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, unsigned int flags)
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 (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,
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, 0);
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 return finish_no_open(file, res);
516 * Code shared between mknod, mkdir, symlink and link
518 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
519 struct inode *dir, struct dentry *entry,
522 struct fuse_entry_out outarg;
525 struct fuse_forget_link *forget;
527 forget = fuse_alloc_forget();
529 fuse_put_request(fc, req);
533 memset(&outarg, 0, sizeof(outarg));
534 req->in.h.nodeid = get_node_id(dir);
535 req->out.numargs = 1;
537 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
539 req->out.args[0].size = sizeof(outarg);
540 req->out.args[0].value = &outarg;
541 fuse_request_send(fc, req);
542 err = req->out.h.error;
543 fuse_put_request(fc, req);
545 goto out_put_forget_req;
548 if (invalid_nodeid(outarg.nodeid))
549 goto out_put_forget_req;
551 if ((outarg.attr.mode ^ mode) & S_IFMT)
552 goto out_put_forget_req;
554 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
555 &outarg.attr, entry_attr_timeout(&outarg), 0);
557 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
562 if (S_ISDIR(inode->i_mode)) {
563 struct dentry *alias;
564 mutex_lock(&fc->inst_mutex);
565 alias = d_find_alias(inode);
567 /* New directory must have moved since mkdir */
568 mutex_unlock(&fc->inst_mutex);
573 d_instantiate(entry, inode);
574 mutex_unlock(&fc->inst_mutex);
576 d_instantiate(entry, inode);
578 fuse_change_entry_timeout(entry, &outarg);
579 fuse_invalidate_attr(dir);
587 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
590 struct fuse_mknod_in inarg;
591 struct fuse_conn *fc = get_fuse_conn(dir);
592 struct fuse_req *req = fuse_get_req(fc);
597 mode &= ~current_umask();
599 memset(&inarg, 0, sizeof(inarg));
601 inarg.rdev = new_encode_dev(rdev);
602 inarg.umask = current_umask();
603 req->in.h.opcode = FUSE_MKNOD;
605 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
607 req->in.args[0].value = &inarg;
608 req->in.args[1].size = entry->d_name.len + 1;
609 req->in.args[1].value = entry->d_name.name;
610 return create_new_entry(fc, req, dir, entry, mode);
613 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
616 return fuse_mknod(dir, entry, mode, 0);
619 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
621 struct fuse_mkdir_in inarg;
622 struct fuse_conn *fc = get_fuse_conn(dir);
623 struct fuse_req *req = fuse_get_req(fc);
628 mode &= ~current_umask();
630 memset(&inarg, 0, sizeof(inarg));
632 inarg.umask = current_umask();
633 req->in.h.opcode = FUSE_MKDIR;
635 req->in.args[0].size = sizeof(inarg);
636 req->in.args[0].value = &inarg;
637 req->in.args[1].size = entry->d_name.len + 1;
638 req->in.args[1].value = entry->d_name.name;
639 return create_new_entry(fc, req, dir, entry, S_IFDIR);
642 static int fuse_symlink(struct inode *dir, struct dentry *entry,
645 struct fuse_conn *fc = get_fuse_conn(dir);
646 unsigned len = strlen(link) + 1;
647 struct fuse_req *req = fuse_get_req(fc);
651 req->in.h.opcode = FUSE_SYMLINK;
653 req->in.args[0].size = entry->d_name.len + 1;
654 req->in.args[0].value = entry->d_name.name;
655 req->in.args[1].size = len;
656 req->in.args[1].value = link;
657 return create_new_entry(fc, req, dir, entry, S_IFLNK);
660 static int fuse_unlink(struct inode *dir, struct dentry *entry)
663 struct fuse_conn *fc = get_fuse_conn(dir);
664 struct fuse_req *req = fuse_get_req(fc);
668 req->in.h.opcode = FUSE_UNLINK;
669 req->in.h.nodeid = get_node_id(dir);
671 req->in.args[0].size = entry->d_name.len + 1;
672 req->in.args[0].value = entry->d_name.name;
673 fuse_request_send(fc, req);
674 err = req->out.h.error;
675 fuse_put_request(fc, req);
677 struct inode *inode = entry->d_inode;
678 struct fuse_inode *fi = get_fuse_inode(inode);
680 spin_lock(&fc->lock);
681 fi->attr_version = ++fc->attr_version;
683 spin_unlock(&fc->lock);
684 fuse_invalidate_attr(inode);
685 fuse_invalidate_attr(dir);
686 fuse_invalidate_entry_cache(entry);
687 } else if (err == -EINTR)
688 fuse_invalidate_entry(entry);
692 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
695 struct fuse_conn *fc = get_fuse_conn(dir);
696 struct fuse_req *req = fuse_get_req(fc);
700 req->in.h.opcode = FUSE_RMDIR;
701 req->in.h.nodeid = get_node_id(dir);
703 req->in.args[0].size = entry->d_name.len + 1;
704 req->in.args[0].value = entry->d_name.name;
705 fuse_request_send(fc, req);
706 err = req->out.h.error;
707 fuse_put_request(fc, req);
709 clear_nlink(entry->d_inode);
710 fuse_invalidate_attr(dir);
711 fuse_invalidate_entry_cache(entry);
712 } else if (err == -EINTR)
713 fuse_invalidate_entry(entry);
717 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
718 struct inode *newdir, struct dentry *newent)
721 struct fuse_rename_in inarg;
722 struct fuse_conn *fc = get_fuse_conn(olddir);
723 struct fuse_req *req = fuse_get_req(fc);
728 memset(&inarg, 0, sizeof(inarg));
729 inarg.newdir = get_node_id(newdir);
730 req->in.h.opcode = FUSE_RENAME;
731 req->in.h.nodeid = get_node_id(olddir);
733 req->in.args[0].size = sizeof(inarg);
734 req->in.args[0].value = &inarg;
735 req->in.args[1].size = oldent->d_name.len + 1;
736 req->in.args[1].value = oldent->d_name.name;
737 req->in.args[2].size = newent->d_name.len + 1;
738 req->in.args[2].value = newent->d_name.name;
739 fuse_request_send(fc, req);
740 err = req->out.h.error;
741 fuse_put_request(fc, req);
744 fuse_invalidate_attr(oldent->d_inode);
746 fuse_invalidate_attr(olddir);
747 if (olddir != newdir)
748 fuse_invalidate_attr(newdir);
750 /* newent will end up negative */
751 if (newent->d_inode) {
752 fuse_invalidate_attr(newent->d_inode);
753 fuse_invalidate_entry_cache(newent);
755 } else if (err == -EINTR) {
756 /* If request was interrupted, DEITY only knows if the
757 rename actually took place. If the invalidation
758 fails (e.g. some process has CWD under the renamed
759 directory), then there can be inconsistency between
760 the dcache and the real filesystem. Tough luck. */
761 fuse_invalidate_entry(oldent);
763 fuse_invalidate_entry(newent);
769 static int fuse_link(struct dentry *entry, struct inode *newdir,
770 struct dentry *newent)
773 struct fuse_link_in inarg;
774 struct inode *inode = entry->d_inode;
775 struct fuse_conn *fc = get_fuse_conn(inode);
776 struct fuse_req *req = fuse_get_req(fc);
780 memset(&inarg, 0, sizeof(inarg));
781 inarg.oldnodeid = get_node_id(inode);
782 req->in.h.opcode = FUSE_LINK;
784 req->in.args[0].size = sizeof(inarg);
785 req->in.args[0].value = &inarg;
786 req->in.args[1].size = newent->d_name.len + 1;
787 req->in.args[1].value = newent->d_name.name;
788 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
789 /* Contrary to "normal" filesystems it can happen that link
790 makes two "logical" inodes point to the same "physical"
791 inode. We invalidate the attributes of the old one, so it
792 will reflect changes in the backing inode (link count,
796 struct fuse_inode *fi = get_fuse_inode(inode);
798 spin_lock(&fc->lock);
799 fi->attr_version = ++fc->attr_version;
801 spin_unlock(&fc->lock);
802 fuse_invalidate_attr(inode);
803 } else if (err == -EINTR) {
804 fuse_invalidate_attr(inode);
809 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
812 unsigned int blkbits;
814 stat->dev = inode->i_sb->s_dev;
815 stat->ino = attr->ino;
816 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
817 stat->nlink = attr->nlink;
818 stat->uid = attr->uid;
819 stat->gid = attr->gid;
820 stat->rdev = inode->i_rdev;
821 stat->atime.tv_sec = attr->atime;
822 stat->atime.tv_nsec = attr->atimensec;
823 stat->mtime.tv_sec = attr->mtime;
824 stat->mtime.tv_nsec = attr->mtimensec;
825 stat->ctime.tv_sec = attr->ctime;
826 stat->ctime.tv_nsec = attr->ctimensec;
827 stat->size = attr->size;
828 stat->blocks = attr->blocks;
830 if (attr->blksize != 0)
831 blkbits = ilog2(attr->blksize);
833 blkbits = inode->i_sb->s_blocksize_bits;
835 stat->blksize = 1 << blkbits;
838 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
842 struct fuse_getattr_in inarg;
843 struct fuse_attr_out outarg;
844 struct fuse_conn *fc = get_fuse_conn(inode);
845 struct fuse_req *req;
848 req = fuse_get_req(fc);
852 attr_version = fuse_get_attr_version(fc);
854 memset(&inarg, 0, sizeof(inarg));
855 memset(&outarg, 0, sizeof(outarg));
856 /* Directories have separate file-handle space */
857 if (file && S_ISREG(inode->i_mode)) {
858 struct fuse_file *ff = file->private_data;
860 inarg.getattr_flags |= FUSE_GETATTR_FH;
863 req->in.h.opcode = FUSE_GETATTR;
864 req->in.h.nodeid = get_node_id(inode);
866 req->in.args[0].size = sizeof(inarg);
867 req->in.args[0].value = &inarg;
868 req->out.numargs = 1;
870 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
872 req->out.args[0].size = sizeof(outarg);
873 req->out.args[0].value = &outarg;
874 fuse_request_send(fc, req);
875 err = req->out.h.error;
876 fuse_put_request(fc, req);
878 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
879 make_bad_inode(inode);
882 fuse_change_attributes(inode, &outarg.attr,
883 attr_timeout(&outarg),
886 fuse_fillattr(inode, &outarg.attr, stat);
892 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
893 struct file *file, bool *refreshed)
895 struct fuse_inode *fi = get_fuse_inode(inode);
899 if (fi->i_time < get_jiffies_64()) {
901 err = fuse_do_getattr(inode, stat, file);
906 generic_fillattr(inode, stat);
907 stat->mode = fi->orig_i_mode;
908 stat->ino = fi->orig_ino;
912 if (refreshed != NULL)
918 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
919 u64 child_nodeid, struct qstr *name)
922 struct inode *parent;
924 struct dentry *entry;
926 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
930 mutex_lock(&parent->i_mutex);
931 if (!S_ISDIR(parent->i_mode))
935 dir = d_find_alias(parent);
939 entry = d_lookup(dir, name);
944 fuse_invalidate_attr(parent);
945 fuse_invalidate_entry(entry);
947 if (child_nodeid != 0 && entry->d_inode) {
948 mutex_lock(&entry->d_inode->i_mutex);
949 if (get_node_id(entry->d_inode) != child_nodeid) {
953 if (d_mountpoint(entry)) {
957 if (S_ISDIR(entry->d_inode->i_mode)) {
958 shrink_dcache_parent(entry);
959 if (!simple_empty(entry)) {
963 entry->d_inode->i_flags |= S_DEAD;
966 clear_nlink(entry->d_inode);
969 mutex_unlock(&entry->d_inode->i_mutex);
978 mutex_unlock(&parent->i_mutex);
984 * Calling into a user-controlled filesystem gives the filesystem
985 * daemon ptrace-like capabilities over the requester process. This
986 * means, that the filesystem daemon is able to record the exact
987 * filesystem operations performed, and can also control the behavior
988 * of the requester process in otherwise impossible ways. For example
989 * it can delay the operation for arbitrary length of time allowing
990 * DoS against the requester.
992 * For this reason only those processes can call into the filesystem,
993 * for which the owner of the mount has ptrace privilege. This
994 * excludes processes started by other users, suid or sgid processes.
996 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
998 const struct cred *cred;
1001 if (fc->flags & FUSE_ALLOW_OTHER)
1006 cred = __task_cred(task);
1007 if (cred->euid == fc->user_id &&
1008 cred->suid == fc->user_id &&
1009 cred->uid == fc->user_id &&
1010 cred->egid == fc->group_id &&
1011 cred->sgid == fc->group_id &&
1012 cred->gid == fc->group_id)
1019 static int fuse_access(struct inode *inode, int mask)
1021 struct fuse_conn *fc = get_fuse_conn(inode);
1022 struct fuse_req *req;
1023 struct fuse_access_in inarg;
1029 req = fuse_get_req(fc);
1031 return PTR_ERR(req);
1033 memset(&inarg, 0, sizeof(inarg));
1034 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1035 req->in.h.opcode = FUSE_ACCESS;
1036 req->in.h.nodeid = get_node_id(inode);
1037 req->in.numargs = 1;
1038 req->in.args[0].size = sizeof(inarg);
1039 req->in.args[0].value = &inarg;
1040 fuse_request_send(fc, req);
1041 err = req->out.h.error;
1042 fuse_put_request(fc, req);
1043 if (err == -ENOSYS) {
1050 static int fuse_perm_getattr(struct inode *inode, int mask)
1052 if (mask & MAY_NOT_BLOCK)
1055 return fuse_do_getattr(inode, NULL, NULL);
1059 * Check permission. The two basic access models of FUSE are:
1061 * 1) Local access checking ('default_permissions' mount option) based
1062 * on file mode. This is the plain old disk filesystem permission
1065 * 2) "Remote" access checking, where server is responsible for
1066 * checking permission in each inode operation. An exception to this
1067 * is if ->permission() was invoked from sys_access() in which case an
1068 * access request is sent. Execute permission is still checked
1069 * locally based on file mode.
1071 static int fuse_permission(struct inode *inode, int mask)
1073 struct fuse_conn *fc = get_fuse_conn(inode);
1074 bool refreshed = false;
1077 if (!fuse_allow_task(fc, current))
1081 * If attributes are needed, refresh them before proceeding
1083 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1084 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1085 struct fuse_inode *fi = get_fuse_inode(inode);
1087 if (fi->i_time < get_jiffies_64()) {
1090 err = fuse_perm_getattr(inode, mask);
1096 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1097 err = generic_permission(inode, mask);
1099 /* If permission is denied, try to refresh file
1100 attributes. This is also needed, because the root
1101 node will at first have no permissions */
1102 if (err == -EACCES && !refreshed) {
1103 err = fuse_perm_getattr(inode, mask);
1105 err = generic_permission(inode, mask);
1108 /* Note: the opposite of the above test does not
1109 exist. So if permissions are revoked this won't be
1110 noticed immediately, only after the attribute
1111 timeout has expired */
1112 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1113 if (mask & MAY_NOT_BLOCK)
1116 err = fuse_access(inode, mask);
1117 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1118 if (!(inode->i_mode & S_IXUGO)) {
1122 err = fuse_perm_getattr(inode, mask);
1123 if (!err && !(inode->i_mode & S_IXUGO))
1130 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1131 void *dstbuf, filldir_t filldir)
1133 while (nbytes >= FUSE_NAME_OFFSET) {
1134 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1135 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1137 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1139 if (reclen > nbytes)
1142 over = filldir(dstbuf, dirent->name, dirent->namelen,
1143 file->f_pos, dirent->ino, dirent->type);
1149 file->f_pos = dirent->off;
1155 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1160 struct inode *inode = file->f_path.dentry->d_inode;
1161 struct fuse_conn *fc = get_fuse_conn(inode);
1162 struct fuse_req *req;
1164 if (is_bad_inode(inode))
1167 req = fuse_get_req(fc);
1169 return PTR_ERR(req);
1171 page = alloc_page(GFP_KERNEL);
1173 fuse_put_request(fc, req);
1176 req->out.argpages = 1;
1178 req->pages[0] = page;
1179 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1180 fuse_request_send(fc, req);
1181 nbytes = req->out.args[0].size;
1182 err = req->out.h.error;
1183 fuse_put_request(fc, req);
1185 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1189 fuse_invalidate_attr(inode); /* atime changed */
1193 static char *read_link(struct dentry *dentry)
1195 struct inode *inode = dentry->d_inode;
1196 struct fuse_conn *fc = get_fuse_conn(inode);
1197 struct fuse_req *req = fuse_get_req(fc);
1201 return ERR_CAST(req);
1203 link = (char *) __get_free_page(GFP_KERNEL);
1205 link = ERR_PTR(-ENOMEM);
1208 req->in.h.opcode = FUSE_READLINK;
1209 req->in.h.nodeid = get_node_id(inode);
1210 req->out.argvar = 1;
1211 req->out.numargs = 1;
1212 req->out.args[0].size = PAGE_SIZE - 1;
1213 req->out.args[0].value = link;
1214 fuse_request_send(fc, req);
1215 if (req->out.h.error) {
1216 free_page((unsigned long) link);
1217 link = ERR_PTR(req->out.h.error);
1219 link[req->out.args[0].size] = '\0';
1221 fuse_put_request(fc, req);
1222 fuse_invalidate_attr(inode); /* atime changed */
1226 static void free_link(char *link)
1229 free_page((unsigned long) link);
1232 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1234 nd_set_link(nd, read_link(dentry));
1238 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1240 free_link(nd_get_link(nd));
1243 static int fuse_dir_open(struct inode *inode, struct file *file)
1245 return fuse_open_common(inode, file, true);
1248 static int fuse_dir_release(struct inode *inode, struct file *file)
1250 fuse_release_common(file, FUSE_RELEASEDIR);
1255 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1258 return fuse_fsync_common(file, start, end, datasync, 1);
1261 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1264 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1266 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1270 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1273 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1276 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1281 return fuse_ioctl_common(file, cmd, arg,
1282 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1285 static bool update_mtime(unsigned ivalid)
1287 /* Always update if mtime is explicitly set */
1288 if (ivalid & ATTR_MTIME_SET)
1291 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1292 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1295 /* In all other cases update */
1299 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1301 unsigned ivalid = iattr->ia_valid;
1303 if (ivalid & ATTR_MODE)
1304 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1305 if (ivalid & ATTR_UID)
1306 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1307 if (ivalid & ATTR_GID)
1308 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1309 if (ivalid & ATTR_SIZE)
1310 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1311 if (ivalid & ATTR_ATIME) {
1312 arg->valid |= FATTR_ATIME;
1313 arg->atime = iattr->ia_atime.tv_sec;
1314 arg->atimensec = iattr->ia_atime.tv_nsec;
1315 if (!(ivalid & ATTR_ATIME_SET))
1316 arg->valid |= FATTR_ATIME_NOW;
1318 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1319 arg->valid |= FATTR_MTIME;
1320 arg->mtime = iattr->ia_mtime.tv_sec;
1321 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1322 if (!(ivalid & ATTR_MTIME_SET))
1323 arg->valid |= FATTR_MTIME_NOW;
1328 * Prevent concurrent writepages on inode
1330 * This is done by adding a negative bias to the inode write counter
1331 * and waiting for all pending writes to finish.
1333 void fuse_set_nowrite(struct inode *inode)
1335 struct fuse_conn *fc = get_fuse_conn(inode);
1336 struct fuse_inode *fi = get_fuse_inode(inode);
1338 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1340 spin_lock(&fc->lock);
1341 BUG_ON(fi->writectr < 0);
1342 fi->writectr += FUSE_NOWRITE;
1343 spin_unlock(&fc->lock);
1344 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1348 * Allow writepages on inode
1350 * Remove the bias from the writecounter and send any queued
1353 static void __fuse_release_nowrite(struct inode *inode)
1355 struct fuse_inode *fi = get_fuse_inode(inode);
1357 BUG_ON(fi->writectr != FUSE_NOWRITE);
1359 fuse_flush_writepages(inode);
1362 void fuse_release_nowrite(struct inode *inode)
1364 struct fuse_conn *fc = get_fuse_conn(inode);
1366 spin_lock(&fc->lock);
1367 __fuse_release_nowrite(inode);
1368 spin_unlock(&fc->lock);
1372 * Set attributes, and at the same time refresh them.
1374 * Truncation is slightly complicated, because the 'truncate' request
1375 * may fail, in which case we don't want to touch the mapping.
1376 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1377 * and the actual truncation by hand.
1379 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1382 struct inode *inode = entry->d_inode;
1383 struct fuse_conn *fc = get_fuse_conn(inode);
1384 struct fuse_req *req;
1385 struct fuse_setattr_in inarg;
1386 struct fuse_attr_out outarg;
1387 bool is_truncate = false;
1391 if (!fuse_allow_task(fc, current))
1394 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1395 attr->ia_valid |= ATTR_FORCE;
1397 err = inode_change_ok(inode, attr);
1401 if (attr->ia_valid & ATTR_OPEN) {
1402 if (fc->atomic_o_trunc)
1407 if (attr->ia_valid & ATTR_SIZE)
1410 req = fuse_get_req(fc);
1412 return PTR_ERR(req);
1415 fuse_set_nowrite(inode);
1417 memset(&inarg, 0, sizeof(inarg));
1418 memset(&outarg, 0, sizeof(outarg));
1419 iattr_to_fattr(attr, &inarg);
1421 struct fuse_file *ff = file->private_data;
1422 inarg.valid |= FATTR_FH;
1425 if (attr->ia_valid & ATTR_SIZE) {
1426 /* For mandatory locking in truncate */
1427 inarg.valid |= FATTR_LOCKOWNER;
1428 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1430 req->in.h.opcode = FUSE_SETATTR;
1431 req->in.h.nodeid = get_node_id(inode);
1432 req->in.numargs = 1;
1433 req->in.args[0].size = sizeof(inarg);
1434 req->in.args[0].value = &inarg;
1435 req->out.numargs = 1;
1437 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1439 req->out.args[0].size = sizeof(outarg);
1440 req->out.args[0].value = &outarg;
1441 fuse_request_send(fc, req);
1442 err = req->out.h.error;
1443 fuse_put_request(fc, req);
1446 fuse_invalidate_attr(inode);
1450 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1451 make_bad_inode(inode);
1456 spin_lock(&fc->lock);
1457 fuse_change_attributes_common(inode, &outarg.attr,
1458 attr_timeout(&outarg));
1459 oldsize = inode->i_size;
1460 i_size_write(inode, outarg.attr.size);
1463 /* NOTE: this may release/reacquire fc->lock */
1464 __fuse_release_nowrite(inode);
1466 spin_unlock(&fc->lock);
1469 * Only call invalidate_inode_pages2() after removing
1470 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1472 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1473 truncate_pagecache(inode, oldsize, outarg.attr.size);
1474 invalidate_inode_pages2(inode->i_mapping);
1481 fuse_release_nowrite(inode);
1486 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1488 if (attr->ia_valid & ATTR_FILE)
1489 return fuse_do_setattr(entry, attr, attr->ia_file);
1491 return fuse_do_setattr(entry, attr, NULL);
1494 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1497 struct inode *inode = entry->d_inode;
1498 struct fuse_conn *fc = get_fuse_conn(inode);
1500 if (!fuse_allow_task(fc, current))
1503 return fuse_update_attributes(inode, stat, NULL, NULL);
1506 static int fuse_setxattr(struct dentry *entry, const char *name,
1507 const void *value, size_t size, int flags)
1509 struct inode *inode = entry->d_inode;
1510 struct fuse_conn *fc = get_fuse_conn(inode);
1511 struct fuse_req *req;
1512 struct fuse_setxattr_in inarg;
1515 if (fc->no_setxattr)
1518 req = fuse_get_req(fc);
1520 return PTR_ERR(req);
1522 memset(&inarg, 0, sizeof(inarg));
1524 inarg.flags = flags;
1525 req->in.h.opcode = FUSE_SETXATTR;
1526 req->in.h.nodeid = get_node_id(inode);
1527 req->in.numargs = 3;
1528 req->in.args[0].size = sizeof(inarg);
1529 req->in.args[0].value = &inarg;
1530 req->in.args[1].size = strlen(name) + 1;
1531 req->in.args[1].value = name;
1532 req->in.args[2].size = size;
1533 req->in.args[2].value = value;
1534 fuse_request_send(fc, req);
1535 err = req->out.h.error;
1536 fuse_put_request(fc, req);
1537 if (err == -ENOSYS) {
1538 fc->no_setxattr = 1;
1544 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1545 void *value, size_t size)
1547 struct inode *inode = entry->d_inode;
1548 struct fuse_conn *fc = get_fuse_conn(inode);
1549 struct fuse_req *req;
1550 struct fuse_getxattr_in inarg;
1551 struct fuse_getxattr_out outarg;
1554 if (fc->no_getxattr)
1557 req = fuse_get_req(fc);
1559 return PTR_ERR(req);
1561 memset(&inarg, 0, sizeof(inarg));
1563 req->in.h.opcode = FUSE_GETXATTR;
1564 req->in.h.nodeid = get_node_id(inode);
1565 req->in.numargs = 2;
1566 req->in.args[0].size = sizeof(inarg);
1567 req->in.args[0].value = &inarg;
1568 req->in.args[1].size = strlen(name) + 1;
1569 req->in.args[1].value = name;
1570 /* This is really two different operations rolled into one */
1571 req->out.numargs = 1;
1573 req->out.argvar = 1;
1574 req->out.args[0].size = size;
1575 req->out.args[0].value = value;
1577 req->out.args[0].size = sizeof(outarg);
1578 req->out.args[0].value = &outarg;
1580 fuse_request_send(fc, req);
1581 ret = req->out.h.error;
1583 ret = size ? req->out.args[0].size : outarg.size;
1585 if (ret == -ENOSYS) {
1586 fc->no_getxattr = 1;
1590 fuse_put_request(fc, req);
1594 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1596 struct inode *inode = entry->d_inode;
1597 struct fuse_conn *fc = get_fuse_conn(inode);
1598 struct fuse_req *req;
1599 struct fuse_getxattr_in inarg;
1600 struct fuse_getxattr_out outarg;
1603 if (!fuse_allow_task(fc, current))
1606 if (fc->no_listxattr)
1609 req = fuse_get_req(fc);
1611 return PTR_ERR(req);
1613 memset(&inarg, 0, sizeof(inarg));
1615 req->in.h.opcode = FUSE_LISTXATTR;
1616 req->in.h.nodeid = get_node_id(inode);
1617 req->in.numargs = 1;
1618 req->in.args[0].size = sizeof(inarg);
1619 req->in.args[0].value = &inarg;
1620 /* This is really two different operations rolled into one */
1621 req->out.numargs = 1;
1623 req->out.argvar = 1;
1624 req->out.args[0].size = size;
1625 req->out.args[0].value = list;
1627 req->out.args[0].size = sizeof(outarg);
1628 req->out.args[0].value = &outarg;
1630 fuse_request_send(fc, req);
1631 ret = req->out.h.error;
1633 ret = size ? req->out.args[0].size : outarg.size;
1635 if (ret == -ENOSYS) {
1636 fc->no_listxattr = 1;
1640 fuse_put_request(fc, req);
1644 static int fuse_removexattr(struct dentry *entry, const char *name)
1646 struct inode *inode = entry->d_inode;
1647 struct fuse_conn *fc = get_fuse_conn(inode);
1648 struct fuse_req *req;
1651 if (fc->no_removexattr)
1654 req = fuse_get_req(fc);
1656 return PTR_ERR(req);
1658 req->in.h.opcode = FUSE_REMOVEXATTR;
1659 req->in.h.nodeid = get_node_id(inode);
1660 req->in.numargs = 1;
1661 req->in.args[0].size = strlen(name) + 1;
1662 req->in.args[0].value = name;
1663 fuse_request_send(fc, req);
1664 err = req->out.h.error;
1665 fuse_put_request(fc, req);
1666 if (err == -ENOSYS) {
1667 fc->no_removexattr = 1;
1673 static const struct inode_operations fuse_dir_inode_operations = {
1674 .lookup = fuse_lookup,
1675 .mkdir = fuse_mkdir,
1676 .symlink = fuse_symlink,
1677 .unlink = fuse_unlink,
1678 .rmdir = fuse_rmdir,
1679 .rename = fuse_rename,
1681 .setattr = fuse_setattr,
1682 .create = fuse_create,
1683 .atomic_open = fuse_atomic_open,
1684 .mknod = fuse_mknod,
1685 .permission = fuse_permission,
1686 .getattr = fuse_getattr,
1687 .setxattr = fuse_setxattr,
1688 .getxattr = fuse_getxattr,
1689 .listxattr = fuse_listxattr,
1690 .removexattr = fuse_removexattr,
1693 static const struct file_operations fuse_dir_operations = {
1694 .llseek = generic_file_llseek,
1695 .read = generic_read_dir,
1696 .readdir = fuse_readdir,
1697 .open = fuse_dir_open,
1698 .release = fuse_dir_release,
1699 .fsync = fuse_dir_fsync,
1700 .unlocked_ioctl = fuse_dir_ioctl,
1701 .compat_ioctl = fuse_dir_compat_ioctl,
1704 static const struct inode_operations fuse_common_inode_operations = {
1705 .setattr = fuse_setattr,
1706 .permission = fuse_permission,
1707 .getattr = fuse_getattr,
1708 .setxattr = fuse_setxattr,
1709 .getxattr = fuse_getxattr,
1710 .listxattr = fuse_listxattr,
1711 .removexattr = fuse_removexattr,
1714 static const struct inode_operations fuse_symlink_inode_operations = {
1715 .setattr = fuse_setattr,
1716 .follow_link = fuse_follow_link,
1717 .put_link = fuse_put_link,
1718 .readlink = generic_readlink,
1719 .getattr = fuse_getattr,
1720 .setxattr = fuse_setxattr,
1721 .getxattr = fuse_getxattr,
1722 .listxattr = fuse_listxattr,
1723 .removexattr = fuse_removexattr,
1726 void fuse_init_common(struct inode *inode)
1728 inode->i_op = &fuse_common_inode_operations;
1731 void fuse_init_dir(struct inode *inode)
1733 inode->i_op = &fuse_dir_inode_operations;
1734 inode->i_fop = &fuse_dir_operations;
1737 void fuse_init_symlink(struct inode *inode)
1739 inode->i_op = &fuse_symlink_inode_operations;