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.
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>
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 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_conn *fc, struct fuse_req *req,
116 u64 nodeid, struct qstr *name,
117 struct fuse_entry_out *outarg)
119 memset(outarg, 0, sizeof(struct fuse_entry_out));
120 req->in.h.opcode = FUSE_LOOKUP;
121 req->in.h.nodeid = nodeid;
123 req->in.args[0].size = name->len + 1;
124 req->in.args[0].value = name->name;
125 req->out.numargs = 1;
127 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
129 req->out.args[0].size = sizeof(struct fuse_entry_out);
130 req->out.args[0].value = outarg;
133 u64 fuse_get_attr_version(struct fuse_conn *fc)
138 * The spin lock isn't actually needed on 64bit archs, but we
139 * don't yet care too much about such optimizations.
141 spin_lock(&fc->lock);
142 curr_version = fc->attr_version;
143 spin_unlock(&fc->lock);
149 * Check whether the dentry is still valid
151 * If the entry validity timeout has expired and the dentry is
152 * positive, try to redo the lookup. If the lookup results in a
153 * different inode, then let the VFS invalidate the dentry and redo
154 * the lookup once more. If the lookup results in the same inode,
155 * then refresh the attributes, timeouts and mark the dentry valid.
157 static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
161 inode = ACCESS_ONCE(entry->d_inode);
162 if (inode && is_bad_inode(inode))
164 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
166 struct fuse_entry_out outarg;
167 struct fuse_conn *fc;
168 struct fuse_req *req;
169 struct fuse_forget_link *forget;
170 struct dentry *parent;
173 /* For negative dentries, always do a fresh lookup */
177 if (nd && (nd->flags & LOOKUP_RCU))
180 fc = get_fuse_conn(inode);
181 req = fuse_get_req(fc);
185 forget = fuse_alloc_forget();
187 fuse_put_request(fc, req);
191 attr_version = fuse_get_attr_version(fc);
193 parent = dget_parent(entry);
194 fuse_lookup_init(fc, req, get_node_id(parent->d_inode),
195 &entry->d_name, &outarg);
196 fuse_request_send(fc, req);
198 err = req->out.h.error;
199 fuse_put_request(fc, req);
200 /* Zero nodeid is same as -ENOENT */
201 if (!err && !outarg.nodeid)
204 struct fuse_inode *fi = get_fuse_inode(inode);
205 if (outarg.nodeid != get_node_id(inode)) {
206 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
209 spin_lock(&fc->lock);
211 spin_unlock(&fc->lock);
214 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
217 fuse_change_attributes(inode, &outarg.attr,
218 entry_attr_timeout(&outarg),
220 fuse_change_entry_timeout(entry, &outarg);
225 static int invalid_nodeid(u64 nodeid)
227 return !nodeid || nodeid == FUSE_ROOT_ID;
230 const struct dentry_operations fuse_dentry_operations = {
231 .d_revalidate = fuse_dentry_revalidate,
234 int fuse_valid_type(int m)
236 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
237 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
241 * Add a directory inode to a dentry, ensuring that no other dentry
242 * refers to this inode. Called with fc->inst_mutex.
244 static struct dentry *fuse_d_add_directory(struct dentry *entry,
247 struct dentry *alias = d_find_alias(inode);
248 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
249 /* This tries to shrink the subtree below alias */
250 fuse_invalidate_entry(alias);
252 if (!hlist_empty(&inode->i_dentry))
253 return ERR_PTR(-EBUSY);
257 return d_splice_alias(inode, entry);
260 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
261 struct fuse_entry_out *outarg, struct inode **inode)
263 struct fuse_conn *fc = get_fuse_conn_super(sb);
264 struct fuse_req *req;
265 struct fuse_forget_link *forget;
271 if (name->len > FUSE_NAME_MAX)
274 req = fuse_get_req(fc);
279 forget = fuse_alloc_forget();
282 fuse_put_request(fc, req);
286 attr_version = fuse_get_attr_version(fc);
288 fuse_lookup_init(fc, req, nodeid, name, outarg);
289 fuse_request_send(fc, req);
290 err = req->out.h.error;
291 fuse_put_request(fc, req);
292 /* Zero nodeid is same as -ENOENT, but with valid timeout */
293 if (err || !outarg->nodeid)
299 if (!fuse_valid_type(outarg->attr.mode))
302 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
303 &outarg->attr, entry_attr_timeout(outarg),
307 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
318 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
319 struct nameidata *nd)
322 struct fuse_entry_out outarg;
324 struct dentry *newent;
325 struct fuse_conn *fc = get_fuse_conn(dir);
326 bool outarg_valid = true;
328 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
330 if (err == -ENOENT) {
331 outarg_valid = false;
338 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
341 if (inode && S_ISDIR(inode->i_mode)) {
342 mutex_lock(&fc->inst_mutex);
343 newent = fuse_d_add_directory(entry, inode);
344 mutex_unlock(&fc->inst_mutex);
345 err = PTR_ERR(newent);
349 newent = d_splice_alias(inode, entry);
352 entry = newent ? newent : entry;
354 fuse_change_entry_timeout(entry, &outarg);
356 fuse_invalidate_entry_cache(entry);
367 * Atomic create+open operation
369 * If the filesystem doesn't support this, then fall back to separate
370 * 'mknod' + 'open' requests.
372 static int fuse_create_open(struct inode *dir, struct dentry *entry,
373 struct file *file, unsigned flags,
374 umode_t mode, int *opened)
378 struct fuse_conn *fc = get_fuse_conn(dir);
379 struct fuse_req *req;
380 struct fuse_forget_link *forget;
381 struct fuse_create_in inarg;
382 struct fuse_open_out outopen;
383 struct fuse_entry_out outentry;
384 struct fuse_file *ff;
386 forget = fuse_alloc_forget();
391 req = fuse_get_req(fc);
394 goto out_put_forget_req;
397 ff = fuse_file_alloc(fc);
399 goto out_put_request;
402 mode &= ~current_umask();
405 memset(&inarg, 0, sizeof(inarg));
406 memset(&outentry, 0, sizeof(outentry));
409 inarg.umask = current_umask();
410 req->in.h.opcode = FUSE_CREATE;
411 req->in.h.nodeid = get_node_id(dir);
413 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
415 req->in.args[0].value = &inarg;
416 req->in.args[1].size = entry->d_name.len + 1;
417 req->in.args[1].value = entry->d_name.name;
418 req->out.numargs = 2;
420 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
422 req->out.args[0].size = sizeof(outentry);
423 req->out.args[0].value = &outentry;
424 req->out.args[1].size = sizeof(outopen);
425 req->out.args[1].value = &outopen;
426 fuse_request_send(fc, req);
427 err = req->out.h.error;
432 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
435 fuse_put_request(fc, req);
437 ff->nodeid = outentry.nodeid;
438 ff->open_flags = outopen.open_flags;
439 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
440 &outentry.attr, entry_attr_timeout(&outentry), 0);
442 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
443 fuse_sync_release(ff, flags);
444 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
449 d_instantiate(entry, inode);
450 fuse_change_entry_timeout(entry, &outentry);
451 fuse_invalidate_attr(dir);
452 err = finish_open(file, entry, generic_file_open, opened);
454 fuse_sync_release(ff, flags);
456 file->private_data = fuse_file_get(ff);
457 fuse_finish_open(inode, file);
464 fuse_put_request(fc, req);
471 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
472 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
473 struct file *file, unsigned flags,
474 umode_t mode, int *opened)
477 struct fuse_conn *fc = get_fuse_conn(dir);
478 struct dentry *res = NULL;
480 if (d_unhashed(entry)) {
481 res = fuse_lookup(dir, entry, NULL);
489 if (!(flags & O_CREAT) || entry->d_inode)
493 *opened |= FILE_CREATED;
498 err = fuse_create_open(dir, entry, file, flags, mode, opened);
499 if (err == -ENOSYS) {
508 err = fuse_mknod(dir, entry, mode, 0);
512 finish_no_open(file, res);
517 * Code shared between mknod, mkdir, symlink and link
519 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
520 struct inode *dir, struct dentry *entry,
523 struct fuse_entry_out outarg;
526 struct fuse_forget_link *forget;
528 forget = fuse_alloc_forget();
530 fuse_put_request(fc, req);
534 memset(&outarg, 0, sizeof(outarg));
535 req->in.h.nodeid = get_node_id(dir);
536 req->out.numargs = 1;
538 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
540 req->out.args[0].size = sizeof(outarg);
541 req->out.args[0].value = &outarg;
542 fuse_request_send(fc, req);
543 err = req->out.h.error;
544 fuse_put_request(fc, req);
546 goto out_put_forget_req;
549 if (invalid_nodeid(outarg.nodeid))
550 goto out_put_forget_req;
552 if ((outarg.attr.mode ^ mode) & S_IFMT)
553 goto out_put_forget_req;
555 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
556 &outarg.attr, entry_attr_timeout(&outarg), 0);
558 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
563 if (S_ISDIR(inode->i_mode)) {
564 struct dentry *alias;
565 mutex_lock(&fc->inst_mutex);
566 alias = d_find_alias(inode);
568 /* New directory must have moved since mkdir */
569 mutex_unlock(&fc->inst_mutex);
574 d_instantiate(entry, inode);
575 mutex_unlock(&fc->inst_mutex);
577 d_instantiate(entry, inode);
579 fuse_change_entry_timeout(entry, &outarg);
580 fuse_invalidate_attr(dir);
588 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
591 struct fuse_mknod_in inarg;
592 struct fuse_conn *fc = get_fuse_conn(dir);
593 struct fuse_req *req = fuse_get_req(fc);
598 mode &= ~current_umask();
600 memset(&inarg, 0, sizeof(inarg));
602 inarg.rdev = new_encode_dev(rdev);
603 inarg.umask = current_umask();
604 req->in.h.opcode = FUSE_MKNOD;
606 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
608 req->in.args[0].value = &inarg;
609 req->in.args[1].size = entry->d_name.len + 1;
610 req->in.args[1].value = entry->d_name.name;
611 return create_new_entry(fc, req, dir, entry, mode);
614 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
615 struct nameidata *nd)
617 return fuse_mknod(dir, entry, mode, 0);
620 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
622 struct fuse_mkdir_in inarg;
623 struct fuse_conn *fc = get_fuse_conn(dir);
624 struct fuse_req *req = fuse_get_req(fc);
629 mode &= ~current_umask();
631 memset(&inarg, 0, sizeof(inarg));
633 inarg.umask = current_umask();
634 req->in.h.opcode = FUSE_MKDIR;
636 req->in.args[0].size = sizeof(inarg);
637 req->in.args[0].value = &inarg;
638 req->in.args[1].size = entry->d_name.len + 1;
639 req->in.args[1].value = entry->d_name.name;
640 return create_new_entry(fc, req, dir, entry, S_IFDIR);
643 static int fuse_symlink(struct inode *dir, struct dentry *entry,
646 struct fuse_conn *fc = get_fuse_conn(dir);
647 unsigned len = strlen(link) + 1;
648 struct fuse_req *req = fuse_get_req(fc);
652 req->in.h.opcode = FUSE_SYMLINK;
654 req->in.args[0].size = entry->d_name.len + 1;
655 req->in.args[0].value = entry->d_name.name;
656 req->in.args[1].size = len;
657 req->in.args[1].value = link;
658 return create_new_entry(fc, req, dir, entry, S_IFLNK);
661 static int fuse_unlink(struct inode *dir, struct dentry *entry)
664 struct fuse_conn *fc = get_fuse_conn(dir);
665 struct fuse_req *req = fuse_get_req(fc);
669 req->in.h.opcode = FUSE_UNLINK;
670 req->in.h.nodeid = get_node_id(dir);
672 req->in.args[0].size = entry->d_name.len + 1;
673 req->in.args[0].value = entry->d_name.name;
674 fuse_request_send(fc, req);
675 err = req->out.h.error;
676 fuse_put_request(fc, req);
678 struct inode *inode = entry->d_inode;
679 struct fuse_inode *fi = get_fuse_inode(inode);
681 spin_lock(&fc->lock);
682 fi->attr_version = ++fc->attr_version;
684 spin_unlock(&fc->lock);
685 fuse_invalidate_attr(inode);
686 fuse_invalidate_attr(dir);
687 fuse_invalidate_entry_cache(entry);
688 } else if (err == -EINTR)
689 fuse_invalidate_entry(entry);
693 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
696 struct fuse_conn *fc = get_fuse_conn(dir);
697 struct fuse_req *req = fuse_get_req(fc);
701 req->in.h.opcode = FUSE_RMDIR;
702 req->in.h.nodeid = get_node_id(dir);
704 req->in.args[0].size = entry->d_name.len + 1;
705 req->in.args[0].value = entry->d_name.name;
706 fuse_request_send(fc, req);
707 err = req->out.h.error;
708 fuse_put_request(fc, req);
710 clear_nlink(entry->d_inode);
711 fuse_invalidate_attr(dir);
712 fuse_invalidate_entry_cache(entry);
713 } else if (err == -EINTR)
714 fuse_invalidate_entry(entry);
718 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
719 struct inode *newdir, struct dentry *newent)
722 struct fuse_rename_in inarg;
723 struct fuse_conn *fc = get_fuse_conn(olddir);
724 struct fuse_req *req = fuse_get_req(fc);
729 memset(&inarg, 0, sizeof(inarg));
730 inarg.newdir = get_node_id(newdir);
731 req->in.h.opcode = FUSE_RENAME;
732 req->in.h.nodeid = get_node_id(olddir);
734 req->in.args[0].size = sizeof(inarg);
735 req->in.args[0].value = &inarg;
736 req->in.args[1].size = oldent->d_name.len + 1;
737 req->in.args[1].value = oldent->d_name.name;
738 req->in.args[2].size = newent->d_name.len + 1;
739 req->in.args[2].value = newent->d_name.name;
740 fuse_request_send(fc, req);
741 err = req->out.h.error;
742 fuse_put_request(fc, req);
745 fuse_invalidate_attr(oldent->d_inode);
747 fuse_invalidate_attr(olddir);
748 if (olddir != newdir)
749 fuse_invalidate_attr(newdir);
751 /* newent will end up negative */
752 if (newent->d_inode) {
753 fuse_invalidate_attr(newent->d_inode);
754 fuse_invalidate_entry_cache(newent);
756 } else if (err == -EINTR) {
757 /* If request was interrupted, DEITY only knows if the
758 rename actually took place. If the invalidation
759 fails (e.g. some process has CWD under the renamed
760 directory), then there can be inconsistency between
761 the dcache and the real filesystem. Tough luck. */
762 fuse_invalidate_entry(oldent);
764 fuse_invalidate_entry(newent);
770 static int fuse_link(struct dentry *entry, struct inode *newdir,
771 struct dentry *newent)
774 struct fuse_link_in inarg;
775 struct inode *inode = entry->d_inode;
776 struct fuse_conn *fc = get_fuse_conn(inode);
777 struct fuse_req *req = fuse_get_req(fc);
781 memset(&inarg, 0, sizeof(inarg));
782 inarg.oldnodeid = get_node_id(inode);
783 req->in.h.opcode = FUSE_LINK;
785 req->in.args[0].size = sizeof(inarg);
786 req->in.args[0].value = &inarg;
787 req->in.args[1].size = newent->d_name.len + 1;
788 req->in.args[1].value = newent->d_name.name;
789 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
790 /* Contrary to "normal" filesystems it can happen that link
791 makes two "logical" inodes point to the same "physical"
792 inode. We invalidate the attributes of the old one, so it
793 will reflect changes in the backing inode (link count,
797 struct fuse_inode *fi = get_fuse_inode(inode);
799 spin_lock(&fc->lock);
800 fi->attr_version = ++fc->attr_version;
802 spin_unlock(&fc->lock);
803 fuse_invalidate_attr(inode);
804 } else if (err == -EINTR) {
805 fuse_invalidate_attr(inode);
810 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
813 unsigned int blkbits;
815 stat->dev = inode->i_sb->s_dev;
816 stat->ino = attr->ino;
817 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
818 stat->nlink = attr->nlink;
819 stat->uid = attr->uid;
820 stat->gid = attr->gid;
821 stat->rdev = inode->i_rdev;
822 stat->atime.tv_sec = attr->atime;
823 stat->atime.tv_nsec = attr->atimensec;
824 stat->mtime.tv_sec = attr->mtime;
825 stat->mtime.tv_nsec = attr->mtimensec;
826 stat->ctime.tv_sec = attr->ctime;
827 stat->ctime.tv_nsec = attr->ctimensec;
828 stat->size = attr->size;
829 stat->blocks = attr->blocks;
831 if (attr->blksize != 0)
832 blkbits = ilog2(attr->blksize);
834 blkbits = inode->i_sb->s_blocksize_bits;
836 stat->blksize = 1 << blkbits;
839 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
843 struct fuse_getattr_in inarg;
844 struct fuse_attr_out outarg;
845 struct fuse_conn *fc = get_fuse_conn(inode);
846 struct fuse_req *req;
849 req = fuse_get_req(fc);
853 attr_version = fuse_get_attr_version(fc);
855 memset(&inarg, 0, sizeof(inarg));
856 memset(&outarg, 0, sizeof(outarg));
857 /* Directories have separate file-handle space */
858 if (file && S_ISREG(inode->i_mode)) {
859 struct fuse_file *ff = file->private_data;
861 inarg.getattr_flags |= FUSE_GETATTR_FH;
864 req->in.h.opcode = FUSE_GETATTR;
865 req->in.h.nodeid = get_node_id(inode);
867 req->in.args[0].size = sizeof(inarg);
868 req->in.args[0].value = &inarg;
869 req->out.numargs = 1;
871 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
873 req->out.args[0].size = sizeof(outarg);
874 req->out.args[0].value = &outarg;
875 fuse_request_send(fc, req);
876 err = req->out.h.error;
877 fuse_put_request(fc, req);
879 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
880 make_bad_inode(inode);
883 fuse_change_attributes(inode, &outarg.attr,
884 attr_timeout(&outarg),
887 fuse_fillattr(inode, &outarg.attr, stat);
893 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
894 struct file *file, bool *refreshed)
896 struct fuse_inode *fi = get_fuse_inode(inode);
900 if (fi->i_time < get_jiffies_64()) {
902 err = fuse_do_getattr(inode, stat, file);
907 generic_fillattr(inode, stat);
908 stat->mode = fi->orig_i_mode;
909 stat->ino = fi->orig_ino;
913 if (refreshed != NULL)
919 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
920 u64 child_nodeid, struct qstr *name)
923 struct inode *parent;
925 struct dentry *entry;
927 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
931 mutex_lock(&parent->i_mutex);
932 if (!S_ISDIR(parent->i_mode))
936 dir = d_find_alias(parent);
940 entry = d_lookup(dir, name);
945 fuse_invalidate_attr(parent);
946 fuse_invalidate_entry(entry);
948 if (child_nodeid != 0 && entry->d_inode) {
949 mutex_lock(&entry->d_inode->i_mutex);
950 if (get_node_id(entry->d_inode) != child_nodeid) {
954 if (d_mountpoint(entry)) {
958 if (S_ISDIR(entry->d_inode->i_mode)) {
959 shrink_dcache_parent(entry);
960 if (!simple_empty(entry)) {
964 entry->d_inode->i_flags |= S_DEAD;
967 clear_nlink(entry->d_inode);
970 mutex_unlock(&entry->d_inode->i_mutex);
979 mutex_unlock(&parent->i_mutex);
985 * Calling into a user-controlled filesystem gives the filesystem
986 * daemon ptrace-like capabilities over the requester process. This
987 * means, that the filesystem daemon is able to record the exact
988 * filesystem operations performed, and can also control the behavior
989 * of the requester process in otherwise impossible ways. For example
990 * it can delay the operation for arbitrary length of time allowing
991 * DoS against the requester.
993 * For this reason only those processes can call into the filesystem,
994 * for which the owner of the mount has ptrace privilege. This
995 * excludes processes started by other users, suid or sgid processes.
997 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
999 const struct cred *cred;
1002 if (fc->flags & FUSE_ALLOW_OTHER)
1007 cred = __task_cred(task);
1008 if (cred->euid == fc->user_id &&
1009 cred->suid == fc->user_id &&
1010 cred->uid == fc->user_id &&
1011 cred->egid == fc->group_id &&
1012 cred->sgid == fc->group_id &&
1013 cred->gid == fc->group_id)
1020 static int fuse_access(struct inode *inode, int mask)
1022 struct fuse_conn *fc = get_fuse_conn(inode);
1023 struct fuse_req *req;
1024 struct fuse_access_in inarg;
1030 req = fuse_get_req(fc);
1032 return PTR_ERR(req);
1034 memset(&inarg, 0, sizeof(inarg));
1035 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1036 req->in.h.opcode = FUSE_ACCESS;
1037 req->in.h.nodeid = get_node_id(inode);
1038 req->in.numargs = 1;
1039 req->in.args[0].size = sizeof(inarg);
1040 req->in.args[0].value = &inarg;
1041 fuse_request_send(fc, req);
1042 err = req->out.h.error;
1043 fuse_put_request(fc, req);
1044 if (err == -ENOSYS) {
1051 static int fuse_perm_getattr(struct inode *inode, int mask)
1053 if (mask & MAY_NOT_BLOCK)
1056 return fuse_do_getattr(inode, NULL, NULL);
1060 * Check permission. The two basic access models of FUSE are:
1062 * 1) Local access checking ('default_permissions' mount option) based
1063 * on file mode. This is the plain old disk filesystem permission
1066 * 2) "Remote" access checking, where server is responsible for
1067 * checking permission in each inode operation. An exception to this
1068 * is if ->permission() was invoked from sys_access() in which case an
1069 * access request is sent. Execute permission is still checked
1070 * locally based on file mode.
1072 static int fuse_permission(struct inode *inode, int mask)
1074 struct fuse_conn *fc = get_fuse_conn(inode);
1075 bool refreshed = false;
1078 if (!fuse_allow_task(fc, current))
1082 * If attributes are needed, refresh them before proceeding
1084 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1085 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1086 struct fuse_inode *fi = get_fuse_inode(inode);
1088 if (fi->i_time < get_jiffies_64()) {
1091 err = fuse_perm_getattr(inode, mask);
1097 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1098 err = generic_permission(inode, mask);
1100 /* If permission is denied, try to refresh file
1101 attributes. This is also needed, because the root
1102 node will at first have no permissions */
1103 if (err == -EACCES && !refreshed) {
1104 err = fuse_perm_getattr(inode, mask);
1106 err = generic_permission(inode, mask);
1109 /* Note: the opposite of the above test does not
1110 exist. So if permissions are revoked this won't be
1111 noticed immediately, only after the attribute
1112 timeout has expired */
1113 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1114 if (mask & MAY_NOT_BLOCK)
1117 err = fuse_access(inode, mask);
1118 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1119 if (!(inode->i_mode & S_IXUGO)) {
1123 err = fuse_perm_getattr(inode, mask);
1124 if (!err && !(inode->i_mode & S_IXUGO))
1131 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1132 void *dstbuf, filldir_t filldir)
1134 while (nbytes >= FUSE_NAME_OFFSET) {
1135 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1136 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1138 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1140 if (reclen > nbytes)
1143 over = filldir(dstbuf, dirent->name, dirent->namelen,
1144 file->f_pos, dirent->ino, dirent->type);
1150 file->f_pos = dirent->off;
1156 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1161 struct inode *inode = file->f_path.dentry->d_inode;
1162 struct fuse_conn *fc = get_fuse_conn(inode);
1163 struct fuse_req *req;
1165 if (is_bad_inode(inode))
1168 req = fuse_get_req(fc);
1170 return PTR_ERR(req);
1172 page = alloc_page(GFP_KERNEL);
1174 fuse_put_request(fc, req);
1177 req->out.argpages = 1;
1179 req->pages[0] = page;
1180 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1181 fuse_request_send(fc, req);
1182 nbytes = req->out.args[0].size;
1183 err = req->out.h.error;
1184 fuse_put_request(fc, req);
1186 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1190 fuse_invalidate_attr(inode); /* atime changed */
1194 static char *read_link(struct dentry *dentry)
1196 struct inode *inode = dentry->d_inode;
1197 struct fuse_conn *fc = get_fuse_conn(inode);
1198 struct fuse_req *req = fuse_get_req(fc);
1202 return ERR_CAST(req);
1204 link = (char *) __get_free_page(GFP_KERNEL);
1206 link = ERR_PTR(-ENOMEM);
1209 req->in.h.opcode = FUSE_READLINK;
1210 req->in.h.nodeid = get_node_id(inode);
1211 req->out.argvar = 1;
1212 req->out.numargs = 1;
1213 req->out.args[0].size = PAGE_SIZE - 1;
1214 req->out.args[0].value = link;
1215 fuse_request_send(fc, req);
1216 if (req->out.h.error) {
1217 free_page((unsigned long) link);
1218 link = ERR_PTR(req->out.h.error);
1220 link[req->out.args[0].size] = '\0';
1222 fuse_put_request(fc, req);
1223 fuse_invalidate_attr(inode); /* atime changed */
1227 static void free_link(char *link)
1230 free_page((unsigned long) link);
1233 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1235 nd_set_link(nd, read_link(dentry));
1239 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1241 free_link(nd_get_link(nd));
1244 static int fuse_dir_open(struct inode *inode, struct file *file)
1246 return fuse_open_common(inode, file, true);
1249 static int fuse_dir_release(struct inode *inode, struct file *file)
1251 fuse_release_common(file, FUSE_RELEASEDIR);
1256 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1259 return fuse_fsync_common(file, start, end, datasync, 1);
1262 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1265 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1267 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1271 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1274 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1277 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1282 return fuse_ioctl_common(file, cmd, arg,
1283 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1286 static bool update_mtime(unsigned ivalid)
1288 /* Always update if mtime is explicitly set */
1289 if (ivalid & ATTR_MTIME_SET)
1292 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1293 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1296 /* In all other cases update */
1300 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1302 unsigned ivalid = iattr->ia_valid;
1304 if (ivalid & ATTR_MODE)
1305 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1306 if (ivalid & ATTR_UID)
1307 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1308 if (ivalid & ATTR_GID)
1309 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1310 if (ivalid & ATTR_SIZE)
1311 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1312 if (ivalid & ATTR_ATIME) {
1313 arg->valid |= FATTR_ATIME;
1314 arg->atime = iattr->ia_atime.tv_sec;
1315 arg->atimensec = iattr->ia_atime.tv_nsec;
1316 if (!(ivalid & ATTR_ATIME_SET))
1317 arg->valid |= FATTR_ATIME_NOW;
1319 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1320 arg->valid |= FATTR_MTIME;
1321 arg->mtime = iattr->ia_mtime.tv_sec;
1322 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1323 if (!(ivalid & ATTR_MTIME_SET))
1324 arg->valid |= FATTR_MTIME_NOW;
1329 * Prevent concurrent writepages on inode
1331 * This is done by adding a negative bias to the inode write counter
1332 * and waiting for all pending writes to finish.
1334 void fuse_set_nowrite(struct inode *inode)
1336 struct fuse_conn *fc = get_fuse_conn(inode);
1337 struct fuse_inode *fi = get_fuse_inode(inode);
1339 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1341 spin_lock(&fc->lock);
1342 BUG_ON(fi->writectr < 0);
1343 fi->writectr += FUSE_NOWRITE;
1344 spin_unlock(&fc->lock);
1345 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1349 * Allow writepages on inode
1351 * Remove the bias from the writecounter and send any queued
1354 static void __fuse_release_nowrite(struct inode *inode)
1356 struct fuse_inode *fi = get_fuse_inode(inode);
1358 BUG_ON(fi->writectr != FUSE_NOWRITE);
1360 fuse_flush_writepages(inode);
1363 void fuse_release_nowrite(struct inode *inode)
1365 struct fuse_conn *fc = get_fuse_conn(inode);
1367 spin_lock(&fc->lock);
1368 __fuse_release_nowrite(inode);
1369 spin_unlock(&fc->lock);
1373 * Set attributes, and at the same time refresh them.
1375 * Truncation is slightly complicated, because the 'truncate' request
1376 * may fail, in which case we don't want to touch the mapping.
1377 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1378 * and the actual truncation by hand.
1380 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1383 struct inode *inode = entry->d_inode;
1384 struct fuse_conn *fc = get_fuse_conn(inode);
1385 struct fuse_req *req;
1386 struct fuse_setattr_in inarg;
1387 struct fuse_attr_out outarg;
1388 bool is_truncate = false;
1392 if (!fuse_allow_task(fc, current))
1395 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1396 attr->ia_valid |= ATTR_FORCE;
1398 err = inode_change_ok(inode, attr);
1402 if (attr->ia_valid & ATTR_OPEN) {
1403 if (fc->atomic_o_trunc)
1408 if (attr->ia_valid & ATTR_SIZE)
1411 req = fuse_get_req(fc);
1413 return PTR_ERR(req);
1416 fuse_set_nowrite(inode);
1418 memset(&inarg, 0, sizeof(inarg));
1419 memset(&outarg, 0, sizeof(outarg));
1420 iattr_to_fattr(attr, &inarg);
1422 struct fuse_file *ff = file->private_data;
1423 inarg.valid |= FATTR_FH;
1426 if (attr->ia_valid & ATTR_SIZE) {
1427 /* For mandatory locking in truncate */
1428 inarg.valid |= FATTR_LOCKOWNER;
1429 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1431 req->in.h.opcode = FUSE_SETATTR;
1432 req->in.h.nodeid = get_node_id(inode);
1433 req->in.numargs = 1;
1434 req->in.args[0].size = sizeof(inarg);
1435 req->in.args[0].value = &inarg;
1436 req->out.numargs = 1;
1438 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1440 req->out.args[0].size = sizeof(outarg);
1441 req->out.args[0].value = &outarg;
1442 fuse_request_send(fc, req);
1443 err = req->out.h.error;
1444 fuse_put_request(fc, req);
1447 fuse_invalidate_attr(inode);
1451 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1452 make_bad_inode(inode);
1457 spin_lock(&fc->lock);
1458 fuse_change_attributes_common(inode, &outarg.attr,
1459 attr_timeout(&outarg));
1460 oldsize = inode->i_size;
1461 i_size_write(inode, outarg.attr.size);
1464 /* NOTE: this may release/reacquire fc->lock */
1465 __fuse_release_nowrite(inode);
1467 spin_unlock(&fc->lock);
1470 * Only call invalidate_inode_pages2() after removing
1471 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1473 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1474 truncate_pagecache(inode, oldsize, outarg.attr.size);
1475 invalidate_inode_pages2(inode->i_mapping);
1482 fuse_release_nowrite(inode);
1487 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1489 if (attr->ia_valid & ATTR_FILE)
1490 return fuse_do_setattr(entry, attr, attr->ia_file);
1492 return fuse_do_setattr(entry, attr, NULL);
1495 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1498 struct inode *inode = entry->d_inode;
1499 struct fuse_conn *fc = get_fuse_conn(inode);
1501 if (!fuse_allow_task(fc, current))
1504 return fuse_update_attributes(inode, stat, NULL, NULL);
1507 static int fuse_setxattr(struct dentry *entry, const char *name,
1508 const void *value, size_t size, int flags)
1510 struct inode *inode = entry->d_inode;
1511 struct fuse_conn *fc = get_fuse_conn(inode);
1512 struct fuse_req *req;
1513 struct fuse_setxattr_in inarg;
1516 if (fc->no_setxattr)
1519 req = fuse_get_req(fc);
1521 return PTR_ERR(req);
1523 memset(&inarg, 0, sizeof(inarg));
1525 inarg.flags = flags;
1526 req->in.h.opcode = FUSE_SETXATTR;
1527 req->in.h.nodeid = get_node_id(inode);
1528 req->in.numargs = 3;
1529 req->in.args[0].size = sizeof(inarg);
1530 req->in.args[0].value = &inarg;
1531 req->in.args[1].size = strlen(name) + 1;
1532 req->in.args[1].value = name;
1533 req->in.args[2].size = size;
1534 req->in.args[2].value = value;
1535 fuse_request_send(fc, req);
1536 err = req->out.h.error;
1537 fuse_put_request(fc, req);
1538 if (err == -ENOSYS) {
1539 fc->no_setxattr = 1;
1545 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1546 void *value, size_t size)
1548 struct inode *inode = entry->d_inode;
1549 struct fuse_conn *fc = get_fuse_conn(inode);
1550 struct fuse_req *req;
1551 struct fuse_getxattr_in inarg;
1552 struct fuse_getxattr_out outarg;
1555 if (fc->no_getxattr)
1558 req = fuse_get_req(fc);
1560 return PTR_ERR(req);
1562 memset(&inarg, 0, sizeof(inarg));
1564 req->in.h.opcode = FUSE_GETXATTR;
1565 req->in.h.nodeid = get_node_id(inode);
1566 req->in.numargs = 2;
1567 req->in.args[0].size = sizeof(inarg);
1568 req->in.args[0].value = &inarg;
1569 req->in.args[1].size = strlen(name) + 1;
1570 req->in.args[1].value = name;
1571 /* This is really two different operations rolled into one */
1572 req->out.numargs = 1;
1574 req->out.argvar = 1;
1575 req->out.args[0].size = size;
1576 req->out.args[0].value = value;
1578 req->out.args[0].size = sizeof(outarg);
1579 req->out.args[0].value = &outarg;
1581 fuse_request_send(fc, req);
1582 ret = req->out.h.error;
1584 ret = size ? req->out.args[0].size : outarg.size;
1586 if (ret == -ENOSYS) {
1587 fc->no_getxattr = 1;
1591 fuse_put_request(fc, req);
1595 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1597 struct inode *inode = entry->d_inode;
1598 struct fuse_conn *fc = get_fuse_conn(inode);
1599 struct fuse_req *req;
1600 struct fuse_getxattr_in inarg;
1601 struct fuse_getxattr_out outarg;
1604 if (!fuse_allow_task(fc, current))
1607 if (fc->no_listxattr)
1610 req = fuse_get_req(fc);
1612 return PTR_ERR(req);
1614 memset(&inarg, 0, sizeof(inarg));
1616 req->in.h.opcode = FUSE_LISTXATTR;
1617 req->in.h.nodeid = get_node_id(inode);
1618 req->in.numargs = 1;
1619 req->in.args[0].size = sizeof(inarg);
1620 req->in.args[0].value = &inarg;
1621 /* This is really two different operations rolled into one */
1622 req->out.numargs = 1;
1624 req->out.argvar = 1;
1625 req->out.args[0].size = size;
1626 req->out.args[0].value = list;
1628 req->out.args[0].size = sizeof(outarg);
1629 req->out.args[0].value = &outarg;
1631 fuse_request_send(fc, req);
1632 ret = req->out.h.error;
1634 ret = size ? req->out.args[0].size : outarg.size;
1636 if (ret == -ENOSYS) {
1637 fc->no_listxattr = 1;
1641 fuse_put_request(fc, req);
1645 static int fuse_removexattr(struct dentry *entry, const char *name)
1647 struct inode *inode = entry->d_inode;
1648 struct fuse_conn *fc = get_fuse_conn(inode);
1649 struct fuse_req *req;
1652 if (fc->no_removexattr)
1655 req = fuse_get_req(fc);
1657 return PTR_ERR(req);
1659 req->in.h.opcode = FUSE_REMOVEXATTR;
1660 req->in.h.nodeid = get_node_id(inode);
1661 req->in.numargs = 1;
1662 req->in.args[0].size = strlen(name) + 1;
1663 req->in.args[0].value = name;
1664 fuse_request_send(fc, req);
1665 err = req->out.h.error;
1666 fuse_put_request(fc, req);
1667 if (err == -ENOSYS) {
1668 fc->no_removexattr = 1;
1674 static const struct inode_operations fuse_dir_inode_operations = {
1675 .lookup = fuse_lookup,
1676 .mkdir = fuse_mkdir,
1677 .symlink = fuse_symlink,
1678 .unlink = fuse_unlink,
1679 .rmdir = fuse_rmdir,
1680 .rename = fuse_rename,
1682 .setattr = fuse_setattr,
1683 .create = fuse_create,
1684 .atomic_open = fuse_atomic_open,
1685 .mknod = fuse_mknod,
1686 .permission = fuse_permission,
1687 .getattr = fuse_getattr,
1688 .setxattr = fuse_setxattr,
1689 .getxattr = fuse_getxattr,
1690 .listxattr = fuse_listxattr,
1691 .removexattr = fuse_removexattr,
1694 static const struct file_operations fuse_dir_operations = {
1695 .llseek = generic_file_llseek,
1696 .read = generic_read_dir,
1697 .readdir = fuse_readdir,
1698 .open = fuse_dir_open,
1699 .release = fuse_dir_release,
1700 .fsync = fuse_dir_fsync,
1701 .unlocked_ioctl = fuse_dir_ioctl,
1702 .compat_ioctl = fuse_dir_compat_ioctl,
1705 static const struct inode_operations fuse_common_inode_operations = {
1706 .setattr = fuse_setattr,
1707 .permission = fuse_permission,
1708 .getattr = fuse_getattr,
1709 .setxattr = fuse_setxattr,
1710 .getxattr = fuse_getxattr,
1711 .listxattr = fuse_listxattr,
1712 .removexattr = fuse_removexattr,
1715 static const struct inode_operations fuse_symlink_inode_operations = {
1716 .setattr = fuse_setattr,
1717 .follow_link = fuse_follow_link,
1718 .put_link = fuse_put_link,
1719 .readlink = generic_readlink,
1720 .getattr = fuse_getattr,
1721 .setxattr = fuse_setxattr,
1722 .getxattr = fuse_getxattr,
1723 .listxattr = fuse_listxattr,
1724 .removexattr = fuse_removexattr,
1727 void fuse_init_common(struct inode *inode)
1729 inode->i_op = &fuse_common_inode_operations;
1732 void fuse_init_dir(struct inode *inode)
1734 inode->i_op = &fuse_dir_inode_operations;
1735 inode->i_fop = &fuse_dir_operations;
1738 void fuse_init_symlink(struct inode *inode)
1740 inode->i_op = &fuse_symlink_inode_operations;