2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/gfp.h>
14 #include <linux/sched.h>
15 #include <linux/namei.h>
17 #if BITS_PER_LONG >= 64
18 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
23 static inline u64 fuse_dentry_time(struct dentry *entry)
29 * On 32 bit archs store the high 32 bits of time in d_fsdata
31 static void fuse_dentry_settime(struct dentry *entry, u64 time)
34 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
37 static u64 fuse_dentry_time(struct dentry *entry)
39 return (u64) entry->d_time +
40 ((u64) (unsigned long) entry->d_fsdata << 32);
45 * FUSE caches dentries and attributes with separate timeout. The
46 * time in jiffies until the dentry/attributes are valid is stored in
47 * dentry->d_time and fuse_inode->i_time respectively.
51 * Calculate the time in jiffies until a dentry/attributes are valid
53 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
56 struct timespec ts = {sec, nsec};
57 return get_jiffies_64() + timespec_to_jiffies(&ts);
63 * Set dentry and possibly attribute timeouts from the lookup/mk*
66 static void fuse_change_entry_timeout(struct dentry *entry,
67 struct fuse_entry_out *o)
69 fuse_dentry_settime(entry,
70 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
73 static u64 attr_timeout(struct fuse_attr_out *o)
75 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
78 static u64 entry_attr_timeout(struct fuse_entry_out *o)
80 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
84 * Mark the attributes as stale, so that at the next call to
85 * ->getattr() they will be fetched from userspace
87 void fuse_invalidate_attr(struct inode *inode)
89 get_fuse_inode(inode)->i_time = 0;
93 * Just mark the entry as stale, so that a next attempt to look it up
94 * will result in a new lookup call to userspace
96 * This is called when a dentry is about to become negative and the
97 * timeout is unknown (unlink, rmdir, rename and in some cases
100 static void fuse_invalidate_entry_cache(struct dentry *entry)
102 fuse_dentry_settime(entry, 0);
106 * Same as fuse_invalidate_entry_cache(), but also try to remove the
107 * dentry from the hash
109 static void fuse_invalidate_entry(struct dentry *entry)
112 fuse_invalidate_entry_cache(entry);
115 static void fuse_lookup_init(struct fuse_req *req, struct inode *dir,
116 struct dentry *entry,
117 struct fuse_entry_out *outarg)
119 struct fuse_conn *fc = get_fuse_conn(dir);
121 memset(outarg, 0, sizeof(struct fuse_entry_out));
122 req->in.h.opcode = FUSE_LOOKUP;
123 req->in.h.nodeid = get_node_id(dir);
125 req->in.args[0].size = entry->d_name.len + 1;
126 req->in.args[0].value = entry->d_name.name;
127 req->out.numargs = 1;
129 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
131 req->out.args[0].size = sizeof(struct fuse_entry_out);
132 req->out.args[0].value = outarg;
135 u64 fuse_get_attr_version(struct fuse_conn *fc)
140 * The spin lock isn't actually needed on 64bit archs, but we
141 * don't yet care too much about such optimizations.
143 spin_lock(&fc->lock);
144 curr_version = fc->attr_version;
145 spin_unlock(&fc->lock);
151 * Check whether the dentry is still valid
153 * If the entry validity timeout has expired and the dentry is
154 * positive, try to redo the lookup. If the lookup results in a
155 * different inode, then let the VFS invalidate the dentry and redo
156 * the lookup once more. If the lookup results in the same inode,
157 * then refresh the attributes, timeouts and mark the dentry valid.
159 static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
161 struct inode *inode = entry->d_inode;
163 if (inode && is_bad_inode(inode))
165 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
167 struct fuse_entry_out outarg;
168 struct fuse_conn *fc;
169 struct fuse_req *req;
170 struct fuse_req *forget_req;
171 struct dentry *parent;
174 /* For negative dentries, always do a fresh lookup */
178 fc = get_fuse_conn(inode);
179 req = fuse_get_req(fc);
183 forget_req = fuse_get_req(fc);
184 if (IS_ERR(forget_req)) {
185 fuse_put_request(fc, req);
189 attr_version = fuse_get_attr_version(fc);
191 parent = dget_parent(entry);
192 fuse_lookup_init(req, parent->d_inode, entry, &outarg);
193 request_send(fc, req);
195 err = req->out.h.error;
196 fuse_put_request(fc, req);
197 /* Zero nodeid is same as -ENOENT */
198 if (!err && !outarg.nodeid)
201 struct fuse_inode *fi = get_fuse_inode(inode);
202 if (outarg.nodeid != get_node_id(inode)) {
203 fuse_send_forget(fc, forget_req,
207 spin_lock(&fc->lock);
209 spin_unlock(&fc->lock);
211 fuse_put_request(fc, forget_req);
212 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
215 fuse_change_attributes(inode, &outarg.attr,
216 entry_attr_timeout(&outarg),
218 fuse_change_entry_timeout(entry, &outarg);
223 static int invalid_nodeid(u64 nodeid)
225 return !nodeid || nodeid == FUSE_ROOT_ID;
228 static struct dentry_operations fuse_dentry_operations = {
229 .d_revalidate = fuse_dentry_revalidate,
232 int fuse_valid_type(int m)
234 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
235 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
239 * Add a directory inode to a dentry, ensuring that no other dentry
240 * refers to this inode. Called with fc->inst_mutex.
242 static int fuse_d_add_directory(struct dentry *entry, struct inode *inode)
244 struct dentry *alias = d_find_alias(inode);
246 /* This tries to shrink the subtree below alias */
247 fuse_invalidate_entry(alias);
249 if (!list_empty(&inode->i_dentry))
256 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
257 struct nameidata *nd)
260 struct fuse_entry_out outarg;
261 struct inode *inode = NULL;
262 struct fuse_conn *fc = get_fuse_conn(dir);
263 struct fuse_req *req;
264 struct fuse_req *forget_req;
267 if (entry->d_name.len > FUSE_NAME_MAX)
268 return ERR_PTR(-ENAMETOOLONG);
270 req = fuse_get_req(fc);
272 return ERR_CAST(req);
274 forget_req = fuse_get_req(fc);
275 if (IS_ERR(forget_req)) {
276 fuse_put_request(fc, req);
277 return ERR_CAST(forget_req);
280 attr_version = fuse_get_attr_version(fc);
282 fuse_lookup_init(req, dir, entry, &outarg);
283 request_send(fc, req);
284 err = req->out.h.error;
285 fuse_put_request(fc, req);
286 /* Zero nodeid is same as -ENOENT, but with valid timeout */
287 if (!err && outarg.nodeid &&
288 (invalid_nodeid(outarg.nodeid) ||
289 !fuse_valid_type(outarg.attr.mode)))
291 if (!err && outarg.nodeid) {
292 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
293 &outarg.attr, entry_attr_timeout(&outarg),
296 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
297 return ERR_PTR(-ENOMEM);
300 fuse_put_request(fc, forget_req);
301 if (err && err != -ENOENT)
304 if (inode && S_ISDIR(inode->i_mode)) {
305 mutex_lock(&fc->inst_mutex);
306 err = fuse_d_add_directory(entry, inode);
307 mutex_unlock(&fc->inst_mutex);
315 entry->d_op = &fuse_dentry_operations;
317 fuse_change_entry_timeout(entry, &outarg);
319 fuse_invalidate_entry_cache(entry);
324 * Synchronous release for the case when something goes wrong in CREATE_OPEN
326 static void fuse_sync_release(struct fuse_conn *fc, struct fuse_file *ff,
327 u64 nodeid, int flags)
329 fuse_release_fill(ff, nodeid, flags, FUSE_RELEASE);
330 ff->reserved_req->force = 1;
331 request_send(fc, ff->reserved_req);
332 fuse_put_request(fc, ff->reserved_req);
337 * Atomic create+open operation
339 * If the filesystem doesn't support this, then fall back to separate
340 * 'mknod' + 'open' requests.
342 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
343 struct nameidata *nd)
347 struct fuse_conn *fc = get_fuse_conn(dir);
348 struct fuse_req *req;
349 struct fuse_req *forget_req;
350 struct fuse_open_in inarg;
351 struct fuse_open_out outopen;
352 struct fuse_entry_out outentry;
353 struct fuse_file *ff;
355 int flags = nd->intent.open.flags - 1;
360 forget_req = fuse_get_req(fc);
361 if (IS_ERR(forget_req))
362 return PTR_ERR(forget_req);
364 req = fuse_get_req(fc);
367 goto out_put_forget_req;
370 ff = fuse_file_alloc();
372 goto out_put_request;
375 memset(&inarg, 0, sizeof(inarg));
376 memset(&outentry, 0, sizeof(outentry));
379 req->in.h.opcode = FUSE_CREATE;
380 req->in.h.nodeid = get_node_id(dir);
382 req->in.args[0].size = sizeof(inarg);
383 req->in.args[0].value = &inarg;
384 req->in.args[1].size = entry->d_name.len + 1;
385 req->in.args[1].value = entry->d_name.name;
386 req->out.numargs = 2;
388 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
390 req->out.args[0].size = sizeof(outentry);
391 req->out.args[0].value = &outentry;
392 req->out.args[1].size = sizeof(outopen);
393 req->out.args[1].value = &outopen;
394 request_send(fc, req);
395 err = req->out.h.error;
403 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
406 fuse_put_request(fc, req);
407 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
408 &outentry.attr, entry_attr_timeout(&outentry), 0);
410 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
412 fuse_sync_release(fc, ff, outentry.nodeid, flags);
413 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
416 fuse_put_request(fc, forget_req);
417 d_instantiate(entry, inode);
418 fuse_change_entry_timeout(entry, &outentry);
419 fuse_invalidate_attr(dir);
420 file = lookup_instantiate_filp(nd, entry, generic_file_open);
423 fuse_sync_release(fc, ff, outentry.nodeid, flags);
424 return PTR_ERR(file);
426 fuse_finish_open(inode, file, ff, &outopen);
432 fuse_put_request(fc, req);
434 fuse_put_request(fc, forget_req);
439 * Code shared between mknod, mkdir, symlink and link
441 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
442 struct inode *dir, struct dentry *entry,
445 struct fuse_entry_out outarg;
448 struct fuse_req *forget_req;
450 forget_req = fuse_get_req(fc);
451 if (IS_ERR(forget_req)) {
452 fuse_put_request(fc, req);
453 return PTR_ERR(forget_req);
456 memset(&outarg, 0, sizeof(outarg));
457 req->in.h.nodeid = get_node_id(dir);
458 req->out.numargs = 1;
460 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
462 req->out.args[0].size = sizeof(outarg);
463 req->out.args[0].value = &outarg;
464 request_send(fc, req);
465 err = req->out.h.error;
466 fuse_put_request(fc, req);
468 goto out_put_forget_req;
471 if (invalid_nodeid(outarg.nodeid))
472 goto out_put_forget_req;
474 if ((outarg.attr.mode ^ mode) & S_IFMT)
475 goto out_put_forget_req;
477 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
478 &outarg.attr, entry_attr_timeout(&outarg), 0);
480 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
483 fuse_put_request(fc, forget_req);
485 if (S_ISDIR(inode->i_mode)) {
486 struct dentry *alias;
487 mutex_lock(&fc->inst_mutex);
488 alias = d_find_alias(inode);
490 /* New directory must have moved since mkdir */
491 mutex_unlock(&fc->inst_mutex);
496 d_instantiate(entry, inode);
497 mutex_unlock(&fc->inst_mutex);
499 d_instantiate(entry, inode);
501 fuse_change_entry_timeout(entry, &outarg);
502 fuse_invalidate_attr(dir);
506 fuse_put_request(fc, forget_req);
510 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
513 struct fuse_mknod_in inarg;
514 struct fuse_conn *fc = get_fuse_conn(dir);
515 struct fuse_req *req = fuse_get_req(fc);
519 memset(&inarg, 0, sizeof(inarg));
521 inarg.rdev = new_encode_dev(rdev);
522 req->in.h.opcode = FUSE_MKNOD;
524 req->in.args[0].size = sizeof(inarg);
525 req->in.args[0].value = &inarg;
526 req->in.args[1].size = entry->d_name.len + 1;
527 req->in.args[1].value = entry->d_name.name;
528 return create_new_entry(fc, req, dir, entry, mode);
531 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
532 struct nameidata *nd)
534 if (nd && (nd->flags & LOOKUP_OPEN)) {
535 int err = fuse_create_open(dir, entry, mode, nd);
538 /* Fall back on mknod */
540 return fuse_mknod(dir, entry, mode, 0);
543 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
545 struct fuse_mkdir_in inarg;
546 struct fuse_conn *fc = get_fuse_conn(dir);
547 struct fuse_req *req = fuse_get_req(fc);
551 memset(&inarg, 0, sizeof(inarg));
553 req->in.h.opcode = FUSE_MKDIR;
555 req->in.args[0].size = sizeof(inarg);
556 req->in.args[0].value = &inarg;
557 req->in.args[1].size = entry->d_name.len + 1;
558 req->in.args[1].value = entry->d_name.name;
559 return create_new_entry(fc, req, dir, entry, S_IFDIR);
562 static int fuse_symlink(struct inode *dir, struct dentry *entry,
565 struct fuse_conn *fc = get_fuse_conn(dir);
566 unsigned len = strlen(link) + 1;
567 struct fuse_req *req = fuse_get_req(fc);
571 req->in.h.opcode = FUSE_SYMLINK;
573 req->in.args[0].size = entry->d_name.len + 1;
574 req->in.args[0].value = entry->d_name.name;
575 req->in.args[1].size = len;
576 req->in.args[1].value = link;
577 return create_new_entry(fc, req, dir, entry, S_IFLNK);
580 static int fuse_unlink(struct inode *dir, struct dentry *entry)
583 struct fuse_conn *fc = get_fuse_conn(dir);
584 struct fuse_req *req = fuse_get_req(fc);
588 req->in.h.opcode = FUSE_UNLINK;
589 req->in.h.nodeid = get_node_id(dir);
591 req->in.args[0].size = entry->d_name.len + 1;
592 req->in.args[0].value = entry->d_name.name;
593 request_send(fc, req);
594 err = req->out.h.error;
595 fuse_put_request(fc, req);
597 struct inode *inode = entry->d_inode;
599 /* Set nlink to zero so the inode can be cleared, if
600 the inode does have more links this will be
601 discovered at the next lookup/getattr */
603 fuse_invalidate_attr(inode);
604 fuse_invalidate_attr(dir);
605 fuse_invalidate_entry_cache(entry);
606 } else if (err == -EINTR)
607 fuse_invalidate_entry(entry);
611 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
614 struct fuse_conn *fc = get_fuse_conn(dir);
615 struct fuse_req *req = fuse_get_req(fc);
619 req->in.h.opcode = FUSE_RMDIR;
620 req->in.h.nodeid = get_node_id(dir);
622 req->in.args[0].size = entry->d_name.len + 1;
623 req->in.args[0].value = entry->d_name.name;
624 request_send(fc, req);
625 err = req->out.h.error;
626 fuse_put_request(fc, req);
628 clear_nlink(entry->d_inode);
629 fuse_invalidate_attr(dir);
630 fuse_invalidate_entry_cache(entry);
631 } else if (err == -EINTR)
632 fuse_invalidate_entry(entry);
636 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
637 struct inode *newdir, struct dentry *newent)
640 struct fuse_rename_in inarg;
641 struct fuse_conn *fc = get_fuse_conn(olddir);
642 struct fuse_req *req = fuse_get_req(fc);
646 memset(&inarg, 0, sizeof(inarg));
647 inarg.newdir = get_node_id(newdir);
648 req->in.h.opcode = FUSE_RENAME;
649 req->in.h.nodeid = get_node_id(olddir);
651 req->in.args[0].size = sizeof(inarg);
652 req->in.args[0].value = &inarg;
653 req->in.args[1].size = oldent->d_name.len + 1;
654 req->in.args[1].value = oldent->d_name.name;
655 req->in.args[2].size = newent->d_name.len + 1;
656 req->in.args[2].value = newent->d_name.name;
657 request_send(fc, req);
658 err = req->out.h.error;
659 fuse_put_request(fc, req);
662 fuse_invalidate_attr(oldent->d_inode);
664 fuse_invalidate_attr(olddir);
665 if (olddir != newdir)
666 fuse_invalidate_attr(newdir);
668 /* newent will end up negative */
670 fuse_invalidate_entry_cache(newent);
671 } else if (err == -EINTR) {
672 /* If request was interrupted, DEITY only knows if the
673 rename actually took place. If the invalidation
674 fails (e.g. some process has CWD under the renamed
675 directory), then there can be inconsistency between
676 the dcache and the real filesystem. Tough luck. */
677 fuse_invalidate_entry(oldent);
679 fuse_invalidate_entry(newent);
685 static int fuse_link(struct dentry *entry, struct inode *newdir,
686 struct dentry *newent)
689 struct fuse_link_in inarg;
690 struct inode *inode = entry->d_inode;
691 struct fuse_conn *fc = get_fuse_conn(inode);
692 struct fuse_req *req = fuse_get_req(fc);
696 memset(&inarg, 0, sizeof(inarg));
697 inarg.oldnodeid = get_node_id(inode);
698 req->in.h.opcode = FUSE_LINK;
700 req->in.args[0].size = sizeof(inarg);
701 req->in.args[0].value = &inarg;
702 req->in.args[1].size = newent->d_name.len + 1;
703 req->in.args[1].value = newent->d_name.name;
704 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
705 /* Contrary to "normal" filesystems it can happen that link
706 makes two "logical" inodes point to the same "physical"
707 inode. We invalidate the attributes of the old one, so it
708 will reflect changes in the backing inode (link count,
711 if (!err || err == -EINTR)
712 fuse_invalidate_attr(inode);
716 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
719 stat->dev = inode->i_sb->s_dev;
720 stat->ino = attr->ino;
721 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
722 stat->nlink = attr->nlink;
723 stat->uid = attr->uid;
724 stat->gid = attr->gid;
725 stat->rdev = inode->i_rdev;
726 stat->atime.tv_sec = attr->atime;
727 stat->atime.tv_nsec = attr->atimensec;
728 stat->mtime.tv_sec = attr->mtime;
729 stat->mtime.tv_nsec = attr->mtimensec;
730 stat->ctime.tv_sec = attr->ctime;
731 stat->ctime.tv_nsec = attr->ctimensec;
732 stat->size = attr->size;
733 stat->blocks = attr->blocks;
734 stat->blksize = (1 << inode->i_blkbits);
737 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
741 struct fuse_getattr_in inarg;
742 struct fuse_attr_out outarg;
743 struct fuse_conn *fc = get_fuse_conn(inode);
744 struct fuse_req *req;
747 req = fuse_get_req(fc);
751 attr_version = fuse_get_attr_version(fc);
753 memset(&inarg, 0, sizeof(inarg));
754 memset(&outarg, 0, sizeof(outarg));
755 /* Directories have separate file-handle space */
756 if (file && S_ISREG(inode->i_mode)) {
757 struct fuse_file *ff = file->private_data;
759 inarg.getattr_flags |= FUSE_GETATTR_FH;
762 req->in.h.opcode = FUSE_GETATTR;
763 req->in.h.nodeid = get_node_id(inode);
765 req->in.args[0].size = sizeof(inarg);
766 req->in.args[0].value = &inarg;
767 req->out.numargs = 1;
769 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
771 req->out.args[0].size = sizeof(outarg);
772 req->out.args[0].value = &outarg;
773 request_send(fc, req);
774 err = req->out.h.error;
775 fuse_put_request(fc, req);
777 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
778 make_bad_inode(inode);
781 fuse_change_attributes(inode, &outarg.attr,
782 attr_timeout(&outarg),
785 fuse_fillattr(inode, &outarg.attr, stat);
791 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
792 struct file *file, bool *refreshed)
794 struct fuse_inode *fi = get_fuse_inode(inode);
798 if (fi->i_time < get_jiffies_64()) {
800 err = fuse_do_getattr(inode, stat, file);
805 generic_fillattr(inode, stat);
806 stat->mode = fi->orig_i_mode;
810 if (refreshed != NULL)
817 * Calling into a user-controlled filesystem gives the filesystem
818 * daemon ptrace-like capabilities over the requester process. This
819 * means, that the filesystem daemon is able to record the exact
820 * filesystem operations performed, and can also control the behavior
821 * of the requester process in otherwise impossible ways. For example
822 * it can delay the operation for arbitrary length of time allowing
823 * DoS against the requester.
825 * For this reason only those processes can call into the filesystem,
826 * for which the owner of the mount has ptrace privilege. This
827 * excludes processes started by other users, suid or sgid processes.
829 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
831 if (fc->flags & FUSE_ALLOW_OTHER)
834 if (task->euid == fc->user_id &&
835 task->suid == fc->user_id &&
836 task->uid == fc->user_id &&
837 task->egid == fc->group_id &&
838 task->sgid == fc->group_id &&
839 task->gid == fc->group_id)
845 static int fuse_access(struct inode *inode, int mask)
847 struct fuse_conn *fc = get_fuse_conn(inode);
848 struct fuse_req *req;
849 struct fuse_access_in inarg;
855 req = fuse_get_req(fc);
859 memset(&inarg, 0, sizeof(inarg));
861 req->in.h.opcode = FUSE_ACCESS;
862 req->in.h.nodeid = get_node_id(inode);
864 req->in.args[0].size = sizeof(inarg);
865 req->in.args[0].value = &inarg;
866 request_send(fc, req);
867 err = req->out.h.error;
868 fuse_put_request(fc, req);
869 if (err == -ENOSYS) {
877 * Check permission. The two basic access models of FUSE are:
879 * 1) Local access checking ('default_permissions' mount option) based
880 * on file mode. This is the plain old disk filesystem permission
883 * 2) "Remote" access checking, where server is responsible for
884 * checking permission in each inode operation. An exception to this
885 * is if ->permission() was invoked from sys_access() in which case an
886 * access request is sent. Execute permission is still checked
887 * locally based on file mode.
889 static int fuse_permission(struct inode *inode, int mask, struct nameidata *nd)
891 struct fuse_conn *fc = get_fuse_conn(inode);
892 bool refreshed = false;
895 if (!fuse_allow_task(fc, current))
899 * If attributes are needed, refresh them before proceeding
901 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
902 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
903 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
908 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
909 err = generic_permission(inode, mask, NULL);
911 /* If permission is denied, try to refresh file
912 attributes. This is also needed, because the root
913 node will at first have no permissions */
914 if (err == -EACCES && !refreshed) {
915 err = fuse_do_getattr(inode, NULL, NULL);
917 err = generic_permission(inode, mask, NULL);
920 /* Note: the opposite of the above test does not
921 exist. So if permissions are revoked this won't be
922 noticed immediately, only after the attribute
923 timeout has expired */
924 } else if (nd && (nd->flags & (LOOKUP_ACCESS | LOOKUP_CHDIR))) {
925 err = fuse_access(inode, mask);
926 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
927 if (!(inode->i_mode & S_IXUGO)) {
931 err = fuse_do_getattr(inode, NULL, NULL);
932 if (!err && !(inode->i_mode & S_IXUGO))
939 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
940 void *dstbuf, filldir_t filldir)
942 while (nbytes >= FUSE_NAME_OFFSET) {
943 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
944 size_t reclen = FUSE_DIRENT_SIZE(dirent);
946 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
951 over = filldir(dstbuf, dirent->name, dirent->namelen,
952 file->f_pos, dirent->ino, dirent->type);
958 file->f_pos = dirent->off;
964 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
969 struct inode *inode = file->f_path.dentry->d_inode;
970 struct fuse_conn *fc = get_fuse_conn(inode);
971 struct fuse_req *req;
973 if (is_bad_inode(inode))
976 req = fuse_get_req(fc);
980 page = alloc_page(GFP_KERNEL);
982 fuse_put_request(fc, req);
986 req->pages[0] = page;
987 fuse_read_fill(req, file, inode, file->f_pos, PAGE_SIZE, FUSE_READDIR);
988 request_send(fc, req);
989 nbytes = req->out.args[0].size;
990 err = req->out.h.error;
991 fuse_put_request(fc, req);
993 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
997 fuse_invalidate_attr(inode); /* atime changed */
1001 static char *read_link(struct dentry *dentry)
1003 struct inode *inode = dentry->d_inode;
1004 struct fuse_conn *fc = get_fuse_conn(inode);
1005 struct fuse_req *req = fuse_get_req(fc);
1009 return ERR_CAST(req);
1011 link = (char *) __get_free_page(GFP_KERNEL);
1013 link = ERR_PTR(-ENOMEM);
1016 req->in.h.opcode = FUSE_READLINK;
1017 req->in.h.nodeid = get_node_id(inode);
1018 req->out.argvar = 1;
1019 req->out.numargs = 1;
1020 req->out.args[0].size = PAGE_SIZE - 1;
1021 req->out.args[0].value = link;
1022 request_send(fc, req);
1023 if (req->out.h.error) {
1024 free_page((unsigned long) link);
1025 link = ERR_PTR(req->out.h.error);
1027 link[req->out.args[0].size] = '\0';
1029 fuse_put_request(fc, req);
1030 fuse_invalidate_attr(inode); /* atime changed */
1034 static void free_link(char *link)
1037 free_page((unsigned long) link);
1040 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1042 nd_set_link(nd, read_link(dentry));
1046 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1048 free_link(nd_get_link(nd));
1051 static int fuse_dir_open(struct inode *inode, struct file *file)
1053 return fuse_open_common(inode, file, 1);
1056 static int fuse_dir_release(struct inode *inode, struct file *file)
1058 return fuse_release_common(inode, file, 1);
1061 static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
1063 /* nfsd can call this with no file */
1064 return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
1067 static bool update_mtime(unsigned ivalid)
1069 /* Always update if mtime is explicitly set */
1070 if (ivalid & ATTR_MTIME_SET)
1073 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1074 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1077 /* In all other cases update */
1081 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1083 unsigned ivalid = iattr->ia_valid;
1085 if (ivalid & ATTR_MODE)
1086 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1087 if (ivalid & ATTR_UID)
1088 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1089 if (ivalid & ATTR_GID)
1090 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1091 if (ivalid & ATTR_SIZE)
1092 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1093 if (ivalid & ATTR_ATIME) {
1094 arg->valid |= FATTR_ATIME;
1095 arg->atime = iattr->ia_atime.tv_sec;
1096 arg->atimensec = iattr->ia_atime.tv_nsec;
1097 if (!(ivalid & ATTR_ATIME_SET))
1098 arg->valid |= FATTR_ATIME_NOW;
1100 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1101 arg->valid |= FATTR_MTIME;
1102 arg->mtime = iattr->ia_mtime.tv_sec;
1103 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1104 if (!(ivalid & ATTR_MTIME_SET))
1105 arg->valid |= FATTR_MTIME_NOW;
1110 * Prevent concurrent writepages on inode
1112 * This is done by adding a negative bias to the inode write counter
1113 * and waiting for all pending writes to finish.
1115 void fuse_set_nowrite(struct inode *inode)
1117 struct fuse_conn *fc = get_fuse_conn(inode);
1118 struct fuse_inode *fi = get_fuse_inode(inode);
1120 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1122 spin_lock(&fc->lock);
1123 BUG_ON(fi->writectr < 0);
1124 fi->writectr += FUSE_NOWRITE;
1125 spin_unlock(&fc->lock);
1126 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1130 * Allow writepages on inode
1132 * Remove the bias from the writecounter and send any queued
1135 static void __fuse_release_nowrite(struct inode *inode)
1137 struct fuse_inode *fi = get_fuse_inode(inode);
1139 BUG_ON(fi->writectr != FUSE_NOWRITE);
1141 fuse_flush_writepages(inode);
1144 void fuse_release_nowrite(struct inode *inode)
1146 struct fuse_conn *fc = get_fuse_conn(inode);
1148 spin_lock(&fc->lock);
1149 __fuse_release_nowrite(inode);
1150 spin_unlock(&fc->lock);
1154 * Set attributes, and at the same time refresh them.
1156 * Truncation is slightly complicated, because the 'truncate' request
1157 * may fail, in which case we don't want to touch the mapping.
1158 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1159 * and the actual truncation by hand.
1161 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1164 struct inode *inode = entry->d_inode;
1165 struct fuse_conn *fc = get_fuse_conn(inode);
1166 struct fuse_req *req;
1167 struct fuse_setattr_in inarg;
1168 struct fuse_attr_out outarg;
1169 bool is_truncate = false;
1173 if (!fuse_allow_task(fc, current))
1176 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1177 err = inode_change_ok(inode, attr);
1182 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1185 if (attr->ia_valid & ATTR_SIZE) {
1186 unsigned long limit;
1187 if (IS_SWAPFILE(inode))
1189 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1190 if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
1191 send_sig(SIGXFSZ, current, 0);
1197 req = fuse_get_req(fc);
1199 return PTR_ERR(req);
1202 fuse_set_nowrite(inode);
1204 memset(&inarg, 0, sizeof(inarg));
1205 memset(&outarg, 0, sizeof(outarg));
1206 iattr_to_fattr(attr, &inarg);
1208 struct fuse_file *ff = file->private_data;
1209 inarg.valid |= FATTR_FH;
1212 if (attr->ia_valid & ATTR_SIZE) {
1213 /* For mandatory locking in truncate */
1214 inarg.valid |= FATTR_LOCKOWNER;
1215 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1217 req->in.h.opcode = FUSE_SETATTR;
1218 req->in.h.nodeid = get_node_id(inode);
1219 req->in.numargs = 1;
1220 req->in.args[0].size = sizeof(inarg);
1221 req->in.args[0].value = &inarg;
1222 req->out.numargs = 1;
1224 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1226 req->out.args[0].size = sizeof(outarg);
1227 req->out.args[0].value = &outarg;
1228 request_send(fc, req);
1229 err = req->out.h.error;
1230 fuse_put_request(fc, req);
1233 fuse_invalidate_attr(inode);
1237 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1238 make_bad_inode(inode);
1243 spin_lock(&fc->lock);
1244 fuse_change_attributes_common(inode, &outarg.attr,
1245 attr_timeout(&outarg));
1246 oldsize = inode->i_size;
1247 i_size_write(inode, outarg.attr.size);
1250 /* NOTE: this may release/reacquire fc->lock */
1251 __fuse_release_nowrite(inode);
1253 spin_unlock(&fc->lock);
1256 * Only call invalidate_inode_pages2() after removing
1257 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1259 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1260 if (outarg.attr.size < oldsize)
1261 fuse_truncate(inode->i_mapping, outarg.attr.size);
1262 invalidate_inode_pages2(inode->i_mapping);
1269 fuse_release_nowrite(inode);
1274 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1276 if (attr->ia_valid & ATTR_FILE)
1277 return fuse_do_setattr(entry, attr, attr->ia_file);
1279 return fuse_do_setattr(entry, attr, NULL);
1282 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1285 struct inode *inode = entry->d_inode;
1286 struct fuse_conn *fc = get_fuse_conn(inode);
1288 if (!fuse_allow_task(fc, current))
1291 return fuse_update_attributes(inode, stat, NULL, NULL);
1294 static int fuse_setxattr(struct dentry *entry, const char *name,
1295 const void *value, size_t size, int flags)
1297 struct inode *inode = entry->d_inode;
1298 struct fuse_conn *fc = get_fuse_conn(inode);
1299 struct fuse_req *req;
1300 struct fuse_setxattr_in inarg;
1303 if (fc->no_setxattr)
1306 req = fuse_get_req(fc);
1308 return PTR_ERR(req);
1310 memset(&inarg, 0, sizeof(inarg));
1312 inarg.flags = flags;
1313 req->in.h.opcode = FUSE_SETXATTR;
1314 req->in.h.nodeid = get_node_id(inode);
1315 req->in.numargs = 3;
1316 req->in.args[0].size = sizeof(inarg);
1317 req->in.args[0].value = &inarg;
1318 req->in.args[1].size = strlen(name) + 1;
1319 req->in.args[1].value = name;
1320 req->in.args[2].size = size;
1321 req->in.args[2].value = value;
1322 request_send(fc, req);
1323 err = req->out.h.error;
1324 fuse_put_request(fc, req);
1325 if (err == -ENOSYS) {
1326 fc->no_setxattr = 1;
1332 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1333 void *value, size_t size)
1335 struct inode *inode = entry->d_inode;
1336 struct fuse_conn *fc = get_fuse_conn(inode);
1337 struct fuse_req *req;
1338 struct fuse_getxattr_in inarg;
1339 struct fuse_getxattr_out outarg;
1342 if (fc->no_getxattr)
1345 req = fuse_get_req(fc);
1347 return PTR_ERR(req);
1349 memset(&inarg, 0, sizeof(inarg));
1351 req->in.h.opcode = FUSE_GETXATTR;
1352 req->in.h.nodeid = get_node_id(inode);
1353 req->in.numargs = 2;
1354 req->in.args[0].size = sizeof(inarg);
1355 req->in.args[0].value = &inarg;
1356 req->in.args[1].size = strlen(name) + 1;
1357 req->in.args[1].value = name;
1358 /* This is really two different operations rolled into one */
1359 req->out.numargs = 1;
1361 req->out.argvar = 1;
1362 req->out.args[0].size = size;
1363 req->out.args[0].value = value;
1365 req->out.args[0].size = sizeof(outarg);
1366 req->out.args[0].value = &outarg;
1368 request_send(fc, req);
1369 ret = req->out.h.error;
1371 ret = size ? req->out.args[0].size : outarg.size;
1373 if (ret == -ENOSYS) {
1374 fc->no_getxattr = 1;
1378 fuse_put_request(fc, req);
1382 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1384 struct inode *inode = entry->d_inode;
1385 struct fuse_conn *fc = get_fuse_conn(inode);
1386 struct fuse_req *req;
1387 struct fuse_getxattr_in inarg;
1388 struct fuse_getxattr_out outarg;
1391 if (!fuse_allow_task(fc, current))
1394 if (fc->no_listxattr)
1397 req = fuse_get_req(fc);
1399 return PTR_ERR(req);
1401 memset(&inarg, 0, sizeof(inarg));
1403 req->in.h.opcode = FUSE_LISTXATTR;
1404 req->in.h.nodeid = get_node_id(inode);
1405 req->in.numargs = 1;
1406 req->in.args[0].size = sizeof(inarg);
1407 req->in.args[0].value = &inarg;
1408 /* This is really two different operations rolled into one */
1409 req->out.numargs = 1;
1411 req->out.argvar = 1;
1412 req->out.args[0].size = size;
1413 req->out.args[0].value = list;
1415 req->out.args[0].size = sizeof(outarg);
1416 req->out.args[0].value = &outarg;
1418 request_send(fc, req);
1419 ret = req->out.h.error;
1421 ret = size ? req->out.args[0].size : outarg.size;
1423 if (ret == -ENOSYS) {
1424 fc->no_listxattr = 1;
1428 fuse_put_request(fc, req);
1432 static int fuse_removexattr(struct dentry *entry, const char *name)
1434 struct inode *inode = entry->d_inode;
1435 struct fuse_conn *fc = get_fuse_conn(inode);
1436 struct fuse_req *req;
1439 if (fc->no_removexattr)
1442 req = fuse_get_req(fc);
1444 return PTR_ERR(req);
1446 req->in.h.opcode = FUSE_REMOVEXATTR;
1447 req->in.h.nodeid = get_node_id(inode);
1448 req->in.numargs = 1;
1449 req->in.args[0].size = strlen(name) + 1;
1450 req->in.args[0].value = name;
1451 request_send(fc, req);
1452 err = req->out.h.error;
1453 fuse_put_request(fc, req);
1454 if (err == -ENOSYS) {
1455 fc->no_removexattr = 1;
1461 static const struct inode_operations fuse_dir_inode_operations = {
1462 .lookup = fuse_lookup,
1463 .mkdir = fuse_mkdir,
1464 .symlink = fuse_symlink,
1465 .unlink = fuse_unlink,
1466 .rmdir = fuse_rmdir,
1467 .rename = fuse_rename,
1469 .setattr = fuse_setattr,
1470 .create = fuse_create,
1471 .mknod = fuse_mknod,
1472 .permission = fuse_permission,
1473 .getattr = fuse_getattr,
1474 .setxattr = fuse_setxattr,
1475 .getxattr = fuse_getxattr,
1476 .listxattr = fuse_listxattr,
1477 .removexattr = fuse_removexattr,
1480 static const struct file_operations fuse_dir_operations = {
1481 .llseek = generic_file_llseek,
1482 .read = generic_read_dir,
1483 .readdir = fuse_readdir,
1484 .open = fuse_dir_open,
1485 .release = fuse_dir_release,
1486 .fsync = fuse_dir_fsync,
1489 static const struct inode_operations fuse_common_inode_operations = {
1490 .setattr = fuse_setattr,
1491 .permission = fuse_permission,
1492 .getattr = fuse_getattr,
1493 .setxattr = fuse_setxattr,
1494 .getxattr = fuse_getxattr,
1495 .listxattr = fuse_listxattr,
1496 .removexattr = fuse_removexattr,
1499 static const struct inode_operations fuse_symlink_inode_operations = {
1500 .setattr = fuse_setattr,
1501 .follow_link = fuse_follow_link,
1502 .put_link = fuse_put_link,
1503 .readlink = generic_readlink,
1504 .getattr = fuse_getattr,
1505 .setxattr = fuse_setxattr,
1506 .getxattr = fuse_getxattr,
1507 .listxattr = fuse_listxattr,
1508 .removexattr = fuse_removexattr,
1511 void fuse_init_common(struct inode *inode)
1513 inode->i_op = &fuse_common_inode_operations;
1516 void fuse_init_dir(struct inode *inode)
1518 inode->i_op = &fuse_dir_inode_operations;
1519 inode->i_fop = &fuse_dir_operations;
1522 void fuse_init_symlink(struct inode *inode)
1524 inode->i_op = &fuse_symlink_inode_operations;