Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu
[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 (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
202 (flags & LOOKUP_REVAL)) {
203 int err;
204 struct fuse_entry_out outarg;
205 struct fuse_req *req;
206 struct fuse_forget_link *forget;
207 u64 attr_version;
208
209 /* For negative dentries, always do a fresh lookup */
210 if (!inode)
211 goto invalid;
212
213 ret = -ECHILD;
214 if (flags & LOOKUP_RCU)
215 goto out;
216
217 fc = get_fuse_conn(inode);
218 req = fuse_get_req_nopages(fc);
219 ret = PTR_ERR(req);
220 if (IS_ERR(req))
221 goto out;
222
223 forget = fuse_alloc_forget();
224 if (!forget) {
225 fuse_put_request(fc, req);
226 ret = -ENOMEM;
227 goto out;
228 }
229
230 attr_version = fuse_get_attr_version(fc);
231
232 parent = dget_parent(entry);
233 fuse_lookup_init(fc, req, get_node_id(parent->d_inode),
234 &entry->d_name, &outarg);
235 fuse_request_send(fc, req);
236 dput(parent);
237 err = req->out.h.error;
238 fuse_put_request(fc, req);
239 /* Zero nodeid is same as -ENOENT */
240 if (!err && !outarg.nodeid)
241 err = -ENOENT;
242 if (!err) {
243 fi = get_fuse_inode(inode);
244 if (outarg.nodeid != get_node_id(inode)) {
245 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
246 goto invalid;
247 }
248 spin_lock(&fc->lock);
249 fi->nlookup++;
250 spin_unlock(&fc->lock);
251 }
252 kfree(forget);
253 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
254 goto invalid;
255
256 fuse_change_attributes(inode, &outarg.attr,
257 entry_attr_timeout(&outarg),
258 attr_version);
259 fuse_change_entry_timeout(entry, &outarg);
260 } else if (inode) {
261 fi = get_fuse_inode(inode);
262 if (flags & LOOKUP_RCU) {
263 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
264 return -ECHILD;
265 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
266 parent = dget_parent(entry);
267 fuse_advise_use_readdirplus(parent->d_inode);
268 dput(parent);
269 }
270 }
271 ret = 1;
272 out:
273 return ret;
274
275 invalid:
276 ret = 0;
277 goto out;
278 }
279
280 static int invalid_nodeid(u64 nodeid)
281 {
282 return !nodeid || nodeid == FUSE_ROOT_ID;
283 }
284
285 const struct dentry_operations fuse_dentry_operations = {
286 .d_revalidate = fuse_dentry_revalidate,
287 };
288
289 int fuse_valid_type(int m)
290 {
291 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
292 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
293 }
294
295 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
296 struct fuse_entry_out *outarg, struct inode **inode)
297 {
298 struct fuse_conn *fc = get_fuse_conn_super(sb);
299 struct fuse_req *req;
300 struct fuse_forget_link *forget;
301 u64 attr_version;
302 int err;
303
304 *inode = NULL;
305 err = -ENAMETOOLONG;
306 if (name->len > FUSE_NAME_MAX)
307 goto out;
308
309 req = fuse_get_req_nopages(fc);
310 err = PTR_ERR(req);
311 if (IS_ERR(req))
312 goto out;
313
314 forget = fuse_alloc_forget();
315 err = -ENOMEM;
316 if (!forget) {
317 fuse_put_request(fc, req);
318 goto out;
319 }
320
321 attr_version = fuse_get_attr_version(fc);
322
323 fuse_lookup_init(fc, req, nodeid, name, outarg);
324 fuse_request_send(fc, req);
325 err = req->out.h.error;
326 fuse_put_request(fc, req);
327 /* Zero nodeid is same as -ENOENT, but with valid timeout */
328 if (err || !outarg->nodeid)
329 goto out_put_forget;
330
331 err = -EIO;
332 if (!outarg->nodeid)
333 goto out_put_forget;
334 if (!fuse_valid_type(outarg->attr.mode))
335 goto out_put_forget;
336
337 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
338 &outarg->attr, entry_attr_timeout(outarg),
339 attr_version);
340 err = -ENOMEM;
341 if (!*inode) {
342 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
343 goto out;
344 }
345 err = 0;
346
347 out_put_forget:
348 kfree(forget);
349 out:
350 return err;
351 }
352
353 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
354 unsigned int flags)
355 {
356 int err;
357 struct fuse_entry_out outarg;
358 struct inode *inode;
359 struct dentry *newent;
360 bool outarg_valid = true;
361
362 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
363 &outarg, &inode);
364 if (err == -ENOENT) {
365 outarg_valid = false;
366 err = 0;
367 }
368 if (err)
369 goto out_err;
370
371 err = -EIO;
372 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
373 goto out_iput;
374
375 newent = d_materialise_unique(entry, inode);
376 err = PTR_ERR(newent);
377 if (IS_ERR(newent))
378 goto out_err;
379
380 entry = newent ? newent : entry;
381 if (outarg_valid)
382 fuse_change_entry_timeout(entry, &outarg);
383 else
384 fuse_invalidate_entry_cache(entry);
385
386 fuse_advise_use_readdirplus(dir);
387 return newent;
388
389 out_iput:
390 iput(inode);
391 out_err:
392 return ERR_PTR(err);
393 }
394
395 /*
396 * Atomic create+open operation
397 *
398 * If the filesystem doesn't support this, then fall back to separate
399 * 'mknod' + 'open' requests.
400 */
401 static int fuse_create_open(struct inode *dir, struct dentry *entry,
402 struct file *file, unsigned flags,
403 umode_t mode, int *opened)
404 {
405 int err;
406 struct inode *inode;
407 struct fuse_conn *fc = get_fuse_conn(dir);
408 struct fuse_req *req;
409 struct fuse_forget_link *forget;
410 struct fuse_create_in inarg;
411 struct fuse_open_out outopen;
412 struct fuse_entry_out outentry;
413 struct fuse_file *ff;
414
415 /* Userspace expects S_IFREG in create mode */
416 BUG_ON((mode & S_IFMT) != S_IFREG);
417
418 forget = fuse_alloc_forget();
419 err = -ENOMEM;
420 if (!forget)
421 goto out_err;
422
423 req = fuse_get_req_nopages(fc);
424 err = PTR_ERR(req);
425 if (IS_ERR(req))
426 goto out_put_forget_req;
427
428 err = -ENOMEM;
429 ff = fuse_file_alloc(fc);
430 if (!ff)
431 goto out_put_request;
432
433 if (!fc->dont_mask)
434 mode &= ~current_umask();
435
436 flags &= ~O_NOCTTY;
437 memset(&inarg, 0, sizeof(inarg));
438 memset(&outentry, 0, sizeof(outentry));
439 inarg.flags = flags;
440 inarg.mode = mode;
441 inarg.umask = current_umask();
442 req->in.h.opcode = FUSE_CREATE;
443 req->in.h.nodeid = get_node_id(dir);
444 req->in.numargs = 2;
445 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
446 sizeof(inarg);
447 req->in.args[0].value = &inarg;
448 req->in.args[1].size = entry->d_name.len + 1;
449 req->in.args[1].value = entry->d_name.name;
450 req->out.numargs = 2;
451 if (fc->minor < 9)
452 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
453 else
454 req->out.args[0].size = sizeof(outentry);
455 req->out.args[0].value = &outentry;
456 req->out.args[1].size = sizeof(outopen);
457 req->out.args[1].value = &outopen;
458 fuse_request_send(fc, req);
459 err = req->out.h.error;
460 if (err)
461 goto out_free_ff;
462
463 err = -EIO;
464 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
465 goto out_free_ff;
466
467 fuse_put_request(fc, req);
468 ff->fh = outopen.fh;
469 ff->nodeid = outentry.nodeid;
470 ff->open_flags = outopen.open_flags;
471 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
472 &outentry.attr, entry_attr_timeout(&outentry), 0);
473 if (!inode) {
474 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
475 fuse_sync_release(ff, flags);
476 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
477 err = -ENOMEM;
478 goto out_err;
479 }
480 kfree(forget);
481 d_instantiate(entry, inode);
482 fuse_change_entry_timeout(entry, &outentry);
483 fuse_invalidate_attr(dir);
484 err = finish_open(file, entry, generic_file_open, opened);
485 if (err) {
486 fuse_sync_release(ff, flags);
487 } else {
488 file->private_data = fuse_file_get(ff);
489 fuse_finish_open(inode, file);
490 }
491 return err;
492
493 out_free_ff:
494 fuse_file_free(ff);
495 out_put_request:
496 fuse_put_request(fc, req);
497 out_put_forget_req:
498 kfree(forget);
499 out_err:
500 return err;
501 }
502
503 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
504 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
505 struct file *file, unsigned flags,
506 umode_t mode, int *opened)
507 {
508 int err;
509 struct fuse_conn *fc = get_fuse_conn(dir);
510 struct dentry *res = NULL;
511
512 if (d_unhashed(entry)) {
513 res = fuse_lookup(dir, entry, 0);
514 if (IS_ERR(res))
515 return PTR_ERR(res);
516
517 if (res)
518 entry = res;
519 }
520
521 if (!(flags & O_CREAT) || entry->d_inode)
522 goto no_open;
523
524 /* Only creates */
525 *opened |= FILE_CREATED;
526
527 if (fc->no_create)
528 goto mknod;
529
530 err = fuse_create_open(dir, entry, file, flags, mode, opened);
531 if (err == -ENOSYS) {
532 fc->no_create = 1;
533 goto mknod;
534 }
535 out_dput:
536 dput(res);
537 return err;
538
539 mknod:
540 err = fuse_mknod(dir, entry, mode, 0);
541 if (err)
542 goto out_dput;
543 no_open:
544 return finish_no_open(file, res);
545 }
546
547 /*
548 * Code shared between mknod, mkdir, symlink and link
549 */
550 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
551 struct inode *dir, struct dentry *entry,
552 umode_t mode)
553 {
554 struct fuse_entry_out outarg;
555 struct inode *inode;
556 int err;
557 struct fuse_forget_link *forget;
558
559 forget = fuse_alloc_forget();
560 if (!forget) {
561 fuse_put_request(fc, req);
562 return -ENOMEM;
563 }
564
565 memset(&outarg, 0, sizeof(outarg));
566 req->in.h.nodeid = get_node_id(dir);
567 req->out.numargs = 1;
568 if (fc->minor < 9)
569 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
570 else
571 req->out.args[0].size = sizeof(outarg);
572 req->out.args[0].value = &outarg;
573 fuse_request_send(fc, req);
574 err = req->out.h.error;
575 fuse_put_request(fc, req);
576 if (err)
577 goto out_put_forget_req;
578
579 err = -EIO;
580 if (invalid_nodeid(outarg.nodeid))
581 goto out_put_forget_req;
582
583 if ((outarg.attr.mode ^ mode) & S_IFMT)
584 goto out_put_forget_req;
585
586 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
587 &outarg.attr, entry_attr_timeout(&outarg), 0);
588 if (!inode) {
589 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
590 return -ENOMEM;
591 }
592 kfree(forget);
593
594 err = d_instantiate_no_diralias(entry, inode);
595 if (err)
596 return err;
597
598 fuse_change_entry_timeout(entry, &outarg);
599 fuse_invalidate_attr(dir);
600 return 0;
601
602 out_put_forget_req:
603 kfree(forget);
604 return err;
605 }
606
607 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
608 dev_t rdev)
609 {
610 struct fuse_mknod_in inarg;
611 struct fuse_conn *fc = get_fuse_conn(dir);
612 struct fuse_req *req = fuse_get_req_nopages(fc);
613 if (IS_ERR(req))
614 return PTR_ERR(req);
615
616 if (!fc->dont_mask)
617 mode &= ~current_umask();
618
619 memset(&inarg, 0, sizeof(inarg));
620 inarg.mode = mode;
621 inarg.rdev = new_encode_dev(rdev);
622 inarg.umask = current_umask();
623 req->in.h.opcode = FUSE_MKNOD;
624 req->in.numargs = 2;
625 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
626 sizeof(inarg);
627 req->in.args[0].value = &inarg;
628 req->in.args[1].size = entry->d_name.len + 1;
629 req->in.args[1].value = entry->d_name.name;
630 return create_new_entry(fc, req, dir, entry, mode);
631 }
632
633 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
634 bool excl)
635 {
636 return fuse_mknod(dir, entry, mode, 0);
637 }
638
639 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
640 {
641 struct fuse_mkdir_in inarg;
642 struct fuse_conn *fc = get_fuse_conn(dir);
643 struct fuse_req *req = fuse_get_req_nopages(fc);
644 if (IS_ERR(req))
645 return PTR_ERR(req);
646
647 if (!fc->dont_mask)
648 mode &= ~current_umask();
649
650 memset(&inarg, 0, sizeof(inarg));
651 inarg.mode = mode;
652 inarg.umask = current_umask();
653 req->in.h.opcode = FUSE_MKDIR;
654 req->in.numargs = 2;
655 req->in.args[0].size = sizeof(inarg);
656 req->in.args[0].value = &inarg;
657 req->in.args[1].size = entry->d_name.len + 1;
658 req->in.args[1].value = entry->d_name.name;
659 return create_new_entry(fc, req, dir, entry, S_IFDIR);
660 }
661
662 static int fuse_symlink(struct inode *dir, struct dentry *entry,
663 const char *link)
664 {
665 struct fuse_conn *fc = get_fuse_conn(dir);
666 unsigned len = strlen(link) + 1;
667 struct fuse_req *req = fuse_get_req_nopages(fc);
668 if (IS_ERR(req))
669 return PTR_ERR(req);
670
671 req->in.h.opcode = FUSE_SYMLINK;
672 req->in.numargs = 2;
673 req->in.args[0].size = entry->d_name.len + 1;
674 req->in.args[0].value = entry->d_name.name;
675 req->in.args[1].size = len;
676 req->in.args[1].value = link;
677 return create_new_entry(fc, req, dir, entry, S_IFLNK);
678 }
679
680 static inline void fuse_update_ctime(struct inode *inode)
681 {
682 if (!IS_NOCMTIME(inode)) {
683 inode->i_ctime = current_fs_time(inode->i_sb);
684 mark_inode_dirty_sync(inode);
685 }
686 }
687
688 static int fuse_unlink(struct inode *dir, struct dentry *entry)
689 {
690 int err;
691 struct fuse_conn *fc = get_fuse_conn(dir);
692 struct fuse_req *req = fuse_get_req_nopages(fc);
693 if (IS_ERR(req))
694 return PTR_ERR(req);
695
696 req->in.h.opcode = FUSE_UNLINK;
697 req->in.h.nodeid = get_node_id(dir);
698 req->in.numargs = 1;
699 req->in.args[0].size = entry->d_name.len + 1;
700 req->in.args[0].value = entry->d_name.name;
701 fuse_request_send(fc, req);
702 err = req->out.h.error;
703 fuse_put_request(fc, req);
704 if (!err) {
705 struct inode *inode = entry->d_inode;
706 struct fuse_inode *fi = get_fuse_inode(inode);
707
708 spin_lock(&fc->lock);
709 fi->attr_version = ++fc->attr_version;
710 /*
711 * If i_nlink == 0 then unlink doesn't make sense, yet this can
712 * happen if userspace filesystem is careless. It would be
713 * difficult to enforce correct nlink usage so just ignore this
714 * condition here
715 */
716 if (inode->i_nlink > 0)
717 drop_nlink(inode);
718 spin_unlock(&fc->lock);
719 fuse_invalidate_attr(inode);
720 fuse_invalidate_attr(dir);
721 fuse_invalidate_entry_cache(entry);
722 fuse_update_ctime(inode);
723 } else if (err == -EINTR)
724 fuse_invalidate_entry(entry);
725 return err;
726 }
727
728 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
729 {
730 int err;
731 struct fuse_conn *fc = get_fuse_conn(dir);
732 struct fuse_req *req = fuse_get_req_nopages(fc);
733 if (IS_ERR(req))
734 return PTR_ERR(req);
735
736 req->in.h.opcode = FUSE_RMDIR;
737 req->in.h.nodeid = get_node_id(dir);
738 req->in.numargs = 1;
739 req->in.args[0].size = entry->d_name.len + 1;
740 req->in.args[0].value = entry->d_name.name;
741 fuse_request_send(fc, req);
742 err = req->out.h.error;
743 fuse_put_request(fc, req);
744 if (!err) {
745 clear_nlink(entry->d_inode);
746 fuse_invalidate_attr(dir);
747 fuse_invalidate_entry_cache(entry);
748 } else if (err == -EINTR)
749 fuse_invalidate_entry(entry);
750 return err;
751 }
752
753 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
754 struct inode *newdir, struct dentry *newent,
755 unsigned int flags, int opcode, size_t argsize)
756 {
757 int err;
758 struct fuse_rename2_in inarg;
759 struct fuse_conn *fc = get_fuse_conn(olddir);
760 struct fuse_req *req;
761
762 req = fuse_get_req_nopages(fc);
763 if (IS_ERR(req))
764 return PTR_ERR(req);
765
766 memset(&inarg, 0, argsize);
767 inarg.newdir = get_node_id(newdir);
768 inarg.flags = flags;
769 req->in.h.opcode = opcode;
770 req->in.h.nodeid = get_node_id(olddir);
771 req->in.numargs = 3;
772 req->in.args[0].size = argsize;
773 req->in.args[0].value = &inarg;
774 req->in.args[1].size = oldent->d_name.len + 1;
775 req->in.args[1].value = oldent->d_name.name;
776 req->in.args[2].size = newent->d_name.len + 1;
777 req->in.args[2].value = newent->d_name.name;
778 fuse_request_send(fc, req);
779 err = req->out.h.error;
780 fuse_put_request(fc, req);
781 if (!err) {
782 /* ctime changes */
783 fuse_invalidate_attr(oldent->d_inode);
784 fuse_update_ctime(oldent->d_inode);
785
786 if (flags & RENAME_EXCHANGE) {
787 fuse_invalidate_attr(newent->d_inode);
788 fuse_update_ctime(newent->d_inode);
789 }
790
791 fuse_invalidate_attr(olddir);
792 if (olddir != newdir)
793 fuse_invalidate_attr(newdir);
794
795 /* newent will end up negative */
796 if (!(flags & RENAME_EXCHANGE) && newent->d_inode) {
797 fuse_invalidate_attr(newent->d_inode);
798 fuse_invalidate_entry_cache(newent);
799 fuse_update_ctime(newent->d_inode);
800 }
801 } else if (err == -EINTR) {
802 /* If request was interrupted, DEITY only knows if the
803 rename actually took place. If the invalidation
804 fails (e.g. some process has CWD under the renamed
805 directory), then there can be inconsistency between
806 the dcache and the real filesystem. Tough luck. */
807 fuse_invalidate_entry(oldent);
808 if (newent->d_inode)
809 fuse_invalidate_entry(newent);
810 }
811
812 return err;
813 }
814
815 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
816 struct inode *newdir, struct dentry *newent,
817 unsigned int flags)
818 {
819 struct fuse_conn *fc = get_fuse_conn(olddir);
820 int err;
821
822 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
823 return -EINVAL;
824
825 if (flags) {
826 if (fc->no_rename2 || fc->minor < 23)
827 return -EINVAL;
828
829 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
830 FUSE_RENAME2,
831 sizeof(struct fuse_rename2_in));
832 if (err == -ENOSYS) {
833 fc->no_rename2 = 1;
834 err = -EINVAL;
835 }
836 } else {
837 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
838 FUSE_RENAME,
839 sizeof(struct fuse_rename_in));
840 }
841
842 return err;
843 }
844
845 static int fuse_link(struct dentry *entry, struct inode *newdir,
846 struct dentry *newent)
847 {
848 int err;
849 struct fuse_link_in inarg;
850 struct inode *inode = entry->d_inode;
851 struct fuse_conn *fc = get_fuse_conn(inode);
852 struct fuse_req *req = fuse_get_req_nopages(fc);
853 if (IS_ERR(req))
854 return PTR_ERR(req);
855
856 memset(&inarg, 0, sizeof(inarg));
857 inarg.oldnodeid = get_node_id(inode);
858 req->in.h.opcode = FUSE_LINK;
859 req->in.numargs = 2;
860 req->in.args[0].size = sizeof(inarg);
861 req->in.args[0].value = &inarg;
862 req->in.args[1].size = newent->d_name.len + 1;
863 req->in.args[1].value = newent->d_name.name;
864 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
865 /* Contrary to "normal" filesystems it can happen that link
866 makes two "logical" inodes point to the same "physical"
867 inode. We invalidate the attributes of the old one, so it
868 will reflect changes in the backing inode (link count,
869 etc.)
870 */
871 if (!err) {
872 struct fuse_inode *fi = get_fuse_inode(inode);
873
874 spin_lock(&fc->lock);
875 fi->attr_version = ++fc->attr_version;
876 inc_nlink(inode);
877 spin_unlock(&fc->lock);
878 fuse_invalidate_attr(inode);
879 fuse_update_ctime(inode);
880 } else if (err == -EINTR) {
881 fuse_invalidate_attr(inode);
882 }
883 return err;
884 }
885
886 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
887 struct kstat *stat)
888 {
889 unsigned int blkbits;
890 struct fuse_conn *fc = get_fuse_conn(inode);
891
892 /* see the comment in fuse_change_attributes() */
893 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
894 attr->size = i_size_read(inode);
895 attr->mtime = inode->i_mtime.tv_sec;
896 attr->mtimensec = inode->i_mtime.tv_nsec;
897 attr->ctime = inode->i_ctime.tv_sec;
898 attr->ctimensec = inode->i_ctime.tv_nsec;
899 }
900
901 stat->dev = inode->i_sb->s_dev;
902 stat->ino = attr->ino;
903 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
904 stat->nlink = attr->nlink;
905 stat->uid = make_kuid(&init_user_ns, attr->uid);
906 stat->gid = make_kgid(&init_user_ns, attr->gid);
907 stat->rdev = inode->i_rdev;
908 stat->atime.tv_sec = attr->atime;
909 stat->atime.tv_nsec = attr->atimensec;
910 stat->mtime.tv_sec = attr->mtime;
911 stat->mtime.tv_nsec = attr->mtimensec;
912 stat->ctime.tv_sec = attr->ctime;
913 stat->ctime.tv_nsec = attr->ctimensec;
914 stat->size = attr->size;
915 stat->blocks = attr->blocks;
916
917 if (attr->blksize != 0)
918 blkbits = ilog2(attr->blksize);
919 else
920 blkbits = inode->i_sb->s_blocksize_bits;
921
922 stat->blksize = 1 << blkbits;
923 }
924
925 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
926 struct file *file)
927 {
928 int err;
929 struct fuse_getattr_in inarg;
930 struct fuse_attr_out outarg;
931 struct fuse_conn *fc = get_fuse_conn(inode);
932 struct fuse_req *req;
933 u64 attr_version;
934
935 req = fuse_get_req_nopages(fc);
936 if (IS_ERR(req))
937 return PTR_ERR(req);
938
939 attr_version = fuse_get_attr_version(fc);
940
941 memset(&inarg, 0, sizeof(inarg));
942 memset(&outarg, 0, sizeof(outarg));
943 /* Directories have separate file-handle space */
944 if (file && S_ISREG(inode->i_mode)) {
945 struct fuse_file *ff = file->private_data;
946
947 inarg.getattr_flags |= FUSE_GETATTR_FH;
948 inarg.fh = ff->fh;
949 }
950 req->in.h.opcode = FUSE_GETATTR;
951 req->in.h.nodeid = get_node_id(inode);
952 req->in.numargs = 1;
953 req->in.args[0].size = sizeof(inarg);
954 req->in.args[0].value = &inarg;
955 req->out.numargs = 1;
956 if (fc->minor < 9)
957 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
958 else
959 req->out.args[0].size = sizeof(outarg);
960 req->out.args[0].value = &outarg;
961 fuse_request_send(fc, req);
962 err = req->out.h.error;
963 fuse_put_request(fc, req);
964 if (!err) {
965 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
966 make_bad_inode(inode);
967 err = -EIO;
968 } else {
969 fuse_change_attributes(inode, &outarg.attr,
970 attr_timeout(&outarg),
971 attr_version);
972 if (stat)
973 fuse_fillattr(inode, &outarg.attr, stat);
974 }
975 }
976 return err;
977 }
978
979 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
980 struct file *file, bool *refreshed)
981 {
982 struct fuse_inode *fi = get_fuse_inode(inode);
983 int err;
984 bool r;
985
986 if (time_before64(fi->i_time, get_jiffies_64())) {
987 r = true;
988 err = fuse_do_getattr(inode, stat, file);
989 } else {
990 r = false;
991 err = 0;
992 if (stat) {
993 generic_fillattr(inode, stat);
994 stat->mode = fi->orig_i_mode;
995 stat->ino = fi->orig_ino;
996 }
997 }
998
999 if (refreshed != NULL)
1000 *refreshed = r;
1001
1002 return err;
1003 }
1004
1005 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
1006 u64 child_nodeid, struct qstr *name)
1007 {
1008 int err = -ENOTDIR;
1009 struct inode *parent;
1010 struct dentry *dir;
1011 struct dentry *entry;
1012
1013 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
1014 if (!parent)
1015 return -ENOENT;
1016
1017 mutex_lock(&parent->i_mutex);
1018 if (!S_ISDIR(parent->i_mode))
1019 goto unlock;
1020
1021 err = -ENOENT;
1022 dir = d_find_alias(parent);
1023 if (!dir)
1024 goto unlock;
1025
1026 entry = d_lookup(dir, name);
1027 dput(dir);
1028 if (!entry)
1029 goto unlock;
1030
1031 fuse_invalidate_attr(parent);
1032 fuse_invalidate_entry(entry);
1033
1034 if (child_nodeid != 0 && entry->d_inode) {
1035 mutex_lock(&entry->d_inode->i_mutex);
1036 if (get_node_id(entry->d_inode) != child_nodeid) {
1037 err = -ENOENT;
1038 goto badentry;
1039 }
1040 if (d_mountpoint(entry)) {
1041 err = -EBUSY;
1042 goto badentry;
1043 }
1044 if (S_ISDIR(entry->d_inode->i_mode)) {
1045 shrink_dcache_parent(entry);
1046 if (!simple_empty(entry)) {
1047 err = -ENOTEMPTY;
1048 goto badentry;
1049 }
1050 entry->d_inode->i_flags |= S_DEAD;
1051 }
1052 dont_mount(entry);
1053 clear_nlink(entry->d_inode);
1054 err = 0;
1055 badentry:
1056 mutex_unlock(&entry->d_inode->i_mutex);
1057 if (!err)
1058 d_delete(entry);
1059 } else {
1060 err = 0;
1061 }
1062 dput(entry);
1063
1064 unlock:
1065 mutex_unlock(&parent->i_mutex);
1066 iput(parent);
1067 return err;
1068 }
1069
1070 /*
1071 * Calling into a user-controlled filesystem gives the filesystem
1072 * daemon ptrace-like capabilities over the current process. This
1073 * means, that the filesystem daemon is able to record the exact
1074 * filesystem operations performed, and can also control the behavior
1075 * of the requester process in otherwise impossible ways. For example
1076 * it can delay the operation for arbitrary length of time allowing
1077 * DoS against the requester.
1078 *
1079 * For this reason only those processes can call into the filesystem,
1080 * for which the owner of the mount has ptrace privilege. This
1081 * excludes processes started by other users, suid or sgid processes.
1082 */
1083 int fuse_allow_current_process(struct fuse_conn *fc)
1084 {
1085 const struct cred *cred;
1086
1087 if (fc->flags & FUSE_ALLOW_OTHER)
1088 return 1;
1089
1090 cred = current_cred();
1091 if (uid_eq(cred->euid, fc->user_id) &&
1092 uid_eq(cred->suid, fc->user_id) &&
1093 uid_eq(cred->uid, fc->user_id) &&
1094 gid_eq(cred->egid, fc->group_id) &&
1095 gid_eq(cred->sgid, fc->group_id) &&
1096 gid_eq(cred->gid, fc->group_id))
1097 return 1;
1098
1099 return 0;
1100 }
1101
1102 static int fuse_access(struct inode *inode, int mask)
1103 {
1104 struct fuse_conn *fc = get_fuse_conn(inode);
1105 struct fuse_req *req;
1106 struct fuse_access_in inarg;
1107 int err;
1108
1109 BUG_ON(mask & MAY_NOT_BLOCK);
1110
1111 if (fc->no_access)
1112 return 0;
1113
1114 req = fuse_get_req_nopages(fc);
1115 if (IS_ERR(req))
1116 return PTR_ERR(req);
1117
1118 memset(&inarg, 0, sizeof(inarg));
1119 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1120 req->in.h.opcode = FUSE_ACCESS;
1121 req->in.h.nodeid = get_node_id(inode);
1122 req->in.numargs = 1;
1123 req->in.args[0].size = sizeof(inarg);
1124 req->in.args[0].value = &inarg;
1125 fuse_request_send(fc, req);
1126 err = req->out.h.error;
1127 fuse_put_request(fc, req);
1128 if (err == -ENOSYS) {
1129 fc->no_access = 1;
1130 err = 0;
1131 }
1132 return err;
1133 }
1134
1135 static int fuse_perm_getattr(struct inode *inode, int mask)
1136 {
1137 if (mask & MAY_NOT_BLOCK)
1138 return -ECHILD;
1139
1140 return fuse_do_getattr(inode, NULL, NULL);
1141 }
1142
1143 /*
1144 * Check permission. The two basic access models of FUSE are:
1145 *
1146 * 1) Local access checking ('default_permissions' mount option) based
1147 * on file mode. This is the plain old disk filesystem permission
1148 * modell.
1149 *
1150 * 2) "Remote" access checking, where server is responsible for
1151 * checking permission in each inode operation. An exception to this
1152 * is if ->permission() was invoked from sys_access() in which case an
1153 * access request is sent. Execute permission is still checked
1154 * locally based on file mode.
1155 */
1156 static int fuse_permission(struct inode *inode, int mask)
1157 {
1158 struct fuse_conn *fc = get_fuse_conn(inode);
1159 bool refreshed = false;
1160 int err = 0;
1161
1162 if (!fuse_allow_current_process(fc))
1163 return -EACCES;
1164
1165 /*
1166 * If attributes are needed, refresh them before proceeding
1167 */
1168 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1169 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1170 struct fuse_inode *fi = get_fuse_inode(inode);
1171
1172 if (time_before64(fi->i_time, get_jiffies_64())) {
1173 refreshed = true;
1174
1175 err = fuse_perm_getattr(inode, mask);
1176 if (err)
1177 return err;
1178 }
1179 }
1180
1181 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1182 err = generic_permission(inode, mask);
1183
1184 /* If permission is denied, try to refresh file
1185 attributes. This is also needed, because the root
1186 node will at first have no permissions */
1187 if (err == -EACCES && !refreshed) {
1188 err = fuse_perm_getattr(inode, mask);
1189 if (!err)
1190 err = generic_permission(inode, mask);
1191 }
1192
1193 /* Note: the opposite of the above test does not
1194 exist. So if permissions are revoked this won't be
1195 noticed immediately, only after the attribute
1196 timeout has expired */
1197 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1198 err = fuse_access(inode, mask);
1199 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1200 if (!(inode->i_mode & S_IXUGO)) {
1201 if (refreshed)
1202 return -EACCES;
1203
1204 err = fuse_perm_getattr(inode, mask);
1205 if (!err && !(inode->i_mode & S_IXUGO))
1206 return -EACCES;
1207 }
1208 }
1209 return err;
1210 }
1211
1212 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1213 struct dir_context *ctx)
1214 {
1215 while (nbytes >= FUSE_NAME_OFFSET) {
1216 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1217 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1218 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1219 return -EIO;
1220 if (reclen > nbytes)
1221 break;
1222 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1223 return -EIO;
1224
1225 if (!dir_emit(ctx, dirent->name, dirent->namelen,
1226 dirent->ino, dirent->type))
1227 break;
1228
1229 buf += reclen;
1230 nbytes -= reclen;
1231 ctx->pos = dirent->off;
1232 }
1233
1234 return 0;
1235 }
1236
1237 static int fuse_direntplus_link(struct file *file,
1238 struct fuse_direntplus *direntplus,
1239 u64 attr_version)
1240 {
1241 int err;
1242 struct fuse_entry_out *o = &direntplus->entry_out;
1243 struct fuse_dirent *dirent = &direntplus->dirent;
1244 struct dentry *parent = file->f_path.dentry;
1245 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1246 struct dentry *dentry;
1247 struct dentry *alias;
1248 struct inode *dir = parent->d_inode;
1249 struct fuse_conn *fc;
1250 struct inode *inode;
1251
1252 if (!o->nodeid) {
1253 /*
1254 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1255 * ENOENT. Instead, it only means the userspace filesystem did
1256 * not want to return attributes/handle for this entry.
1257 *
1258 * So do nothing.
1259 */
1260 return 0;
1261 }
1262
1263 if (name.name[0] == '.') {
1264 /*
1265 * We could potentially refresh the attributes of the directory
1266 * and its parent?
1267 */
1268 if (name.len == 1)
1269 return 0;
1270 if (name.name[1] == '.' && name.len == 2)
1271 return 0;
1272 }
1273
1274 if (invalid_nodeid(o->nodeid))
1275 return -EIO;
1276 if (!fuse_valid_type(o->attr.mode))
1277 return -EIO;
1278
1279 fc = get_fuse_conn(dir);
1280
1281 name.hash = full_name_hash(name.name, name.len);
1282 dentry = d_lookup(parent, &name);
1283 if (dentry) {
1284 inode = dentry->d_inode;
1285 if (!inode) {
1286 d_drop(dentry);
1287 } else if (get_node_id(inode) != o->nodeid ||
1288 ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1289 d_invalidate(dentry);
1290 } else if (is_bad_inode(inode)) {
1291 err = -EIO;
1292 goto out;
1293 } else {
1294 struct fuse_inode *fi;
1295 fi = get_fuse_inode(inode);
1296 spin_lock(&fc->lock);
1297 fi->nlookup++;
1298 spin_unlock(&fc->lock);
1299
1300 fuse_change_attributes(inode, &o->attr,
1301 entry_attr_timeout(o),
1302 attr_version);
1303
1304 /*
1305 * The other branch to 'found' comes via fuse_iget()
1306 * which bumps nlookup inside
1307 */
1308 goto found;
1309 }
1310 dput(dentry);
1311 }
1312
1313 dentry = d_alloc(parent, &name);
1314 err = -ENOMEM;
1315 if (!dentry)
1316 goto out;
1317
1318 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1319 &o->attr, entry_attr_timeout(o), attr_version);
1320 if (!inode)
1321 goto out;
1322
1323 alias = d_materialise_unique(dentry, inode);
1324 err = PTR_ERR(alias);
1325 if (IS_ERR(alias))
1326 goto out;
1327
1328 if (alias) {
1329 dput(dentry);
1330 dentry = alias;
1331 }
1332
1333 found:
1334 if (fc->readdirplus_auto)
1335 set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1336 fuse_change_entry_timeout(dentry, o);
1337
1338 err = 0;
1339 out:
1340 dput(dentry);
1341 return err;
1342 }
1343
1344 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1345 struct dir_context *ctx, u64 attr_version)
1346 {
1347 struct fuse_direntplus *direntplus;
1348 struct fuse_dirent *dirent;
1349 size_t reclen;
1350 int over = 0;
1351 int ret;
1352
1353 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1354 direntplus = (struct fuse_direntplus *) buf;
1355 dirent = &direntplus->dirent;
1356 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1357
1358 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1359 return -EIO;
1360 if (reclen > nbytes)
1361 break;
1362 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1363 return -EIO;
1364
1365 if (!over) {
1366 /* We fill entries into dstbuf only as much as
1367 it can hold. But we still continue iterating
1368 over remaining entries to link them. If not,
1369 we need to send a FORGET for each of those
1370 which we did not link.
1371 */
1372 over = !dir_emit(ctx, dirent->name, dirent->namelen,
1373 dirent->ino, dirent->type);
1374 ctx->pos = dirent->off;
1375 }
1376
1377 buf += reclen;
1378 nbytes -= reclen;
1379
1380 ret = fuse_direntplus_link(file, direntplus, attr_version);
1381 if (ret)
1382 fuse_force_forget(file, direntplus->entry_out.nodeid);
1383 }
1384
1385 return 0;
1386 }
1387
1388 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1389 {
1390 int plus, err;
1391 size_t nbytes;
1392 struct page *page;
1393 struct inode *inode = file_inode(file);
1394 struct fuse_conn *fc = get_fuse_conn(inode);
1395 struct fuse_req *req;
1396 u64 attr_version = 0;
1397
1398 if (is_bad_inode(inode))
1399 return -EIO;
1400
1401 req = fuse_get_req(fc, 1);
1402 if (IS_ERR(req))
1403 return PTR_ERR(req);
1404
1405 page = alloc_page(GFP_KERNEL);
1406 if (!page) {
1407 fuse_put_request(fc, req);
1408 return -ENOMEM;
1409 }
1410
1411 plus = fuse_use_readdirplus(inode, ctx);
1412 req->out.argpages = 1;
1413 req->num_pages = 1;
1414 req->pages[0] = page;
1415 req->page_descs[0].length = PAGE_SIZE;
1416 if (plus) {
1417 attr_version = fuse_get_attr_version(fc);
1418 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1419 FUSE_READDIRPLUS);
1420 } else {
1421 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1422 FUSE_READDIR);
1423 }
1424 fuse_request_send(fc, req);
1425 nbytes = req->out.args[0].size;
1426 err = req->out.h.error;
1427 fuse_put_request(fc, req);
1428 if (!err) {
1429 if (plus) {
1430 err = parse_dirplusfile(page_address(page), nbytes,
1431 file, ctx,
1432 attr_version);
1433 } else {
1434 err = parse_dirfile(page_address(page), nbytes, file,
1435 ctx);
1436 }
1437 }
1438
1439 __free_page(page);
1440 fuse_invalidate_atime(inode);
1441 return err;
1442 }
1443
1444 static char *read_link(struct dentry *dentry)
1445 {
1446 struct inode *inode = dentry->d_inode;
1447 struct fuse_conn *fc = get_fuse_conn(inode);
1448 struct fuse_req *req = fuse_get_req_nopages(fc);
1449 char *link;
1450
1451 if (IS_ERR(req))
1452 return ERR_CAST(req);
1453
1454 link = (char *) __get_free_page(GFP_KERNEL);
1455 if (!link) {
1456 link = ERR_PTR(-ENOMEM);
1457 goto out;
1458 }
1459 req->in.h.opcode = FUSE_READLINK;
1460 req->in.h.nodeid = get_node_id(inode);
1461 req->out.argvar = 1;
1462 req->out.numargs = 1;
1463 req->out.args[0].size = PAGE_SIZE - 1;
1464 req->out.args[0].value = link;
1465 fuse_request_send(fc, req);
1466 if (req->out.h.error) {
1467 free_page((unsigned long) link);
1468 link = ERR_PTR(req->out.h.error);
1469 } else
1470 link[req->out.args[0].size] = '\0';
1471 out:
1472 fuse_put_request(fc, req);
1473 fuse_invalidate_atime(inode);
1474 return link;
1475 }
1476
1477 static void free_link(char *link)
1478 {
1479 if (!IS_ERR(link))
1480 free_page((unsigned long) link);
1481 }
1482
1483 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1484 {
1485 nd_set_link(nd, read_link(dentry));
1486 return NULL;
1487 }
1488
1489 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1490 {
1491 free_link(nd_get_link(nd));
1492 }
1493
1494 static int fuse_dir_open(struct inode *inode, struct file *file)
1495 {
1496 return fuse_open_common(inode, file, true);
1497 }
1498
1499 static int fuse_dir_release(struct inode *inode, struct file *file)
1500 {
1501 fuse_release_common(file, FUSE_RELEASEDIR);
1502
1503 return 0;
1504 }
1505
1506 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1507 int datasync)
1508 {
1509 return fuse_fsync_common(file, start, end, datasync, 1);
1510 }
1511
1512 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1513 unsigned long arg)
1514 {
1515 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1516
1517 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1518 if (fc->minor < 18)
1519 return -ENOTTY;
1520
1521 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1522 }
1523
1524 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1525 unsigned long arg)
1526 {
1527 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1528
1529 if (fc->minor < 18)
1530 return -ENOTTY;
1531
1532 return fuse_ioctl_common(file, cmd, arg,
1533 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1534 }
1535
1536 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1537 {
1538 /* Always update if mtime is explicitly set */
1539 if (ivalid & ATTR_MTIME_SET)
1540 return true;
1541
1542 /* Or if kernel i_mtime is the official one */
1543 if (trust_local_mtime)
1544 return true;
1545
1546 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1547 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1548 return false;
1549
1550 /* In all other cases update */
1551 return true;
1552 }
1553
1554 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
1555 bool trust_local_cmtime)
1556 {
1557 unsigned ivalid = iattr->ia_valid;
1558
1559 if (ivalid & ATTR_MODE)
1560 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1561 if (ivalid & ATTR_UID)
1562 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1563 if (ivalid & ATTR_GID)
1564 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1565 if (ivalid & ATTR_SIZE)
1566 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1567 if (ivalid & ATTR_ATIME) {
1568 arg->valid |= FATTR_ATIME;
1569 arg->atime = iattr->ia_atime.tv_sec;
1570 arg->atimensec = iattr->ia_atime.tv_nsec;
1571 if (!(ivalid & ATTR_ATIME_SET))
1572 arg->valid |= FATTR_ATIME_NOW;
1573 }
1574 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1575 arg->valid |= FATTR_MTIME;
1576 arg->mtime = iattr->ia_mtime.tv_sec;
1577 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1578 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1579 arg->valid |= FATTR_MTIME_NOW;
1580 }
1581 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1582 arg->valid |= FATTR_CTIME;
1583 arg->ctime = iattr->ia_ctime.tv_sec;
1584 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1585 }
1586 }
1587
1588 /*
1589 * Prevent concurrent writepages on inode
1590 *
1591 * This is done by adding a negative bias to the inode write counter
1592 * and waiting for all pending writes to finish.
1593 */
1594 void fuse_set_nowrite(struct inode *inode)
1595 {
1596 struct fuse_conn *fc = get_fuse_conn(inode);
1597 struct fuse_inode *fi = get_fuse_inode(inode);
1598
1599 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1600
1601 spin_lock(&fc->lock);
1602 BUG_ON(fi->writectr < 0);
1603 fi->writectr += FUSE_NOWRITE;
1604 spin_unlock(&fc->lock);
1605 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1606 }
1607
1608 /*
1609 * Allow writepages on inode
1610 *
1611 * Remove the bias from the writecounter and send any queued
1612 * writepages.
1613 */
1614 static void __fuse_release_nowrite(struct inode *inode)
1615 {
1616 struct fuse_inode *fi = get_fuse_inode(inode);
1617
1618 BUG_ON(fi->writectr != FUSE_NOWRITE);
1619 fi->writectr = 0;
1620 fuse_flush_writepages(inode);
1621 }
1622
1623 void fuse_release_nowrite(struct inode *inode)
1624 {
1625 struct fuse_conn *fc = get_fuse_conn(inode);
1626
1627 spin_lock(&fc->lock);
1628 __fuse_release_nowrite(inode);
1629 spin_unlock(&fc->lock);
1630 }
1631
1632 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_req *req,
1633 struct inode *inode,
1634 struct fuse_setattr_in *inarg_p,
1635 struct fuse_attr_out *outarg_p)
1636 {
1637 req->in.h.opcode = FUSE_SETATTR;
1638 req->in.h.nodeid = get_node_id(inode);
1639 req->in.numargs = 1;
1640 req->in.args[0].size = sizeof(*inarg_p);
1641 req->in.args[0].value = inarg_p;
1642 req->out.numargs = 1;
1643 if (fc->minor < 9)
1644 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1645 else
1646 req->out.args[0].size = sizeof(*outarg_p);
1647 req->out.args[0].value = outarg_p;
1648 }
1649
1650 /*
1651 * Flush inode->i_mtime to the server
1652 */
1653 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1654 {
1655 struct fuse_conn *fc = get_fuse_conn(inode);
1656 struct fuse_req *req;
1657 struct fuse_setattr_in inarg;
1658 struct fuse_attr_out outarg;
1659 int err;
1660
1661 req = fuse_get_req_nopages(fc);
1662 if (IS_ERR(req))
1663 return PTR_ERR(req);
1664
1665 memset(&inarg, 0, sizeof(inarg));
1666 memset(&outarg, 0, sizeof(outarg));
1667
1668 inarg.valid = FATTR_MTIME;
1669 inarg.mtime = inode->i_mtime.tv_sec;
1670 inarg.mtimensec = inode->i_mtime.tv_nsec;
1671 if (fc->minor >= 23) {
1672 inarg.valid |= FATTR_CTIME;
1673 inarg.ctime = inode->i_ctime.tv_sec;
1674 inarg.ctimensec = inode->i_ctime.tv_nsec;
1675 }
1676 if (ff) {
1677 inarg.valid |= FATTR_FH;
1678 inarg.fh = ff->fh;
1679 }
1680 fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
1681 fuse_request_send(fc, req);
1682 err = req->out.h.error;
1683 fuse_put_request(fc, req);
1684
1685 return err;
1686 }
1687
1688 /*
1689 * Set attributes, and at the same time refresh them.
1690 *
1691 * Truncation is slightly complicated, because the 'truncate' request
1692 * may fail, in which case we don't want to touch the mapping.
1693 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1694 * and the actual truncation by hand.
1695 */
1696 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
1697 struct file *file)
1698 {
1699 struct fuse_conn *fc = get_fuse_conn(inode);
1700 struct fuse_inode *fi = get_fuse_inode(inode);
1701 struct fuse_req *req;
1702 struct fuse_setattr_in inarg;
1703 struct fuse_attr_out outarg;
1704 bool is_truncate = false;
1705 bool is_wb = fc->writeback_cache;
1706 loff_t oldsize;
1707 int err;
1708 bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1709
1710 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1711 attr->ia_valid |= ATTR_FORCE;
1712
1713 err = inode_change_ok(inode, attr);
1714 if (err)
1715 return err;
1716
1717 if (attr->ia_valid & ATTR_OPEN) {
1718 if (fc->atomic_o_trunc)
1719 return 0;
1720 file = NULL;
1721 }
1722
1723 if (attr->ia_valid & ATTR_SIZE)
1724 is_truncate = true;
1725
1726 req = fuse_get_req_nopages(fc);
1727 if (IS_ERR(req))
1728 return PTR_ERR(req);
1729
1730 if (is_truncate) {
1731 fuse_set_nowrite(inode);
1732 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1733 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1734 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1735 }
1736
1737 memset(&inarg, 0, sizeof(inarg));
1738 memset(&outarg, 0, sizeof(outarg));
1739 iattr_to_fattr(attr, &inarg, trust_local_cmtime);
1740 if (file) {
1741 struct fuse_file *ff = file->private_data;
1742 inarg.valid |= FATTR_FH;
1743 inarg.fh = ff->fh;
1744 }
1745 if (attr->ia_valid & ATTR_SIZE) {
1746 /* For mandatory locking in truncate */
1747 inarg.valid |= FATTR_LOCKOWNER;
1748 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1749 }
1750 fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
1751 fuse_request_send(fc, req);
1752 err = req->out.h.error;
1753 fuse_put_request(fc, req);
1754 if (err) {
1755 if (err == -EINTR)
1756 fuse_invalidate_attr(inode);
1757 goto error;
1758 }
1759
1760 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1761 make_bad_inode(inode);
1762 err = -EIO;
1763 goto error;
1764 }
1765
1766 spin_lock(&fc->lock);
1767 /* the kernel maintains i_mtime locally */
1768 if (trust_local_cmtime) {
1769 if (attr->ia_valid & ATTR_MTIME)
1770 inode->i_mtime = attr->ia_mtime;
1771 if (attr->ia_valid & ATTR_CTIME)
1772 inode->i_ctime = attr->ia_ctime;
1773 /* FIXME: clear I_DIRTY_SYNC? */
1774 }
1775
1776 fuse_change_attributes_common(inode, &outarg.attr,
1777 attr_timeout(&outarg));
1778 oldsize = inode->i_size;
1779 /* see the comment in fuse_change_attributes() */
1780 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1781 i_size_write(inode, outarg.attr.size);
1782
1783 if (is_truncate) {
1784 /* NOTE: this may release/reacquire fc->lock */
1785 __fuse_release_nowrite(inode);
1786 }
1787 spin_unlock(&fc->lock);
1788
1789 /*
1790 * Only call invalidate_inode_pages2() after removing
1791 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1792 */
1793 if ((is_truncate || !is_wb) &&
1794 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1795 truncate_pagecache(inode, outarg.attr.size);
1796 invalidate_inode_pages2(inode->i_mapping);
1797 }
1798
1799 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1800 return 0;
1801
1802 error:
1803 if (is_truncate)
1804 fuse_release_nowrite(inode);
1805
1806 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1807 return err;
1808 }
1809
1810 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1811 {
1812 struct inode *inode = entry->d_inode;
1813
1814 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1815 return -EACCES;
1816
1817 if (attr->ia_valid & ATTR_FILE)
1818 return fuse_do_setattr(inode, attr, attr->ia_file);
1819 else
1820 return fuse_do_setattr(inode, attr, NULL);
1821 }
1822
1823 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1824 struct kstat *stat)
1825 {
1826 struct inode *inode = entry->d_inode;
1827 struct fuse_conn *fc = get_fuse_conn(inode);
1828
1829 if (!fuse_allow_current_process(fc))
1830 return -EACCES;
1831
1832 return fuse_update_attributes(inode, stat, NULL, NULL);
1833 }
1834
1835 static int fuse_setxattr(struct dentry *entry, const char *name,
1836 const void *value, size_t size, int flags)
1837 {
1838 struct inode *inode = entry->d_inode;
1839 struct fuse_conn *fc = get_fuse_conn(inode);
1840 struct fuse_req *req;
1841 struct fuse_setxattr_in inarg;
1842 int err;
1843
1844 if (fc->no_setxattr)
1845 return -EOPNOTSUPP;
1846
1847 req = fuse_get_req_nopages(fc);
1848 if (IS_ERR(req))
1849 return PTR_ERR(req);
1850
1851 memset(&inarg, 0, sizeof(inarg));
1852 inarg.size = size;
1853 inarg.flags = flags;
1854 req->in.h.opcode = FUSE_SETXATTR;
1855 req->in.h.nodeid = get_node_id(inode);
1856 req->in.numargs = 3;
1857 req->in.args[0].size = sizeof(inarg);
1858 req->in.args[0].value = &inarg;
1859 req->in.args[1].size = strlen(name) + 1;
1860 req->in.args[1].value = name;
1861 req->in.args[2].size = size;
1862 req->in.args[2].value = value;
1863 fuse_request_send(fc, req);
1864 err = req->out.h.error;
1865 fuse_put_request(fc, req);
1866 if (err == -ENOSYS) {
1867 fc->no_setxattr = 1;
1868 err = -EOPNOTSUPP;
1869 }
1870 if (!err) {
1871 fuse_invalidate_attr(inode);
1872 fuse_update_ctime(inode);
1873 }
1874 return err;
1875 }
1876
1877 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1878 void *value, size_t size)
1879 {
1880 struct inode *inode = entry->d_inode;
1881 struct fuse_conn *fc = get_fuse_conn(inode);
1882 struct fuse_req *req;
1883 struct fuse_getxattr_in inarg;
1884 struct fuse_getxattr_out outarg;
1885 ssize_t ret;
1886
1887 if (fc->no_getxattr)
1888 return -EOPNOTSUPP;
1889
1890 req = fuse_get_req_nopages(fc);
1891 if (IS_ERR(req))
1892 return PTR_ERR(req);
1893
1894 memset(&inarg, 0, sizeof(inarg));
1895 inarg.size = size;
1896 req->in.h.opcode = FUSE_GETXATTR;
1897 req->in.h.nodeid = get_node_id(inode);
1898 req->in.numargs = 2;
1899 req->in.args[0].size = sizeof(inarg);
1900 req->in.args[0].value = &inarg;
1901 req->in.args[1].size = strlen(name) + 1;
1902 req->in.args[1].value = name;
1903 /* This is really two different operations rolled into one */
1904 req->out.numargs = 1;
1905 if (size) {
1906 req->out.argvar = 1;
1907 req->out.args[0].size = size;
1908 req->out.args[0].value = value;
1909 } else {
1910 req->out.args[0].size = sizeof(outarg);
1911 req->out.args[0].value = &outarg;
1912 }
1913 fuse_request_send(fc, req);
1914 ret = req->out.h.error;
1915 if (!ret)
1916 ret = size ? req->out.args[0].size : outarg.size;
1917 else {
1918 if (ret == -ENOSYS) {
1919 fc->no_getxattr = 1;
1920 ret = -EOPNOTSUPP;
1921 }
1922 }
1923 fuse_put_request(fc, req);
1924 return ret;
1925 }
1926
1927 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1928 {
1929 struct inode *inode = entry->d_inode;
1930 struct fuse_conn *fc = get_fuse_conn(inode);
1931 struct fuse_req *req;
1932 struct fuse_getxattr_in inarg;
1933 struct fuse_getxattr_out outarg;
1934 ssize_t ret;
1935
1936 if (!fuse_allow_current_process(fc))
1937 return -EACCES;
1938
1939 if (fc->no_listxattr)
1940 return -EOPNOTSUPP;
1941
1942 req = fuse_get_req_nopages(fc);
1943 if (IS_ERR(req))
1944 return PTR_ERR(req);
1945
1946 memset(&inarg, 0, sizeof(inarg));
1947 inarg.size = size;
1948 req->in.h.opcode = FUSE_LISTXATTR;
1949 req->in.h.nodeid = get_node_id(inode);
1950 req->in.numargs = 1;
1951 req->in.args[0].size = sizeof(inarg);
1952 req->in.args[0].value = &inarg;
1953 /* This is really two different operations rolled into one */
1954 req->out.numargs = 1;
1955 if (size) {
1956 req->out.argvar = 1;
1957 req->out.args[0].size = size;
1958 req->out.args[0].value = list;
1959 } else {
1960 req->out.args[0].size = sizeof(outarg);
1961 req->out.args[0].value = &outarg;
1962 }
1963 fuse_request_send(fc, req);
1964 ret = req->out.h.error;
1965 if (!ret)
1966 ret = size ? req->out.args[0].size : outarg.size;
1967 else {
1968 if (ret == -ENOSYS) {
1969 fc->no_listxattr = 1;
1970 ret = -EOPNOTSUPP;
1971 }
1972 }
1973 fuse_put_request(fc, req);
1974 return ret;
1975 }
1976
1977 static int fuse_removexattr(struct dentry *entry, const char *name)
1978 {
1979 struct inode *inode = entry->d_inode;
1980 struct fuse_conn *fc = get_fuse_conn(inode);
1981 struct fuse_req *req;
1982 int err;
1983
1984 if (fc->no_removexattr)
1985 return -EOPNOTSUPP;
1986
1987 req = fuse_get_req_nopages(fc);
1988 if (IS_ERR(req))
1989 return PTR_ERR(req);
1990
1991 req->in.h.opcode = FUSE_REMOVEXATTR;
1992 req->in.h.nodeid = get_node_id(inode);
1993 req->in.numargs = 1;
1994 req->in.args[0].size = strlen(name) + 1;
1995 req->in.args[0].value = name;
1996 fuse_request_send(fc, req);
1997 err = req->out.h.error;
1998 fuse_put_request(fc, req);
1999 if (err == -ENOSYS) {
2000 fc->no_removexattr = 1;
2001 err = -EOPNOTSUPP;
2002 }
2003 if (!err) {
2004 fuse_invalidate_attr(inode);
2005 fuse_update_ctime(inode);
2006 }
2007 return err;
2008 }
2009
2010 static const struct inode_operations fuse_dir_inode_operations = {
2011 .lookup = fuse_lookup,
2012 .mkdir = fuse_mkdir,
2013 .symlink = fuse_symlink,
2014 .unlink = fuse_unlink,
2015 .rmdir = fuse_rmdir,
2016 .rename2 = fuse_rename2,
2017 .link = fuse_link,
2018 .setattr = fuse_setattr,
2019 .create = fuse_create,
2020 .atomic_open = fuse_atomic_open,
2021 .mknod = fuse_mknod,
2022 .permission = fuse_permission,
2023 .getattr = fuse_getattr,
2024 .setxattr = fuse_setxattr,
2025 .getxattr = fuse_getxattr,
2026 .listxattr = fuse_listxattr,
2027 .removexattr = fuse_removexattr,
2028 };
2029
2030 static const struct file_operations fuse_dir_operations = {
2031 .llseek = generic_file_llseek,
2032 .read = generic_read_dir,
2033 .iterate = fuse_readdir,
2034 .open = fuse_dir_open,
2035 .release = fuse_dir_release,
2036 .fsync = fuse_dir_fsync,
2037 .unlocked_ioctl = fuse_dir_ioctl,
2038 .compat_ioctl = fuse_dir_compat_ioctl,
2039 };
2040
2041 static const struct inode_operations fuse_common_inode_operations = {
2042 .setattr = fuse_setattr,
2043 .permission = fuse_permission,
2044 .getattr = fuse_getattr,
2045 .setxattr = fuse_setxattr,
2046 .getxattr = fuse_getxattr,
2047 .listxattr = fuse_listxattr,
2048 .removexattr = fuse_removexattr,
2049 };
2050
2051 static const struct inode_operations fuse_symlink_inode_operations = {
2052 .setattr = fuse_setattr,
2053 .follow_link = fuse_follow_link,
2054 .put_link = fuse_put_link,
2055 .readlink = generic_readlink,
2056 .getattr = fuse_getattr,
2057 .setxattr = fuse_setxattr,
2058 .getxattr = fuse_getxattr,
2059 .listxattr = fuse_listxattr,
2060 .removexattr = fuse_removexattr,
2061 };
2062
2063 void fuse_init_common(struct inode *inode)
2064 {
2065 inode->i_op = &fuse_common_inode_operations;
2066 }
2067
2068 void fuse_init_dir(struct inode *inode)
2069 {
2070 inode->i_op = &fuse_dir_inode_operations;
2071 inode->i_fop = &fuse_dir_operations;
2072 }
2073
2074 void fuse_init_symlink(struct inode *inode)
2075 {
2076 inode->i_op = &fuse_symlink_inode_operations;
2077 }
This page took 0.072455 seconds and 5 git commands to generate.