
239
240 return inode;
241 }
242
243 /*
244 * Allocate a new inode, dropping/retaking btree locks if necessary:
245 */
bch2_new_inode(struct btree_trans * trans)246 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
247 {
248 struct bch_inode_info *inode =
249 memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
250 __bch2_new_inode(trans->c));
251
252 if (unlikely(!inode)) {
253 int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c)) ? 0 : -ENOMEM);
254 if (ret && inode) {
255 __destroy_inode(&inode->v);
256 kmem_cache_free(bch2_inode_cache, inode);
257 }
258 if (ret)
259 return ERR_PTR(ret);
260 }
261
262 return inode;
263 }
264
bch2_vfs_inode_get(struct bch_fs * c,subvol_inum inum)265 struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
266 {
267 struct bch_inode_info *inode =
268 to_bch_ei(ilookup5_nowait(c->vfs_sb,
269 bch2_inode_hash(inum),
270 bch2_iget5_test,
271 &inum));
272 if (inode)
273 return &inode->v;
274
275 struct btree_trans *trans = bch2_trans_get(c);
276
277 struct bch_inode_unpacked inode_u;
278 struct bch_subvolume subvol;
279 int ret = lockrestart_do(trans,
280 bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
281 bch2_inode_find_by_inum_trans(trans, inum, &inode_u)) ?:
282 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
283 if (!ret) {
284 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
285 inode = bch2_inode_insert(c, inode);
286 }
287 bch2_trans_put(trans);
288
289 return ret ? ERR_PTR(ret) : &inode->v;
290 }
291
292 struct bch_inode_info *
__bch2_create(struct mnt_idmap * idmap,struct bch_inode_info * dir,struct dentry * dentry,umode_t mode,dev_t rdev,subvol_inum snapshot_src,unsigned flags)293 __bch2_create(struct mnt_idmap *idmap,
294 struct bch_inode_info *dir, struct dentry *dentry,
295 umode_t mode, dev_t rdev, subvol_inum snapshot_src,
296 unsigned flags)
297 {
298 struct bch_fs *c = dir->v.i_sb->s_fs_info;
299 struct btree_trans *trans;
300 struct bch_inode_unpacked dir_u;
301 struct bch_inode_info *inode;
302 struct bch_inode_unpacked inode_u;
303 struct posix_acl *default_acl = NULL, *acl = NULL;
304 subvol_inum inum;
305 struct bch_subvolume subvol;
306 u64 journal_seq = 0;
307 int ret;
308
309 /*
310 * preallocate acls + vfs inode before btree transaction, so that
311 * nothing can fail after the transaction succeeds:
312 */
313 #ifdef CONFIG_BCACHEFS_POSIX_ACL
314 ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
315 if (ret)
316 return ERR_PTR(ret);
317 #endif
318 inode = __bch2_new_inode(c);
319 if (unlikely(!inode)) {
320 inode = ERR_PTR(-ENOMEM);
321 goto err;
322 }
323
324 bch2_inode_init_early(c, &inode_u);
325
326 if (!(flags & BCH_CREATE_TMPFILE))
327 mutex_lock(&dir->ei_update_lock);
328
329 trans = bch2_trans_get(c);
330 retry:
331 bch2_trans_begin(trans);
332
333 ret = bch2_subvol_is_ro_trans(trans, dir->ei_subvol) ?:
334 bch2_create_trans(trans,
335 inode_inum(dir), &dir_u, &inode_u,
336 !(flags & BCH_CREATE_TMPFILE)
337 ? &dentry->d_name : NULL,
338 from_kuid(i_user_ns(&dir->v), current_fsuid()),
339 from_kgid(i_user_ns(&dir->v), current_fsgid()),
340 mode, rdev,
341 default_acl, acl, snapshot_src, flags) ?:
342 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
343 KEY_TYPE_QUOTA_PREALLOC);
344 if (unlikely(ret))
345 goto err_before_quota;
346
347 inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
348 inum.inum = inode_u.bi_inum;
349
350 ret = bch2_subvolume_get(trans, inum.subvol, true,
351 BTREE_ITER_with_updates, &subvol) ?:
352 bch2_trans_commit(trans, NULL, &journal_seq, 0);
353 if (unlikely(ret)) {
354 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
355 KEY_TYPE_QUOTA_WARN);
356 err_before_quota:
357 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
358 goto retry;
359 goto err_trans;
360 }
361
362 if (!(flags & BCH_CREATE_TMPFILE)) {
363 bch2_inode_update_after_write(trans, dir, &dir_u,
364 ATTR_MTIME|ATTR_CTIME);
365 mutex_unlock(&dir->ei_update_lock);
366 }
367
368 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
369
370 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
371 set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
372
373 /*
374 * we must insert the new inode into the inode cache before calling
375 * bch2_trans_exit() and dropping locks, else we could race with another
376 * thread pulling the inode in and modifying it:
377 */
378 inode = bch2_inode_insert(c, inode);
379 bch2_trans_put(trans);
380 err:
381 posix_acl_release(default_acl);
382 posix_acl_release(acl);
383 return inode;
384 err_trans:
385 if (!(flags & BCH_CREATE_TMPFILE))
386 mutex_unlock(&dir->ei_update_lock);
387
388 bch2_trans_put(trans);
389 make_bad_inode(&inode->v);
390 iput(&inode->v);
391 inode = ERR_PTR(ret);
392 goto err;
393 }
394
395 /* methods */
396
bch2_lookup_trans(struct btree_trans * trans,subvol_inum dir,struct bch_hash_info * dir_hash_info,const struct qstr * name)397 static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
398 subvol_inum dir, struct bch_hash_info *dir_hash_info,
399 const struct qstr *name)
400 {
401 struct bch_fs *c = trans->c;
402 struct btree_iter dirent_iter = {};
403 subvol_inum inum = {};
404 struct printbuf buf = PRINTBUF;
405
406 struct bkey_s_c k = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
407 dir_hash_info, dir, name, 0);
408 int ret = bkey_err(k);
409 if (ret)
410 return ERR_PTR(ret);
411
412 ret = bch2_dirent_read_target(trans, dir, bkey_s_c_to_dirent(k), &inum);
413 if (ret > 0)
414 ret = -ENOENT;
415 if (ret)
416 goto err;
417
418 struct bch_inode_info *inode =
419 to_bch_ei(ilookup5_nowait(c->vfs_sb,
420 bch2_inode_hash(inum),
421 bch2_iget5_test,
422 &inum));
423 if (inode)
424 goto out;
425
426 struct bch_subvolume subvol;
427 struct bch_inode_unpacked inode_u;
428 ret = bch2_subvolume_get(trans, inum.subvol, true, 0, &subvol) ?:
429 bch2_inode_find_by_inum_nowarn_trans(trans, inum, &inode_u) ?:
430 PTR_ERR_OR_ZERO(inode = bch2_new_inode(trans));
431
432 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
433 c, "dirent to missing inode:\n %s",
434 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
435 if (ret)
436 goto err;
437
438 /* regular files may have hardlinks: */
439 if (bch2_fs_inconsistent_on(bch2_inode_should_have_bp(&inode_u) &&
440 !bkey_eq(k.k->p, POS(inode_u.bi_dir, inode_u.bi_dir_offset)),
441 c,
442 "dirent points to inode that does not point back:\n %s",
443 (bch2_bkey_val_to_text(&buf, c, k),
444 prt_printf(&buf, "\n "),
445 bch2_inode_unpacked_to_text(&buf, &inode_u),
446 buf.buf))) {
447 ret = -ENOENT;
448 goto err;
449 }
450
451 bch2_vfs_inode_init(trans, inum, inode, &inode_u, &subvol);
452 inode = bch2_inode_insert(c, inode);
453 out:
454 bch2_trans_iter_exit(trans, &dirent_iter);
455 printbuf_exit(&buf);
456 return inode;
457 err:
458 inode = ERR_PTR(ret);
459 goto out;
460 }
461
bch2_lookup(struct inode * vdir,struct dentry * dentry,unsigned int flags)462 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
463 unsigned int flags)
464 {
465 struct bch_fs *c = vdir->i_sb->s_fs_info;
466 struct bch_inode_info *dir = to_bch_ei(vdir);
467 struct bch_hash_info hash = bch2_hash_info_init(c, &dir->ei_inode);
468
469 struct bch_inode_info *inode;
470 bch2_trans_do(c, NULL, NULL, 0,
471 PTR_ERR_OR_ZERO(inode = bch2_lookup_trans(trans, inode_inum(dir),
472 &hash, &dentry->d_name)));
473 if (IS_ERR(inode))
474 inode = NULL;
475
476 return d_splice_alias(&inode->v, dentry);
477 }
478
bch2_mknod(struct mnt_idmap * idmap,struct inode * vdir,struct dentry * dentry,umode_t mode,dev_t rdev)479 static int bch2_mknod(struct mnt_idmap *idmap,
480 struct inode *vdir, struct dentry *dentry,
481 umode_t mode, dev_t rdev)
482 {
483 struct bch_inode_info *inode =
484 __bch2_create(idmap, to_bch_ei(vdir), dentry, mode, rdev,
485 (subvol_inum) { 0 }, 0);
486
487 if (IS_ERR(inode))
488 return bch2_err_class(PTR_ERR(inode));
489
490 d_instantiate(dentry, &inode->v);
491 return 0;
492 }
493
bch2_create(struct mnt_idmap * idmap,struct inode * vdir,struct dentry * dentry,umode_t mode,bool excl)494 static int bch2_create(struct mnt_idmap *idmap,
495 struct inode *vdir, struct dentry *dentry,
496 umode_t mode, bool excl)
497 {
498 return bch2_mknod(idmap, vdir, dentry, mode|S_IFREG, 0);
499 }
500
__bch2_link(struct bch_fs * c,struct bch_inode_info * inode,struct bch_inode_info * dir,struct dentry * dentry)501 static int __bch2_link(struct bch_fs *c,
502 struct bch_inode_info *inode,
503 struct bch_inode_info *dir,
504 struct dentry *dentry)
505 {
506 struct btree_trans *trans = bch2_trans_get(c);
507 struct bch_inode_unpacked dir_u, inode_u;
508 int ret;
509
510 mutex_lock(&inode->ei_update_lock);
511
512 ret = commit_do(trans, NULL, NULL, 0,
513 bch2_link_trans(trans,
514 inode_inum(dir), &dir_u,
515 inode_inum(inode), &inode_u,
516 &dentry->d_name));
517
518 if (likely(!ret)) {
519 bch2_inode_update_after_write(trans, dir, &dir_u,
520 ATTR_MTIME|ATTR_CTIME);
521 bch2_inode_update_after_write(trans, inode, &inode_u, ATTR_CTIME);
522 }
523
524 bch2_trans_put(trans);
525 mutex_unlock(&inode->ei_update_lock);
526 return ret;
527 }
528
bch2_link(struct dentry * old_dentry,struct inode * vdir,struct dentry * dentry)529 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
530 struct dentry *dentry)
531 {
532 struct bch_fs *c = vdir->i_sb->s_fs_info;
533 struct bch_inode_info *dir = to_bch_ei(vdir);
534 struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
535 int ret;
536
537 lockdep_assert_held(&inode->v.i_rwsem);
538
539 ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
540 bch2_subvol_is_ro(c, inode->ei_subvol) ?:
541 __bch2_link(c, inode, dir, dentry);
542 if (unlikely(ret))
543 return bch2_err_class(ret);
544
545 ihold(&inode->v);
546 d_instantiate(dentry, &inode->v);
547 return 0;
548 }
549
__bch2_unlink(struct inode * vdir,struct dentry * dentry,bool deleting_snapshot)550 int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
551 bool deleting_snapshot)
552 {
553 struct bch_fs *c = vdir->i_sb->s_fs_info;
554 struct bch_inode_info *dir = to_bch_ei(vdir);
555 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
556 struct bch_inode_unpacked dir_u, inode_u;
557 struct btree_trans *trans = bch2_trans_get(c);
558 int ret;
559
560 bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
561
562 ret = commit_do(trans, NULL, NULL,
563 BCH_TRANS_COMMIT_no_enospc,
564 bch2_unlink_trans(trans,
565 inode_inum(dir), &dir_u,
566 &inode_u, &dentry->d_name,
567 deleting_snapshot));
568 if (unlikely(ret))
569 goto err;
570
571 bch2_inode_update_after_write(trans, dir, &dir_u,
572 ATTR_MTIME|ATTR_CTIME);
573 bch2_inode_update_after_write(trans, inode, &inode_u,
574 ATTR_MTIME);
575
576 if (inode_u.bi_subvol) {
577 /*
578 * Subvolume deletion is asynchronous, but we still want to tell
579 * the VFS that it's been deleted here:
580 */
581 set_nlink(&inode->v, 0);
582 }
583 err:
584 bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
585 bch2_trans_put(trans);
586
587 return ret;
588 }
589
bch2_unlink(struct inode * vdir,struct dentry * dentry)590 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
591 {
592 struct bch_inode_info *dir= to_bch_ei(vdir);
593 struct bch_fs *c = dir->v.i_sb->s_fs_info;
594
595 int ret = bch2_subvol_is_ro(c, dir->ei_subvol) ?:
596 __bch2_unlink(vdir, dentry, false);
597 return bch2_err_class(ret);
598 }
599
bch2_symlink(struct mnt_idmap * idmap,struct inode * vdir,struct dentry * dentry,const char * symname)600 static int bch2_symlink(struct mnt_idmap *idmap,
601 struct inode *vdir, struct dentry *dentry,
602 const char *symname)
603 {
604 struct bch_fs *c = vdir->i_sb->s_fs_info;
605 struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
606 int ret;
607
608 inode = __bch2_create(idmap, dir, dentry, S_IFLNK|S_IRWXUGO, 0,
609 (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
610 if (IS_ERR(inode))
611 return bch2_err_class(PTR_ERR(inode));
612
613 inode_lock(&inode->v);
614 ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
615 inode_unlock(&inode->v);
616
617 if (unlikely(ret))
618 goto err;
619
620 ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
621 if (unlikely(ret))
622 goto err;
623
624 ret = __bch2_link(c, inode, dir, dentry);
625 if (unlikely(ret))
626 goto err;
627
628 d_instantiate(dentry, &inode->v);
629 return 0;
630 err:
631 iput(&inode->v);
632 return bch2_err_class(ret);
633 }
634
bch2_mkdir(struct mnt_idmap * idmap,struct inode * vdir,struct dentry * dentry,umode_t mode)635 static int bch2_mkdir(struct mnt_idmap *idmap,
636 struct inode *vdir, struct dentry *dentry, umode_t mode)
637 {
638 return bch2_mknod(idmap, vdir, dentry, mode|S_IFDIR, 0);
639 }
640
bch2_rename2(struct mnt_idmap * idmap,struct inode * src_vdir,struct dentry * src_dentry,struct inode * dst_vdir,struct dentry * dst_dentry,unsigned flags)641 static int bch2_rename2(struct mnt_idmap *idmap,
642 struct inode *src_vdir, struct dentry *src_dentry,
643 struct inode *dst_vdir, struct dentry *dst_dentry,
644 unsigned flags)
645 {
646 struct bch_fs *c = src_vdir->i_sb->s_fs_info;
647 struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
648 struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
649 struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
650 struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
651 struct bch_inode_unpacked dst_dir_u, src_dir_u;
652 struct bch_inode_unpacked src_inode_u, dst_inode_u;
653 struct btree_trans *trans;
654 enum bch_rename_mode mode = flags & RENAME_EXCHANGE
655 ? BCH_RENAME_EXCHANGE
656 : dst_dentry->d_inode
657 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
658 int ret;
659
660 if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
661 return -EINVAL;
662
663 if (mode == BCH_RENAME_OVERWRITE) {
664 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
665 0, LLONG_MAX);
666 if (ret)
667 return ret;
668 }
669
670 trans = bch2_trans_get(c);
671
672 bch2_lock_inodes(INODE_UPDATE_LOCK,
673 src_dir,
674 dst_dir,
675 src_inode,
676 dst_inode);
677
678 ret = bch2_subvol_is_ro_trans(trans, src_dir->ei_subvol) ?:
679 bch2_subvol_is_ro_trans(trans, dst_dir->ei_subvol);
680 if (ret)
681 goto err;
682
683 if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
684 ret = bch2_fs_quota_transfer(c, src_inode,
685 dst_dir->ei_qid,
686 1 << QTYP_PRJ,
687 KEY_TYPE_QUOTA_PREALLOC);
688 if (ret)
689 goto err;
690 }
691
692 if (mode == BCH_RENAME_EXCHANGE &&
693 inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
694 ret = bch2_fs_quota_transfer(c, dst_inode,
695 src_dir->ei_qid,
696 1 << QTYP_PRJ,
697 KEY_TYPE_QUOTA_PREALLOC);
698 if (ret)
699 goto err;
700 }
701
702 ret = commit_do(trans, NULL, NULL, 0,
703 bch2_rename_trans(trans,
704 inode_inum(src_dir), &src_dir_u,
705 inode_inum(dst_dir), &dst_dir_u,
706 &src_inode_u,
707 &dst_inode_u,
708 &src_dentry->d_name,
709 &dst_dentry->d_name,
710 mode));
711 if (unlikely(ret))
712 goto err;
713
714 BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
715 BUG_ON(dst_inode &&
716 dst_inode->v.i_ino != dst_inode_u.bi_inum);
717
718 bch2_inode_update_after_write(trans, src_dir, &src_dir_u,
719 ATTR_MTIME|ATTR_CTIME);
720
721 if (src_dir != dst_dir)
722 bch2_inode_update_after_write(trans, dst_dir, &dst_dir_u,
723 ATTR_MTIME|ATTR_CTIME);
724
725 bch2_inode_update_after_write(trans, src_inode, &src_inode_u,
726 ATTR_CTIME);
727
728 if (dst_inode)
729 bch2_inode_update_after_write(trans, dst_inode, &dst_inode_u,
730 ATTR_CTIME);
731 err:
732 bch2_trans_put(trans);
733
734 bch2_fs_quota_transfer(c, src_inode,
735 bch_qid(&src_inode->ei_inode),
736 1 << QTYP_PRJ,
737 KEY_TYPE_QUOTA_NOCHECK);
738 if (dst_inode)
739 bch2_fs_quota_transfer(c, dst_inode,
740 bch_qid(&dst_inode->ei_inode),
741 1 << QTYP_PRJ,
742 KEY_TYPE_QUOTA_NOCHECK);
743
744 bch2_unlock_inodes(INODE_UPDATE_LOCK,
745 src_dir,
746 dst_dir,
747 src_inode,
748 dst_inode);
749
750 return bch2_err_class(ret);
751 }
752
bch2_setattr_copy(struct mnt_idmap * idmap,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,struct iattr * attr)753 static void bch2_setattr_copy(struct mnt_idmap *idmap,
754 struct bch_inode_info *inode,
755 struct bch_inode_unpacked *bi,
756 struct iattr *attr)
757 {
758 struct bch_fs *c = inode->v.i_sb->s_fs_info;
759 unsigned int ia_valid = attr->ia_valid;
760
761 if (ia_valid & ATTR_UID)
762 bi->bi_uid = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
763 if (ia_valid & ATTR_GID)
764 bi->bi_gid = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
765
766 if (ia_valid & ATTR_SIZE)
767 bi->bi_size = attr->ia_size;
768
769 if (ia_valid & ATTR_ATIME)
770 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
771 if (ia_valid & ATTR_MTIME)
772 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
773 if (ia_valid & ATTR_CTIME)
774 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
775
776 if (ia_valid & ATTR_MODE) {
777 umode_t mode = attr->ia_mode;
778 kgid_t gid = ia_valid & ATTR_GID
779 ? attr->ia_gid
780 : inode->v.i_gid;
781
782 if (!in_group_p(gid) &&
783 !capable_wrt_inode_uidgid(idmap, &inode->v, CAP_FSETID))
784 mode &= ~S_ISGID;
785 bi->bi_mode = mode;
786 }
787 }
788
bch2_setattr_nonsize(struct mnt_idmap * idmap,struct bch_inode_info * inode,struct iattr * attr)789 int bch2_setattr_nonsize(struct mnt_idmap *idmap,
790 struct bch_inode_info *inode,
791 struct iattr *attr)
792 {
793 struct bch_fs *c = inode->v.i_sb->s_fs_info;
794 struct bch_qid qid;
795 struct btree_trans *trans;
796 struct btree_iter inode_iter = { NULL };
797 struct bch_inode_unpacked inode_u;
798 struct posix_acl *acl = NULL;
799 int ret;
800
801 mutex_lock(&inode->ei_update_lock);
802
803 qid = inode->ei_qid;
804
805 if (attr->ia_valid & ATTR_UID)
806 qid.q[QTYP_USR] = from_kuid(i_user_ns(&inode->v), attr->ia_uid);
807
808 if (attr->ia_valid & ATTR_GID)
809 qid.q[QTYP_GRP] = from_kgid(i_user_ns(&inode->v), attr->ia_gid);
810
811 ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
812 KEY_TYPE_QUOTA_PREALLOC);
813 if (ret)
814 goto err;
815
816 trans = bch2_trans_get(c);
817 retry:
818 bch2_trans_begin(trans);
819 kfree(acl);
820 acl = NULL;
821
822 ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
823 BTREE_ITER_intent);
824 if (ret)
825 goto btree_err;
826
827 bch2_setattr_copy(idmap, inode, &inode_u, attr);
828
829 if (attr->ia_valid & ATTR_MODE) {
830 ret = bch2_acl_chmod(trans, inode_inum(inode), &inode_u,
831 inode_u.bi_mode, &acl);
832 if (ret)
833 goto btree_err;
834 }
835
836 ret = bch2_inode_write(trans, &inode_iter, &inode_u) ?:
837 bch2_trans_commit(trans, NULL, NULL,
838 BCH_TRANS_COMMIT_no_enospc);
839 btree_err:
840 bch2_trans_iter_exit(trans, &inode_iter);
841
842 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
843 goto retry;
844 if (unlikely(ret))
845 goto err_trans;
846
847 bch2_inode_update_after_write(trans, inode, &inode_u, attr->ia_valid);
848
849 if (acl)
850 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
851 err_trans:
852 bch2_trans_put(trans);
853 err:
854 mutex_unlock(&inode->ei_update_lock);
855
856 return bch2_err_class(ret);
857 }
858
bch2_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned query_flags)859 static int bch2_getattr(struct mnt_idmap *idmap,
860 const struct path *path, struct kstat *stat,
861 u32 request_mask, unsigned query_flags)
862 {
863 struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
864 struct bch_fs *c = inode->v.i_sb->s_fs_info;
865
866 stat->dev = inode->v.i_sb->s_dev;
867 stat->ino = inode->v.i_ino;
868 stat->mode = inode->v.i_mode;
869 stat->nlink = inode->v.i_nlink;
870 stat->uid = inode->v.i_uid;
871 stat->gid = inode->v.i_gid;
872 stat->rdev = inode->v.i_rdev;
873 stat->size = i_size_read(&inode->v);
874 stat->atime = inode_get_atime(&inode->v);
875 stat->mtime = inode_get_mtime(&inode->v);
876 stat->ctime = inode_get_ctime(&inode->v);
877 stat->blksize = block_bytes(c);
878 stat->blocks = inode->v.i_blocks;
879
880 stat->subvol = inode->ei_subvol;
881 stat->result_mask |= STATX_SUBVOL;
882
883 if (request_mask & STATX_BTIME) {
884 stat->result_mask |= STATX_BTIME;
885 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
886 }
887
888 if (inode->ei_inode.bi_flags & BCH_INODE_immutable)
889 stat->attributes |= STATX_ATTR_IMMUTABLE;
890 stat->attributes_mask |= STATX_ATTR_IMMUTABLE;
891
892 if (inode->ei_inode.bi_flags & BCH_INODE_append)
893 stat->attributes |= STATX_ATTR_APPEND;
894 stat->attributes_mask |= STATX_ATTR_APPEND;
895
896 if (inode->ei_inode.bi_flags & BCH_INODE_nodump)
897 stat->attributes |= STATX_ATTR_NODUMP;
898 stat->attributes_mask |= STATX_ATTR_NODUMP;
899
900 return 0;
901 }
902
bch2_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * iattr)903 static int bch2_setattr(struct mnt_idmap *idmap,
904 struct dentry *dentry, struct iattr *iattr)
905 {
906 struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
907 struct bch_fs *c = inode->v.i_sb->s_fs_info;
908 int ret;
909
910 lockdep_assert_held(&inode->v.i_rwsem);
911
912 ret = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
913 setattr_prepare(idmap, dentry, iattr);
914 if (ret)
915 return ret;
916
917 return iattr->ia_valid & ATTR_SIZE
918 ? bchfs_truncate(idmap, inode, iattr)
919 : bch2_setattr_nonsize(idmap, inode, iattr);
920 }
921
bch2_tmpfile(struct mnt_idmap * idmap,struct inode * vdir,struct file * file,umode_t mode)922 static int bch2_tmpfile(struct mnt_idmap *idmap,
923 struct inode *vdir, struct file *file, umode_t mode)
924 {
925 struct bch_inode_info *inode =
926 __bch2_create(idmap, to_bch_ei(vdir),
927 file->f_path.dentry, mode, 0,
928 (subvol_inum) { 0 }, BCH_CREATE_TMPFILE);
929
930 if (IS_ERR(inode))
931 return bch2_err_class(PTR_ERR(inode));
932
933 d_mark_tmpfile(file, &inode->v);
934 d_instantiate(file->f_path.dentry, &inode->v);
935 return finish_open_simple(file, 0);
936 }
937
bch2_fill_extent(struct bch_fs * c,struct fiemap_extent_info * info,struct bkey_s_c k,unsigned flags)938 static int bch2_fill_extent(struct bch_fs *c,
939 struct fiemap_extent_info *info,
940 struct bkey_s_c k, unsigned flags)
941 {
942 if (bkey_extent_is_direct_data(k.k)) {
943 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
944 const union bch_extent_entry *entry;
945 struct extent_ptr_decoded p;
946 int ret;
947
948 if (k.k->type == KEY_TYPE_reflink_v)
949 flags |= FIEMAP_EXTENT_SHARED;
950
951 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
952 int flags2 = 0;
953 u64 offset = p.ptr.offset;
954
955 if (p.ptr.unwritten)
956 flags2 |= FIEMAP_EXTENT_UNWRITTEN;
957
958 if (p.crc.compression_type)
959 flags2 |= FIEMAP_EXTENT_ENCODED;
960 else
961 offset += p.crc.offset;
962
963 if ((offset & (block_sectors(c) - 1)) ||
964 (k.k->size & (block_sectors(c) - 1)))
965 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
966
967 ret = fiemap_fill_next_extent(info,
968 bkey_start_offset(k.k) << 9,
969 offset << 9,
970 k.k->size << 9, flags|flags2);
971 if (ret)
972 return ret;
973 }
974
975 return 0;
976 } else if (bkey_extent_is_inline_data(k.k)) {
977 return fiemap_fill_next_extent(info,
978 bkey_start_offset(k.k) << 9,
979 0, k.k->size << 9,
980 flags|
981 FIEMAP_EXTENT_DATA_INLINE);
982 } else if (k.k->type == KEY_TYPE_reservation) {
983 return fiemap_fill_next_extent(info,
984 bkey_start_offset(k.k) << 9,
985 0, k.k->size << 9,
986 flags|
987 FIEMAP_EXTENT_DELALLOC|
988 FIEMAP_EXTENT_UNWRITTEN);
989 } else {
990 BUG();
991 }
992 }
993
bch2_fiemap(struct inode * vinode,struct fiemap_extent_info * info,u64 start,u64 len)994 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
995 u64 start, u64 len)
996 {
997 struct bch_fs *c = vinode->i_sb->s_fs_info;
998 struct bch_inode_info *ei = to_bch_ei(vinode);
999 struct btree_trans *trans;
1000 struct btree_iter iter;
1001 struct bkey_s_c k;
1002 struct bkey_buf cur, prev;
1003 unsigned offset_into_extent, sectors;
1004 bool have_extent = false;
1005 u32 snapshot;
1006 int ret = 0;
1007
1008 ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
1009 if (ret)
1010 return ret;
1011
1012 struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
1013 if (start + len < start)
1014 return -EINVAL;
1015
1016 start >>= 9;
1017
1018 bch2_bkey_buf_init(&cur);
1019 bch2_bkey_buf_init(&prev);
1020 trans = bch2_trans_get(c);
1021 retry:
1022 bch2_trans_begin(trans);
1023
1024 ret = bch2_subvolume_get_snapshot(trans, ei->ei_subvol, &snapshot);
1025 if (ret)
1026 goto err;
1027
1028 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
1029 SPOS(ei->v.i_ino, start, snapshot), 0);
1030
1031 while (!(ret = btree_trans_too_many_iters(trans)) &&
1032 (k = bch2_btree_iter_peek_upto(&iter, end)).k &&
1033 !(ret = bkey_err(k))) {
1034 enum btree_id data_btree = BTREE_ID_extents;
1035
1036 if (!bkey_extent_is_data(k.k) &&
1037 k.k->type != KEY_TYPE_reservation) {
1038 bch2_btree_iter_advance(&iter);
1039 continue;
1040 }
1041
1042 offset_into_extent = iter.pos.offset -
1043 bkey_start_offset(k.k);
1044 sectors = k.k->size - offset_into_extent;
1045
1046 bch2_bkey_buf_reassemble(&cur, c, k);
1047
1048 ret = bch2_read_indirect_extent(trans, &data_btree,
1049 &offset_into_extent, &cur);
1050 if (ret)
1051 break;
1052
1053 k = bkey_i_to_s_c(cur.k);
1054 bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
1055
1056 sectors = min(sectors, k.k->size - offset_into_extent);
1057
1058 bch2_cut_front(POS(k.k->p.inode,
1059 bkey_start_offset(k.k) +
1060 offset_into_extent),
1061 cur.k);
1062 bch2_key_resize(&cur.k->k, sectors);
1063 cur.k->k.p = iter.pos;
1064 cur.k->k.p.offset += cur.k->k.size;
1065
1066 if (have_extent) {
1067 bch2_trans_unlock(trans);
1068 ret = bch2_fill_extent(c, info,
1069 bkey_i_to_s_c(prev.k), 0);
1070 if (ret)
1071 break;
1072 }
1073
1074 bkey_copy(prev.k, cur.k);
1075 have_extent = true;
1076
1077 bch2_btree_iter_set_pos(&iter,
1078 POS(iter.pos.inode, iter.pos.offset + sectors));
1079
1080 ret = bch2_trans_relock(trans);
1081 if (ret)
1082 break;
1083 }
1084 start = iter.pos.offset;
1085 bch2_trans_iter_exit(trans, &iter);
1086 err:
1087 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1088 goto retry;
1089
1090 if (!ret && have_extent) {
1091 bch2_trans_unlock(trans);
1092 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
1093 FIEMAP_EXTENT_LAST);
1094 }
1095
1096 bch2_trans_put(trans);
1097 bch2_bkey_buf_exit(&cur, c);
1098 bch2_bkey_buf_exit(&prev, c);
1099 return ret < 0 ? ret : 0;
1100 }
1101
1102 static const struct vm_operations_struct bch_vm_ops = {
1103 .fault = bch2_page_fault,
1104 .map_pages = filemap_map_pages,
1105 .page_mkwrite = bch2_page_mkwrite,
1106 };
1107
bch2_mmap(struct file * file,struct vm_area_struct * vma)1108 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
1109 {
1110 file_accessed(file);
1111
1112 vma->vm_ops = &bch_vm_ops;
1113 return 0;
1114 }
1115
1116 /* Directories: */
1117
bch2_dir_llseek(struct file * file,loff_t offset,int whence)1118 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
1119 {
1120 return generic_file_llseek_size(file, offset, whence,
1121 S64_MAX, S64_MAX);
1122 }
1123
bch2_vfs_readdir(struct file * file,struct dir_context * ctx)1124 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
1125 {
1126 struct bch_inode_info *inode = file_bch_inode(file);
1127 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1128
1129 if (!dir_emit_dots(file, ctx))
1130 return 0;
1131
1132 int ret = bch2_readdir(c, inode_inum(inode), ctx);
1133
1134 bch_err_fn(c, ret);
1135 return bch2_err_class(ret);
1136 }
1137
bch2_open(struct inode * vinode,struct file * file)1138 static int bch2_open(struct inode *vinode, struct file *file)
1139 {
1140 if (file->f_flags & (O_WRONLY|O_RDWR)) {
1141 struct bch_inode_info *inode = to_bch_ei(vinode);
1142 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1143
1144 int ret = bch2_subvol_is_ro(c, inode->ei_subvol);
1145 if (ret)
1146 return ret;
1147 }
1148
1149 file->f_mode |= FMODE_CAN_ODIRECT;
1150
1151 return generic_file_open(vinode, file);
1152 }
1153
1154 static const struct file_operations bch_file_operations = {
1155 .open = bch2_open,
1156 .llseek = bch2_llseek,
1157 .read_iter = bch2_read_iter,
1158 .write_iter = bch2_write_iter,
1159 .mmap = bch2_mmap,
1160 .fsync = bch2_fsync,
1161 .splice_read = filemap_splice_read,
1162 .splice_write = iter_file_splice_write,
1163 .fallocate = bch2_fallocate_dispatch,
1164 .unlocked_ioctl = bch2_fs_file_ioctl,
1165 #ifdef CONFIG_COMPAT
1166 .compat_ioctl = bch2_compat_fs_ioctl,
1167 #endif
1168 .remap_file_range = bch2_remap_file_range,
1169 };
1170
1171 static const struct inode_operations bch_file_inode_operations = {
1172 .getattr = bch2_getattr,
1173 .setattr = bch2_setattr,
1174 .fiemap = bch2_fiemap,
1175 .listxattr = bch2_xattr_list,
1176 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1177 .get_acl = bch2_get_acl,
1178 .set_acl = bch2_set_acl,
1179 #endif
1180 };
1181
1182 static const struct inode_operations bch_dir_inode_operations = {
1183 .lookup = bch2_lookup,
1184 .create = bch2_create,
1185 .link = bch2_link,
1186 .unlink = bch2_unlink,
1187 .symlink = bch2_symlink,
1188 .mkdir = bch2_mkdir,
1189 .rmdir = bch2_unlink,
1190 .mknod = bch2_mknod,
1191 .rename = bch2_rename2,
1192 .getattr = bch2_getattr,
1193 .setattr = bch2_setattr,
1194 .tmpfile = bch2_tmpfile,
1195 .listxattr = bch2_xattr_list,
1196 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1197 .get_acl = bch2_get_acl,
1198 .set_acl = bch2_set_acl,
1199 #endif
1200 };
1201
1202 static const struct file_operations bch_dir_file_operations = {
1203 .llseek = bch2_dir_llseek,
1204 .read = generic_read_dir,
1205 .iterate_shared = bch2_vfs_readdir,
1206 .fsync = bch2_fsync,
1207 .unlocked_ioctl = bch2_fs_file_ioctl,
1208 #ifdef CONFIG_COMPAT
1209 .compat_ioctl = bch2_compat_fs_ioctl,
1210 #endif
1211 };
1212
1213 static const struct inode_operations bch_symlink_inode_operations = {
1214 .get_link = page_get_link,
1215 .getattr = bch2_getattr,
1216 .setattr = bch2_setattr,
1217 .listxattr = bch2_xattr_list,
1218 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1219 .get_acl = bch2_get_acl,
1220 .set_acl = bch2_set_acl,
1221 #endif
1222 };
1223
1224 static const struct inode_operations bch_special_inode_operations = {
1225 .getattr = bch2_getattr,
1226 .setattr = bch2_setattr,
1227 .listxattr = bch2_xattr_list,
1228 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1229 .get_acl = bch2_get_acl,
1230 .set_acl = bch2_set_acl,
1231 #endif
1232 };
1233
1234 static const struct address_space_operations bch_address_space_operations = {
1235 .read_folio = bch2_read_folio,
1236 .writepages = bch2_writepages,
1237 .readahead = bch2_readahead,
1238 .dirty_folio = filemap_dirty_folio,
1239 .write_begin = bch2_write_begin,
1240 .write_end = bch2_write_end,
1241 .invalidate_folio = bch2_invalidate_folio,
1242 .release_folio = bch2_release_folio,
1243 #ifdef CONFIG_MIGRATION
1244 .migrate_folio = filemap_migrate_folio,
1245 #endif
1246 .error_remove_folio = generic_error_remove_folio,
1247 };
1248
1249 struct bcachefs_fid {
1250 u64 inum;
1251 u32 subvol;
1252 u32 gen;
1253 } __packed;
1254
1255 struct bcachefs_fid_with_parent {
1256 struct bcachefs_fid fid;
1257 struct bcachefs_fid dir;
1258 } __packed;
1259
bcachefs_fid_valid(int fh_len,int fh_type)1260 static int bcachefs_fid_valid(int fh_len, int fh_type)
1261 {
1262 switch (fh_type) {
1263 case FILEID_BCACHEFS_WITHOUT_PARENT:
1264 return fh_len == sizeof(struct bcachefs_fid) / sizeof(u32);
1265 case FILEID_BCACHEFS_WITH_PARENT:
1266 return fh_len == sizeof(struct bcachefs_fid_with_parent) / sizeof(u32);
1267 default:
1268 return false;
1269 }
1270 }
1271
bch2_inode_to_fid(struct bch_inode_info * inode)1272 static struct bcachefs_fid bch2_inode_to_fid(struct bch_inode_info *inode)
1273 {
1274 return (struct bcachefs_fid) {
1275 .inum = inode->ei_inode.bi_inum,
1276 .subvol = inode->ei_subvol,
1277 .gen = inode->ei_inode.bi_generation,
1278 };
1279 }
1280
bch2_encode_fh(struct inode * vinode,u32 * fh,int * len,struct inode * vdir)1281 static int bch2_encode_fh(struct inode *vinode, u32 *fh, int *len,
1282 struct inode *vdir)
1283 {
1284 struct bch_inode_info *inode = to_bch_ei(vinode);
1285 struct bch_inode_info *dir = to_bch_ei(vdir);
1286 int min_len;
1287
1288 if (!S_ISDIR(inode->v.i_mode) && dir) {
1289 struct bcachefs_fid_with_parent *fid = (void *) fh;
1290
1291 min_len = sizeof(*fid) / sizeof(u32);
1292 if (*len < min_len) {
1293 *len = min_len;
1294 return FILEID_INVALID;
1295 }
1296
1297 fid->fid = bch2_inode_to_fid(inode);
1298 fid->dir = bch2_inode_to_fid(dir);
1299
1300 *len = min_len;
1301 return FILEID_BCACHEFS_WITH_PARENT;
1302 } else {
1303 struct bcachefs_fid *fid = (void *) fh;
1304
1305 min_len = sizeof(*fid) / sizeof(u32);
1306 if (*len < min_len) {
1307 *len = min_len;
1308 return FILEID_INVALID;
1309 }
1310 *fid = bch2_inode_to_fid(inode);
1311
1312 *len = min_len;
1313 return FILEID_BCACHEFS_WITHOUT_PARENT;
1314 }
1315 }
1316
bch2_nfs_get_inode(struct super_block * sb,struct bcachefs_fid fid)1317 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1318 struct bcachefs_fid fid)
1319 {
1320 struct bch_fs *c = sb->s_fs_info;
1321 struct inode *vinode = bch2_vfs_inode_get(c, (subvol_inum) {
1322 .subvol = fid.subvol,
1323 .inum = fid.inum,
1324 });
1325 if (!IS_ERR(vinode) && vinode->i_generation != fid.gen) {
1326 iput(vinode);
1327 vinode = ERR_PTR(-ESTALE);
1328 }
1329 return vinode;
1330 }
1331
bch2_fh_to_dentry(struct super_block * sb,struct fid * _fid,int fh_len,int fh_type)1332 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *_fid,
1333 int fh_len, int fh_type)
1334 {
1335 struct bcachefs_fid *fid = (void *) _fid;
1336
1337 if (!bcachefs_fid_valid(fh_len, fh_type))
1338 return NULL;
1339
1340 return d_obtain_alias(bch2_nfs_get_inode(sb, *fid));
1341 }
1342
bch2_fh_to_parent(struct super_block * sb,struct fid * _fid,int fh_len,int fh_type)1343 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *_fid,
1344 int fh_len, int fh_type)
1345 {
1346 struct bcachefs_fid_with_parent *fid = (void *) _fid;
1347
1348 if (!bcachefs_fid_valid(fh_len, fh_type) ||
1349 fh_type != FILEID_BCACHEFS_WITH_PARENT)
1350 return NULL;
1351
1352 return d_obtain_alias(bch2_nfs_get_inode(sb, fid->dir));
1353 }
1354
bch2_get_parent(struct dentry * child)1355 static struct dentry *bch2_get_parent(struct dentry *child)
1356 {
1357 struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1358 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1359 subvol_inum parent_inum = {
1360 .subvol = inode->ei_inode.bi_parent_subvol ?:
1361 inode->ei_subvol,
1362 .inum = inode->ei_inode.bi_dir,
1363 };
1364
1365 return d_obtain_alias(bch2_vfs_inode_get(c, parent_inum));
1366 }
1367
bch2_get_name(struct dentry * parent,char * name,struct dentry * child)1368 static int bch2_get_name(struct dentry *parent, char *name, struct dentry *child)
1369 {
1370 struct bch_inode_info *inode = to_bch_ei(child->d_inode);
1371 struct bch_inode_info *dir = to_bch_ei(parent->d_inode);
1372 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1373 struct btree_trans *trans;
1374 struct btree_iter iter1;
1375 struct btree_iter iter2;
1376 struct bkey_s_c k;
1377 struct bkey_s_c_dirent d;
1378 struct bch_inode_unpacked inode_u;
1379 subvol_inum target;
1380 u32 snapshot;
1381 struct qstr dirent_name;
1382 unsigned name_len = 0;
1383 int ret;
1384
1385 if (!S_ISDIR(dir->v.i_mode))
1386 return -EINVAL;
1387
1388 trans = bch2_trans_get(c);
1389
1390 bch2_trans_iter_init(trans, &iter1, BTREE_ID_dirents,
1391 POS(dir->ei_inode.bi_inum, 0), 0);
1392 bch2_trans_iter_init(trans, &iter2, BTREE_ID_dirents,
1393 POS(dir->ei_inode.bi_inum, 0), 0);
1394 retry:
1395 bch2_trans_begin(trans);
1396
1397 ret = bch2_subvolume_get_snapshot(trans, dir->ei_subvol, &snapshot);
1398 if (ret)
1399 goto err;
1400
1401 bch2_btree_iter_set_snapshot(&iter1, snapshot);
1402 bch2_btree_iter_set_snapshot(&iter2, snapshot);
1403
1404 ret = bch2_inode_find_by_inum_trans(trans, inode_inum(inode), &inode_u);
1405 if (ret)
1406 goto err;
1407
1408 if (inode_u.bi_dir == dir->ei_inode.bi_inum) {
1409 bch2_btree_iter_set_pos(&iter1, POS(inode_u.bi_dir, inode_u.bi_dir_offset));
1410
1411 k = bch2_btree_iter_peek_slot(&iter1);
1412 ret = bkey_err(k);
1413 if (ret)
1414 goto err;
1415
1416 if (k.k->type != KEY_TYPE_dirent) {
1417 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1418 goto err;
1419 }
1420
1421 d = bkey_s_c_to_dirent(k);
1422 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1423 if (ret > 0)
1424 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
1425 if (ret)
1426 goto err;
1427
1428 if (target.subvol == inode->ei_subvol &&
1429 target.inum == inode->ei_inode.bi_inum)
1430 goto found;
1431 } else {
1432 /*
1433 * File with multiple hardlinks and our backref is to the wrong
1434 * directory - linear search:
1435 */
1436 for_each_btree_key_continue_norestart(iter2, 0, k, ret) {
1437 if (k.k->p.inode > dir->ei_inode.bi_inum)
1438 break;
1439
1440 if (k.k->type != KEY_TYPE_dirent)
1441 continue;
1442
1443 d = bkey_s_c_to_dirent(k);
1444 ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
1445 if (ret < 0)
1446 break;
1447 if (ret)
1448 continue;
1449
1450 if (target.subvol == inode->ei_subvol &&
1451 target.inum == inode->ei_inode.bi_inum)
1452 goto found;
1453 }
1454 }
1455
1456 ret = -ENOENT;
1457 goto err;
1458 found:
1459 dirent_name = bch2_dirent_get_name(d);
1460
1461 name_len = min_t(unsigned, dirent_name.len, NAME_MAX);
1462 memcpy(name, dirent_name.name, name_len);
1463 name[name_len] = '\0';
1464 err:
1465 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1466 goto retry;
1467
1468 bch2_trans_iter_exit(trans, &iter1);
1469 bch2_trans_iter_exit(trans, &iter2);
1470 bch2_trans_put(trans);
1471
1472 return ret;
1473 }
1474
1475 static const struct export_operations bch_export_ops = {
1476 .encode_fh = bch2_encode_fh,
1477 .fh_to_dentry = bch2_fh_to_dentry,
1478 .fh_to_parent = bch2_fh_to_parent,
1479 .get_parent = bch2_get_parent,
1480 .get_name = bch2_get_name,
1481 };
1482
bch2_vfs_inode_init(struct btree_trans * trans,subvol_inum inum,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,struct bch_subvolume * subvol)1483 static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
1484 struct bch_inode_info *inode,
1485 struct bch_inode_unpacked *bi,
1486 struct bch_subvolume *subvol)
1487 {
1488 bch2_iget5_set(&inode->v, &inum);
1489 bch2_inode_update_after_write(trans, inode, bi, ~0);
1490
1491 if (BCH_SUBVOLUME_SNAP(subvol))
1492 set_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1493 else
1494 clear_bit(EI_INODE_SNAPSHOT, &inode->ei_flags);
1495
1496 inode->v.i_blocks = bi->bi_sectors;
1497 inode->v.i_ino = bi->bi_inum;
1498 inode->v.i_rdev = bi->bi_dev;
1499 inode->v.i_generation = bi->bi_generation;
1500 inode->v.i_size = bi->bi_size;
1501
1502 inode->ei_flags = 0;
1503 inode->ei_quota_reserved = 0;
1504 inode->ei_qid = bch_qid(bi);
1505 inode->ei_subvol = inum.subvol;
1506
1507 inode->v.i_mapping->a_ops = &bch_address_space_operations;
1508
1509 switch (inode->v.i_mode & S_IFMT) {
1510 case S_IFREG:
1511 inode->v.i_op = &bch_file_inode_operations;
1512 inode->v.i_fop = &bch_file_operations;
1513 break;
1514 case S_IFDIR:
1515 inode->v.i_op = &bch_dir_inode_operations;
1516 inode->v.i_fop = &bch_dir_file_operations;
1517 break;
1518 case S_IFLNK:
1519 inode_nohighmem(&inode->v);
1520 inode->v.i_op = &bch_symlink_inode_operations;
1521 break;
1522 default:
1523 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1524 inode->v.i_op = &bch_special_inode_operations;
1525 break;
1526 }
1527
1528 mapping_set_large_folios(inode->v.i_mapping);
1529 }
1530
bch2_free_inode(struct inode * vinode)1531 static void bch2_free_inode(struct inode *vinode)
1532 {
1533 kmem_cache_free(bch2_inode_cache, to_bch_ei(vinode));
1534 }
1535
inode_update_times_fn(struct btree_trans * trans,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,void * p)1536 static int inode_update_times_fn(struct btree_trans *trans,
1537 struct bch_inode_info *inode,
1538 struct bch_inode_unpacked *bi,
1539 void *p)
1540 {
1541 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1542
1543 bi->bi_atime = timespec_to_bch2_time(c, inode_get_atime(&inode->v));
1544 bi->bi_mtime = timespec_to_bch2_time(c, inode_get_mtime(&inode->v));
1545 bi->bi_ctime = timespec_to_bch2_time(c, inode_get_ctime(&inode->v));
1546
1547 return 0;
1548 }
1549
bch2_vfs_write_inode(struct inode * vinode,struct writeback_control * wbc)1550 static int bch2_vfs_write_inode(struct inode *vinode,
1551 struct writeback_control *wbc)
1552 {
1553 struct bch_fs *c = vinode->i_sb->s_fs_info;
1554 struct bch_inode_info *inode = to_bch_ei(vinode);
1555 int ret;
1556
1557 mutex_lock(&inode->ei_update_lock);
1558 ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1559 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1560 mutex_unlock(&inode->ei_update_lock);
1561
1562 return bch2_err_class(ret);
1563 }
1564
bch2_evict_inode(struct inode * vinode)1565 static void bch2_evict_inode(struct inode *vinode)
1566 {
1567 struct bch_fs *c = vinode->i_sb->s_fs_info;
1568 struct bch_inode_info *inode = to_bch_ei(vinode);
1569
1570 truncate_inode_pages_final(&inode->v.i_data);
1571
1572 clear_inode(&inode->v);
1573
1574 BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1575
1576 if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1577 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1578 KEY_TYPE_QUOTA_WARN);
1579 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1580 KEY_TYPE_QUOTA_WARN);
1581 bch2_inode_rm(c, inode_inum(inode));
1582 }
1583
1584 mutex_lock(&c->vfs_inodes_lock);
1585 list_del_init(&inode->ei_vfs_inode_list);
1586 mutex_unlock(&c->vfs_inodes_lock);
1587 }
1588
bch2_evict_subvolume_inodes(struct bch_fs * c,snapshot_id_list * s)1589 void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
1590 {
1591 struct bch_inode_info *inode;
1592 DARRAY(struct bch_inode_info *) grabbed;
1593 bool clean_pass = false, this_pass_clean;
1594
1595 /*
1596 * Initially, we scan for inodes without I_DONTCACHE, then mark them to
1597 * be pruned with d_mark_dontcache().
1598 *
1599 * Once we've had a clean pass where we didn't find any inodes without
1600 * I_DONTCACHE, we wait for them to be freed:
1601 */
1602
1603 darray_init(&grabbed);
1604 darray_make_room(&grabbed, 1024);
1605 again:
1606 cond_resched();
1607 this_pass_clean = true;
1608
1609 mutex_lock(&c->vfs_inodes_lock);
1610 list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
1611 if (!snapshot_list_has_id(s, inode->ei_subvol))
1612 continue;
1613
1614 if (!(inode->v.i_state & I_DONTCACHE) &&
1615 !(inode->v.i_state & I_FREEING) &&
1616 igrab(&inode->v)) {
1617 this_pass_clean = false;
1618
1619 if (darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN)) {
1620 iput(&inode->v);
1621 break;
1622 }
1623 } else if (clean_pass && this_pass_clean) {
1624 wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
1625 DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
1626
1627 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1628 mutex_unlock(&c->vfs_inodes_lock);
1629
1630 schedule();
1631 finish_wait(wq, &wait.wq_entry);
1632 goto again;
1633 }
1634 }
1635 mutex_unlock(&c->vfs_inodes_lock);
1636
1637 darray_for_each(grabbed, i) {
1638 inode = *i;
1639 d_mark_dontcache(&inode->v);
1640 d_prune_aliases(&inode->v);
1641 iput(&inode->v);
1642 }
1643 grabbed.nr = 0;
1644
1645 if (!clean_pass || !this_pass_clean) {
1646 clean_pass = this_pass_clean;
1647 goto again;
1648 }
1649
1650 darray_exit(&grabbed);
1651 }
1652
bch2_statfs(struct dentry * dentry,struct kstatfs * buf)1653 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1654 {
1655 struct super_block *sb = dentry->d_sb;
1656 struct bch_fs *c = sb->s_fs_info;
1657 struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1658 unsigned shift = sb->s_blocksize_bits - 9;
1659 /*
1660 * this assumes inodes take up 64 bytes, which is a decent average
1661 * number:
1662 */
1663 u64 avail_inodes = ((usage.capacity - usage.used) << 3);
1664
1665 buf->f_type = BCACHEFS_STATFS_MAGIC;
1666 buf->f_bsize = sb->s_blocksize;
1667 buf->f_blocks = usage.capacity >> shift;
1668 buf->f_bfree = usage.free >> shift;
1669 buf->f_bavail = avail_factor(usage.free) >> shift;
1670
1671 buf->f_files = usage.nr_inodes + avail_inodes;
1672 buf->f_ffree = avail_inodes;
1673
1674 buf->f_fsid = uuid_to_fsid(c->sb.user_uuid.b);
1675 buf->f_namelen = BCH_NAME_MAX;
1676
1677 return 0;
1678 }
1679
bch2_sync_fs(struct super_block * sb,int wait)1680 static int bch2_sync_fs(struct super_block *sb, int wait)
1681 {
1682 struct bch_fs *c = sb->s_fs_info;
1683 int ret;
1684
1685 if (c->opts.journal_flush_disabled)
1686 return 0;
1687
1688 if (!wait) {
1689 bch2_journal_flush_async(&c->journal, NULL);
1690 return 0;
1691 }
1692
1693 ret = bch2_journal_flush(&c->journal);
1694 return bch2_err_class(ret);
1695 }
1696
bch2_path_to_fs(const char * path)1697 static struct bch_fs *bch2_path_to_fs(const char *path)
1698 {
1699 struct bch_fs *c;
1700 dev_t dev;
1701 int ret;
1702
1703 ret = lookup_bdev(path, &dev);
1704 if (ret)
1705 return ERR_PTR(ret);
1706
1707 c = bch2_dev_to_fs(dev);
1708 if (c)
1709 closure_put(&c->cl);
1710 return c ?: ERR_PTR(-ENOENT);
1711 }
1712
bch2_remount(struct super_block * sb,int * flags,char * data)1713 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1714 {
1715 struct bch_fs *c = sb->s_fs_info;
1716 struct bch_opts opts = bch2_opts_empty();
1717 int ret;
1718
1719 ret = bch2_parse_mount_opts(c, &opts, data);
1720 if (ret)
1721 goto err;
1722
1723 opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1724
1725 if (opts.read_only != c->opts.read_only) {
1726 down_write(&c->state_lock);
1727
1728 if (opts.read_only) {
1729 bch2_fs_read_only(c);
1730
1731 sb->s_flags |= SB_RDONLY;
1732 } else {
1733 ret = bch2_fs_read_write(c);
1734 if (ret) {
1735 bch_err(c, "error going rw: %i", ret);
1736 up_write(&c->state_lock);
1737 ret = -EINVAL;
1738 goto err;
1739 }
1740
1741 sb->s_flags &= ~SB_RDONLY;
1742 }
1743
1744 c->opts.read_only = opts.read_only;
1745
1746 up_write(&c->state_lock);
1747 }
1748
1749 if (opt_defined(opts, errors))
1750 c->opts.errors = opts.errors;
1751 err:
1752 return bch2_err_class(ret);
1753 }
1754
bch2_show_devname(struct seq_file * seq,struct dentry * root)1755 static int bch2_show_devname(struct seq_file *seq, struct dentry *root)
1756 {
1757 struct bch_fs *c = root->d_sb->s_fs_info;
1758 bool first = true;
1759
1760 for_each_online_member(c, ca) {
1761 if (!first)
1762 seq_putc(seq, ':');
1763 first = false;
1764 seq_puts(seq, ca->disk_sb.sb_name);
1765 }
1766
1767 return 0;
1768 }
1769
bch2_show_options(struct seq_file * seq,struct dentry * root)1770 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1771 {
1772 struct bch_fs *c = root->d_sb->s_fs_info;
1773 enum bch_opt_id i;
1774 struct printbuf buf = PRINTBUF;
1775 int ret = 0;
1776
1777 for (i = 0; i < bch2_opts_nr; i++) {
1778 const struct bch_option *opt = &bch2_opt_table[i];
1779 u64 v = bch2_opt_get_by_id(&c->opts, i);
1780
1781 if (!(opt->flags & OPT_MOUNT))
1782 continue;
1783
1784 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1785 continue;
1786
1787 printbuf_reset(&buf);
1788 bch2_opt_to_text(&buf, c, c->disk_sb.sb, opt, v,
1789 OPT_SHOW_MOUNT_STYLE);
1790 seq_putc(seq, ',');
1791 seq_puts(seq, buf.buf);
1792 }
1793
1794 if (buf.allocation_failure)
1795 ret = -ENOMEM;
1796 printbuf_exit(&buf);
1797 return ret;
1798 }
1799
bch2_put_super(struct super_block * sb)1800 static void bch2_put_super(struct super_block *sb)
1801 {
1802 struct bch_fs *c = sb->s_fs_info;
1803
1804 __bch2_fs_stop(c);
1805 }
1806
1807 /*
1808 * bcachefs doesn't currently integrate intwrite freeze protection but the
1809 * internal write references serve the same purpose. Therefore reuse the
1810 * read-only transition code to perform the quiesce. The caveat is that we don't
1811 * currently have the ability to block tasks that want a write reference while
1812 * the superblock is frozen. This is fine for now, but we should either add
1813 * blocking support or find a way to integrate sb_start_intwrite() and friends.
1814 */
bch2_freeze(struct super_block * sb)1815 static int bch2_freeze(struct super_block *sb)
1816 {
1817 struct bch_fs *c = sb->s_fs_info;
1818
1819 down_write(&c->state_lock);
1820 bch2_fs_read_only(c);
1821 up_write(&c->state_lock);
1822 return 0;
1823 }
1824
bch2_unfreeze(struct super_block * sb)1825 static int bch2_unfreeze(struct super_block *sb)
1826 {
1827 struct bch_fs *c = sb->s_fs_info;
1828 int ret;
1829
1830 if (test_bit(BCH_FS_emergency_ro, &c->flags))
1831 return 0;
1832
1833 down_write(&c->state_lock);
1834 ret = bch2_fs_read_write(c);
1835 up_write(&c->state_lock);
1836 return ret;
1837 }
1838
1839 static const struct super_operations bch_super_operations = {
1840 .alloc_inode = bch2_alloc_inode,
1841 .free_inode = bch2_free_inode,
1842 .write_inode = bch2_vfs_write_inode,
1843 .evict_inode = bch2_evict_inode,
1844 .sync_fs = bch2_sync_fs,
1845 .statfs = bch2_statfs,
1846 .show_devname = bch2_show_devname,
1847 .show_options = bch2_show_options,
1848 .remount_fs = bch2_remount,
1849 .put_super = bch2_put_super,
1850 .freeze_fs = bch2_freeze,
1851 .unfreeze_fs = bch2_unfreeze,
1852 };
1853
bch2_set_super(struct super_block * s,void * data)1854 static int bch2_set_super(struct super_block *s, void *data)
1855 {
1856 s->s_fs_info = data;
1857 return 0;
1858 }
1859
bch2_noset_super(struct super_block * s,void * data)1860 static int bch2_noset_super(struct super_block *s, void *data)
1861 {
1862 return -EBUSY;
1863 }
1864
1865 typedef DARRAY(struct bch_fs *) darray_fs;
1866
bch2_test_super(struct super_block * s,void * data)1867 static int bch2_test_super(struct super_block *s, void *data)
1868 {
1869 struct bch_fs *c = s->s_fs_info;
1870 darray_fs *d = data;
1871
1872 if (!c)
1873 return false;
1874
1875 darray_for_each(*d, i)
1876 if (c != *i)
1877 return false;
1878 return true;
1879 }
1880
bch2_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1881 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1882 int flags, const char *dev_name, void *data)
1883 {
1884 struct bch_fs *c;
1885 struct super_block *sb;
1886 struct inode *vinode;
1887 struct bch_opts opts = bch2_opts_empty();
1888 int ret;
1889
1890 opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1891
1892 ret = bch2_parse_mount_opts(NULL, &opts, data);
1893 if (ret) {
1894 ret = bch2_err_class(ret);
1895 return ERR_PTR(ret);
1896 }
1897
1898 if (!dev_name || strlen(dev_name) == 0)
1899 return ERR_PTR(-EINVAL);
1900
1901 darray_str devs;
1902 ret = bch2_split_devs(dev_name, &devs);
1903 if (ret)
1904 return ERR_PTR(ret);
1905
1906 darray_fs devs_to_fs = {};
1907 darray_for_each(devs, i) {
1908 ret = darray_push(&devs_to_fs, bch2_path_to_fs(*i));
1909 if (ret) {
1910 sb = ERR_PTR(ret);
1911 goto got_sb;
1912 }
1913 }
1914
1915 sb = sget(fs_type, bch2_test_super, bch2_noset_super, flags|SB_NOSEC, &devs_to_fs);
1916 if (!IS_ERR(sb))
1917 goto got_sb;
1918
1919 c = bch2_fs_open(devs.data, devs.nr, opts);
1920 if (IS_ERR(c)) {
1921 sb = ERR_CAST(c);
1922 goto got_sb;
1923 }
1924
1925 /* Some options can't be parsed until after the fs is started: */
1926 ret = bch2_parse_mount_opts(c, &opts, data);
1927 if (ret) {
1928 bch2_fs_stop(c);
1929 sb = ERR_PTR(ret);
1930 goto got_sb;
1931 }
1932
1933 bch2_opts_apply(&c->opts, opts);
1934
1935 sb = sget(fs_type, NULL, bch2_set_super, flags|SB_NOSEC, c);
1936 if (IS_ERR(sb))
1937 bch2_fs_stop(c);
1938 got_sb:
1939 darray_exit(&devs_to_fs);
1940 bch2_darray_str_exit(&devs);
1941
1942 if (IS_ERR(sb)) {
1943 ret = PTR_ERR(sb);
1944 goto err;
1945 }
1946
1947 c = sb->s_fs_info;
1948
1949 if (sb->s_root) {
1950 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1951 ret = -EBUSY;
1952 goto err_put_super;
1953 }
1954 goto out;
1955 }
1956
1957 sb->s_blocksize = block_bytes(c);
1958 sb->s_blocksize_bits = ilog2(block_bytes(c));
1959 sb->s_maxbytes = MAX_LFS_FILESIZE;
1960 sb->s_op = &bch_super_operations;
1961 sb->s_export_op = &bch_export_ops;
1962 #ifdef CONFIG_BCACHEFS_QUOTA
1963 sb->s_qcop = &bch2_quotactl_operations;
1964 sb->s_quota_types = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1965 #endif
1966 sb->s_xattr = bch2_xattr_handlers;
1967 sb->s_magic = BCACHEFS_STATFS_MAGIC;
1968 sb->s_time_gran = c->sb.nsec_per_time_unit;
1969 sb->s_time_min = div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
1970 sb->s_time_max = div_s64(S64_MAX, c->sb.time_units_per_sec);
1971 sb->s_uuid = c->sb.user_uuid;
1972 sb->s_shrink->seeks = 0;
1973 c->vfs_sb = sb;
1974 strscpy(sb->s_id, c->name, sizeof(sb->s_id));
1975
1976 ret = super_setup_bdi(sb);
1977 if (ret)
1978 goto err_put_super;
1979
1980 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1981
1982 for_each_online_member(c, ca) {
1983 struct block_device *bdev = ca->disk_sb.bdev;
1984
1985 /* XXX: create an anonymous device for multi device filesystems */
1986 sb->s_bdev = bdev;
1987 sb->s_dev = bdev->bd_dev;
1988 percpu_ref_put(&ca->io_ref);
1989 break;
1990 }
1991
1992 c->dev = sb->s_dev;
1993
1994 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1995 if (c->opts.acl)
1996 sb->s_flags |= SB_POSIXACL;
1997 #endif
1998
1999 sb->s_shrink->seeks = 0;
2000
2001 vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
2002 ret = PTR_ERR_OR_ZERO(vinode);
2003 bch_err_msg(c, ret, "mounting: error getting root inode");
2004 if (ret)
2005 goto err_put_super;
2006
2007 sb->s_root = d_make_root(vinode);
2008 if (!sb->s_root) {
2009 bch_err(c, "error mounting: error allocating root dentry");
2010 ret = -ENOMEM;
2011 goto err_put_super;
2012 }
2013
2014 sb->s_flags |= SB_ACTIVE;
2015 out:
2016 return dget(sb->s_root);
2017
2018 err_put_super:
2019 __bch2_fs_stop(c);
2020 deactivate_locked_super(sb);
2021 err:
2022 /*
2023 * On an inconsistency error in recovery we might see an -EROFS derived
2024 * errorcode (from the journal), but we don't want to return that to
2025 * userspace as that causes util-linux to retry the mount RO - which is
2026 * confusing:
2027 */
2028 if (bch2_err_matches(ret, EROFS) && ret != -EROFS)
2029 ret = -EIO;
2030 return ERR_PTR(bch2_err_class(ret));
2031 }
2032
bch2_kill_sb(struct super_block * sb)2033 static void bch2_kill_sb(struct super_block *sb)
2034 {
2035 struct bch_fs *c = sb->s_fs_info;
2036
2037 generic_shutdown_super(sb);
2038 bch2_fs_free(c);
2039 }
2040
2041 static struct file_system_type bcache_fs_type = {
2042 .owner = THIS_MODULE,
2043 .name = "bcachefs",
2044 .mount = bch2_mount,
2045 .kill_sb = bch2_kill_sb,
2046 .fs_flags = FS_REQUIRES_DEV,
2047 };
2048
2049 MODULE_ALIAS_FS("bcachefs");
2050
bch2_vfs_exit(void)2051 void bch2_vfs_exit(void)
2052 {
2053 unregister_filesystem(&bcache_fs_type);
2054 kmem_cache_destroy(bch2_inode_cache);
2055 }
2056
bch2_vfs_init(void)2057 int __init bch2_vfs_init(void)
2058 {
2059 int ret = -ENOMEM;
2060
2061 bch2_inode_cache = KMEM_CACHE(bch_inode_info, SLAB_RECLAIM_ACCOUNT);
2062 if (!bch2_inode_cache)
2063 goto err;
2064
2065 ret = register_filesystem(&bcache_fs_type);
2066 if (ret)
2067 goto err;
2068
2069 return 0;
2070 err:
2071 bch2_vfs_exit();
2072 return ret;
2073 }
2074
2075 #endif /* NO_BCACHEFS_FS */
2076