selftests: ftrace: Introduce TMPDIR for temporary files
[linux-stable.git] / fs / fuse / dir.c
blob6a4d0e5418a179def0e5d4da4d3adbbcf3a82cd0
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
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 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
19 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
21 struct fuse_conn *fc = get_fuse_conn(dir);
22 struct fuse_inode *fi = get_fuse_inode(dir);
24 if (!fc->do_readdirplus)
25 return false;
26 if (!fc->readdirplus_auto)
27 return true;
28 if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
29 return true;
30 if (ctx->pos == 0)
31 return true;
32 return false;
35 static void fuse_advise_use_readdirplus(struct inode *dir)
37 struct fuse_inode *fi = get_fuse_inode(dir);
39 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
42 union fuse_dentry {
43 u64 time;
44 struct rcu_head rcu;
47 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
49 ((union fuse_dentry *) entry->d_fsdata)->time = time;
52 static inline u64 fuse_dentry_time(struct dentry *entry)
54 return ((union fuse_dentry *) entry->d_fsdata)->time;
58 * FUSE caches dentries and attributes with separate timeout. The
59 * time in jiffies until the dentry/attributes are valid is stored in
60 * dentry->d_fsdata and fuse_inode->i_time respectively.
64 * Calculate the time in jiffies until a dentry/attributes are valid
66 static u64 time_to_jiffies(u64 sec, u32 nsec)
68 if (sec || nsec) {
69 struct timespec64 ts = {
70 sec,
71 max_t(u32, nsec, NSEC_PER_SEC - 1)
74 return get_jiffies_64() + timespec64_to_jiffies(&ts);
75 } else
76 return 0;
80 * Set dentry and possibly attribute timeouts from the lookup/mk*
81 * replies
83 static void fuse_change_entry_timeout(struct dentry *entry,
84 struct fuse_entry_out *o)
86 fuse_dentry_settime(entry,
87 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
90 static u64 attr_timeout(struct fuse_attr_out *o)
92 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
95 static u64 entry_attr_timeout(struct fuse_entry_out *o)
97 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
101 * Mark the attributes as stale, so that at the next call to
102 * ->getattr() they will be fetched from userspace
104 void fuse_invalidate_attr(struct inode *inode)
106 get_fuse_inode(inode)->i_time = 0;
110 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
111 * atime is not used.
113 void fuse_invalidate_atime(struct inode *inode)
115 if (!IS_RDONLY(inode))
116 fuse_invalidate_attr(inode);
120 * Just mark the entry as stale, so that a next attempt to look it up
121 * will result in a new lookup call to userspace
123 * This is called when a dentry is about to become negative and the
124 * timeout is unknown (unlink, rmdir, rename and in some cases
125 * lookup)
127 void fuse_invalidate_entry_cache(struct dentry *entry)
129 fuse_dentry_settime(entry, 0);
133 * Same as fuse_invalidate_entry_cache(), but also try to remove the
134 * dentry from the hash
136 static void fuse_invalidate_entry(struct dentry *entry)
138 d_invalidate(entry);
139 fuse_invalidate_entry_cache(entry);
142 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
143 u64 nodeid, const struct qstr *name,
144 struct fuse_entry_out *outarg)
146 memset(outarg, 0, sizeof(struct fuse_entry_out));
147 args->in.h.opcode = FUSE_LOOKUP;
148 args->in.h.nodeid = nodeid;
149 args->in.numargs = 1;
150 args->in.args[0].size = name->len + 1;
151 args->in.args[0].value = name->name;
152 args->out.numargs = 1;
153 args->out.args[0].size = sizeof(struct fuse_entry_out);
154 args->out.args[0].value = outarg;
157 u64 fuse_get_attr_version(struct fuse_conn *fc)
159 u64 curr_version;
162 * The spin lock isn't actually needed on 64bit archs, but we
163 * don't yet care too much about such optimizations.
165 spin_lock(&fc->lock);
166 curr_version = fc->attr_version;
167 spin_unlock(&fc->lock);
169 return curr_version;
173 * Check whether the dentry is still valid
175 * If the entry validity timeout has expired and the dentry is
176 * positive, try to redo the lookup. If the lookup results in a
177 * different inode, then let the VFS invalidate the dentry and redo
178 * the lookup once more. If the lookup results in the same inode,
179 * then refresh the attributes, timeouts and mark the dentry valid.
181 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
183 struct inode *inode;
184 struct dentry *parent;
185 struct fuse_conn *fc;
186 struct fuse_inode *fi;
187 int ret;
189 inode = d_inode_rcu(entry);
190 if (inode && is_bad_inode(inode))
191 goto invalid;
192 else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
193 (flags & LOOKUP_REVAL)) {
194 struct fuse_entry_out outarg;
195 FUSE_ARGS(args);
196 struct fuse_forget_link *forget;
197 u64 attr_version;
199 /* For negative dentries, always do a fresh lookup */
200 if (!inode)
201 goto invalid;
203 ret = -ECHILD;
204 if (flags & LOOKUP_RCU)
205 goto out;
207 fc = get_fuse_conn(inode);
209 forget = fuse_alloc_forget();
210 ret = -ENOMEM;
211 if (!forget)
212 goto out;
214 attr_version = fuse_get_attr_version(fc);
216 parent = dget_parent(entry);
217 fuse_lookup_init(fc, &args, get_node_id(d_inode(parent)),
218 &entry->d_name, &outarg);
219 ret = fuse_simple_request(fc, &args);
220 dput(parent);
221 /* Zero nodeid is same as -ENOENT */
222 if (!ret && !outarg.nodeid)
223 ret = -ENOENT;
224 if (!ret) {
225 fi = get_fuse_inode(inode);
226 if (outarg.nodeid != get_node_id(inode)) {
227 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
228 goto invalid;
230 spin_lock(&fc->lock);
231 fi->nlookup++;
232 spin_unlock(&fc->lock);
234 kfree(forget);
235 if (ret == -ENOMEM)
236 goto out;
237 if (ret || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
238 goto invalid;
240 forget_all_cached_acls(inode);
241 fuse_change_attributes(inode, &outarg.attr,
242 entry_attr_timeout(&outarg),
243 attr_version);
244 fuse_change_entry_timeout(entry, &outarg);
245 } else if (inode) {
246 fi = get_fuse_inode(inode);
247 if (flags & LOOKUP_RCU) {
248 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
249 return -ECHILD;
250 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
251 parent = dget_parent(entry);
252 fuse_advise_use_readdirplus(d_inode(parent));
253 dput(parent);
256 ret = 1;
257 out:
258 return ret;
260 invalid:
261 ret = 0;
262 goto out;
265 static int invalid_nodeid(u64 nodeid)
267 return !nodeid || nodeid == FUSE_ROOT_ID;
270 static int fuse_dentry_init(struct dentry *dentry)
272 dentry->d_fsdata = kzalloc(sizeof(union fuse_dentry), GFP_KERNEL);
274 return dentry->d_fsdata ? 0 : -ENOMEM;
276 static void fuse_dentry_release(struct dentry *dentry)
278 union fuse_dentry *fd = dentry->d_fsdata;
280 kfree_rcu(fd, rcu);
283 const struct dentry_operations fuse_dentry_operations = {
284 .d_revalidate = fuse_dentry_revalidate,
285 .d_init = fuse_dentry_init,
286 .d_release = fuse_dentry_release,
289 int fuse_valid_type(int m)
291 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
292 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
295 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
296 struct fuse_entry_out *outarg, struct inode **inode)
298 struct fuse_conn *fc = get_fuse_conn_super(sb);
299 FUSE_ARGS(args);
300 struct fuse_forget_link *forget;
301 u64 attr_version;
302 int err;
304 *inode = NULL;
305 err = -ENAMETOOLONG;
306 if (name->len > FUSE_NAME_MAX)
307 goto out;
310 forget = fuse_alloc_forget();
311 err = -ENOMEM;
312 if (!forget)
313 goto out;
315 attr_version = fuse_get_attr_version(fc);
317 fuse_lookup_init(fc, &args, nodeid, name, outarg);
318 err = fuse_simple_request(fc, &args);
319 /* Zero nodeid is same as -ENOENT, but with valid timeout */
320 if (err || !outarg->nodeid)
321 goto out_put_forget;
323 err = -EIO;
324 if (!outarg->nodeid)
325 goto out_put_forget;
326 if (!fuse_valid_type(outarg->attr.mode))
327 goto out_put_forget;
329 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
330 &outarg->attr, entry_attr_timeout(outarg),
331 attr_version);
332 err = -ENOMEM;
333 if (!*inode) {
334 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
335 goto out;
337 err = 0;
339 out_put_forget:
340 kfree(forget);
341 out:
342 return err;
345 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
346 unsigned int flags)
348 int err;
349 struct fuse_entry_out outarg;
350 struct inode *inode;
351 struct dentry *newent;
352 bool outarg_valid = true;
354 fuse_lock_inode(dir);
355 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
356 &outarg, &inode);
357 fuse_unlock_inode(dir);
358 if (err == -ENOENT) {
359 outarg_valid = false;
360 err = 0;
362 if (err)
363 goto out_err;
365 err = -EIO;
366 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
367 goto out_iput;
369 newent = d_splice_alias(inode, entry);
370 err = PTR_ERR(newent);
371 if (IS_ERR(newent))
372 goto out_err;
374 entry = newent ? newent : entry;
375 if (outarg_valid)
376 fuse_change_entry_timeout(entry, &outarg);
377 else
378 fuse_invalidate_entry_cache(entry);
380 fuse_advise_use_readdirplus(dir);
381 return newent;
383 out_iput:
384 iput(inode);
385 out_err:
386 return ERR_PTR(err);
390 * Atomic create+open operation
392 * If the filesystem doesn't support this, then fall back to separate
393 * 'mknod' + 'open' requests.
395 static int fuse_create_open(struct inode *dir, struct dentry *entry,
396 struct file *file, unsigned flags,
397 umode_t mode, int *opened)
399 int err;
400 struct inode *inode;
401 struct fuse_conn *fc = get_fuse_conn(dir);
402 FUSE_ARGS(args);
403 struct fuse_forget_link *forget;
404 struct fuse_create_in inarg;
405 struct fuse_open_out outopen;
406 struct fuse_entry_out outentry;
407 struct fuse_file *ff;
409 /* Userspace expects S_IFREG in create mode */
410 BUG_ON((mode & S_IFMT) != S_IFREG);
412 forget = fuse_alloc_forget();
413 err = -ENOMEM;
414 if (!forget)
415 goto out_err;
417 err = -ENOMEM;
418 ff = fuse_file_alloc(fc);
419 if (!ff)
420 goto out_put_forget_req;
422 if (!fc->dont_mask)
423 mode &= ~current_umask();
425 flags &= ~O_NOCTTY;
426 memset(&inarg, 0, sizeof(inarg));
427 memset(&outentry, 0, sizeof(outentry));
428 inarg.flags = flags;
429 inarg.mode = mode;
430 inarg.umask = current_umask();
431 args.in.h.opcode = FUSE_CREATE;
432 args.in.h.nodeid = get_node_id(dir);
433 args.in.numargs = 2;
434 args.in.args[0].size = sizeof(inarg);
435 args.in.args[0].value = &inarg;
436 args.in.args[1].size = entry->d_name.len + 1;
437 args.in.args[1].value = entry->d_name.name;
438 args.out.numargs = 2;
439 args.out.args[0].size = sizeof(outentry);
440 args.out.args[0].value = &outentry;
441 args.out.args[1].size = sizeof(outopen);
442 args.out.args[1].value = &outopen;
443 err = fuse_simple_request(fc, &args);
444 if (err)
445 goto out_free_ff;
447 err = -EIO;
448 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
449 goto out_free_ff;
451 ff->fh = outopen.fh;
452 ff->nodeid = outentry.nodeid;
453 ff->open_flags = outopen.open_flags;
454 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
455 &outentry.attr, entry_attr_timeout(&outentry), 0);
456 if (!inode) {
457 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
458 fuse_sync_release(ff, flags);
459 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
460 err = -ENOMEM;
461 goto out_err;
463 kfree(forget);
464 d_instantiate(entry, inode);
465 fuse_change_entry_timeout(entry, &outentry);
466 fuse_invalidate_attr(dir);
467 err = finish_open(file, entry, generic_file_open, opened);
468 if (err) {
469 fuse_sync_release(ff, flags);
470 } else {
471 file->private_data = fuse_file_get(ff);
472 fuse_finish_open(inode, file);
474 return err;
476 out_free_ff:
477 fuse_file_free(ff);
478 out_put_forget_req:
479 kfree(forget);
480 out_err:
481 return err;
484 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
485 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
486 struct file *file, unsigned flags,
487 umode_t mode, int *opened)
489 int err;
490 struct fuse_conn *fc = get_fuse_conn(dir);
491 struct dentry *res = NULL;
493 if (d_in_lookup(entry)) {
494 res = fuse_lookup(dir, entry, 0);
495 if (IS_ERR(res))
496 return PTR_ERR(res);
498 if (res)
499 entry = res;
502 if (!(flags & O_CREAT) || d_really_is_positive(entry))
503 goto no_open;
505 /* Only creates */
506 *opened |= FILE_CREATED;
508 if (fc->no_create)
509 goto mknod;
511 err = fuse_create_open(dir, entry, file, flags, mode, opened);
512 if (err == -ENOSYS) {
513 fc->no_create = 1;
514 goto mknod;
516 out_dput:
517 dput(res);
518 return err;
520 mknod:
521 err = fuse_mknod(dir, entry, mode, 0);
522 if (err)
523 goto out_dput;
524 no_open:
525 return finish_no_open(file, res);
529 * Code shared between mknod, mkdir, symlink and link
531 static int create_new_entry(struct fuse_conn *fc, struct fuse_args *args,
532 struct inode *dir, struct dentry *entry,
533 umode_t mode)
535 struct fuse_entry_out outarg;
536 struct inode *inode;
537 int err;
538 struct fuse_forget_link *forget;
540 forget = fuse_alloc_forget();
541 if (!forget)
542 return -ENOMEM;
544 memset(&outarg, 0, sizeof(outarg));
545 args->in.h.nodeid = get_node_id(dir);
546 args->out.numargs = 1;
547 args->out.args[0].size = sizeof(outarg);
548 args->out.args[0].value = &outarg;
549 err = fuse_simple_request(fc, args);
550 if (err)
551 goto out_put_forget_req;
553 err = -EIO;
554 if (invalid_nodeid(outarg.nodeid))
555 goto out_put_forget_req;
557 if ((outarg.attr.mode ^ mode) & S_IFMT)
558 goto out_put_forget_req;
560 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
561 &outarg.attr, entry_attr_timeout(&outarg), 0);
562 if (!inode) {
563 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
564 return -ENOMEM;
566 kfree(forget);
568 err = d_instantiate_no_diralias(entry, inode);
569 if (err)
570 return err;
572 fuse_change_entry_timeout(entry, &outarg);
573 fuse_invalidate_attr(dir);
574 return 0;
576 out_put_forget_req:
577 kfree(forget);
578 return err;
581 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
582 dev_t rdev)
584 struct fuse_mknod_in inarg;
585 struct fuse_conn *fc = get_fuse_conn(dir);
586 FUSE_ARGS(args);
588 if (!fc->dont_mask)
589 mode &= ~current_umask();
591 memset(&inarg, 0, sizeof(inarg));
592 inarg.mode = mode;
593 inarg.rdev = new_encode_dev(rdev);
594 inarg.umask = current_umask();
595 args.in.h.opcode = FUSE_MKNOD;
596 args.in.numargs = 2;
597 args.in.args[0].size = sizeof(inarg);
598 args.in.args[0].value = &inarg;
599 args.in.args[1].size = entry->d_name.len + 1;
600 args.in.args[1].value = entry->d_name.name;
601 return create_new_entry(fc, &args, dir, entry, mode);
604 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
605 bool excl)
607 return fuse_mknod(dir, entry, mode, 0);
610 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
612 struct fuse_mkdir_in inarg;
613 struct fuse_conn *fc = get_fuse_conn(dir);
614 FUSE_ARGS(args);
616 if (!fc->dont_mask)
617 mode &= ~current_umask();
619 memset(&inarg, 0, sizeof(inarg));
620 inarg.mode = mode;
621 inarg.umask = current_umask();
622 args.in.h.opcode = FUSE_MKDIR;
623 args.in.numargs = 2;
624 args.in.args[0].size = sizeof(inarg);
625 args.in.args[0].value = &inarg;
626 args.in.args[1].size = entry->d_name.len + 1;
627 args.in.args[1].value = entry->d_name.name;
628 return create_new_entry(fc, &args, dir, entry, S_IFDIR);
631 static int fuse_symlink(struct inode *dir, struct dentry *entry,
632 const char *link)
634 struct fuse_conn *fc = get_fuse_conn(dir);
635 unsigned len = strlen(link) + 1;
636 FUSE_ARGS(args);
638 args.in.h.opcode = FUSE_SYMLINK;
639 args.in.numargs = 2;
640 args.in.args[0].size = entry->d_name.len + 1;
641 args.in.args[0].value = entry->d_name.name;
642 args.in.args[1].size = len;
643 args.in.args[1].value = link;
644 return create_new_entry(fc, &args, dir, entry, S_IFLNK);
647 void fuse_update_ctime(struct inode *inode)
649 if (!IS_NOCMTIME(inode)) {
650 inode->i_ctime = current_time(inode);
651 mark_inode_dirty_sync(inode);
655 static int fuse_unlink(struct inode *dir, struct dentry *entry)
657 int err;
658 struct fuse_conn *fc = get_fuse_conn(dir);
659 FUSE_ARGS(args);
661 args.in.h.opcode = FUSE_UNLINK;
662 args.in.h.nodeid = get_node_id(dir);
663 args.in.numargs = 1;
664 args.in.args[0].size = entry->d_name.len + 1;
665 args.in.args[0].value = entry->d_name.name;
666 err = fuse_simple_request(fc, &args);
667 if (!err) {
668 struct inode *inode = d_inode(entry);
669 struct fuse_inode *fi = get_fuse_inode(inode);
671 spin_lock(&fc->lock);
672 fi->attr_version = ++fc->attr_version;
674 * If i_nlink == 0 then unlink doesn't make sense, yet this can
675 * happen if userspace filesystem is careless. It would be
676 * difficult to enforce correct nlink usage so just ignore this
677 * condition here
679 if (inode->i_nlink > 0)
680 drop_nlink(inode);
681 spin_unlock(&fc->lock);
682 fuse_invalidate_attr(inode);
683 fuse_invalidate_attr(dir);
684 fuse_invalidate_entry_cache(entry);
685 fuse_update_ctime(inode);
686 } else if (err == -EINTR)
687 fuse_invalidate_entry(entry);
688 return err;
691 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
693 int err;
694 struct fuse_conn *fc = get_fuse_conn(dir);
695 FUSE_ARGS(args);
697 args.in.h.opcode = FUSE_RMDIR;
698 args.in.h.nodeid = get_node_id(dir);
699 args.in.numargs = 1;
700 args.in.args[0].size = entry->d_name.len + 1;
701 args.in.args[0].value = entry->d_name.name;
702 err = fuse_simple_request(fc, &args);
703 if (!err) {
704 clear_nlink(d_inode(entry));
705 fuse_invalidate_attr(dir);
706 fuse_invalidate_entry_cache(entry);
707 } else if (err == -EINTR)
708 fuse_invalidate_entry(entry);
709 return err;
712 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
713 struct inode *newdir, struct dentry *newent,
714 unsigned int flags, int opcode, size_t argsize)
716 int err;
717 struct fuse_rename2_in inarg;
718 struct fuse_conn *fc = get_fuse_conn(olddir);
719 FUSE_ARGS(args);
721 memset(&inarg, 0, argsize);
722 inarg.newdir = get_node_id(newdir);
723 inarg.flags = flags;
724 args.in.h.opcode = opcode;
725 args.in.h.nodeid = get_node_id(olddir);
726 args.in.numargs = 3;
727 args.in.args[0].size = argsize;
728 args.in.args[0].value = &inarg;
729 args.in.args[1].size = oldent->d_name.len + 1;
730 args.in.args[1].value = oldent->d_name.name;
731 args.in.args[2].size = newent->d_name.len + 1;
732 args.in.args[2].value = newent->d_name.name;
733 err = fuse_simple_request(fc, &args);
734 if (!err) {
735 /* ctime changes */
736 fuse_invalidate_attr(d_inode(oldent));
737 fuse_update_ctime(d_inode(oldent));
739 if (flags & RENAME_EXCHANGE) {
740 fuse_invalidate_attr(d_inode(newent));
741 fuse_update_ctime(d_inode(newent));
744 fuse_invalidate_attr(olddir);
745 if (olddir != newdir)
746 fuse_invalidate_attr(newdir);
748 /* newent will end up negative */
749 if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
750 fuse_invalidate_attr(d_inode(newent));
751 fuse_invalidate_entry_cache(newent);
752 fuse_update_ctime(d_inode(newent));
754 } else if (err == -EINTR) {
755 /* If request was interrupted, DEITY only knows if the
756 rename actually took place. If the invalidation
757 fails (e.g. some process has CWD under the renamed
758 directory), then there can be inconsistency between
759 the dcache and the real filesystem. Tough luck. */
760 fuse_invalidate_entry(oldent);
761 if (d_really_is_positive(newent))
762 fuse_invalidate_entry(newent);
765 return err;
768 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
769 struct inode *newdir, struct dentry *newent,
770 unsigned int flags)
772 struct fuse_conn *fc = get_fuse_conn(olddir);
773 int err;
775 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
776 return -EINVAL;
778 if (flags) {
779 if (fc->no_rename2 || fc->minor < 23)
780 return -EINVAL;
782 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
783 FUSE_RENAME2,
784 sizeof(struct fuse_rename2_in));
785 if (err == -ENOSYS) {
786 fc->no_rename2 = 1;
787 err = -EINVAL;
789 } else {
790 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
791 FUSE_RENAME,
792 sizeof(struct fuse_rename_in));
795 return err;
798 static int fuse_link(struct dentry *entry, struct inode *newdir,
799 struct dentry *newent)
801 int err;
802 struct fuse_link_in inarg;
803 struct inode *inode = d_inode(entry);
804 struct fuse_conn *fc = get_fuse_conn(inode);
805 FUSE_ARGS(args);
807 memset(&inarg, 0, sizeof(inarg));
808 inarg.oldnodeid = get_node_id(inode);
809 args.in.h.opcode = FUSE_LINK;
810 args.in.numargs = 2;
811 args.in.args[0].size = sizeof(inarg);
812 args.in.args[0].value = &inarg;
813 args.in.args[1].size = newent->d_name.len + 1;
814 args.in.args[1].value = newent->d_name.name;
815 err = create_new_entry(fc, &args, newdir, newent, inode->i_mode);
816 /* Contrary to "normal" filesystems it can happen that link
817 makes two "logical" inodes point to the same "physical"
818 inode. We invalidate the attributes of the old one, so it
819 will reflect changes in the backing inode (link count,
820 etc.)
822 if (!err) {
823 struct fuse_inode *fi = get_fuse_inode(inode);
825 spin_lock(&fc->lock);
826 fi->attr_version = ++fc->attr_version;
827 inc_nlink(inode);
828 spin_unlock(&fc->lock);
829 fuse_invalidate_attr(inode);
830 fuse_update_ctime(inode);
831 } else if (err == -EINTR) {
832 fuse_invalidate_attr(inode);
834 return err;
837 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
838 struct kstat *stat)
840 unsigned int blkbits;
841 struct fuse_conn *fc = get_fuse_conn(inode);
843 /* see the comment in fuse_change_attributes() */
844 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
845 attr->size = i_size_read(inode);
846 attr->mtime = inode->i_mtime.tv_sec;
847 attr->mtimensec = inode->i_mtime.tv_nsec;
848 attr->ctime = inode->i_ctime.tv_sec;
849 attr->ctimensec = inode->i_ctime.tv_nsec;
852 stat->dev = inode->i_sb->s_dev;
853 stat->ino = attr->ino;
854 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
855 stat->nlink = attr->nlink;
856 stat->uid = make_kuid(&init_user_ns, attr->uid);
857 stat->gid = make_kgid(&init_user_ns, attr->gid);
858 stat->rdev = inode->i_rdev;
859 stat->atime.tv_sec = attr->atime;
860 stat->atime.tv_nsec = attr->atimensec;
861 stat->mtime.tv_sec = attr->mtime;
862 stat->mtime.tv_nsec = attr->mtimensec;
863 stat->ctime.tv_sec = attr->ctime;
864 stat->ctime.tv_nsec = attr->ctimensec;
865 stat->size = attr->size;
866 stat->blocks = attr->blocks;
868 if (attr->blksize != 0)
869 blkbits = ilog2(attr->blksize);
870 else
871 blkbits = inode->i_sb->s_blocksize_bits;
873 stat->blksize = 1 << blkbits;
876 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
877 struct file *file)
879 int err;
880 struct fuse_getattr_in inarg;
881 struct fuse_attr_out outarg;
882 struct fuse_conn *fc = get_fuse_conn(inode);
883 FUSE_ARGS(args);
884 u64 attr_version;
886 attr_version = fuse_get_attr_version(fc);
888 memset(&inarg, 0, sizeof(inarg));
889 memset(&outarg, 0, sizeof(outarg));
890 /* Directories have separate file-handle space */
891 if (file && S_ISREG(inode->i_mode)) {
892 struct fuse_file *ff = file->private_data;
894 inarg.getattr_flags |= FUSE_GETATTR_FH;
895 inarg.fh = ff->fh;
897 args.in.h.opcode = FUSE_GETATTR;
898 args.in.h.nodeid = get_node_id(inode);
899 args.in.numargs = 1;
900 args.in.args[0].size = sizeof(inarg);
901 args.in.args[0].value = &inarg;
902 args.out.numargs = 1;
903 args.out.args[0].size = sizeof(outarg);
904 args.out.args[0].value = &outarg;
905 err = fuse_simple_request(fc, &args);
906 if (!err) {
907 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
908 make_bad_inode(inode);
909 err = -EIO;
910 } else {
911 fuse_change_attributes(inode, &outarg.attr,
912 attr_timeout(&outarg),
913 attr_version);
914 if (stat)
915 fuse_fillattr(inode, &outarg.attr, stat);
918 return err;
921 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
922 struct file *file, bool *refreshed)
924 struct fuse_inode *fi = get_fuse_inode(inode);
925 int err;
926 bool r;
928 if (time_before64(fi->i_time, get_jiffies_64())) {
929 r = true;
930 forget_all_cached_acls(inode);
931 err = fuse_do_getattr(inode, stat, file);
932 } else {
933 r = false;
934 err = 0;
935 if (stat) {
936 generic_fillattr(inode, stat);
937 stat->mode = fi->orig_i_mode;
938 stat->ino = fi->orig_ino;
942 if (refreshed != NULL)
943 *refreshed = r;
945 return err;
948 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
949 u64 child_nodeid, struct qstr *name)
951 int err = -ENOTDIR;
952 struct inode *parent;
953 struct dentry *dir;
954 struct dentry *entry;
956 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
957 if (!parent)
958 return -ENOENT;
960 inode_lock(parent);
961 if (!S_ISDIR(parent->i_mode))
962 goto unlock;
964 err = -ENOENT;
965 dir = d_find_alias(parent);
966 if (!dir)
967 goto unlock;
969 name->hash = full_name_hash(dir, name->name, name->len);
970 entry = d_lookup(dir, name);
971 dput(dir);
972 if (!entry)
973 goto unlock;
975 fuse_invalidate_attr(parent);
976 fuse_invalidate_entry(entry);
978 if (child_nodeid != 0 && d_really_is_positive(entry)) {
979 inode_lock(d_inode(entry));
980 if (get_node_id(d_inode(entry)) != child_nodeid) {
981 err = -ENOENT;
982 goto badentry;
984 if (d_mountpoint(entry)) {
985 err = -EBUSY;
986 goto badentry;
988 if (d_is_dir(entry)) {
989 shrink_dcache_parent(entry);
990 if (!simple_empty(entry)) {
991 err = -ENOTEMPTY;
992 goto badentry;
994 d_inode(entry)->i_flags |= S_DEAD;
996 dont_mount(entry);
997 clear_nlink(d_inode(entry));
998 err = 0;
999 badentry:
1000 inode_unlock(d_inode(entry));
1001 if (!err)
1002 d_delete(entry);
1003 } else {
1004 err = 0;
1006 dput(entry);
1008 unlock:
1009 inode_unlock(parent);
1010 iput(parent);
1011 return err;
1015 * Calling into a user-controlled filesystem gives the filesystem
1016 * daemon ptrace-like capabilities over the current process. This
1017 * means, that the filesystem daemon is able to record the exact
1018 * filesystem operations performed, and can also control the behavior
1019 * of the requester process in otherwise impossible ways. For example
1020 * it can delay the operation for arbitrary length of time allowing
1021 * DoS against the requester.
1023 * For this reason only those processes can call into the filesystem,
1024 * for which the owner of the mount has ptrace privilege. This
1025 * excludes processes started by other users, suid or sgid processes.
1027 int fuse_allow_current_process(struct fuse_conn *fc)
1029 const struct cred *cred;
1031 if (fc->allow_other)
1032 return 1;
1034 cred = current_cred();
1035 if (uid_eq(cred->euid, fc->user_id) &&
1036 uid_eq(cred->suid, fc->user_id) &&
1037 uid_eq(cred->uid, fc->user_id) &&
1038 gid_eq(cred->egid, fc->group_id) &&
1039 gid_eq(cred->sgid, fc->group_id) &&
1040 gid_eq(cred->gid, fc->group_id))
1041 return 1;
1043 return 0;
1046 static int fuse_access(struct inode *inode, int mask)
1048 struct fuse_conn *fc = get_fuse_conn(inode);
1049 FUSE_ARGS(args);
1050 struct fuse_access_in inarg;
1051 int err;
1053 BUG_ON(mask & MAY_NOT_BLOCK);
1055 if (fc->no_access)
1056 return 0;
1058 memset(&inarg, 0, sizeof(inarg));
1059 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1060 args.in.h.opcode = FUSE_ACCESS;
1061 args.in.h.nodeid = get_node_id(inode);
1062 args.in.numargs = 1;
1063 args.in.args[0].size = sizeof(inarg);
1064 args.in.args[0].value = &inarg;
1065 err = fuse_simple_request(fc, &args);
1066 if (err == -ENOSYS) {
1067 fc->no_access = 1;
1068 err = 0;
1070 return err;
1073 static int fuse_perm_getattr(struct inode *inode, int mask)
1075 if (mask & MAY_NOT_BLOCK)
1076 return -ECHILD;
1078 forget_all_cached_acls(inode);
1079 return fuse_do_getattr(inode, NULL, NULL);
1083 * Check permission. The two basic access models of FUSE are:
1085 * 1) Local access checking ('default_permissions' mount option) based
1086 * on file mode. This is the plain old disk filesystem permission
1087 * modell.
1089 * 2) "Remote" access checking, where server is responsible for
1090 * checking permission in each inode operation. An exception to this
1091 * is if ->permission() was invoked from sys_access() in which case an
1092 * access request is sent. Execute permission is still checked
1093 * locally based on file mode.
1095 static int fuse_permission(struct inode *inode, int mask)
1097 struct fuse_conn *fc = get_fuse_conn(inode);
1098 bool refreshed = false;
1099 int err = 0;
1101 if (!fuse_allow_current_process(fc))
1102 return -EACCES;
1105 * If attributes are needed, refresh them before proceeding
1107 if (fc->default_permissions ||
1108 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1109 struct fuse_inode *fi = get_fuse_inode(inode);
1111 if (time_before64(fi->i_time, get_jiffies_64())) {
1112 refreshed = true;
1114 err = fuse_perm_getattr(inode, mask);
1115 if (err)
1116 return err;
1120 if (fc->default_permissions) {
1121 err = generic_permission(inode, mask);
1123 /* If permission is denied, try to refresh file
1124 attributes. This is also needed, because the root
1125 node will at first have no permissions */
1126 if (err == -EACCES && !refreshed) {
1127 err = fuse_perm_getattr(inode, mask);
1128 if (!err)
1129 err = generic_permission(inode, mask);
1132 /* Note: the opposite of the above test does not
1133 exist. So if permissions are revoked this won't be
1134 noticed immediately, only after the attribute
1135 timeout has expired */
1136 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1137 err = fuse_access(inode, mask);
1138 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1139 if (!(inode->i_mode & S_IXUGO)) {
1140 if (refreshed)
1141 return -EACCES;
1143 err = fuse_perm_getattr(inode, mask);
1144 if (!err && !(inode->i_mode & S_IXUGO))
1145 return -EACCES;
1148 return err;
1151 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1152 struct dir_context *ctx)
1154 while (nbytes >= FUSE_NAME_OFFSET) {
1155 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1156 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1157 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1158 return -EIO;
1159 if (reclen > nbytes)
1160 break;
1161 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1162 return -EIO;
1164 if (!dir_emit(ctx, dirent->name, dirent->namelen,
1165 dirent->ino, dirent->type))
1166 break;
1168 buf += reclen;
1169 nbytes -= reclen;
1170 ctx->pos = dirent->off;
1173 return 0;
1176 static int fuse_direntplus_link(struct file *file,
1177 struct fuse_direntplus *direntplus,
1178 u64 attr_version)
1180 struct fuse_entry_out *o = &direntplus->entry_out;
1181 struct fuse_dirent *dirent = &direntplus->dirent;
1182 struct dentry *parent = file->f_path.dentry;
1183 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1184 struct dentry *dentry;
1185 struct dentry *alias;
1186 struct inode *dir = d_inode(parent);
1187 struct fuse_conn *fc;
1188 struct inode *inode;
1189 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1191 if (!o->nodeid) {
1193 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1194 * ENOENT. Instead, it only means the userspace filesystem did
1195 * not want to return attributes/handle for this entry.
1197 * So do nothing.
1199 return 0;
1202 if (name.name[0] == '.') {
1204 * We could potentially refresh the attributes of the directory
1205 * and its parent?
1207 if (name.len == 1)
1208 return 0;
1209 if (name.name[1] == '.' && name.len == 2)
1210 return 0;
1213 if (invalid_nodeid(o->nodeid))
1214 return -EIO;
1215 if (!fuse_valid_type(o->attr.mode))
1216 return -EIO;
1218 fc = get_fuse_conn(dir);
1220 name.hash = full_name_hash(parent, name.name, name.len);
1221 dentry = d_lookup(parent, &name);
1222 if (!dentry) {
1223 retry:
1224 dentry = d_alloc_parallel(parent, &name, &wq);
1225 if (IS_ERR(dentry))
1226 return PTR_ERR(dentry);
1228 if (!d_in_lookup(dentry)) {
1229 struct fuse_inode *fi;
1230 inode = d_inode(dentry);
1231 if (!inode ||
1232 get_node_id(inode) != o->nodeid ||
1233 ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1234 d_invalidate(dentry);
1235 dput(dentry);
1236 goto retry;
1238 if (is_bad_inode(inode)) {
1239 dput(dentry);
1240 return -EIO;
1243 fi = get_fuse_inode(inode);
1244 spin_lock(&fc->lock);
1245 fi->nlookup++;
1246 spin_unlock(&fc->lock);
1248 forget_all_cached_acls(inode);
1249 fuse_change_attributes(inode, &o->attr,
1250 entry_attr_timeout(o),
1251 attr_version);
1253 * The other branch comes via fuse_iget()
1254 * which bumps nlookup inside
1256 } else {
1257 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1258 &o->attr, entry_attr_timeout(o),
1259 attr_version);
1260 if (!inode)
1261 inode = ERR_PTR(-ENOMEM);
1263 alias = d_splice_alias(inode, dentry);
1264 d_lookup_done(dentry);
1265 if (alias) {
1266 dput(dentry);
1267 dentry = alias;
1269 if (IS_ERR(dentry))
1270 return PTR_ERR(dentry);
1272 if (fc->readdirplus_auto)
1273 set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1274 fuse_change_entry_timeout(dentry, o);
1276 dput(dentry);
1277 return 0;
1280 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1281 struct dir_context *ctx, u64 attr_version)
1283 struct fuse_direntplus *direntplus;
1284 struct fuse_dirent *dirent;
1285 size_t reclen;
1286 int over = 0;
1287 int ret;
1289 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1290 direntplus = (struct fuse_direntplus *) buf;
1291 dirent = &direntplus->dirent;
1292 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1294 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1295 return -EIO;
1296 if (reclen > nbytes)
1297 break;
1298 if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1299 return -EIO;
1301 if (!over) {
1302 /* We fill entries into dstbuf only as much as
1303 it can hold. But we still continue iterating
1304 over remaining entries to link them. If not,
1305 we need to send a FORGET for each of those
1306 which we did not link.
1308 over = !dir_emit(ctx, dirent->name, dirent->namelen,
1309 dirent->ino, dirent->type);
1310 ctx->pos = dirent->off;
1313 buf += reclen;
1314 nbytes -= reclen;
1316 ret = fuse_direntplus_link(file, direntplus, attr_version);
1317 if (ret)
1318 fuse_force_forget(file, direntplus->entry_out.nodeid);
1321 return 0;
1324 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1326 int plus, err;
1327 size_t nbytes;
1328 struct page *page;
1329 struct inode *inode = file_inode(file);
1330 struct fuse_conn *fc = get_fuse_conn(inode);
1331 struct fuse_req *req;
1332 u64 attr_version = 0;
1334 if (is_bad_inode(inode))
1335 return -EIO;
1337 req = fuse_get_req(fc, 1);
1338 if (IS_ERR(req))
1339 return PTR_ERR(req);
1341 page = alloc_page(GFP_KERNEL);
1342 if (!page) {
1343 fuse_put_request(fc, req);
1344 return -ENOMEM;
1347 plus = fuse_use_readdirplus(inode, ctx);
1348 req->out.argpages = 1;
1349 req->num_pages = 1;
1350 req->pages[0] = page;
1351 req->page_descs[0].length = PAGE_SIZE;
1352 if (plus) {
1353 attr_version = fuse_get_attr_version(fc);
1354 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1355 FUSE_READDIRPLUS);
1356 } else {
1357 fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1358 FUSE_READDIR);
1360 fuse_lock_inode(inode);
1361 fuse_request_send(fc, req);
1362 fuse_unlock_inode(inode);
1363 nbytes = req->out.args[0].size;
1364 err = req->out.h.error;
1365 fuse_put_request(fc, req);
1366 if (!err) {
1367 if (plus) {
1368 err = parse_dirplusfile(page_address(page), nbytes,
1369 file, ctx,
1370 attr_version);
1371 } else {
1372 err = parse_dirfile(page_address(page), nbytes, file,
1373 ctx);
1377 __free_page(page);
1378 fuse_invalidate_atime(inode);
1379 return err;
1382 static const char *fuse_get_link(struct dentry *dentry,
1383 struct inode *inode,
1384 struct delayed_call *done)
1386 struct fuse_conn *fc = get_fuse_conn(inode);
1387 FUSE_ARGS(args);
1388 char *link;
1389 ssize_t ret;
1391 if (!dentry)
1392 return ERR_PTR(-ECHILD);
1394 link = kmalloc(PAGE_SIZE, GFP_KERNEL);
1395 if (!link)
1396 return ERR_PTR(-ENOMEM);
1398 args.in.h.opcode = FUSE_READLINK;
1399 args.in.h.nodeid = get_node_id(inode);
1400 args.out.argvar = 1;
1401 args.out.numargs = 1;
1402 args.out.args[0].size = PAGE_SIZE - 1;
1403 args.out.args[0].value = link;
1404 ret = fuse_simple_request(fc, &args);
1405 if (ret < 0) {
1406 kfree(link);
1407 link = ERR_PTR(ret);
1408 } else {
1409 link[ret] = '\0';
1410 set_delayed_call(done, kfree_link, link);
1412 fuse_invalidate_atime(inode);
1413 return link;
1416 static int fuse_dir_open(struct inode *inode, struct file *file)
1418 return fuse_open_common(inode, file, true);
1421 static int fuse_dir_release(struct inode *inode, struct file *file)
1423 fuse_release_common(file, FUSE_RELEASEDIR);
1425 return 0;
1428 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1429 int datasync)
1431 return fuse_fsync_common(file, start, end, datasync, 1);
1434 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1435 unsigned long arg)
1437 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1439 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1440 if (fc->minor < 18)
1441 return -ENOTTY;
1443 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1446 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1447 unsigned long arg)
1449 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1451 if (fc->minor < 18)
1452 return -ENOTTY;
1454 return fuse_ioctl_common(file, cmd, arg,
1455 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1458 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1460 /* Always update if mtime is explicitly set */
1461 if (ivalid & ATTR_MTIME_SET)
1462 return true;
1464 /* Or if kernel i_mtime is the official one */
1465 if (trust_local_mtime)
1466 return true;
1468 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1469 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1470 return false;
1472 /* In all other cases update */
1473 return true;
1476 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
1477 bool trust_local_cmtime)
1479 unsigned ivalid = iattr->ia_valid;
1481 if (ivalid & ATTR_MODE)
1482 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1483 if (ivalid & ATTR_UID)
1484 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1485 if (ivalid & ATTR_GID)
1486 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1487 if (ivalid & ATTR_SIZE)
1488 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1489 if (ivalid & ATTR_ATIME) {
1490 arg->valid |= FATTR_ATIME;
1491 arg->atime = iattr->ia_atime.tv_sec;
1492 arg->atimensec = iattr->ia_atime.tv_nsec;
1493 if (!(ivalid & ATTR_ATIME_SET))
1494 arg->valid |= FATTR_ATIME_NOW;
1496 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1497 arg->valid |= FATTR_MTIME;
1498 arg->mtime = iattr->ia_mtime.tv_sec;
1499 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1500 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1501 arg->valid |= FATTR_MTIME_NOW;
1503 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1504 arg->valid |= FATTR_CTIME;
1505 arg->ctime = iattr->ia_ctime.tv_sec;
1506 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1511 * Prevent concurrent writepages on inode
1513 * This is done by adding a negative bias to the inode write counter
1514 * and waiting for all pending writes to finish.
1516 void fuse_set_nowrite(struct inode *inode)
1518 struct fuse_conn *fc = get_fuse_conn(inode);
1519 struct fuse_inode *fi = get_fuse_inode(inode);
1521 BUG_ON(!inode_is_locked(inode));
1523 spin_lock(&fc->lock);
1524 BUG_ON(fi->writectr < 0);
1525 fi->writectr += FUSE_NOWRITE;
1526 spin_unlock(&fc->lock);
1527 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1531 * Allow writepages on inode
1533 * Remove the bias from the writecounter and send any queued
1534 * writepages.
1536 static void __fuse_release_nowrite(struct inode *inode)
1538 struct fuse_inode *fi = get_fuse_inode(inode);
1540 BUG_ON(fi->writectr != FUSE_NOWRITE);
1541 fi->writectr = 0;
1542 fuse_flush_writepages(inode);
1545 void fuse_release_nowrite(struct inode *inode)
1547 struct fuse_conn *fc = get_fuse_conn(inode);
1549 spin_lock(&fc->lock);
1550 __fuse_release_nowrite(inode);
1551 spin_unlock(&fc->lock);
1554 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1555 struct inode *inode,
1556 struct fuse_setattr_in *inarg_p,
1557 struct fuse_attr_out *outarg_p)
1559 args->in.h.opcode = FUSE_SETATTR;
1560 args->in.h.nodeid = get_node_id(inode);
1561 args->in.numargs = 1;
1562 args->in.args[0].size = sizeof(*inarg_p);
1563 args->in.args[0].value = inarg_p;
1564 args->out.numargs = 1;
1565 args->out.args[0].size = sizeof(*outarg_p);
1566 args->out.args[0].value = outarg_p;
1570 * Flush inode->i_mtime to the server
1572 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1574 struct fuse_conn *fc = get_fuse_conn(inode);
1575 FUSE_ARGS(args);
1576 struct fuse_setattr_in inarg;
1577 struct fuse_attr_out outarg;
1579 memset(&inarg, 0, sizeof(inarg));
1580 memset(&outarg, 0, sizeof(outarg));
1582 inarg.valid = FATTR_MTIME;
1583 inarg.mtime = inode->i_mtime.tv_sec;
1584 inarg.mtimensec = inode->i_mtime.tv_nsec;
1585 if (fc->minor >= 23) {
1586 inarg.valid |= FATTR_CTIME;
1587 inarg.ctime = inode->i_ctime.tv_sec;
1588 inarg.ctimensec = inode->i_ctime.tv_nsec;
1590 if (ff) {
1591 inarg.valid |= FATTR_FH;
1592 inarg.fh = ff->fh;
1594 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1596 return fuse_simple_request(fc, &args);
1600 * Set attributes, and at the same time refresh them.
1602 * Truncation is slightly complicated, because the 'truncate' request
1603 * may fail, in which case we don't want to touch the mapping.
1604 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1605 * and the actual truncation by hand.
1607 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1608 struct file *file)
1610 struct inode *inode = d_inode(dentry);
1611 struct fuse_conn *fc = get_fuse_conn(inode);
1612 struct fuse_inode *fi = get_fuse_inode(inode);
1613 FUSE_ARGS(args);
1614 struct fuse_setattr_in inarg;
1615 struct fuse_attr_out outarg;
1616 bool is_truncate = false;
1617 bool is_wb = fc->writeback_cache;
1618 loff_t oldsize;
1619 int err;
1620 bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1622 if (!fc->default_permissions)
1623 attr->ia_valid |= ATTR_FORCE;
1625 err = setattr_prepare(dentry, attr);
1626 if (err)
1627 return err;
1629 if (attr->ia_valid & ATTR_OPEN) {
1630 if (fc->atomic_o_trunc)
1631 return 0;
1632 file = NULL;
1635 if (attr->ia_valid & ATTR_SIZE)
1636 is_truncate = true;
1638 if (is_truncate) {
1639 fuse_set_nowrite(inode);
1640 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1641 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1642 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1645 memset(&inarg, 0, sizeof(inarg));
1646 memset(&outarg, 0, sizeof(outarg));
1647 iattr_to_fattr(attr, &inarg, trust_local_cmtime);
1648 if (file) {
1649 struct fuse_file *ff = file->private_data;
1650 inarg.valid |= FATTR_FH;
1651 inarg.fh = ff->fh;
1653 if (attr->ia_valid & ATTR_SIZE) {
1654 /* For mandatory locking in truncate */
1655 inarg.valid |= FATTR_LOCKOWNER;
1656 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1658 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1659 err = fuse_simple_request(fc, &args);
1660 if (err) {
1661 if (err == -EINTR)
1662 fuse_invalidate_attr(inode);
1663 goto error;
1666 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1667 make_bad_inode(inode);
1668 err = -EIO;
1669 goto error;
1672 spin_lock(&fc->lock);
1673 /* the kernel maintains i_mtime locally */
1674 if (trust_local_cmtime) {
1675 if (attr->ia_valid & ATTR_MTIME)
1676 inode->i_mtime = attr->ia_mtime;
1677 if (attr->ia_valid & ATTR_CTIME)
1678 inode->i_ctime = attr->ia_ctime;
1679 /* FIXME: clear I_DIRTY_SYNC? */
1682 fuse_change_attributes_common(inode, &outarg.attr,
1683 attr_timeout(&outarg));
1684 oldsize = inode->i_size;
1685 /* see the comment in fuse_change_attributes() */
1686 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1687 i_size_write(inode, outarg.attr.size);
1689 if (is_truncate) {
1690 /* NOTE: this may release/reacquire fc->lock */
1691 __fuse_release_nowrite(inode);
1693 spin_unlock(&fc->lock);
1696 * Only call invalidate_inode_pages2() after removing
1697 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1699 if ((is_truncate || !is_wb) &&
1700 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1701 truncate_pagecache(inode, outarg.attr.size);
1702 invalidate_inode_pages2(inode->i_mapping);
1705 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1706 return 0;
1708 error:
1709 if (is_truncate)
1710 fuse_release_nowrite(inode);
1712 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1713 return err;
1716 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1718 struct inode *inode = d_inode(entry);
1719 struct fuse_conn *fc = get_fuse_conn(inode);
1720 struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1721 int ret;
1723 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1724 return -EACCES;
1726 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1727 attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1728 ATTR_MODE);
1731 * The only sane way to reliably kill suid/sgid is to do it in
1732 * the userspace filesystem
1734 * This should be done on write(), truncate() and chown().
1736 if (!fc->handle_killpriv) {
1737 int kill;
1740 * ia_mode calculation may have used stale i_mode.
1741 * Refresh and recalculate.
1743 ret = fuse_do_getattr(inode, NULL, file);
1744 if (ret)
1745 return ret;
1747 attr->ia_mode = inode->i_mode;
1748 kill = should_remove_suid(entry);
1749 if (kill & ATTR_KILL_SUID) {
1750 attr->ia_valid |= ATTR_MODE;
1751 attr->ia_mode &= ~S_ISUID;
1753 if (kill & ATTR_KILL_SGID) {
1754 attr->ia_valid |= ATTR_MODE;
1755 attr->ia_mode &= ~S_ISGID;
1759 if (!attr->ia_valid)
1760 return 0;
1762 ret = fuse_do_setattr(entry, attr, file);
1763 if (!ret) {
1765 * If filesystem supports acls it may have updated acl xattrs in
1766 * the filesystem, so forget cached acls for the inode.
1768 if (fc->posix_acl)
1769 forget_all_cached_acls(inode);
1771 /* Directory mode changed, may need to revalidate access */
1772 if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1773 fuse_invalidate_entry_cache(entry);
1775 return ret;
1778 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1779 struct kstat *stat)
1781 struct inode *inode = d_inode(entry);
1782 struct fuse_conn *fc = get_fuse_conn(inode);
1784 if (!fuse_allow_current_process(fc))
1785 return -EACCES;
1787 return fuse_update_attributes(inode, stat, NULL, NULL);
1790 static const struct inode_operations fuse_dir_inode_operations = {
1791 .lookup = fuse_lookup,
1792 .mkdir = fuse_mkdir,
1793 .symlink = fuse_symlink,
1794 .unlink = fuse_unlink,
1795 .rmdir = fuse_rmdir,
1796 .rename = fuse_rename2,
1797 .link = fuse_link,
1798 .setattr = fuse_setattr,
1799 .create = fuse_create,
1800 .atomic_open = fuse_atomic_open,
1801 .mknod = fuse_mknod,
1802 .permission = fuse_permission,
1803 .getattr = fuse_getattr,
1804 .listxattr = fuse_listxattr,
1805 .get_acl = fuse_get_acl,
1806 .set_acl = fuse_set_acl,
1809 static const struct file_operations fuse_dir_operations = {
1810 .llseek = generic_file_llseek,
1811 .read = generic_read_dir,
1812 .iterate_shared = fuse_readdir,
1813 .open = fuse_dir_open,
1814 .release = fuse_dir_release,
1815 .fsync = fuse_dir_fsync,
1816 .unlocked_ioctl = fuse_dir_ioctl,
1817 .compat_ioctl = fuse_dir_compat_ioctl,
1820 static const struct inode_operations fuse_common_inode_operations = {
1821 .setattr = fuse_setattr,
1822 .permission = fuse_permission,
1823 .getattr = fuse_getattr,
1824 .listxattr = fuse_listxattr,
1825 .get_acl = fuse_get_acl,
1826 .set_acl = fuse_set_acl,
1829 static const struct inode_operations fuse_symlink_inode_operations = {
1830 .setattr = fuse_setattr,
1831 .get_link = fuse_get_link,
1832 .readlink = generic_readlink,
1833 .getattr = fuse_getattr,
1834 .listxattr = fuse_listxattr,
1837 void fuse_init_common(struct inode *inode)
1839 inode->i_op = &fuse_common_inode_operations;
1842 void fuse_init_dir(struct inode *inode)
1844 inode->i_op = &fuse_dir_inode_operations;
1845 inode->i_fop = &fuse_dir_operations;
1848 void fuse_init_symlink(struct inode *inode)
1850 inode->i_op = &fuse_symlink_inode_operations;