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 /* Userspace expects S_IFREG in create mode */
387 BUG_ON((mode & S_IFMT) != S_IFREG);
389 forget = fuse_alloc_forget();
394 req = fuse_get_req(fc);
397 goto out_put_forget_req;
400 ff = fuse_file_alloc(fc);
402 goto out_put_request;
405 mode &= ~current_umask();
408 memset(&inarg, 0, sizeof(inarg));
409 memset(&outentry, 0, sizeof(outentry));
412 inarg.umask = current_umask();
413 req->in.h.opcode = FUSE_CREATE;
414 req->in.h.nodeid = get_node_id(dir);
416 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
418 req->in.args[0].value = &inarg;
419 req->in.args[1].size = entry->d_name.len + 1;
420 req->in.args[1].value = entry->d_name.name;
421 req->out.numargs = 2;
423 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
425 req->out.args[0].size = sizeof(outentry);
426 req->out.args[0].value = &outentry;
427 req->out.args[1].size = sizeof(outopen);
428 req->out.args[1].value = &outopen;
429 fuse_request_send(fc, req);
430 err = req->out.h.error;
435 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
438 fuse_put_request(fc, req);
440 ff->nodeid = outentry.nodeid;
441 ff->open_flags = outopen.open_flags;
442 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
443 &outentry.attr, entry_attr_timeout(&outentry), 0);
445 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
446 fuse_sync_release(ff, flags);
447 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
452 d_instantiate(entry, inode);
453 fuse_change_entry_timeout(entry, &outentry);
454 fuse_invalidate_attr(dir);
455 err = finish_open(file, entry, generic_file_open, opened);
457 fuse_sync_release(ff, flags);
459 file->private_data = fuse_file_get(ff);
460 fuse_finish_open(inode, file);
467 fuse_put_request(fc, req);
474 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
475 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
476 struct file *file, unsigned flags,
477 umode_t mode, int *opened)
480 struct fuse_conn *fc = get_fuse_conn(dir);
481 struct dentry *res = NULL;
483 if (d_unhashed(entry)) {
484 res = fuse_lookup(dir, entry, 0);
492 if (!(flags & O_CREAT) || entry->d_inode)
496 *opened |= FILE_CREATED;
501 err = fuse_create_open(dir, entry, file, flags, mode, opened);
502 if (err == -ENOSYS) {
511 err = fuse_mknod(dir, entry, mode, 0);
515 return finish_no_open(file, res);
519 * Code shared between mknod, mkdir, symlink and link
521 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
522 struct inode *dir, struct dentry *entry,
525 struct fuse_entry_out outarg;
528 struct fuse_forget_link *forget;
530 forget = fuse_alloc_forget();
532 fuse_put_request(fc, req);
536 memset(&outarg, 0, sizeof(outarg));
537 req->in.h.nodeid = get_node_id(dir);
538 req->out.numargs = 1;
540 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
542 req->out.args[0].size = sizeof(outarg);
543 req->out.args[0].value = &outarg;
544 fuse_request_send(fc, req);
545 err = req->out.h.error;
546 fuse_put_request(fc, req);
548 goto out_put_forget_req;
551 if (invalid_nodeid(outarg.nodeid))
552 goto out_put_forget_req;
554 if ((outarg.attr.mode ^ mode) & S_IFMT)
555 goto out_put_forget_req;
557 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
558 &outarg.attr, entry_attr_timeout(&outarg), 0);
560 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
565 if (S_ISDIR(inode->i_mode)) {
566 struct dentry *alias;
567 mutex_lock(&fc->inst_mutex);
568 alias = d_find_alias(inode);
570 /* New directory must have moved since mkdir */
571 mutex_unlock(&fc->inst_mutex);
576 d_instantiate(entry, inode);
577 mutex_unlock(&fc->inst_mutex);
579 d_instantiate(entry, inode);
581 fuse_change_entry_timeout(entry, &outarg);
582 fuse_invalidate_attr(dir);
590 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
593 struct fuse_mknod_in inarg;
594 struct fuse_conn *fc = get_fuse_conn(dir);
595 struct fuse_req *req = fuse_get_req(fc);
600 mode &= ~current_umask();
602 memset(&inarg, 0, sizeof(inarg));
604 inarg.rdev = new_encode_dev(rdev);
605 inarg.umask = current_umask();
606 req->in.h.opcode = FUSE_MKNOD;
608 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
610 req->in.args[0].value = &inarg;
611 req->in.args[1].size = entry->d_name.len + 1;
612 req->in.args[1].value = entry->d_name.name;
613 return create_new_entry(fc, req, dir, entry, mode);
616 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
619 return fuse_mknod(dir, entry, mode, 0);
622 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
624 struct fuse_mkdir_in inarg;
625 struct fuse_conn *fc = get_fuse_conn(dir);
626 struct fuse_req *req = fuse_get_req(fc);
631 mode &= ~current_umask();
633 memset(&inarg, 0, sizeof(inarg));
635 inarg.umask = current_umask();
636 req->in.h.opcode = FUSE_MKDIR;
638 req->in.args[0].size = sizeof(inarg);
639 req->in.args[0].value = &inarg;
640 req->in.args[1].size = entry->d_name.len + 1;
641 req->in.args[1].value = entry->d_name.name;
642 return create_new_entry(fc, req, dir, entry, S_IFDIR);
645 static int fuse_symlink(struct inode *dir, struct dentry *entry,
648 struct fuse_conn *fc = get_fuse_conn(dir);
649 unsigned len = strlen(link) + 1;
650 struct fuse_req *req = fuse_get_req(fc);
654 req->in.h.opcode = FUSE_SYMLINK;
656 req->in.args[0].size = entry->d_name.len + 1;
657 req->in.args[0].value = entry->d_name.name;
658 req->in.args[1].size = len;
659 req->in.args[1].value = link;
660 return create_new_entry(fc, req, dir, entry, S_IFLNK);
663 static int fuse_unlink(struct inode *dir, struct dentry *entry)
666 struct fuse_conn *fc = get_fuse_conn(dir);
667 struct fuse_req *req = fuse_get_req(fc);
671 req->in.h.opcode = FUSE_UNLINK;
672 req->in.h.nodeid = get_node_id(dir);
674 req->in.args[0].size = entry->d_name.len + 1;
675 req->in.args[0].value = entry->d_name.name;
676 fuse_request_send(fc, req);
677 err = req->out.h.error;
678 fuse_put_request(fc, req);
680 struct inode *inode = entry->d_inode;
681 struct fuse_inode *fi = get_fuse_inode(inode);
683 spin_lock(&fc->lock);
684 fi->attr_version = ++fc->attr_version;
686 spin_unlock(&fc->lock);
687 fuse_invalidate_attr(inode);
688 fuse_invalidate_attr(dir);
689 fuse_invalidate_entry_cache(entry);
690 } else if (err == -EINTR)
691 fuse_invalidate_entry(entry);
695 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
698 struct fuse_conn *fc = get_fuse_conn(dir);
699 struct fuse_req *req = fuse_get_req(fc);
703 req->in.h.opcode = FUSE_RMDIR;
704 req->in.h.nodeid = get_node_id(dir);
706 req->in.args[0].size = entry->d_name.len + 1;
707 req->in.args[0].value = entry->d_name.name;
708 fuse_request_send(fc, req);
709 err = req->out.h.error;
710 fuse_put_request(fc, req);
712 clear_nlink(entry->d_inode);
713 fuse_invalidate_attr(dir);
714 fuse_invalidate_entry_cache(entry);
715 } else if (err == -EINTR)
716 fuse_invalidate_entry(entry);
720 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
721 struct inode *newdir, struct dentry *newent)
724 struct fuse_rename_in inarg;
725 struct fuse_conn *fc = get_fuse_conn(olddir);
726 struct fuse_req *req = fuse_get_req(fc);
731 memset(&inarg, 0, sizeof(inarg));
732 inarg.newdir = get_node_id(newdir);
733 req->in.h.opcode = FUSE_RENAME;
734 req->in.h.nodeid = get_node_id(olddir);
736 req->in.args[0].size = sizeof(inarg);
737 req->in.args[0].value = &inarg;
738 req->in.args[1].size = oldent->d_name.len + 1;
739 req->in.args[1].value = oldent->d_name.name;
740 req->in.args[2].size = newent->d_name.len + 1;
741 req->in.args[2].value = newent->d_name.name;
742 fuse_request_send(fc, req);
743 err = req->out.h.error;
744 fuse_put_request(fc, req);
747 fuse_invalidate_attr(oldent->d_inode);
749 fuse_invalidate_attr(olddir);
750 if (olddir != newdir)
751 fuse_invalidate_attr(newdir);
753 /* newent will end up negative */
754 if (newent->d_inode) {
755 fuse_invalidate_attr(newent->d_inode);
756 fuse_invalidate_entry_cache(newent);
758 } else if (err == -EINTR) {
759 /* If request was interrupted, DEITY only knows if the
760 rename actually took place. If the invalidation
761 fails (e.g. some process has CWD under the renamed
762 directory), then there can be inconsistency between
763 the dcache and the real filesystem. Tough luck. */
764 fuse_invalidate_entry(oldent);
766 fuse_invalidate_entry(newent);
772 static int fuse_link(struct dentry *entry, struct inode *newdir,
773 struct dentry *newent)
776 struct fuse_link_in inarg;
777 struct inode *inode = entry->d_inode;
778 struct fuse_conn *fc = get_fuse_conn(inode);
779 struct fuse_req *req = fuse_get_req(fc);
783 memset(&inarg, 0, sizeof(inarg));
784 inarg.oldnodeid = get_node_id(inode);
785 req->in.h.opcode = FUSE_LINK;
787 req->in.args[0].size = sizeof(inarg);
788 req->in.args[0].value = &inarg;
789 req->in.args[1].size = newent->d_name.len + 1;
790 req->in.args[1].value = newent->d_name.name;
791 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
792 /* Contrary to "normal" filesystems it can happen that link
793 makes two "logical" inodes point to the same "physical"
794 inode. We invalidate the attributes of the old one, so it
795 will reflect changes in the backing inode (link count,
799 struct fuse_inode *fi = get_fuse_inode(inode);
801 spin_lock(&fc->lock);
802 fi->attr_version = ++fc->attr_version;
804 spin_unlock(&fc->lock);
805 fuse_invalidate_attr(inode);
806 } else if (err == -EINTR) {
807 fuse_invalidate_attr(inode);
812 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
815 unsigned int blkbits;
817 stat->dev = inode->i_sb->s_dev;
818 stat->ino = attr->ino;
819 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
820 stat->nlink = attr->nlink;
821 stat->uid = make_kuid(&init_user_ns, attr->uid);
822 stat->gid = make_kgid(&init_user_ns, attr->gid);
823 stat->rdev = inode->i_rdev;
824 stat->atime.tv_sec = attr->atime;
825 stat->atime.tv_nsec = attr->atimensec;
826 stat->mtime.tv_sec = attr->mtime;
827 stat->mtime.tv_nsec = attr->mtimensec;
828 stat->ctime.tv_sec = attr->ctime;
829 stat->ctime.tv_nsec = attr->ctimensec;
830 stat->size = attr->size;
831 stat->blocks = attr->blocks;
833 if (attr->blksize != 0)
834 blkbits = ilog2(attr->blksize);
836 blkbits = inode->i_sb->s_blocksize_bits;
838 stat->blksize = 1 << blkbits;
841 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
845 struct fuse_getattr_in inarg;
846 struct fuse_attr_out outarg;
847 struct fuse_conn *fc = get_fuse_conn(inode);
848 struct fuse_req *req;
851 req = fuse_get_req(fc);
855 attr_version = fuse_get_attr_version(fc);
857 memset(&inarg, 0, sizeof(inarg));
858 memset(&outarg, 0, sizeof(outarg));
859 /* Directories have separate file-handle space */
860 if (file && S_ISREG(inode->i_mode)) {
861 struct fuse_file *ff = file->private_data;
863 inarg.getattr_flags |= FUSE_GETATTR_FH;
866 req->in.h.opcode = FUSE_GETATTR;
867 req->in.h.nodeid = get_node_id(inode);
869 req->in.args[0].size = sizeof(inarg);
870 req->in.args[0].value = &inarg;
871 req->out.numargs = 1;
873 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
875 req->out.args[0].size = sizeof(outarg);
876 req->out.args[0].value = &outarg;
877 fuse_request_send(fc, req);
878 err = req->out.h.error;
879 fuse_put_request(fc, req);
881 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
882 make_bad_inode(inode);
885 fuse_change_attributes(inode, &outarg.attr,
886 attr_timeout(&outarg),
889 fuse_fillattr(inode, &outarg.attr, stat);
895 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
896 struct file *file, bool *refreshed)
898 struct fuse_inode *fi = get_fuse_inode(inode);
902 if (fi->i_time < get_jiffies_64()) {
904 err = fuse_do_getattr(inode, stat, file);
909 generic_fillattr(inode, stat);
910 stat->mode = fi->orig_i_mode;
911 stat->ino = fi->orig_ino;
915 if (refreshed != NULL)
921 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
922 u64 child_nodeid, struct qstr *name)
925 struct inode *parent;
927 struct dentry *entry;
929 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
933 mutex_lock(&parent->i_mutex);
934 if (!S_ISDIR(parent->i_mode))
938 dir = d_find_alias(parent);
942 entry = d_lookup(dir, name);
947 fuse_invalidate_attr(parent);
948 fuse_invalidate_entry(entry);
950 if (child_nodeid != 0 && entry->d_inode) {
951 mutex_lock(&entry->d_inode->i_mutex);
952 if (get_node_id(entry->d_inode) != child_nodeid) {
956 if (d_mountpoint(entry)) {
960 if (S_ISDIR(entry->d_inode->i_mode)) {
961 shrink_dcache_parent(entry);
962 if (!simple_empty(entry)) {
966 entry->d_inode->i_flags |= S_DEAD;
969 clear_nlink(entry->d_inode);
972 mutex_unlock(&entry->d_inode->i_mutex);
981 mutex_unlock(&parent->i_mutex);
987 * Calling into a user-controlled filesystem gives the filesystem
988 * daemon ptrace-like capabilities over the requester process. This
989 * means, that the filesystem daemon is able to record the exact
990 * filesystem operations performed, and can also control the behavior
991 * of the requester process in otherwise impossible ways. For example
992 * it can delay the operation for arbitrary length of time allowing
993 * DoS against the requester.
995 * For this reason only those processes can call into the filesystem,
996 * for which the owner of the mount has ptrace privilege. This
997 * excludes processes started by other users, suid or sgid processes.
999 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
1001 const struct cred *cred;
1004 if (fc->flags & FUSE_ALLOW_OTHER)
1009 cred = __task_cred(task);
1010 if (uid_eq(cred->euid, fc->user_id) &&
1011 uid_eq(cred->suid, fc->user_id) &&
1012 uid_eq(cred->uid, fc->user_id) &&
1013 gid_eq(cred->egid, fc->group_id) &&
1014 gid_eq(cred->sgid, fc->group_id) &&
1015 gid_eq(cred->gid, fc->group_id))
1022 static int fuse_access(struct inode *inode, int mask)
1024 struct fuse_conn *fc = get_fuse_conn(inode);
1025 struct fuse_req *req;
1026 struct fuse_access_in inarg;
1032 req = fuse_get_req(fc);
1034 return PTR_ERR(req);
1036 memset(&inarg, 0, sizeof(inarg));
1037 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1038 req->in.h.opcode = FUSE_ACCESS;
1039 req->in.h.nodeid = get_node_id(inode);
1040 req->in.numargs = 1;
1041 req->in.args[0].size = sizeof(inarg);
1042 req->in.args[0].value = &inarg;
1043 fuse_request_send(fc, req);
1044 err = req->out.h.error;
1045 fuse_put_request(fc, req);
1046 if (err == -ENOSYS) {
1053 static int fuse_perm_getattr(struct inode *inode, int mask)
1055 if (mask & MAY_NOT_BLOCK)
1058 return fuse_do_getattr(inode, NULL, NULL);
1062 * Check permission. The two basic access models of FUSE are:
1064 * 1) Local access checking ('default_permissions' mount option) based
1065 * on file mode. This is the plain old disk filesystem permission
1068 * 2) "Remote" access checking, where server is responsible for
1069 * checking permission in each inode operation. An exception to this
1070 * is if ->permission() was invoked from sys_access() in which case an
1071 * access request is sent. Execute permission is still checked
1072 * locally based on file mode.
1074 static int fuse_permission(struct inode *inode, int mask)
1076 struct fuse_conn *fc = get_fuse_conn(inode);
1077 bool refreshed = false;
1080 if (!fuse_allow_task(fc, current))
1084 * If attributes are needed, refresh them before proceeding
1086 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1087 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1088 struct fuse_inode *fi = get_fuse_inode(inode);
1090 if (fi->i_time < get_jiffies_64()) {
1093 err = fuse_perm_getattr(inode, mask);
1099 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1100 err = generic_permission(inode, mask);
1102 /* If permission is denied, try to refresh file
1103 attributes. This is also needed, because the root
1104 node will at first have no permissions */
1105 if (err == -EACCES && !refreshed) {
1106 err = fuse_perm_getattr(inode, mask);
1108 err = generic_permission(inode, mask);
1111 /* Note: the opposite of the above test does not
1112 exist. So if permissions are revoked this won't be
1113 noticed immediately, only after the attribute
1114 timeout has expired */
1115 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1116 if (mask & MAY_NOT_BLOCK)
1119 err = fuse_access(inode, mask);
1120 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1121 if (!(inode->i_mode & S_IXUGO)) {
1125 err = fuse_perm_getattr(inode, mask);
1126 if (!err && !(inode->i_mode & S_IXUGO))
1133 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1134 void *dstbuf, filldir_t filldir)
1136 while (nbytes >= FUSE_NAME_OFFSET) {
1137 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1138 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1140 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1142 if (reclen > nbytes)
1145 over = filldir(dstbuf, dirent->name, dirent->namelen,
1146 file->f_pos, dirent->ino, dirent->type);
1152 file->f_pos = dirent->off;
1158 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1163 struct inode *inode = file->f_path.dentry->d_inode;
1164 struct fuse_conn *fc = get_fuse_conn(inode);
1165 struct fuse_req *req;
1167 if (is_bad_inode(inode))
1170 req = fuse_get_req(fc);
1172 return PTR_ERR(req);
1174 page = alloc_page(GFP_KERNEL);
1176 fuse_put_request(fc, req);
1179 req->out.argpages = 1;
1181 req->pages[0] = page;
1182 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1183 fuse_request_send(fc, req);
1184 nbytes = req->out.args[0].size;
1185 err = req->out.h.error;
1186 fuse_put_request(fc, req);
1188 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1192 fuse_invalidate_attr(inode); /* atime changed */
1196 static char *read_link(struct dentry *dentry)
1198 struct inode *inode = dentry->d_inode;
1199 struct fuse_conn *fc = get_fuse_conn(inode);
1200 struct fuse_req *req = fuse_get_req(fc);
1204 return ERR_CAST(req);
1206 link = (char *) __get_free_page(GFP_KERNEL);
1208 link = ERR_PTR(-ENOMEM);
1211 req->in.h.opcode = FUSE_READLINK;
1212 req->in.h.nodeid = get_node_id(inode);
1213 req->out.argvar = 1;
1214 req->out.numargs = 1;
1215 req->out.args[0].size = PAGE_SIZE - 1;
1216 req->out.args[0].value = link;
1217 fuse_request_send(fc, req);
1218 if (req->out.h.error) {
1219 free_page((unsigned long) link);
1220 link = ERR_PTR(req->out.h.error);
1222 link[req->out.args[0].size] = '\0';
1224 fuse_put_request(fc, req);
1225 fuse_invalidate_attr(inode); /* atime changed */
1229 static void free_link(char *link)
1232 free_page((unsigned long) link);
1235 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1237 nd_set_link(nd, read_link(dentry));
1241 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1243 free_link(nd_get_link(nd));
1246 static int fuse_dir_open(struct inode *inode, struct file *file)
1248 return fuse_open_common(inode, file, true);
1251 static int fuse_dir_release(struct inode *inode, struct file *file)
1253 fuse_release_common(file, FUSE_RELEASEDIR);
1258 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1261 return fuse_fsync_common(file, start, end, datasync, 1);
1264 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1267 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1269 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1273 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1276 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1279 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1284 return fuse_ioctl_common(file, cmd, arg,
1285 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1288 static bool update_mtime(unsigned ivalid)
1290 /* Always update if mtime is explicitly set */
1291 if (ivalid & ATTR_MTIME_SET)
1294 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1295 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1298 /* In all other cases update */
1302 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1304 unsigned ivalid = iattr->ia_valid;
1306 if (ivalid & ATTR_MODE)
1307 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1308 if (ivalid & ATTR_UID)
1309 arg->valid |= FATTR_UID, arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1310 if (ivalid & ATTR_GID)
1311 arg->valid |= FATTR_GID, arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1312 if (ivalid & ATTR_SIZE)
1313 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1314 if (ivalid & ATTR_ATIME) {
1315 arg->valid |= FATTR_ATIME;
1316 arg->atime = iattr->ia_atime.tv_sec;
1317 arg->atimensec = iattr->ia_atime.tv_nsec;
1318 if (!(ivalid & ATTR_ATIME_SET))
1319 arg->valid |= FATTR_ATIME_NOW;
1321 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1322 arg->valid |= FATTR_MTIME;
1323 arg->mtime = iattr->ia_mtime.tv_sec;
1324 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1325 if (!(ivalid & ATTR_MTIME_SET))
1326 arg->valid |= FATTR_MTIME_NOW;
1331 * Prevent concurrent writepages on inode
1333 * This is done by adding a negative bias to the inode write counter
1334 * and waiting for all pending writes to finish.
1336 void fuse_set_nowrite(struct inode *inode)
1338 struct fuse_conn *fc = get_fuse_conn(inode);
1339 struct fuse_inode *fi = get_fuse_inode(inode);
1341 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1343 spin_lock(&fc->lock);
1344 BUG_ON(fi->writectr < 0);
1345 fi->writectr += FUSE_NOWRITE;
1346 spin_unlock(&fc->lock);
1347 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1351 * Allow writepages on inode
1353 * Remove the bias from the writecounter and send any queued
1356 static void __fuse_release_nowrite(struct inode *inode)
1358 struct fuse_inode *fi = get_fuse_inode(inode);
1360 BUG_ON(fi->writectr != FUSE_NOWRITE);
1362 fuse_flush_writepages(inode);
1365 void fuse_release_nowrite(struct inode *inode)
1367 struct fuse_conn *fc = get_fuse_conn(inode);
1369 spin_lock(&fc->lock);
1370 __fuse_release_nowrite(inode);
1371 spin_unlock(&fc->lock);
1375 * Set attributes, and at the same time refresh them.
1377 * Truncation is slightly complicated, because the 'truncate' request
1378 * may fail, in which case we don't want to touch the mapping.
1379 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1380 * and the actual truncation by hand.
1382 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1385 struct inode *inode = entry->d_inode;
1386 struct fuse_conn *fc = get_fuse_conn(inode);
1387 struct fuse_req *req;
1388 struct fuse_setattr_in inarg;
1389 struct fuse_attr_out outarg;
1390 bool is_truncate = false;
1394 if (!fuse_allow_task(fc, current))
1397 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1398 attr->ia_valid |= ATTR_FORCE;
1400 err = inode_change_ok(inode, attr);
1404 if (attr->ia_valid & ATTR_OPEN) {
1405 if (fc->atomic_o_trunc)
1410 if (attr->ia_valid & ATTR_SIZE)
1413 req = fuse_get_req(fc);
1415 return PTR_ERR(req);
1418 fuse_set_nowrite(inode);
1420 memset(&inarg, 0, sizeof(inarg));
1421 memset(&outarg, 0, sizeof(outarg));
1422 iattr_to_fattr(attr, &inarg);
1424 struct fuse_file *ff = file->private_data;
1425 inarg.valid |= FATTR_FH;
1428 if (attr->ia_valid & ATTR_SIZE) {
1429 /* For mandatory locking in truncate */
1430 inarg.valid |= FATTR_LOCKOWNER;
1431 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1433 req->in.h.opcode = FUSE_SETATTR;
1434 req->in.h.nodeid = get_node_id(inode);
1435 req->in.numargs = 1;
1436 req->in.args[0].size = sizeof(inarg);
1437 req->in.args[0].value = &inarg;
1438 req->out.numargs = 1;
1440 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1442 req->out.args[0].size = sizeof(outarg);
1443 req->out.args[0].value = &outarg;
1444 fuse_request_send(fc, req);
1445 err = req->out.h.error;
1446 fuse_put_request(fc, req);
1449 fuse_invalidate_attr(inode);
1453 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1454 make_bad_inode(inode);
1459 spin_lock(&fc->lock);
1460 fuse_change_attributes_common(inode, &outarg.attr,
1461 attr_timeout(&outarg));
1462 oldsize = inode->i_size;
1463 i_size_write(inode, outarg.attr.size);
1466 /* NOTE: this may release/reacquire fc->lock */
1467 __fuse_release_nowrite(inode);
1469 spin_unlock(&fc->lock);
1472 * Only call invalidate_inode_pages2() after removing
1473 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1475 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1476 truncate_pagecache(inode, oldsize, outarg.attr.size);
1477 invalidate_inode_pages2(inode->i_mapping);
1484 fuse_release_nowrite(inode);
1489 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1491 if (attr->ia_valid & ATTR_FILE)
1492 return fuse_do_setattr(entry, attr, attr->ia_file);
1494 return fuse_do_setattr(entry, attr, NULL);
1497 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1500 struct inode *inode = entry->d_inode;
1501 struct fuse_conn *fc = get_fuse_conn(inode);
1503 if (!fuse_allow_task(fc, current))
1506 return fuse_update_attributes(inode, stat, NULL, NULL);
1509 static int fuse_setxattr(struct dentry *entry, const char *name,
1510 const void *value, size_t size, int flags)
1512 struct inode *inode = entry->d_inode;
1513 struct fuse_conn *fc = get_fuse_conn(inode);
1514 struct fuse_req *req;
1515 struct fuse_setxattr_in inarg;
1518 if (fc->no_setxattr)
1521 req = fuse_get_req(fc);
1523 return PTR_ERR(req);
1525 memset(&inarg, 0, sizeof(inarg));
1527 inarg.flags = flags;
1528 req->in.h.opcode = FUSE_SETXATTR;
1529 req->in.h.nodeid = get_node_id(inode);
1530 req->in.numargs = 3;
1531 req->in.args[0].size = sizeof(inarg);
1532 req->in.args[0].value = &inarg;
1533 req->in.args[1].size = strlen(name) + 1;
1534 req->in.args[1].value = name;
1535 req->in.args[2].size = size;
1536 req->in.args[2].value = value;
1537 fuse_request_send(fc, req);
1538 err = req->out.h.error;
1539 fuse_put_request(fc, req);
1540 if (err == -ENOSYS) {
1541 fc->no_setxattr = 1;
1547 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1548 void *value, size_t size)
1550 struct inode *inode = entry->d_inode;
1551 struct fuse_conn *fc = get_fuse_conn(inode);
1552 struct fuse_req *req;
1553 struct fuse_getxattr_in inarg;
1554 struct fuse_getxattr_out outarg;
1557 if (fc->no_getxattr)
1560 req = fuse_get_req(fc);
1562 return PTR_ERR(req);
1564 memset(&inarg, 0, sizeof(inarg));
1566 req->in.h.opcode = FUSE_GETXATTR;
1567 req->in.h.nodeid = get_node_id(inode);
1568 req->in.numargs = 2;
1569 req->in.args[0].size = sizeof(inarg);
1570 req->in.args[0].value = &inarg;
1571 req->in.args[1].size = strlen(name) + 1;
1572 req->in.args[1].value = name;
1573 /* This is really two different operations rolled into one */
1574 req->out.numargs = 1;
1576 req->out.argvar = 1;
1577 req->out.args[0].size = size;
1578 req->out.args[0].value = value;
1580 req->out.args[0].size = sizeof(outarg);
1581 req->out.args[0].value = &outarg;
1583 fuse_request_send(fc, req);
1584 ret = req->out.h.error;
1586 ret = size ? req->out.args[0].size : outarg.size;
1588 if (ret == -ENOSYS) {
1589 fc->no_getxattr = 1;
1593 fuse_put_request(fc, req);
1597 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1599 struct inode *inode = entry->d_inode;
1600 struct fuse_conn *fc = get_fuse_conn(inode);
1601 struct fuse_req *req;
1602 struct fuse_getxattr_in inarg;
1603 struct fuse_getxattr_out outarg;
1606 if (!fuse_allow_task(fc, current))
1609 if (fc->no_listxattr)
1612 req = fuse_get_req(fc);
1614 return PTR_ERR(req);
1616 memset(&inarg, 0, sizeof(inarg));
1618 req->in.h.opcode = FUSE_LISTXATTR;
1619 req->in.h.nodeid = get_node_id(inode);
1620 req->in.numargs = 1;
1621 req->in.args[0].size = sizeof(inarg);
1622 req->in.args[0].value = &inarg;
1623 /* This is really two different operations rolled into one */
1624 req->out.numargs = 1;
1626 req->out.argvar = 1;
1627 req->out.args[0].size = size;
1628 req->out.args[0].value = list;
1630 req->out.args[0].size = sizeof(outarg);
1631 req->out.args[0].value = &outarg;
1633 fuse_request_send(fc, req);
1634 ret = req->out.h.error;
1636 ret = size ? req->out.args[0].size : outarg.size;
1638 if (ret == -ENOSYS) {
1639 fc->no_listxattr = 1;
1643 fuse_put_request(fc, req);
1647 static int fuse_removexattr(struct dentry *entry, const char *name)
1649 struct inode *inode = entry->d_inode;
1650 struct fuse_conn *fc = get_fuse_conn(inode);
1651 struct fuse_req *req;
1654 if (fc->no_removexattr)
1657 req = fuse_get_req(fc);
1659 return PTR_ERR(req);
1661 req->in.h.opcode = FUSE_REMOVEXATTR;
1662 req->in.h.nodeid = get_node_id(inode);
1663 req->in.numargs = 1;
1664 req->in.args[0].size = strlen(name) + 1;
1665 req->in.args[0].value = name;
1666 fuse_request_send(fc, req);
1667 err = req->out.h.error;
1668 fuse_put_request(fc, req);
1669 if (err == -ENOSYS) {
1670 fc->no_removexattr = 1;
1676 static const struct inode_operations fuse_dir_inode_operations = {
1677 .lookup = fuse_lookup,
1678 .mkdir = fuse_mkdir,
1679 .symlink = fuse_symlink,
1680 .unlink = fuse_unlink,
1681 .rmdir = fuse_rmdir,
1682 .rename = fuse_rename,
1684 .setattr = fuse_setattr,
1685 .create = fuse_create,
1686 .atomic_open = fuse_atomic_open,
1687 .mknod = fuse_mknod,
1688 .permission = fuse_permission,
1689 .getattr = fuse_getattr,
1690 .setxattr = fuse_setxattr,
1691 .getxattr = fuse_getxattr,
1692 .listxattr = fuse_listxattr,
1693 .removexattr = fuse_removexattr,
1696 static const struct file_operations fuse_dir_operations = {
1697 .llseek = generic_file_llseek,
1698 .read = generic_read_dir,
1699 .readdir = fuse_readdir,
1700 .open = fuse_dir_open,
1701 .release = fuse_dir_release,
1702 .fsync = fuse_dir_fsync,
1703 .unlocked_ioctl = fuse_dir_ioctl,
1704 .compat_ioctl = fuse_dir_compat_ioctl,
1707 static const struct inode_operations fuse_common_inode_operations = {
1708 .setattr = fuse_setattr,
1709 .permission = fuse_permission,
1710 .getattr = fuse_getattr,
1711 .setxattr = fuse_setxattr,
1712 .getxattr = fuse_getxattr,
1713 .listxattr = fuse_listxattr,
1714 .removexattr = fuse_removexattr,
1717 static const struct inode_operations fuse_symlink_inode_operations = {
1718 .setattr = fuse_setattr,
1719 .follow_link = fuse_follow_link,
1720 .put_link = fuse_put_link,
1721 .readlink = generic_readlink,
1722 .getattr = fuse_getattr,
1723 .setxattr = fuse_setxattr,
1724 .getxattr = fuse_getxattr,
1725 .listxattr = fuse_listxattr,
1726 .removexattr = fuse_removexattr,
1729 void fuse_init_common(struct inode *inode)
1731 inode->i_op = &fuse_common_inode_operations;
1734 void fuse_init_dir(struct inode *inode)
1736 inode->i_op = &fuse_dir_inode_operations;
1737 inode->i_fop = &fuse_dir_operations;
1740 void fuse_init_symlink(struct inode *inode)
1742 inode->i_op = &fuse_symlink_inode_operations;