fuse: add .write_inode
[deliverable/linux.git] / fs / fuse / dir.c
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/sched.h>
14 #include <linux/namei.h>
15 #include <linux/slab.h>
16
17 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
18 {
19 struct fuse_conn *fc = get_fuse_conn(dir);
20 struct fuse_inode *fi = get_fuse_inode(dir);
21
22 if (!fc->do_readdirplus)
23 return false;
24 if (!fc->readdirplus_auto)
25 return true;
26 if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
27 return true;
28 if (ctx->pos == 0)
29 return true;
30 return false;
31 }
32
33 static void fuse_advise_use_readdirplus(struct inode *dir)
34 {
35 struct fuse_inode *fi = get_fuse_inode(dir);
36
37 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
38 }
39
40 #if BITS_PER_LONG >= 64
41 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
42 {
43 entry->d_time = time;
44 }
45
46 static inline u64 fuse_dentry_time(struct dentry *entry)
47 {
48 return entry->d_time;
49 }
50 #else
51 /*
52 * On 32 bit archs store the high 32 bits of time in d_fsdata
53 */
54 static void fuse_dentry_settime(struct dentry *entry, u64 time)
55 {
56 entry->d_time = time;
57 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
58 }
59
60 static u64 fuse_dentry_time(struct dentry *entry)
61 {
62 return (u64) entry->d_time +
63 ((u64) (unsigned long) entry->d_fsdata << 32);
64 }
65 #endif
66
67 /*
68 * FUSE caches dentries and attributes with separate timeout. The
69 * time in jiffies until the dentry/attributes are valid is stored in
70 * dentry->d_time and fuse_inode->i_time respectively.
71 */
72
73 /*
74 * Calculate the time in jiffies until a dentry/attributes are valid
75 */
76 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
77 {
78 if (sec || nsec) {
79 struct timespec ts = {sec, nsec};
80 return get_jiffies_64() + timespec_to_jiffies(&ts);
81 } else
82 return 0;
83 }
84
85 /*
86 * Set dentry and possibly attribute timeouts from the lookup/mk*
87 * replies
88 */
89 static void fuse_change_entry_timeout(struct dentry *entry,
90 struct fuse_entry_out *o)
91 {
92 fuse_dentry_settime(entry,
93 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
94 }
95
96 static u64 attr_timeout(struct fuse_attr_out *o)
97 {
98 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
99 }
100
101 static u64 entry_attr_timeout(struct fuse_entry_out *o)
102 {
103 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
104 }
105
106 /*
107 * Mark the attributes as stale, so that at the next call to
108 * ->getattr() they will be fetched from userspace
109 */
110 void fuse_invalidate_attr(struct inode *inode)
111 {
112 get_fuse_inode(inode)->i_time = 0;
113 }
114
115 /**
116 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
117 * atime is not used.
118 */
119 void fuse_invalidate_atime(struct inode *inode)
120 {
121 if (!IS_RDONLY(inode))
122 fuse_invalidate_attr(inode);
123 }
124
125 /*
126 * Just mark the entry as stale, so that a next attempt to look it up
127 * will result in a new lookup call to userspace
128 *
129 * This is called when a dentry is about to become negative and the
130 * timeout is unknown (unlink, rmdir, rename and in some cases
131 * lookup)
132 */
133 void fuse_invalidate_entry_cache(struct dentry *entry)
134 {
135 fuse_dentry_settime(entry, 0);
136 }
137
138 /*
139 * Same as fuse_invalidate_entry_cache(), but also try to remove the
140 * dentry from the hash
141 */
142 static void fuse_invalidate_entry(struct dentry *entry)
143 {
144 d_invalidate(entry);
145 fuse_invalidate_entry_cache(entry);
146 }
147
148 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_req *req,
149 u64 nodeid, struct qstr *name,
150 struct fuse_entry_out *outarg)
151 {
152 memset(outarg, 0, sizeof(struct fuse_entry_out));
153 req->in.h.opcode = FUSE_LOOKUP;
154 req->in.h.nodeid = nodeid;
155 req->in.numargs = 1;
156 req->in.args[0].size = name->len + 1;
157 req->in.args[0].value = name->name;
158 req->out.numargs = 1;
159 if (fc->minor < 9)
160 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
161 else
162 req->out.args[0].size = sizeof(struct fuse_entry_out);
163 req->out.args[0].value = outarg;
164 }
165
166 u64 fuse_get_attr_version(struct fuse_conn *fc)
167 {
168 u64 curr_version;
169
170 /*
171 * The spin lock isn't actually needed on 64bit archs, but we
172 * don't yet care too much about such optimizations.
173 */
174 spin_lock(&fc->lock);
175 curr_version = fc->attr_version;
176 spin_unlock(&fc->lock);
177
178 return curr_version;
179 }
180
181 /*
182 * Check whether the dentry is still valid
183 *
184 * If the entry validity timeout has expired and the dentry is
185 * positive, try to redo the lookup. If the lookup results in a
186 * different inode, then let the VFS invalidate the dentry and redo
187 * the lookup once more. If the lookup results in the same inode,
188 * then refresh the attributes, timeouts and mark the dentry valid.
189 */
190 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
191 {
192 struct inode *inode;
193 struct dentry *parent;
194 struct fuse_conn *fc;
195 struct fuse_inode *fi;
196 int ret;
197
198 inode = ACCESS_ONCE(entry->d_inode);
199 if (inode && is_bad_inode(inode))
200 goto invalid;
201 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
202 int err;
203 struct fuse_entry_out outarg;
204 struct fuse_req *req;
205 struct fuse_forget_link *forget;
206 u64 attr_version;
207
208 /* For negative dentries, always do a fresh lookup */
209 if (!inode)
210 goto invalid;
211
212 ret = -ECHILD;
213 if (flags & LOOKUP_RCU)
214 goto out;
215
216 fc = get_fuse_conn(inode);
217 req = fuse_get_req_nopages(fc);
218 ret = PTR_ERR(req);
219 if (IS_ERR(req))
220 goto out;
221
222 forget = fuse_alloc_forget();
223 if (!forget) {
224 fuse_put_request(fc, req);
225 ret = -ENOMEM;
226 goto out;
227 }
228
229 attr_version = fuse_get_attr_version(fc);
230
231 parent = dget_parent(entry);
232 fuse_lookup_init(fc, req, get_node_id(parent->d_inode),
233 &entry->d_name, &outarg);
234 fuse_request_send(fc, req);
235 dput(parent);
236 err = req->out.h.error;
237 fuse_put_request(fc, req);
238 /* Zero nodeid is same as -ENOENT */
239 if (!err && !outarg.nodeid)
240 err = -ENOENT;
241 if (!err) {
242 fi = get_fuse_inode(inode);
243 if (outarg.nodeid != get_node_id(inode)) {
244 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
245 goto invalid;
246 }
247 spin_lock(&fc->lock);
248 fi->nlookup++;
249 spin_unlock(&fc->lock);
250 }
251 kfree(forget);
252 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
253 goto invalid;
254
255 fuse_change_attributes(inode, &outarg.attr,
256 entry_attr_timeout(&outarg),
257 attr_version);
258 fuse_change_entry_timeout(entry, &outarg);
259 } else if (inode) {
260 fi = get_fuse_inode(inode);
261 if (flags & LOOKUP_RCU) {
262 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
263 return -ECHILD;
264 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
265 parent = dget_parent(entry);
266 fuse_advise_use_readdirplus(parent->d_inode);
267 dput(parent);
268 }
269 }
270 ret = 1;
271 out:
272 return ret;
273
274 invalid:
275 ret = 0;
276
277 if (!(flags & LOOKUP_RCU) && check_submounts_and_drop(entry) != 0)
278 ret = 1;
279 goto out;
280 }
281
282 static int invalid_nodeid(u64 nodeid)
283 {
284 return !nodeid || nodeid == FUSE_ROOT_ID;
285 }
286
287 const struct dentry_operations fuse_dentry_operations = {
288 .d_revalidate = fuse_dentry_revalidate,
289 };
290
291 int fuse_valid_type(int m)
292 {
293 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
294 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
295 }
296
297 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
298 struct fuse_entry_out *outarg, struct inode **inode)
299 {
300 struct fuse_conn *fc = get_fuse_conn_super(sb);
301 struct fuse_req *req;
302 struct fuse_forget_link *forget;
303 u64 attr_version;
304 int err;
305
306 *inode = NULL;
307 err = -ENAMETOOLONG;
308 if (name->len > FUSE_NAME_MAX)
309 goto out;
310
311 req = fuse_get_req_nopages(fc);
312 err = PTR_ERR(req);
313 if (IS_ERR(req))
314 goto out;
315
316 forget = fuse_alloc_forget();
317 err = -ENOMEM;
318 if (!forget) {
319 fuse_put_request(fc, req);
320 goto out;
321 }
322
323 attr_version = fuse_get_attr_version(fc);
324
325 fuse_lookup_init(fc, req, nodeid, name, outarg);
326 fuse_request_send(fc, req);
327 err = req->out.h.error;
328 fuse_put_request(fc, req);
329 /* Zero nodeid is same as -ENOENT, but with valid timeout */
330 if (err || !outarg->nodeid)
331 goto out_put_forget;
332
333 err = -EIO;
334 if (!outarg->nodeid)
335 goto out_put_forget;
336 if (!fuse_valid_type(outarg->attr.mode))
337 goto out_put_forget;
338
339 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
340 &outarg->attr, entry_attr_timeout(outarg),
341 attr_version);
342 err = -ENOMEM;
343 if (!*inode) {
344 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
345 goto out;
346 }
347 err = 0;
348
349 out_put_forget:
350 kfree(forget);
351 out:
352 return err;
353 }
354
355 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
356 unsigned int flags)
357 {
358 int err;
359 struct fuse_entry_out outarg;
360 struct inode *inode;
361 struct dentry *newent;
362 bool outarg_valid = true;
363
364 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
365 &outarg, &inode);
366 if (err == -ENOENT) {
367 outarg_valid = false;
368 err = 0;
369 }
370 if (err)
371 goto out_err;
372
373 err = -EIO;
374 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
375 goto out_iput;
376
377 newent = d_materialise_unique(entry, inode);
378 err = PTR_ERR(newent);
379 if (IS_ERR(newent))
380 goto out_err;
381
382 entry = newent ? newent : entry;
383 if (outarg_valid)
384 fuse_change_entry_timeout(entry, &outarg);
385 else
386 fuse_invalidate_entry_cache(entry);
387
388 fuse_advise_use_readdirplus(dir);
389 return newent;
390
391 out_iput:
392 iput(inode);
393 out_err:
394 return ERR_PTR(err);
395 }
396
397 /*
398 * Atomic create+open operation
399 *
400 * If the filesystem doesn't support this, then fall back to separate
401 * 'mknod' + 'open' requests.
402 */
403 static int fuse_create_open(struct inode *dir, struct dentry *entry,
404 struct file *file, unsigned flags,
405 umode_t mode, int *opened)
406 {
407 int err;
408 struct inode *inode;
409 struct fuse_conn *fc = get_fuse_conn(dir);
410 struct fuse_req *req;
411 struct fuse_forget_link *forget;
412 struct fuse_create_in inarg;
413 struct fuse_open_out outopen;
414 struct fuse_entry_out outentry;
415 struct fuse_file *ff;
416
417 /* Userspace expects S_IFREG in create mode */
418 BUG_ON((mode & S_IFMT) != S_IFREG);
419
420 forget = fuse_alloc_forget();
421 err = -ENOMEM;
422 if (!forget)
423 goto out_err;
424
425 req = fuse_get_req_nopages(fc);
426 err = PTR_ERR(req);
427 if (IS_ERR(req))
428 goto out_put_forget_req;
429
430 err = -ENOMEM;
431 ff = fuse_file_alloc(fc);
432 if (!ff)
433 goto out_put_request;
434
435 if (!fc->dont_mask)
436 mode &= ~current_umask();
437
438 flags &= ~O_NOCTTY;
439 memset(&inarg, 0, sizeof(inarg));
440 memset(&outentry, 0, sizeof(outentry));
441 inarg.flags = flags;
442 inarg.mode = mode;
443 inarg.umask = current_umask();
444 req->in.h.opcode = FUSE_CREATE;
445 req->in.h.nodeid = get_node_id(dir);
446 req->in.numargs = 2;
447 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
448 sizeof(inarg);
449 req->in.args[0].value = &inarg;
450 req->in.args[1].size = entry->d_name.len + 1;
451 req->in.args[1].value = entry->d_name.name;
452 req->out.numargs = 2;
453 if (fc->minor < 9)
454 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
455 else
456 req->out.args[0].size = sizeof(outentry);
457 req->out.args[0].value = &outentry;
458 req->out.args[1].size = sizeof(outopen);
459 req->out.args[1].value = &outopen;
460 fuse_request_send(fc, req);
461 err = req->out.h.error;
462 if (err)
463 goto out_free_ff;
464
465 err = -EIO;
466 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
467 goto out_free_ff;
468
469 fuse_put_request(fc, req);
470 ff->fh = outopen.fh;
471 ff->nodeid = outentry.nodeid;
472 ff->open_flags = outopen.open_flags;
473 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
474 &outentry.attr, entry_attr_timeout(&outentry), 0);
475 if (!inode) {
476 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
477 fuse_sync_release(ff, flags);
478 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
479 err = -ENOMEM;
480 goto out_err;
481 }
482 kfree(forget);
483 d_instantiate(entry, inode);
484 fuse_change_entry_timeout(entry, &outentry);
485 fuse_invalidate_attr(dir);
486 err = finish_open(file, entry, generic_file_open, opened);
487 if (err) {
488 fuse_sync_release(ff, flags);
489 } else {
490 file->private_data = fuse_file_get(ff);
491 fuse_finish_open(inode, file);
492 }
493 return err;
494
495 out_free_ff:
496 fuse_file_free(ff);
497 out_put_request:
498 fuse_put_request(fc, req);
499 out_put_forget_req:
500 kfree(forget);
501 out_err:
502 return err;
503 }
504
505 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
506 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
507 struct file *file, unsigned flags,
508 umode_t mode, int *opened)
509 {
510 int err;
511 struct fuse_conn *fc = get_fuse_conn(dir);
512 struct dentry *res = NULL;
513
514 if (d_unhashed(entry)) {
515 res = fuse_lookup(dir, entry, 0);
516 if (IS_ERR(res))
517 return PTR_ERR(res);
518
519 if (res)
520 entry = res;
521 }
522
523 if (!(flags & O_CREAT) || entry->d_inode)
524 goto no_open;
525
526 /* Only creates */
527 *opened |= FILE_CREATED;
528
529 if (fc->no_create)
530 goto mknod;
531
532 err = fuse_create_open(dir, entry, file, flags, mode, opened);
533 if (err == -ENOSYS) {
534 fc->no_create = 1;
535 goto mknod;
536 }
537 out_dput:
538 dput(res);
539 return err;
540
541 mknod:
542 err = fuse_mknod(dir, entry, mode, 0);
543 if (err)
544 goto out_dput;
545 no_open:
546 return finish_no_open(file, res);
547 }
548
549 /*
550 * Code shared between mknod, mkdir, symlink and link
551 */
552 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
553 struct inode *dir, struct dentry *entry,
554 umode_t mode)
555 {
556 struct fuse_entry_out outarg;
557 struct inode *inode;
558 int err;
559 struct fuse_forget_link *forget;
560
561 forget = fuse_alloc_forget();
562 if (!forget) {
563 fuse_put_request(fc, req);
564 return -ENOMEM;
565 }
566
567 memset(&outarg, 0, sizeof(outarg));
568 req->in.h.nodeid = get_node_id(dir);
569 req->out.numargs = 1;
570 if (fc->minor < 9)
571 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
572 else
573 req->out.args[0].size = sizeof(outarg);
574 req->out.args[0].value = &outarg;
575 fuse_request_send(fc, req);
576 err = req->out.h.error;
577 fuse_put_request(fc, req);
578 if (err)
579 goto out_put_forget_req;
580
581 err = -EIO;
582 if (invalid_nodeid(outarg.nodeid))
583 goto out_put_forget_req;
584
585 if ((outarg.attr.mode ^ mode) & S_IFMT)
586 goto out_put_forget_req;
587
588 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
589 &outarg.attr, entry_attr_timeout(&outarg), 0);
590 if (!inode) {
591 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
592 return -ENOMEM;
593 }
594 kfree(forget);
595
596 err = d_instantiate_no_diralias(entry, inode);
597 if (err)
598 return err;
599
600 fuse_change_entry_timeout(entry, &outarg);
601 fuse_invalidate_attr(dir);
602 return 0;
603
604 out_put_forget_req:
605 kfree(forget);
606 return err;
607 }
608
609 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
610 dev_t rdev)
611 {
612 struct fuse_mknod_in inarg;
613 struct fuse_conn *fc = get_fuse_conn(dir);
614 struct fuse_req *req = fuse_get_req_nopages(fc);
615 if (IS_ERR(req))
616 return PTR_ERR(req);
617
618 if (!fc->dont_mask)
619 mode &= ~current_umask();
620
621 memset(&inarg, 0, sizeof(inarg));
622 inarg.mode = mode;
623 inarg.rdev = new_encode_dev(rdev);
624 inarg.umask = current_umask();
625 req->in.h.opcode = FUSE_MKNOD;
626 req->in.numargs = 2;
627 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
628 sizeof(inarg);
629 req->in.args[0].value = &inarg;
630 req->in.args[1].size = entry->d_name.len + 1;
631 req->in.args[1].value = entry->d_name.name;
632 return create_new_entry(fc, req, dir, entry, mode);
633 }
634
635 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
636 bool excl)
637 {
638 return fuse_mknod(dir, entry, mode, 0);
639 }
640
641 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
642 {
643 struct fuse_mkdir_in inarg;
644 struct fuse_conn *fc = get_fuse_conn(dir);
645 struct fuse_req *req = fuse_get_req_nopages(fc);
646 if (IS_ERR(req))
647 return PTR_ERR(req);
648
649 if (!fc->dont_mask)
650 mode &= ~current_umask();
651
652 memset(&inarg, 0, sizeof(inarg));
653 inarg.mode = mode;
654 inarg.umask = current_umask();
655 req->in.h.opcode = FUSE_MKDIR;
656 req->in.numargs = 2;
657 req->in.args[0].size = sizeof(inarg);
658 req->in.args[0].value = &inarg;
659 req->in.args[1].size = entry->d_name.len + 1;
660 req->in.args[1].value = entry->d_name.name;
661 return create_new_entry(fc, req, dir, entry, S_IFDIR);
662 }
663
664 static int fuse_symlink(struct inode *dir, struct dentry *entry,
665 const char *link)
666 {
667 struct fuse_conn *fc = get_fuse_conn(dir);
668 unsigned len = strlen(link) + 1;
669 struct fuse_req *req = fuse_get_req_nopages(fc);
670 if (IS_ERR(req))
671 return PTR_ERR(req);
672
673 req->in.h.opcode = FUSE_SYMLINK;
674 req->in.numargs = 2;
675 req->in.args[0].size = entry->d_name.len + 1;
676 req->in.args[0].value = entry->d_name.name;
677 req->in.args[1].size = len;
678 req->in.args[1].value = link;
679 return create_new_entry(fc, req, dir, entry, S_IFLNK);
680 }
681
682 static int fuse_unlink(struct inode *dir, struct dentry *entry)
683 {
684 int err;
685 struct fuse_conn *fc = get_fuse_conn(dir);
686 struct fuse_req *req = fuse_get_req_nopages(fc);
687 if (IS_ERR(req))
688 return PTR_ERR(req);
689
690 req->in.h.opcode = FUSE_UNLINK;
691 req->in.h.nodeid = get_node_id(dir);
692 req->in.numargs = 1;
693 req->in.args[0].size = entry->d_name.len + 1;
694 req->in.args[0].value = entry->d_name.name;
695 fuse_request_send(fc, req);
696 err = req->out.h.error;
697 fuse_put_request(fc, req);
698 if (!err) {
699 struct inode *inode = entry->d_inode;
700 struct fuse_inode *fi = get_fuse_inode(inode);
701
702 spin_lock(&fc->lock);
703 fi->attr_version = ++fc->attr_version;
704 /*
705 * If i_nlink == 0 then unlink doesn't make sense, yet this can
706 * happen if userspace filesystem is careless. It would be
707 * difficult to enforce correct nlink usage so just ignore this
708 * condition here
709 */
710 if (inode->i_nlink > 0)
711 drop_nlink(inode);
712 spin_unlock(&fc->lock);
713 fuse_invalidate_attr(inode);
714 fuse_invalidate_attr(dir);
715 fuse_invalidate_entry_cache(entry);
716 } else if (err == -EINTR)
717 fuse_invalidate_entry(entry);
718 return err;
719 }
720
721 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
722 {
723 int err;
724 struct fuse_conn *fc = get_fuse_conn(dir);
725 struct fuse_req *req = fuse_get_req_nopages(fc);
726 if (IS_ERR(req))
727 return PTR_ERR(req);
728
729 req->in.h.opcode = FUSE_RMDIR;
730 req->in.h.nodeid = get_node_id(dir);
731 req->in.numargs = 1;
732 req->in.args[0].size = entry->d_name.len + 1;
733 req->in.args[0].value = entry->d_name.name;
734 fuse_request_send(fc, req);
735 err = req->out.h.error;
736 fuse_put_request(fc, req);
737 if (!err) {
738 clear_nlink(entry->d_inode);
739 fuse_invalidate_attr(dir);
740 fuse_invalidate_entry_cache(entry);
741 } else if (err == -EINTR)
742 fuse_invalidate_entry(entry);
743 return err;
744 }
745
746 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
747 struct inode *newdir, struct dentry *newent)
748 {
749 int err;
750 struct fuse_rename_in inarg;
751 struct fuse_conn *fc = get_fuse_conn(olddir);
752 struct fuse_req *req = fuse_get_req_nopages(fc);
753
754 if (IS_ERR(req))
755 return PTR_ERR(req);
756
757 memset(&inarg, 0, sizeof(inarg));
758 inarg.newdir = get_node_id(newdir);
759 req->in.h.opcode = FUSE_RENAME;
760 req->in.h.nodeid = get_node_id(olddir);
761 req->in.numargs = 3;
762 req->in.args[0].size = sizeof(inarg);
763 req->in.args[0].value = &inarg;
764 req->in.args[1].size = oldent->d_name.len + 1;
765 req->in.args[1].value = oldent->d_name.name;
766 req->in.args[2].size = newent->d_name.len + 1;
767 req->in.args[2].value = newent->d_name.name;
768 fuse_request_send(fc, req);
769 err = req->out.h.error;
770 fuse_put_request(fc, req);
771 if (!err) {
772 /* ctime changes */
773 fuse_invalidate_attr(oldent->d_inode);
774
775 fuse_invalidate_attr(olddir);
776 if (olddir != newdir)
777 fuse_invalidate_attr(newdir);
778
779 /* newent will end up negative */
780 if (newent->d_inode) {
781 fuse_invalidate_attr(newent->d_inode);
782 fuse_invalidate_entry_cache(newent);
783 }
784 } else if (err == -EINTR) {
785 /* If request was interrupted, DEITY only knows if the
786 rename actually took place. If the invalidation
787 fails (e.g. some process has CWD under the renamed
788 directory), then there can be inconsistency between
789 the dcache and the real filesystem. Tough luck. */
790 fuse_invalidate_entry(oldent);
791 if (newent->d_inode)
792 fuse_invalidate_entry(newent);
793 }
794
795 return err;
796 }
797
798 static int fuse_link(struct dentry *entry, struct inode *newdir,
799 struct dentry *newent)
800 {
801 int err;
802 struct fuse_link_in inarg;
803 struct inode *inode = entry->d_inode;
804 struct fuse_conn *fc = get_fuse_conn(inode);
805 struct fuse_req *req = fuse_get_req_nopages(fc);
806 if (IS_ERR(req))
807 return PTR_ERR(req);
808
809 memset(&inarg, 0, sizeof(inarg));
810 inarg.oldnodeid = get_node_id(inode);
811 req->in.h.opcode = FUSE_LINK;
812 req->in.numargs = 2;
813 req->in.args[0].size = sizeof(inarg);
814 req->in.args[0].value = &inarg;
815 req->in.args[1].size = newent->d_name.len + 1;
816 req->in.args[1].value = newent->d_name.name;
817 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
818 /* Contrary to "normal" filesystems it can happen that link
819 makes two "logical" inodes point to the same "physical"
820 inode. We invalidate the attributes of the old one, so it
821 will reflect changes in the backing inode (link count,
822 etc.)
823 */
824 if (!err) {
825 struct fuse_inode *fi = get_fuse_inode(inode);
826
827 spin_lock(&fc->lock);
828 fi->attr_version = ++fc->attr_version;
829 inc_nlink(inode);
830 spin_unlock(&fc->lock);
831 fuse_invalidate_attr(inode);
832 } else if (err == -EINTR) {
833 fuse_invalidate_attr(inode);
834 }
835 return err;
836 }
837
838 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
839 struct kstat *stat)
840 {
841 unsigned int blkbits;
842 struct fuse_conn *fc = get_fuse_conn(inode);
843
844 /* see the comment in fuse_change_attributes() */
845 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
846 attr->size = i_size_read(inode);
847 attr->mtime = inode->i_mtime.tv_sec;
848 attr->mtimensec = inode->i_mtime.tv_nsec;
849 }
850
851 stat->dev = inode->i_sb->s_dev;
852 stat->ino = attr->ino;
853 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
854 stat->nlink = attr->nlink;
855 stat->uid = make_kuid(&init_user_ns, attr->uid);
856 stat->gid = make_kgid(&init_user_ns, attr->gid);
857 stat->rdev = inode->i_rdev;
858 stat->atime.tv_sec = attr->atime;
859 stat->atime.tv_nsec = attr->atimensec;
860 stat->mtime.tv_sec = attr->mtime;
861 stat->mtime.tv_nsec = attr->mtimensec;
862 stat->ctime.tv_sec = attr->ctime;
863 stat->ctime.tv_nsec = attr->ctimensec;
864 stat->size = attr->size;
865 stat->blocks = attr->blocks;
866
867 if (attr->blksize != 0)
868 blkbits = ilog2(attr->blksize);
869 else
870 blkbits = inode->i_sb->s_blocksize_bits;
871
872 stat->blksize = 1 << blkbits;
873 }
874
875 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
876 struct file *file)
877 {
878 int err;
879 struct fuse_getattr_in inarg;
880 struct fuse_attr_out outarg;
881 struct fuse_conn *fc = get_fuse_conn(inode);
882 struct fuse_req *req;
883 u64 attr_version;
884
885 req = fuse_get_req_nopages(fc);
886 if (IS_ERR(req))
887 return PTR_ERR(req);
888
889 attr_version = fuse_get_attr_version(fc);
890
891 memset(&inarg, 0, sizeof(inarg));
892 memset(&outarg, 0, sizeof(outarg));
893 /* Directories have separate file-handle space */
894 if (file && S_ISREG(inode->i_mode)) {
895 struct fuse_file *ff = file->private_data;
896
897 inarg.getattr_flags |= FUSE_GETATTR_FH;
898 inarg.fh = ff->fh;
899 }
900 req->in.h.opcode = FUSE_GETATTR;
901 req->in.h.nodeid = get_node_id(inode);
902 req->in.numargs = 1;
903 req->in.args[0].size = sizeof(inarg);
904 req->in.args[0].value = &inarg;
905 req->out.numargs = 1;
906 if (fc->minor < 9)
907 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
908 else
909 req->out.args[0].size = sizeof(outarg);
910 req->out.args[0].value = &outarg;
911 fuse_request_send(fc, req);
912 err = req->out.h.error;
913 fuse_put_request(fc, req);
914 if (!err) {
915 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
916 make_bad_inode(inode);
917 err = -EIO;
918 } else {
919 fuse_change_attributes(inode, &outarg.attr,
920 attr_timeout(&outarg),
921 attr_version);
922 if (stat)
923 fuse_fillattr(inode, &outarg.attr, stat);
924 }
925 }
926 return err;
927 }
928
929 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
930 struct file *file, bool *refreshed)
931 {
932 struct fuse_inode *fi = get_fuse_inode(inode);
933 int err;
934 bool r;
935
936 if (fi->i_time < get_jiffies_64()) {
937 r = true;
938 err = fuse_do_getattr(inode, stat, file);
939 } else {
940 r = false;
941 err = 0;
942 if (stat) {
943 generic_fillattr(inode, stat);
944 stat->mode = fi->orig_i_mode;
945 stat->ino = fi->orig_ino;
946 }
947 }
948
949 if (refreshed != NULL)
950 *refreshed = r;
951
952 return err;
953 }
954
955 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
956 u64 child_nodeid, struct qstr *name)
957 {
958 int err = -ENOTDIR;
959 struct inode *parent;
960 struct dentry *dir;
961 struct dentry *entry;
962
963 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
964 if (!parent)
965 return -ENOENT;
966
967 mutex_lock(&parent->i_mutex);
968 if (!S_ISDIR(parent->i_mode))
969 goto unlock;
970
971 err = -ENOENT;
972 dir = d_find_alias(parent);
973 if (!dir)
974 goto unlock;
975
976 entry = d_lookup(dir, name);
977 dput(dir);
978 if (!entry)
979 goto unlock;
980
981 fuse_invalidate_attr(parent);
982 fuse_invalidate_entry(entry);
983
984 if (child_nodeid != 0 && entry->d_inode) {
985 mutex_lock(&entry->d_inode->i_mutex);
986 if (get_node_id(entry->d_inode) != child_nodeid) {
987 err = -ENOENT;
988 goto badentry;
989 }
990 if (d_mountpoint(entry)) {
991 err = -EBUSY;
992 goto badentry;
993 }
994 if (S_ISDIR(entry->d_inode->i_mode)) {
995 shrink_dcache_parent(entry);
996 if (!simple_empty(entry)) {
997 err = -ENOTEMPTY;
998 goto badentry;
999 }
1000 entry->d_inode->i_flags |= S_DEAD;
1001 }
1002 dont_mount(entry);
1003 clear_nlink(entry->d_inode);
1004 err = 0;
1005 badentry:
1006 mutex_unlock(&entry->d_inode->i_mutex);
1007 if (!err)
1008 d_delete(entry);
1009 } else {
1010 err = 0;
1011 }
1012 dput(entry);
1013
1014 unlock:
1015 mutex_unlock(&parent->i_mutex);
1016 iput(parent);
1017 return err;
1018 }
1019
1020 /*
1021 * Calling into a user-controlled filesystem gives the filesystem
1022 * daemon ptrace-like capabilities over the current process. This
1023 * means, that the filesystem daemon is able to record the exact
1024 * filesystem operations performed, and can also control the behavior
1025 * of the requester process in otherwise impossible ways. For example
1026 * it can delay the operation for arbitrary length of time allowing
1027 * DoS against the requester.
1028 *
1029 * For this reason only those processes can call into the filesystem,
1030 * for which the owner of the mount has ptrace privilege. This
1031 * excludes processes started by other users, suid or sgid processes.
1032 */
1033 int fuse_allow_current_process(struct fuse_conn *fc)
1034 {
1035 const struct cred *cred;
1036
1037 if (fc->flags & FUSE_ALLOW_OTHER)
1038 return 1;
1039
1040 cred = current_cred();
1041 if (uid_eq(cred->euid, fc->user_id) &&
1042 uid_eq(cred->suid, fc->user_id) &&
1043 uid_eq(cred->uid, fc->user_id) &&
1044 gid_eq(cred->egid, fc->group_id) &&
1045 gid_eq(cred->sgid, fc->group_id) &&
1046 gid_eq(cred->gid, fc->group_id))
1047 return 1;
1048
1049 return 0;
1050 }
1051
1052 static int fuse_access(struct inode *inode, int mask)
1053 {
1054 struct fuse_conn *fc = get_fuse_conn(inode);
1055 struct fuse_req *req;
1056 struct fuse_access_in inarg;
1057 int err;
1058
1059 BUG_ON(mask & MAY_NOT_BLOCK);
1060
1061 if (fc->no_access)
1062 return 0;
1063
1064 req = fuse_get_req_nopages(fc);
1065 if (IS_ERR(req))
1066 return PTR_ERR(req);
1067
1068 memset(&inarg, 0, sizeof(inarg));
1069 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1070 req->in.h.opcode = FUSE_ACCESS;
1071 req->in.h.nodeid = get_node_id(inode);
1072 req->in.numargs = 1;
1073 req->in.args[0].size = sizeof(inarg);
1074 req->in.args[0].value = &inarg;
1075 fuse_request_send(fc, req);
1076 err = req->out.h.error;
1077 fuse_put_request(fc, req);
1078 if (err == -ENOSYS) {
1079 fc->no_access = 1;
1080 err = 0;
1081 }
1082 return err;
1083 }
1084
1085 static int fuse_perm_getattr(struct inode *inode, int mask)
1086 {
1087 if (mask & MAY_NOT_BLOCK)
1088 return -ECHILD;
1089
1090 return fuse_do_getattr(inode, NULL, NULL);
1091 }
1092
1093 /*
1094 * Check permission. The two basic access models of FUSE are:
1095 *
1096 * 1) Local access checking ('default_permissions' mount option) based
1097 * on file mode. This is the plain old disk filesystem permission
1098 * modell.
1099 *
1100 * 2) "Remote" access checking, where server is responsible for
1101 * checking permission in each inode operation. An exception to this
1102 * is if ->permission() was invoked from sys_access() in which case an
1103 * access request is sent. Execute permission is still checked
1104 * locally based on file mode.
1105 */
1106 static int fuse_permission(struct inode *inode, int mask)
1107 {
1108 struct fuse_conn *fc = get_fuse_conn(inode);
1109 bool refreshed = false;
1110 int err = 0;
1111
1112 if (!fuse_allow_current_process(fc))
1113 return -EACCES;
1114
1115 /*
1116 * If attributes are needed, refresh them before proceeding
1117 */
1118 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1119 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1120 struct fuse_inode *fi = get_fuse_inode(inode);
1121
1122 if (fi->i_time < get_jiffies_64()) {
1123 refreshed = true;
1124
1125 err = fuse_perm_getattr(inode, mask);
1126 if (err)
1127 return err;
1128 }
1129 }
1130
1131 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1132 err = generic_permission(inode, mask);
1133
1134 /* If permission is denied, try to refresh file
1135 attributes. This is also needed, because the root
1136 node will at first have no permissions */
1137 if (err == -EACCES && !refreshed) {
1138 err = fuse_perm_getattr(inode, mask);
1139 if (!err)
1140 err = generic_permission(inode, mask);
1141 }
1142
1143 /* Note: the opposite of the above test does not
1144 exist. So if permissions are revoked this won't be
1145 noticed immediately, only after the attribute
1146 timeout has expired */
1147 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1148 err = fuse_access(inode, mask);
1149 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1150 if (!(inode->i_mode & S_IXUGO)) {
1151 if (refreshed)
1152 return -EACCES;
1153
1154 err = fuse_perm_getattr(inode, mask);
1155 if (!err && !(inode->i_mode & S_IXUGO))
1156 return -EACCES;
1157 }
1158 }
1159 return err;
1160 }
1161
1162 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1163 struct dir_context *ctx)
1164 {
1165 while (nbytes >= FUSE_NAME_OFFSET) {
1166 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1167 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1168 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1169 return -EIO;
1170 if (reclen > nbytes)
1171 break;
1172 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1173 return -EIO;
1174
1175 if (!dir_emit(ctx, dirent->name, dirent->namelen,
1176 dirent->ino, dirent->type))
1177 break;
1178
1179 buf += reclen;
1180 nbytes -= reclen;
1181 ctx->pos = dirent->off;
1182 }
1183
1184 return 0;
1185 }
1186
1187 static int fuse_direntplus_link(struct file *file,
1188 struct fuse_direntplus *direntplus,
1189 u64 attr_version)
1190 {
1191 int err;
1192 struct fuse_entry_out *o = &direntplus->entry_out;
1193 struct fuse_dirent *dirent = &direntplus->dirent;
1194 struct dentry *parent = file->f_path.dentry;
1195 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1196 struct dentry *dentry;
1197 struct dentry *alias;
1198 struct inode *dir = parent->d_inode;
1199 struct fuse_conn *fc;
1200 struct inode *inode;
1201
1202 if (!o->nodeid) {
1203 /*
1204 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1205 * ENOENT. Instead, it only means the userspace filesystem did
1206 * not want to return attributes/handle for this entry.
1207 *
1208 * So do nothing.
1209 */
1210 return 0;
1211 }
1212
1213 if (name.name[0] == '.') {
1214 /*
1215 * We could potentially refresh the attributes of the directory
1216 * and its parent?
1217 */
1218 if (name.len == 1)
1219 return 0;
1220 if (name.name[1] == '.' && name.len == 2)
1221 return 0;
1222 }
1223
1224 if (invalid_nodeid(o->nodeid))
1225 return -EIO;
1226 if (!fuse_valid_type(o->attr.mode))
1227 return -EIO;
1228
1229 fc = get_fuse_conn(dir);
1230
1231 name.hash = full_name_hash(name.name, name.len);
1232 dentry = d_lookup(parent, &name);
1233 if (dentry) {
1234 inode = dentry->d_inode;
1235 if (!inode) {
1236 d_drop(dentry);
1237 } else if (get_node_id(inode) != o->nodeid ||
1238 ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1239 err = d_invalidate(dentry);
1240 if (err)
1241 goto out;
1242 } else if (is_bad_inode(inode)) {
1243 err = -EIO;
1244 goto out;
1245 } else {
1246 struct fuse_inode *fi;
1247 fi = get_fuse_inode(inode);
1248 spin_lock(&fc->lock);
1249 fi->nlookup++;
1250 spin_unlock(&fc->lock);
1251
1252 fuse_change_attributes(inode, &o->attr,
1253 entry_attr_timeout(o),
1254 attr_version);
1255
1256 /*
1257 * The other branch to 'found' comes via fuse_iget()
1258 * which bumps nlookup inside
1259 */
1260 goto found;
1261 }
1262 dput(dentry);
1263 }
1264
1265 dentry = d_alloc(parent, &name);
1266 err = -ENOMEM;
1267 if (!dentry)
1268 goto out;
1269
1270 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1271 &o->attr, entry_attr_timeout(o), attr_version);
1272 if (!inode)
1273 goto out;
1274
1275 alias = d_materialise_unique(dentry, inode);
1276 err = PTR_ERR(alias);
1277 if (IS_ERR(alias))
1278 goto out;
1279
1280 if (alias) {
1281 dput(dentry);
1282 dentry = alias;
1283 }
1284
1285 found:
1286 if (fc->readdirplus_auto)
1287 set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1288 fuse_change_entry_timeout(dentry, o);
1289
1290 err = 0;
1291 out:
1292 dput(dentry);
1293 return err;
1294 }
1295
1296 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1297 struct dir_context *ctx, u64 attr_version)
1298 {
1299 struct fuse_direntplus *direntplus;
1300 struct fuse_dirent *dirent;
1301 size_t reclen;
1302 int over = 0;
1303 int ret;
1304
1305 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1306 direntplus = (struct fuse_direntplus *) buf;
1307 dirent = &direntplus->dirent;
1308 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1309
1310 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1311 return -EIO;
1312 if (reclen > nbytes)
1313 break;
1314 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1315 return -EIO;
1316
1317 if (!over) {
1318 /* We fill entries into dstbuf only as much as
1319 it can hold. But we still continue iterating
1320 over remaining entries to link them. If not,
1321 we need to send a FORGET for each of those
1322 which we did not link.
1323 */
1324 over = !dir_emit(ctx, dirent->name, dirent->namelen,
1325 dirent->ino, dirent->type);
1326 ctx->pos = dirent->off;
1327 }
1328
1329 buf += reclen;
1330 nbytes -= reclen;
1331
1332 ret = fuse_direntplus_link(file, direntplus, attr_version);
1333 if (ret)
1334 fuse_force_forget(file, direntplus->entry_out.nodeid);
1335 }
1336
1337 return 0;
1338 }
1339
1340 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1341 {
1342 int plus, err;
1343 size_t nbytes;
1344 struct page *page;
1345 struct inode *inode = file_inode(file);
1346 struct fuse_conn *fc = get_fuse_conn(inode);
1347 struct fuse_req *req;
1348 u64 attr_version = 0;
1349
1350 if (is_bad_inode(inode))
1351 return -EIO;
1352
1353 req = fuse_get_req(fc, 1);
1354 if (IS_ERR(req))
1355 return PTR_ERR(req);
1356
1357 page = alloc_page(GFP_KERNEL);
1358 if (!page) {
1359 fuse_put_request(fc, req);
1360 return -ENOMEM;
1361 }
1362
1363 plus = fuse_use_readdirplus(inode, ctx);
1364 req->out.argpages = 1;
1365 req->num_pages = 1;
1366 req->pages[0] = page;
1367 req->page_descs[0].length = PAGE_SIZE;
1368 if (plus) {
1369 attr_version = fuse_get_attr_version(fc);
1370 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1371 FUSE_READDIRPLUS);
1372 } else {
1373 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1374 FUSE_READDIR);
1375 }
1376 fuse_request_send(fc, req);
1377 nbytes = req->out.args[0].size;
1378 err = req->out.h.error;
1379 fuse_put_request(fc, req);
1380 if (!err) {
1381 if (plus) {
1382 err = parse_dirplusfile(page_address(page), nbytes,
1383 file, ctx,
1384 attr_version);
1385 } else {
1386 err = parse_dirfile(page_address(page), nbytes, file,
1387 ctx);
1388 }
1389 }
1390
1391 __free_page(page);
1392 fuse_invalidate_atime(inode);
1393 return err;
1394 }
1395
1396 static char *read_link(struct dentry *dentry)
1397 {
1398 struct inode *inode = dentry->d_inode;
1399 struct fuse_conn *fc = get_fuse_conn(inode);
1400 struct fuse_req *req = fuse_get_req_nopages(fc);
1401 char *link;
1402
1403 if (IS_ERR(req))
1404 return ERR_CAST(req);
1405
1406 link = (char *) __get_free_page(GFP_KERNEL);
1407 if (!link) {
1408 link = ERR_PTR(-ENOMEM);
1409 goto out;
1410 }
1411 req->in.h.opcode = FUSE_READLINK;
1412 req->in.h.nodeid = get_node_id(inode);
1413 req->out.argvar = 1;
1414 req->out.numargs = 1;
1415 req->out.args[0].size = PAGE_SIZE - 1;
1416 req->out.args[0].value = link;
1417 fuse_request_send(fc, req);
1418 if (req->out.h.error) {
1419 free_page((unsigned long) link);
1420 link = ERR_PTR(req->out.h.error);
1421 } else
1422 link[req->out.args[0].size] = '\0';
1423 out:
1424 fuse_put_request(fc, req);
1425 fuse_invalidate_atime(inode);
1426 return link;
1427 }
1428
1429 static void free_link(char *link)
1430 {
1431 if (!IS_ERR(link))
1432 free_page((unsigned long) link);
1433 }
1434
1435 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1436 {
1437 nd_set_link(nd, read_link(dentry));
1438 return NULL;
1439 }
1440
1441 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1442 {
1443 free_link(nd_get_link(nd));
1444 }
1445
1446 static int fuse_dir_open(struct inode *inode, struct file *file)
1447 {
1448 return fuse_open_common(inode, file, true);
1449 }
1450
1451 static int fuse_dir_release(struct inode *inode, struct file *file)
1452 {
1453 fuse_release_common(file, FUSE_RELEASEDIR);
1454
1455 return 0;
1456 }
1457
1458 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1459 int datasync)
1460 {
1461 return fuse_fsync_common(file, start, end, datasync, 1);
1462 }
1463
1464 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1465 unsigned long arg)
1466 {
1467 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1468
1469 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1470 if (fc->minor < 18)
1471 return -ENOTTY;
1472
1473 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1474 }
1475
1476 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1477 unsigned long arg)
1478 {
1479 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1480
1481 if (fc->minor < 18)
1482 return -ENOTTY;
1483
1484 return fuse_ioctl_common(file, cmd, arg,
1485 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1486 }
1487
1488 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1489 {
1490 /* Always update if mtime is explicitly set */
1491 if (ivalid & ATTR_MTIME_SET)
1492 return true;
1493
1494 /* Or if kernel i_mtime is the official one */
1495 if (trust_local_mtime)
1496 return true;
1497
1498 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1499 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1500 return false;
1501
1502 /* In all other cases update */
1503 return true;
1504 }
1505
1506 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
1507 bool trust_local_mtime)
1508 {
1509 unsigned ivalid = iattr->ia_valid;
1510
1511 if (ivalid & ATTR_MODE)
1512 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1513 if (ivalid & ATTR_UID)
1514 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1515 if (ivalid & ATTR_GID)
1516 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1517 if (ivalid & ATTR_SIZE)
1518 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1519 if (ivalid & ATTR_ATIME) {
1520 arg->valid |= FATTR_ATIME;
1521 arg->atime = iattr->ia_atime.tv_sec;
1522 arg->atimensec = iattr->ia_atime.tv_nsec;
1523 if (!(ivalid & ATTR_ATIME_SET))
1524 arg->valid |= FATTR_ATIME_NOW;
1525 }
1526 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_mtime)) {
1527 arg->valid |= FATTR_MTIME;
1528 arg->mtime = iattr->ia_mtime.tv_sec;
1529 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1530 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_mtime)
1531 arg->valid |= FATTR_MTIME_NOW;
1532 }
1533 }
1534
1535 /*
1536 * Prevent concurrent writepages on inode
1537 *
1538 * This is done by adding a negative bias to the inode write counter
1539 * and waiting for all pending writes to finish.
1540 */
1541 void fuse_set_nowrite(struct inode *inode)
1542 {
1543 struct fuse_conn *fc = get_fuse_conn(inode);
1544 struct fuse_inode *fi = get_fuse_inode(inode);
1545
1546 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1547
1548 spin_lock(&fc->lock);
1549 BUG_ON(fi->writectr < 0);
1550 fi->writectr += FUSE_NOWRITE;
1551 spin_unlock(&fc->lock);
1552 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1553 }
1554
1555 /*
1556 * Allow writepages on inode
1557 *
1558 * Remove the bias from the writecounter and send any queued
1559 * writepages.
1560 */
1561 static void __fuse_release_nowrite(struct inode *inode)
1562 {
1563 struct fuse_inode *fi = get_fuse_inode(inode);
1564
1565 BUG_ON(fi->writectr != FUSE_NOWRITE);
1566 fi->writectr = 0;
1567 fuse_flush_writepages(inode);
1568 }
1569
1570 void fuse_release_nowrite(struct inode *inode)
1571 {
1572 struct fuse_conn *fc = get_fuse_conn(inode);
1573
1574 spin_lock(&fc->lock);
1575 __fuse_release_nowrite(inode);
1576 spin_unlock(&fc->lock);
1577 }
1578
1579 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_req *req,
1580 struct inode *inode,
1581 struct fuse_setattr_in *inarg_p,
1582 struct fuse_attr_out *outarg_p)
1583 {
1584 req->in.h.opcode = FUSE_SETATTR;
1585 req->in.h.nodeid = get_node_id(inode);
1586 req->in.numargs = 1;
1587 req->in.args[0].size = sizeof(*inarg_p);
1588 req->in.args[0].value = inarg_p;
1589 req->out.numargs = 1;
1590 if (fc->minor < 9)
1591 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1592 else
1593 req->out.args[0].size = sizeof(*outarg_p);
1594 req->out.args[0].value = outarg_p;
1595 }
1596
1597 /*
1598 * Flush inode->i_mtime to the server
1599 */
1600 int fuse_flush_mtime(struct inode *inode, struct fuse_file *ff)
1601 {
1602 struct fuse_conn *fc = get_fuse_conn(inode);
1603 struct fuse_req *req;
1604 struct fuse_setattr_in inarg;
1605 struct fuse_attr_out outarg;
1606 int err;
1607
1608 req = fuse_get_req_nopages(fc);
1609 if (IS_ERR(req))
1610 return PTR_ERR(req);
1611
1612 memset(&inarg, 0, sizeof(inarg));
1613 memset(&outarg, 0, sizeof(outarg));
1614
1615 inarg.valid |= FATTR_MTIME;
1616 inarg.mtime = inode->i_mtime.tv_sec;
1617 inarg.mtimensec = inode->i_mtime.tv_nsec;
1618 if (ff) {
1619 inarg.valid |= FATTR_FH;
1620 inarg.fh = ff->fh;
1621 }
1622 fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
1623 fuse_request_send(fc, req);
1624 err = req->out.h.error;
1625 fuse_put_request(fc, req);
1626
1627 return err;
1628 }
1629
1630 /*
1631 * Set attributes, and at the same time refresh them.
1632 *
1633 * Truncation is slightly complicated, because the 'truncate' request
1634 * may fail, in which case we don't want to touch the mapping.
1635 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1636 * and the actual truncation by hand.
1637 */
1638 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
1639 struct file *file)
1640 {
1641 struct fuse_conn *fc = get_fuse_conn(inode);
1642 struct fuse_inode *fi = get_fuse_inode(inode);
1643 struct fuse_req *req;
1644 struct fuse_setattr_in inarg;
1645 struct fuse_attr_out outarg;
1646 bool is_truncate = false;
1647 bool is_wb = fc->writeback_cache;
1648 loff_t oldsize;
1649 int err;
1650 bool trust_local_mtime = is_wb && S_ISREG(inode->i_mode);
1651
1652 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1653 attr->ia_valid |= ATTR_FORCE;
1654
1655 err = inode_change_ok(inode, attr);
1656 if (err)
1657 return err;
1658
1659 if (attr->ia_valid & ATTR_OPEN) {
1660 if (fc->atomic_o_trunc)
1661 return 0;
1662 file = NULL;
1663 }
1664
1665 if (attr->ia_valid & ATTR_SIZE)
1666 is_truncate = true;
1667
1668 req = fuse_get_req_nopages(fc);
1669 if (IS_ERR(req))
1670 return PTR_ERR(req);
1671
1672 if (is_truncate) {
1673 fuse_set_nowrite(inode);
1674 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1675 if (trust_local_mtime && attr->ia_size != inode->i_size)
1676 attr->ia_valid |= ATTR_MTIME;
1677 }
1678
1679 memset(&inarg, 0, sizeof(inarg));
1680 memset(&outarg, 0, sizeof(outarg));
1681 iattr_to_fattr(attr, &inarg, trust_local_mtime);
1682 if (file) {
1683 struct fuse_file *ff = file->private_data;
1684 inarg.valid |= FATTR_FH;
1685 inarg.fh = ff->fh;
1686 }
1687 if (attr->ia_valid & ATTR_SIZE) {
1688 /* For mandatory locking in truncate */
1689 inarg.valid |= FATTR_LOCKOWNER;
1690 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1691 }
1692 fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
1693 fuse_request_send(fc, req);
1694 err = req->out.h.error;
1695 fuse_put_request(fc, req);
1696 if (err) {
1697 if (err == -EINTR)
1698 fuse_invalidate_attr(inode);
1699 goto error;
1700 }
1701
1702 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1703 make_bad_inode(inode);
1704 err = -EIO;
1705 goto error;
1706 }
1707
1708 spin_lock(&fc->lock);
1709 /* the kernel maintains i_mtime locally */
1710 if (trust_local_mtime && (attr->ia_valid & ATTR_MTIME)) {
1711 inode->i_mtime = attr->ia_mtime;
1712 /* FIXME: clear I_DIRTY_SYNC? */
1713 }
1714
1715 fuse_change_attributes_common(inode, &outarg.attr,
1716 attr_timeout(&outarg));
1717 oldsize = inode->i_size;
1718 /* see the comment in fuse_change_attributes() */
1719 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1720 i_size_write(inode, outarg.attr.size);
1721
1722 if (is_truncate) {
1723 /* NOTE: this may release/reacquire fc->lock */
1724 __fuse_release_nowrite(inode);
1725 }
1726 spin_unlock(&fc->lock);
1727
1728 /*
1729 * Only call invalidate_inode_pages2() after removing
1730 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1731 */
1732 if ((is_truncate || !is_wb) &&
1733 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1734 truncate_pagecache(inode, outarg.attr.size);
1735 invalidate_inode_pages2(inode->i_mapping);
1736 }
1737
1738 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1739 return 0;
1740
1741 error:
1742 if (is_truncate)
1743 fuse_release_nowrite(inode);
1744
1745 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1746 return err;
1747 }
1748
1749 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1750 {
1751 struct inode *inode = entry->d_inode;
1752
1753 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1754 return -EACCES;
1755
1756 if (attr->ia_valid & ATTR_FILE)
1757 return fuse_do_setattr(inode, attr, attr->ia_file);
1758 else
1759 return fuse_do_setattr(inode, attr, NULL);
1760 }
1761
1762 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1763 struct kstat *stat)
1764 {
1765 struct inode *inode = entry->d_inode;
1766 struct fuse_conn *fc = get_fuse_conn(inode);
1767
1768 if (!fuse_allow_current_process(fc))
1769 return -EACCES;
1770
1771 return fuse_update_attributes(inode, stat, NULL, NULL);
1772 }
1773
1774 static int fuse_setxattr(struct dentry *entry, const char *name,
1775 const void *value, size_t size, int flags)
1776 {
1777 struct inode *inode = entry->d_inode;
1778 struct fuse_conn *fc = get_fuse_conn(inode);
1779 struct fuse_req *req;
1780 struct fuse_setxattr_in inarg;
1781 int err;
1782
1783 if (fc->no_setxattr)
1784 return -EOPNOTSUPP;
1785
1786 req = fuse_get_req_nopages(fc);
1787 if (IS_ERR(req))
1788 return PTR_ERR(req);
1789
1790 memset(&inarg, 0, sizeof(inarg));
1791 inarg.size = size;
1792 inarg.flags = flags;
1793 req->in.h.opcode = FUSE_SETXATTR;
1794 req->in.h.nodeid = get_node_id(inode);
1795 req->in.numargs = 3;
1796 req->in.args[0].size = sizeof(inarg);
1797 req->in.args[0].value = &inarg;
1798 req->in.args[1].size = strlen(name) + 1;
1799 req->in.args[1].value = name;
1800 req->in.args[2].size = size;
1801 req->in.args[2].value = value;
1802 fuse_request_send(fc, req);
1803 err = req->out.h.error;
1804 fuse_put_request(fc, req);
1805 if (err == -ENOSYS) {
1806 fc->no_setxattr = 1;
1807 err = -EOPNOTSUPP;
1808 }
1809 if (!err)
1810 fuse_invalidate_attr(inode);
1811 return err;
1812 }
1813
1814 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1815 void *value, size_t size)
1816 {
1817 struct inode *inode = entry->d_inode;
1818 struct fuse_conn *fc = get_fuse_conn(inode);
1819 struct fuse_req *req;
1820 struct fuse_getxattr_in inarg;
1821 struct fuse_getxattr_out outarg;
1822 ssize_t ret;
1823
1824 if (fc->no_getxattr)
1825 return -EOPNOTSUPP;
1826
1827 req = fuse_get_req_nopages(fc);
1828 if (IS_ERR(req))
1829 return PTR_ERR(req);
1830
1831 memset(&inarg, 0, sizeof(inarg));
1832 inarg.size = size;
1833 req->in.h.opcode = FUSE_GETXATTR;
1834 req->in.h.nodeid = get_node_id(inode);
1835 req->in.numargs = 2;
1836 req->in.args[0].size = sizeof(inarg);
1837 req->in.args[0].value = &inarg;
1838 req->in.args[1].size = strlen(name) + 1;
1839 req->in.args[1].value = name;
1840 /* This is really two different operations rolled into one */
1841 req->out.numargs = 1;
1842 if (size) {
1843 req->out.argvar = 1;
1844 req->out.args[0].size = size;
1845 req->out.args[0].value = value;
1846 } else {
1847 req->out.args[0].size = sizeof(outarg);
1848 req->out.args[0].value = &outarg;
1849 }
1850 fuse_request_send(fc, req);
1851 ret = req->out.h.error;
1852 if (!ret)
1853 ret = size ? req->out.args[0].size : outarg.size;
1854 else {
1855 if (ret == -ENOSYS) {
1856 fc->no_getxattr = 1;
1857 ret = -EOPNOTSUPP;
1858 }
1859 }
1860 fuse_put_request(fc, req);
1861 return ret;
1862 }
1863
1864 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1865 {
1866 struct inode *inode = entry->d_inode;
1867 struct fuse_conn *fc = get_fuse_conn(inode);
1868 struct fuse_req *req;
1869 struct fuse_getxattr_in inarg;
1870 struct fuse_getxattr_out outarg;
1871 ssize_t ret;
1872
1873 if (!fuse_allow_current_process(fc))
1874 return -EACCES;
1875
1876 if (fc->no_listxattr)
1877 return -EOPNOTSUPP;
1878
1879 req = fuse_get_req_nopages(fc);
1880 if (IS_ERR(req))
1881 return PTR_ERR(req);
1882
1883 memset(&inarg, 0, sizeof(inarg));
1884 inarg.size = size;
1885 req->in.h.opcode = FUSE_LISTXATTR;
1886 req->in.h.nodeid = get_node_id(inode);
1887 req->in.numargs = 1;
1888 req->in.args[0].size = sizeof(inarg);
1889 req->in.args[0].value = &inarg;
1890 /* This is really two different operations rolled into one */
1891 req->out.numargs = 1;
1892 if (size) {
1893 req->out.argvar = 1;
1894 req->out.args[0].size = size;
1895 req->out.args[0].value = list;
1896 } else {
1897 req->out.args[0].size = sizeof(outarg);
1898 req->out.args[0].value = &outarg;
1899 }
1900 fuse_request_send(fc, req);
1901 ret = req->out.h.error;
1902 if (!ret)
1903 ret = size ? req->out.args[0].size : outarg.size;
1904 else {
1905 if (ret == -ENOSYS) {
1906 fc->no_listxattr = 1;
1907 ret = -EOPNOTSUPP;
1908 }
1909 }
1910 fuse_put_request(fc, req);
1911 return ret;
1912 }
1913
1914 static int fuse_removexattr(struct dentry *entry, const char *name)
1915 {
1916 struct inode *inode = entry->d_inode;
1917 struct fuse_conn *fc = get_fuse_conn(inode);
1918 struct fuse_req *req;
1919 int err;
1920
1921 if (fc->no_removexattr)
1922 return -EOPNOTSUPP;
1923
1924 req = fuse_get_req_nopages(fc);
1925 if (IS_ERR(req))
1926 return PTR_ERR(req);
1927
1928 req->in.h.opcode = FUSE_REMOVEXATTR;
1929 req->in.h.nodeid = get_node_id(inode);
1930 req->in.numargs = 1;
1931 req->in.args[0].size = strlen(name) + 1;
1932 req->in.args[0].value = name;
1933 fuse_request_send(fc, req);
1934 err = req->out.h.error;
1935 fuse_put_request(fc, req);
1936 if (err == -ENOSYS) {
1937 fc->no_removexattr = 1;
1938 err = -EOPNOTSUPP;
1939 }
1940 if (!err)
1941 fuse_invalidate_attr(inode);
1942 return err;
1943 }
1944
1945 static int fuse_update_time(struct inode *inode, struct timespec *now,
1946 int flags)
1947 {
1948 if (flags & S_MTIME) {
1949 inode->i_mtime = *now;
1950 mark_inode_dirty_sync(inode);
1951 BUG_ON(!S_ISREG(inode->i_mode));
1952 }
1953 return 0;
1954 }
1955
1956 static const struct inode_operations fuse_dir_inode_operations = {
1957 .lookup = fuse_lookup,
1958 .mkdir = fuse_mkdir,
1959 .symlink = fuse_symlink,
1960 .unlink = fuse_unlink,
1961 .rmdir = fuse_rmdir,
1962 .rename = fuse_rename,
1963 .link = fuse_link,
1964 .setattr = fuse_setattr,
1965 .create = fuse_create,
1966 .atomic_open = fuse_atomic_open,
1967 .mknod = fuse_mknod,
1968 .permission = fuse_permission,
1969 .getattr = fuse_getattr,
1970 .setxattr = fuse_setxattr,
1971 .getxattr = fuse_getxattr,
1972 .listxattr = fuse_listxattr,
1973 .removexattr = fuse_removexattr,
1974 };
1975
1976 static const struct file_operations fuse_dir_operations = {
1977 .llseek = generic_file_llseek,
1978 .read = generic_read_dir,
1979 .iterate = fuse_readdir,
1980 .open = fuse_dir_open,
1981 .release = fuse_dir_release,
1982 .fsync = fuse_dir_fsync,
1983 .unlocked_ioctl = fuse_dir_ioctl,
1984 .compat_ioctl = fuse_dir_compat_ioctl,
1985 };
1986
1987 static const struct inode_operations fuse_common_inode_operations = {
1988 .setattr = fuse_setattr,
1989 .permission = fuse_permission,
1990 .getattr = fuse_getattr,
1991 .setxattr = fuse_setxattr,
1992 .getxattr = fuse_getxattr,
1993 .listxattr = fuse_listxattr,
1994 .removexattr = fuse_removexattr,
1995 .update_time = fuse_update_time,
1996 };
1997
1998 static const struct inode_operations fuse_symlink_inode_operations = {
1999 .setattr = fuse_setattr,
2000 .follow_link = fuse_follow_link,
2001 .put_link = fuse_put_link,
2002 .readlink = generic_readlink,
2003 .getattr = fuse_getattr,
2004 .setxattr = fuse_setxattr,
2005 .getxattr = fuse_getxattr,
2006 .listxattr = fuse_listxattr,
2007 .removexattr = fuse_removexattr,
2008 };
2009
2010 void fuse_init_common(struct inode *inode)
2011 {
2012 inode->i_op = &fuse_common_inode_operations;
2013 }
2014
2015 void fuse_init_dir(struct inode *inode)
2016 {
2017 inode->i_op = &fuse_dir_inode_operations;
2018 inode->i_fop = &fuse_dir_operations;
2019 }
2020
2021 void fuse_init_symlink(struct inode *inode)
2022 {
2023 inode->i_op = &fuse_symlink_inode_operations;
2024 }
This page took 0.08248 seconds and 5 git commands to generate.