2 FUSE: Filesystem in Userspace
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/gfp.h>
14 #include <linux/sched.h>
15 #include <linux/namei.h>
17 #if BITS_PER_LONG >= 64
18 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
23 static inline u64 fuse_dentry_time(struct dentry *entry)
29 * On 32 bit archs store the high 32 bits of time in d_fsdata
31 static void fuse_dentry_settime(struct dentry *entry, u64 time)
34 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
37 static u64 fuse_dentry_time(struct dentry *entry)
39 return (u64) entry->d_time +
40 ((u64) (unsigned long) entry->d_fsdata << 32);
45 * FUSE caches dentries and attributes with separate timeout. The
46 * time in jiffies until the dentry/attributes are valid is stored in
47 * dentry->d_time and fuse_inode->i_time respectively.
51 * Calculate the time in jiffies until a dentry/attributes are valid
53 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
56 struct timespec ts = {sec, nsec};
57 return get_jiffies_64() + timespec_to_jiffies(&ts);
63 * Set dentry and possibly attribute timeouts from the lookup/mk*
66 static void fuse_change_entry_timeout(struct dentry *entry,
67 struct fuse_entry_out *o)
69 fuse_dentry_settime(entry,
70 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
73 static u64 attr_timeout(struct fuse_attr_out *o)
75 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
78 static u64 entry_attr_timeout(struct fuse_entry_out *o)
80 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
84 * Mark the attributes as stale, so that at the next call to
85 * ->getattr() they will be fetched from userspace
87 void fuse_invalidate_attr(struct inode *inode)
89 get_fuse_inode(inode)->i_time = 0;
93 * Just mark the entry as stale, so that a next attempt to look it up
94 * will result in a new lookup call to userspace
96 * This is called when a dentry is about to become negative and the
97 * timeout is unknown (unlink, rmdir, rename and in some cases
100 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 if (nd->flags & LOOKUP_RCU)
164 inode = entry->d_inode;
165 if (inode && is_bad_inode(inode))
167 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
169 struct fuse_entry_out outarg;
170 struct fuse_conn *fc;
171 struct fuse_req *req;
172 struct fuse_req *forget_req;
173 struct dentry *parent;
176 /* For negative dentries, always do a fresh lookup */
180 fc = get_fuse_conn(inode);
181 req = fuse_get_req(fc);
185 forget_req = fuse_get_req(fc);
186 if (IS_ERR(forget_req)) {
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_send_forget(fc, forget_req,
210 spin_lock(&fc->lock);
212 spin_unlock(&fc->lock);
214 fuse_put_request(fc, forget_req);
215 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
218 fuse_change_attributes(inode, &outarg.attr,
219 entry_attr_timeout(&outarg),
221 fuse_change_entry_timeout(entry, &outarg);
226 static int invalid_nodeid(u64 nodeid)
228 return !nodeid || nodeid == FUSE_ROOT_ID;
231 const struct dentry_operations fuse_dentry_operations = {
232 .d_revalidate = fuse_dentry_revalidate,
235 int fuse_valid_type(int m)
237 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
238 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
242 * Add a directory inode to a dentry, ensuring that no other dentry
243 * refers to this inode. Called with fc->inst_mutex.
245 static struct dentry *fuse_d_add_directory(struct dentry *entry,
248 struct dentry *alias = d_find_alias(inode);
249 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
250 /* This tries to shrink the subtree below alias */
251 fuse_invalidate_entry(alias);
253 if (!list_empty(&inode->i_dentry))
254 return ERR_PTR(-EBUSY);
258 return d_splice_alias(inode, entry);
261 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
262 struct fuse_entry_out *outarg, struct inode **inode)
264 struct fuse_conn *fc = get_fuse_conn_super(sb);
265 struct fuse_req *req;
266 struct fuse_req *forget_req;
272 if (name->len > FUSE_NAME_MAX)
275 req = fuse_get_req(fc);
280 forget_req = fuse_get_req(fc);
281 err = PTR_ERR(forget_req);
282 if (IS_ERR(forget_req)) {
283 fuse_put_request(fc, req);
287 attr_version = fuse_get_attr_version(fc);
289 fuse_lookup_init(fc, req, nodeid, name, outarg);
290 fuse_request_send(fc, req);
291 err = req->out.h.error;
292 fuse_put_request(fc, req);
293 /* Zero nodeid is same as -ENOENT, but with valid timeout */
294 if (err || !outarg->nodeid)
300 if (!fuse_valid_type(outarg->attr.mode))
303 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
304 &outarg->attr, entry_attr_timeout(outarg),
308 fuse_send_forget(fc, forget_req, outarg->nodeid, 1);
314 fuse_put_request(fc, forget_req);
319 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
320 struct nameidata *nd)
323 struct fuse_entry_out outarg;
325 struct dentry *newent;
326 struct fuse_conn *fc = get_fuse_conn(dir);
327 bool outarg_valid = true;
329 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
331 if (err == -ENOENT) {
332 outarg_valid = false;
339 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
342 if (inode && S_ISDIR(inode->i_mode)) {
343 mutex_lock(&fc->inst_mutex);
344 newent = fuse_d_add_directory(entry, inode);
345 mutex_unlock(&fc->inst_mutex);
346 err = PTR_ERR(newent);
350 newent = d_splice_alias(inode, entry);
353 entry = newent ? newent : entry;
354 d_set_d_op(entry, &fuse_dentry_operations);
356 fuse_change_entry_timeout(entry, &outarg);
358 fuse_invalidate_entry_cache(entry);
369 * Atomic create+open operation
371 * If the filesystem doesn't support this, then fall back to separate
372 * 'mknod' + 'open' requests.
374 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
375 struct nameidata *nd)
379 struct fuse_conn *fc = get_fuse_conn(dir);
380 struct fuse_req *req;
381 struct fuse_req *forget_req;
382 struct fuse_create_in inarg;
383 struct fuse_open_out outopen;
384 struct fuse_entry_out outentry;
385 struct fuse_file *ff;
387 int flags = nd->intent.open.flags - 1;
392 if (flags & O_DIRECT)
395 forget_req = fuse_get_req(fc);
396 if (IS_ERR(forget_req))
397 return PTR_ERR(forget_req);
399 req = fuse_get_req(fc);
402 goto out_put_forget_req;
405 ff = fuse_file_alloc(fc);
407 goto out_put_request;
410 mode &= ~current_umask();
413 memset(&inarg, 0, sizeof(inarg));
414 memset(&outentry, 0, sizeof(outentry));
417 inarg.umask = current_umask();
418 req->in.h.opcode = FUSE_CREATE;
419 req->in.h.nodeid = get_node_id(dir);
421 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
423 req->in.args[0].value = &inarg;
424 req->in.args[1].size = entry->d_name.len + 1;
425 req->in.args[1].value = entry->d_name.name;
426 req->out.numargs = 2;
428 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
430 req->out.args[0].size = sizeof(outentry);
431 req->out.args[0].value = &outentry;
432 req->out.args[1].size = sizeof(outopen);
433 req->out.args[1].value = &outopen;
434 fuse_request_send(fc, req);
435 err = req->out.h.error;
443 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
446 fuse_put_request(fc, req);
448 ff->nodeid = outentry.nodeid;
449 ff->open_flags = outopen.open_flags;
450 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
451 &outentry.attr, entry_attr_timeout(&outentry), 0);
453 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
454 fuse_sync_release(ff, flags);
455 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
458 fuse_put_request(fc, forget_req);
459 d_instantiate(entry, inode);
460 fuse_change_entry_timeout(entry, &outentry);
461 fuse_invalidate_attr(dir);
462 file = lookup_instantiate_filp(nd, entry, generic_file_open);
464 fuse_sync_release(ff, flags);
465 return PTR_ERR(file);
467 file->private_data = fuse_file_get(ff);
468 fuse_finish_open(inode, file);
474 fuse_put_request(fc, req);
476 fuse_put_request(fc, forget_req);
481 * Code shared between mknod, mkdir, symlink and link
483 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
484 struct inode *dir, struct dentry *entry,
487 struct fuse_entry_out outarg;
490 struct fuse_req *forget_req;
492 forget_req = fuse_get_req(fc);
493 if (IS_ERR(forget_req)) {
494 fuse_put_request(fc, req);
495 return PTR_ERR(forget_req);
498 memset(&outarg, 0, sizeof(outarg));
499 req->in.h.nodeid = get_node_id(dir);
500 req->out.numargs = 1;
502 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
504 req->out.args[0].size = sizeof(outarg);
505 req->out.args[0].value = &outarg;
506 fuse_request_send(fc, req);
507 err = req->out.h.error;
508 fuse_put_request(fc, req);
510 goto out_put_forget_req;
513 if (invalid_nodeid(outarg.nodeid))
514 goto out_put_forget_req;
516 if ((outarg.attr.mode ^ mode) & S_IFMT)
517 goto out_put_forget_req;
519 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
520 &outarg.attr, entry_attr_timeout(&outarg), 0);
522 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
525 fuse_put_request(fc, forget_req);
527 if (S_ISDIR(inode->i_mode)) {
528 struct dentry *alias;
529 mutex_lock(&fc->inst_mutex);
530 alias = d_find_alias(inode);
532 /* New directory must have moved since mkdir */
533 mutex_unlock(&fc->inst_mutex);
538 d_instantiate(entry, inode);
539 mutex_unlock(&fc->inst_mutex);
541 d_instantiate(entry, inode);
543 fuse_change_entry_timeout(entry, &outarg);
544 fuse_invalidate_attr(dir);
548 fuse_put_request(fc, forget_req);
552 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
555 struct fuse_mknod_in inarg;
556 struct fuse_conn *fc = get_fuse_conn(dir);
557 struct fuse_req *req = fuse_get_req(fc);
562 mode &= ~current_umask();
564 memset(&inarg, 0, sizeof(inarg));
566 inarg.rdev = new_encode_dev(rdev);
567 inarg.umask = current_umask();
568 req->in.h.opcode = FUSE_MKNOD;
570 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
572 req->in.args[0].value = &inarg;
573 req->in.args[1].size = entry->d_name.len + 1;
574 req->in.args[1].value = entry->d_name.name;
575 return create_new_entry(fc, req, dir, entry, mode);
578 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
579 struct nameidata *nd)
581 if (nd && (nd->flags & LOOKUP_OPEN)) {
582 int err = fuse_create_open(dir, entry, mode, nd);
585 /* Fall back on mknod */
587 return fuse_mknod(dir, entry, mode, 0);
590 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
592 struct fuse_mkdir_in inarg;
593 struct fuse_conn *fc = get_fuse_conn(dir);
594 struct fuse_req *req = fuse_get_req(fc);
599 mode &= ~current_umask();
601 memset(&inarg, 0, sizeof(inarg));
603 inarg.umask = current_umask();
604 req->in.h.opcode = FUSE_MKDIR;
606 req->in.args[0].size = sizeof(inarg);
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, S_IFDIR);
613 static int fuse_symlink(struct inode *dir, struct dentry *entry,
616 struct fuse_conn *fc = get_fuse_conn(dir);
617 unsigned len = strlen(link) + 1;
618 struct fuse_req *req = fuse_get_req(fc);
622 req->in.h.opcode = FUSE_SYMLINK;
624 req->in.args[0].size = entry->d_name.len + 1;
625 req->in.args[0].value = entry->d_name.name;
626 req->in.args[1].size = len;
627 req->in.args[1].value = link;
628 return create_new_entry(fc, req, dir, entry, S_IFLNK);
631 static int fuse_unlink(struct inode *dir, struct dentry *entry)
634 struct fuse_conn *fc = get_fuse_conn(dir);
635 struct fuse_req *req = fuse_get_req(fc);
639 req->in.h.opcode = FUSE_UNLINK;
640 req->in.h.nodeid = get_node_id(dir);
642 req->in.args[0].size = entry->d_name.len + 1;
643 req->in.args[0].value = entry->d_name.name;
644 fuse_request_send(fc, req);
645 err = req->out.h.error;
646 fuse_put_request(fc, req);
648 struct inode *inode = entry->d_inode;
651 * Set nlink to zero so the inode can be cleared, if the inode
652 * does have more links this will be discovered at the next
656 fuse_invalidate_attr(inode);
657 fuse_invalidate_attr(dir);
658 fuse_invalidate_entry_cache(entry);
659 } else if (err == -EINTR)
660 fuse_invalidate_entry(entry);
664 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
667 struct fuse_conn *fc = get_fuse_conn(dir);
668 struct fuse_req *req = fuse_get_req(fc);
672 req->in.h.opcode = FUSE_RMDIR;
673 req->in.h.nodeid = get_node_id(dir);
675 req->in.args[0].size = entry->d_name.len + 1;
676 req->in.args[0].value = entry->d_name.name;
677 fuse_request_send(fc, req);
678 err = req->out.h.error;
679 fuse_put_request(fc, req);
681 clear_nlink(entry->d_inode);
682 fuse_invalidate_attr(dir);
683 fuse_invalidate_entry_cache(entry);
684 } else if (err == -EINTR)
685 fuse_invalidate_entry(entry);
689 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
690 struct inode *newdir, struct dentry *newent)
693 struct fuse_rename_in inarg;
694 struct fuse_conn *fc = get_fuse_conn(olddir);
695 struct fuse_req *req = fuse_get_req(fc);
699 memset(&inarg, 0, sizeof(inarg));
700 inarg.newdir = get_node_id(newdir);
701 req->in.h.opcode = FUSE_RENAME;
702 req->in.h.nodeid = get_node_id(olddir);
704 req->in.args[0].size = sizeof(inarg);
705 req->in.args[0].value = &inarg;
706 req->in.args[1].size = oldent->d_name.len + 1;
707 req->in.args[1].value = oldent->d_name.name;
708 req->in.args[2].size = newent->d_name.len + 1;
709 req->in.args[2].value = newent->d_name.name;
710 fuse_request_send(fc, req);
711 err = req->out.h.error;
712 fuse_put_request(fc, req);
715 fuse_invalidate_attr(oldent->d_inode);
717 fuse_invalidate_attr(olddir);
718 if (olddir != newdir)
719 fuse_invalidate_attr(newdir);
721 /* newent will end up negative */
722 if (newent->d_inode) {
723 fuse_invalidate_attr(newent->d_inode);
724 fuse_invalidate_entry_cache(newent);
726 } else if (err == -EINTR) {
727 /* If request was interrupted, DEITY only knows if the
728 rename actually took place. If the invalidation
729 fails (e.g. some process has CWD under the renamed
730 directory), then there can be inconsistency between
731 the dcache and the real filesystem. Tough luck. */
732 fuse_invalidate_entry(oldent);
734 fuse_invalidate_entry(newent);
740 static int fuse_link(struct dentry *entry, struct inode *newdir,
741 struct dentry *newent)
744 struct fuse_link_in inarg;
745 struct inode *inode = entry->d_inode;
746 struct fuse_conn *fc = get_fuse_conn(inode);
747 struct fuse_req *req = fuse_get_req(fc);
751 memset(&inarg, 0, sizeof(inarg));
752 inarg.oldnodeid = get_node_id(inode);
753 req->in.h.opcode = FUSE_LINK;
755 req->in.args[0].size = sizeof(inarg);
756 req->in.args[0].value = &inarg;
757 req->in.args[1].size = newent->d_name.len + 1;
758 req->in.args[1].value = newent->d_name.name;
759 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
760 /* Contrary to "normal" filesystems it can happen that link
761 makes two "logical" inodes point to the same "physical"
762 inode. We invalidate the attributes of the old one, so it
763 will reflect changes in the backing inode (link count,
766 if (!err || err == -EINTR)
767 fuse_invalidate_attr(inode);
771 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
774 stat->dev = inode->i_sb->s_dev;
775 stat->ino = attr->ino;
776 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
777 stat->nlink = attr->nlink;
778 stat->uid = attr->uid;
779 stat->gid = attr->gid;
780 stat->rdev = inode->i_rdev;
781 stat->atime.tv_sec = attr->atime;
782 stat->atime.tv_nsec = attr->atimensec;
783 stat->mtime.tv_sec = attr->mtime;
784 stat->mtime.tv_nsec = attr->mtimensec;
785 stat->ctime.tv_sec = attr->ctime;
786 stat->ctime.tv_nsec = attr->ctimensec;
787 stat->size = attr->size;
788 stat->blocks = attr->blocks;
789 stat->blksize = (1 << inode->i_blkbits);
792 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
796 struct fuse_getattr_in inarg;
797 struct fuse_attr_out outarg;
798 struct fuse_conn *fc = get_fuse_conn(inode);
799 struct fuse_req *req;
802 req = fuse_get_req(fc);
806 attr_version = fuse_get_attr_version(fc);
808 memset(&inarg, 0, sizeof(inarg));
809 memset(&outarg, 0, sizeof(outarg));
810 /* Directories have separate file-handle space */
811 if (file && S_ISREG(inode->i_mode)) {
812 struct fuse_file *ff = file->private_data;
814 inarg.getattr_flags |= FUSE_GETATTR_FH;
817 req->in.h.opcode = FUSE_GETATTR;
818 req->in.h.nodeid = get_node_id(inode);
820 req->in.args[0].size = sizeof(inarg);
821 req->in.args[0].value = &inarg;
822 req->out.numargs = 1;
824 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
826 req->out.args[0].size = sizeof(outarg);
827 req->out.args[0].value = &outarg;
828 fuse_request_send(fc, req);
829 err = req->out.h.error;
830 fuse_put_request(fc, req);
832 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
833 make_bad_inode(inode);
836 fuse_change_attributes(inode, &outarg.attr,
837 attr_timeout(&outarg),
840 fuse_fillattr(inode, &outarg.attr, stat);
846 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
847 struct file *file, bool *refreshed)
849 struct fuse_inode *fi = get_fuse_inode(inode);
853 if (fi->i_time < get_jiffies_64()) {
855 err = fuse_do_getattr(inode, stat, file);
860 generic_fillattr(inode, stat);
861 stat->mode = fi->orig_i_mode;
865 if (refreshed != NULL)
871 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
875 struct inode *parent;
877 struct dentry *entry;
879 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
883 mutex_lock(&parent->i_mutex);
884 if (!S_ISDIR(parent->i_mode))
888 dir = d_find_alias(parent);
892 entry = d_lookup(dir, name);
897 fuse_invalidate_attr(parent);
898 fuse_invalidate_entry(entry);
903 mutex_unlock(&parent->i_mutex);
909 * Calling into a user-controlled filesystem gives the filesystem
910 * daemon ptrace-like capabilities over the requester process. This
911 * means, that the filesystem daemon is able to record the exact
912 * filesystem operations performed, and can also control the behavior
913 * of the requester process in otherwise impossible ways. For example
914 * it can delay the operation for arbitrary length of time allowing
915 * DoS against the requester.
917 * For this reason only those processes can call into the filesystem,
918 * for which the owner of the mount has ptrace privilege. This
919 * excludes processes started by other users, suid or sgid processes.
921 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
923 const struct cred *cred;
926 if (fc->flags & FUSE_ALLOW_OTHER)
931 cred = __task_cred(task);
932 if (cred->euid == fc->user_id &&
933 cred->suid == fc->user_id &&
934 cred->uid == fc->user_id &&
935 cred->egid == fc->group_id &&
936 cred->sgid == fc->group_id &&
937 cred->gid == fc->group_id)
944 static int fuse_access(struct inode *inode, int mask)
946 struct fuse_conn *fc = get_fuse_conn(inode);
947 struct fuse_req *req;
948 struct fuse_access_in inarg;
954 req = fuse_get_req(fc);
958 memset(&inarg, 0, sizeof(inarg));
959 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
960 req->in.h.opcode = FUSE_ACCESS;
961 req->in.h.nodeid = get_node_id(inode);
963 req->in.args[0].size = sizeof(inarg);
964 req->in.args[0].value = &inarg;
965 fuse_request_send(fc, req);
966 err = req->out.h.error;
967 fuse_put_request(fc, req);
968 if (err == -ENOSYS) {
976 * Check permission. The two basic access models of FUSE are:
978 * 1) Local access checking ('default_permissions' mount option) based
979 * on file mode. This is the plain old disk filesystem permission
982 * 2) "Remote" access checking, where server is responsible for
983 * checking permission in each inode operation. An exception to this
984 * is if ->permission() was invoked from sys_access() in which case an
985 * access request is sent. Execute permission is still checked
986 * locally based on file mode.
988 static int fuse_permission(struct inode *inode, int mask)
990 struct fuse_conn *fc = get_fuse_conn(inode);
991 bool refreshed = false;
994 if (!fuse_allow_task(fc, current))
998 * If attributes are needed, refresh them before proceeding
1000 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1001 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1002 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
1007 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1008 err = generic_permission(inode, mask, NULL);
1010 /* If permission is denied, try to refresh file
1011 attributes. This is also needed, because the root
1012 node will at first have no permissions */
1013 if (err == -EACCES && !refreshed) {
1014 err = fuse_do_getattr(inode, NULL, NULL);
1016 err = generic_permission(inode, mask, NULL);
1019 /* Note: the opposite of the above test does not
1020 exist. So if permissions are revoked this won't be
1021 noticed immediately, only after the attribute
1022 timeout has expired */
1023 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1024 err = fuse_access(inode, mask);
1025 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1026 if (!(inode->i_mode & S_IXUGO)) {
1030 err = fuse_do_getattr(inode, NULL, NULL);
1031 if (!err && !(inode->i_mode & S_IXUGO))
1038 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1039 void *dstbuf, filldir_t filldir)
1041 while (nbytes >= FUSE_NAME_OFFSET) {
1042 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1043 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1045 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1047 if (reclen > nbytes)
1050 over = filldir(dstbuf, dirent->name, dirent->namelen,
1051 file->f_pos, dirent->ino, dirent->type);
1057 file->f_pos = dirent->off;
1063 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1068 struct inode *inode = file->f_path.dentry->d_inode;
1069 struct fuse_conn *fc = get_fuse_conn(inode);
1070 struct fuse_req *req;
1072 if (is_bad_inode(inode))
1075 req = fuse_get_req(fc);
1077 return PTR_ERR(req);
1079 page = alloc_page(GFP_KERNEL);
1081 fuse_put_request(fc, req);
1084 req->out.argpages = 1;
1086 req->pages[0] = page;
1087 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1088 fuse_request_send(fc, req);
1089 nbytes = req->out.args[0].size;
1090 err = req->out.h.error;
1091 fuse_put_request(fc, req);
1093 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1097 fuse_invalidate_attr(inode); /* atime changed */
1101 static char *read_link(struct dentry *dentry)
1103 struct inode *inode = dentry->d_inode;
1104 struct fuse_conn *fc = get_fuse_conn(inode);
1105 struct fuse_req *req = fuse_get_req(fc);
1109 return ERR_CAST(req);
1111 link = (char *) __get_free_page(GFP_KERNEL);
1113 link = ERR_PTR(-ENOMEM);
1116 req->in.h.opcode = FUSE_READLINK;
1117 req->in.h.nodeid = get_node_id(inode);
1118 req->out.argvar = 1;
1119 req->out.numargs = 1;
1120 req->out.args[0].size = PAGE_SIZE - 1;
1121 req->out.args[0].value = link;
1122 fuse_request_send(fc, req);
1123 if (req->out.h.error) {
1124 free_page((unsigned long) link);
1125 link = ERR_PTR(req->out.h.error);
1127 link[req->out.args[0].size] = '\0';
1129 fuse_put_request(fc, req);
1130 fuse_invalidate_attr(inode); /* atime changed */
1134 static void free_link(char *link)
1137 free_page((unsigned long) link);
1140 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1142 nd_set_link(nd, read_link(dentry));
1146 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1148 free_link(nd_get_link(nd));
1151 static int fuse_dir_open(struct inode *inode, struct file *file)
1153 return fuse_open_common(inode, file, true);
1156 static int fuse_dir_release(struct inode *inode, struct file *file)
1158 fuse_release_common(file, FUSE_RELEASEDIR);
1163 static int fuse_dir_fsync(struct file *file, int datasync)
1165 return fuse_fsync_common(file, datasync, 1);
1168 static bool update_mtime(unsigned ivalid)
1170 /* Always update if mtime is explicitly set */
1171 if (ivalid & ATTR_MTIME_SET)
1174 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1175 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1178 /* In all other cases update */
1182 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1184 unsigned ivalid = iattr->ia_valid;
1186 if (ivalid & ATTR_MODE)
1187 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1188 if (ivalid & ATTR_UID)
1189 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1190 if (ivalid & ATTR_GID)
1191 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1192 if (ivalid & ATTR_SIZE)
1193 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1194 if (ivalid & ATTR_ATIME) {
1195 arg->valid |= FATTR_ATIME;
1196 arg->atime = iattr->ia_atime.tv_sec;
1197 arg->atimensec = iattr->ia_atime.tv_nsec;
1198 if (!(ivalid & ATTR_ATIME_SET))
1199 arg->valid |= FATTR_ATIME_NOW;
1201 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1202 arg->valid |= FATTR_MTIME;
1203 arg->mtime = iattr->ia_mtime.tv_sec;
1204 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1205 if (!(ivalid & ATTR_MTIME_SET))
1206 arg->valid |= FATTR_MTIME_NOW;
1211 * Prevent concurrent writepages on inode
1213 * This is done by adding a negative bias to the inode write counter
1214 * and waiting for all pending writes to finish.
1216 void fuse_set_nowrite(struct inode *inode)
1218 struct fuse_conn *fc = get_fuse_conn(inode);
1219 struct fuse_inode *fi = get_fuse_inode(inode);
1221 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1223 spin_lock(&fc->lock);
1224 BUG_ON(fi->writectr < 0);
1225 fi->writectr += FUSE_NOWRITE;
1226 spin_unlock(&fc->lock);
1227 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1231 * Allow writepages on inode
1233 * Remove the bias from the writecounter and send any queued
1236 static void __fuse_release_nowrite(struct inode *inode)
1238 struct fuse_inode *fi = get_fuse_inode(inode);
1240 BUG_ON(fi->writectr != FUSE_NOWRITE);
1242 fuse_flush_writepages(inode);
1245 void fuse_release_nowrite(struct inode *inode)
1247 struct fuse_conn *fc = get_fuse_conn(inode);
1249 spin_lock(&fc->lock);
1250 __fuse_release_nowrite(inode);
1251 spin_unlock(&fc->lock);
1255 * Set attributes, and at the same time refresh them.
1257 * Truncation is slightly complicated, because the 'truncate' request
1258 * may fail, in which case we don't want to touch the mapping.
1259 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1260 * and the actual truncation by hand.
1262 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1265 struct inode *inode = entry->d_inode;
1266 struct fuse_conn *fc = get_fuse_conn(inode);
1267 struct fuse_req *req;
1268 struct fuse_setattr_in inarg;
1269 struct fuse_attr_out outarg;
1270 bool is_truncate = false;
1274 if (!fuse_allow_task(fc, current))
1277 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1278 attr->ia_valid |= ATTR_FORCE;
1280 err = inode_change_ok(inode, attr);
1284 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1287 if (attr->ia_valid & ATTR_SIZE)
1290 req = fuse_get_req(fc);
1292 return PTR_ERR(req);
1295 fuse_set_nowrite(inode);
1297 memset(&inarg, 0, sizeof(inarg));
1298 memset(&outarg, 0, sizeof(outarg));
1299 iattr_to_fattr(attr, &inarg);
1301 struct fuse_file *ff = file->private_data;
1302 inarg.valid |= FATTR_FH;
1305 if (attr->ia_valid & ATTR_SIZE) {
1306 /* For mandatory locking in truncate */
1307 inarg.valid |= FATTR_LOCKOWNER;
1308 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1310 req->in.h.opcode = FUSE_SETATTR;
1311 req->in.h.nodeid = get_node_id(inode);
1312 req->in.numargs = 1;
1313 req->in.args[0].size = sizeof(inarg);
1314 req->in.args[0].value = &inarg;
1315 req->out.numargs = 1;
1317 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1319 req->out.args[0].size = sizeof(outarg);
1320 req->out.args[0].value = &outarg;
1321 fuse_request_send(fc, req);
1322 err = req->out.h.error;
1323 fuse_put_request(fc, req);
1326 fuse_invalidate_attr(inode);
1330 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1331 make_bad_inode(inode);
1336 spin_lock(&fc->lock);
1337 fuse_change_attributes_common(inode, &outarg.attr,
1338 attr_timeout(&outarg));
1339 oldsize = inode->i_size;
1340 i_size_write(inode, outarg.attr.size);
1343 /* NOTE: this may release/reacquire fc->lock */
1344 __fuse_release_nowrite(inode);
1346 spin_unlock(&fc->lock);
1349 * Only call invalidate_inode_pages2() after removing
1350 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1352 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1353 truncate_pagecache(inode, oldsize, outarg.attr.size);
1354 invalidate_inode_pages2(inode->i_mapping);
1361 fuse_release_nowrite(inode);
1366 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1368 if (attr->ia_valid & ATTR_FILE)
1369 return fuse_do_setattr(entry, attr, attr->ia_file);
1371 return fuse_do_setattr(entry, attr, NULL);
1374 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1377 struct inode *inode = entry->d_inode;
1378 struct fuse_conn *fc = get_fuse_conn(inode);
1380 if (!fuse_allow_task(fc, current))
1383 return fuse_update_attributes(inode, stat, NULL, NULL);
1386 static int fuse_setxattr(struct dentry *entry, const char *name,
1387 const void *value, size_t size, int flags)
1389 struct inode *inode = entry->d_inode;
1390 struct fuse_conn *fc = get_fuse_conn(inode);
1391 struct fuse_req *req;
1392 struct fuse_setxattr_in inarg;
1395 if (fc->no_setxattr)
1398 req = fuse_get_req(fc);
1400 return PTR_ERR(req);
1402 memset(&inarg, 0, sizeof(inarg));
1404 inarg.flags = flags;
1405 req->in.h.opcode = FUSE_SETXATTR;
1406 req->in.h.nodeid = get_node_id(inode);
1407 req->in.numargs = 3;
1408 req->in.args[0].size = sizeof(inarg);
1409 req->in.args[0].value = &inarg;
1410 req->in.args[1].size = strlen(name) + 1;
1411 req->in.args[1].value = name;
1412 req->in.args[2].size = size;
1413 req->in.args[2].value = value;
1414 fuse_request_send(fc, req);
1415 err = req->out.h.error;
1416 fuse_put_request(fc, req);
1417 if (err == -ENOSYS) {
1418 fc->no_setxattr = 1;
1424 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1425 void *value, size_t size)
1427 struct inode *inode = entry->d_inode;
1428 struct fuse_conn *fc = get_fuse_conn(inode);
1429 struct fuse_req *req;
1430 struct fuse_getxattr_in inarg;
1431 struct fuse_getxattr_out outarg;
1434 if (fc->no_getxattr)
1437 req = fuse_get_req(fc);
1439 return PTR_ERR(req);
1441 memset(&inarg, 0, sizeof(inarg));
1443 req->in.h.opcode = FUSE_GETXATTR;
1444 req->in.h.nodeid = get_node_id(inode);
1445 req->in.numargs = 2;
1446 req->in.args[0].size = sizeof(inarg);
1447 req->in.args[0].value = &inarg;
1448 req->in.args[1].size = strlen(name) + 1;
1449 req->in.args[1].value = name;
1450 /* This is really two different operations rolled into one */
1451 req->out.numargs = 1;
1453 req->out.argvar = 1;
1454 req->out.args[0].size = size;
1455 req->out.args[0].value = value;
1457 req->out.args[0].size = sizeof(outarg);
1458 req->out.args[0].value = &outarg;
1460 fuse_request_send(fc, req);
1461 ret = req->out.h.error;
1463 ret = size ? req->out.args[0].size : outarg.size;
1465 if (ret == -ENOSYS) {
1466 fc->no_getxattr = 1;
1470 fuse_put_request(fc, req);
1474 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1476 struct inode *inode = entry->d_inode;
1477 struct fuse_conn *fc = get_fuse_conn(inode);
1478 struct fuse_req *req;
1479 struct fuse_getxattr_in inarg;
1480 struct fuse_getxattr_out outarg;
1483 if (!fuse_allow_task(fc, current))
1486 if (fc->no_listxattr)
1489 req = fuse_get_req(fc);
1491 return PTR_ERR(req);
1493 memset(&inarg, 0, sizeof(inarg));
1495 req->in.h.opcode = FUSE_LISTXATTR;
1496 req->in.h.nodeid = get_node_id(inode);
1497 req->in.numargs = 1;
1498 req->in.args[0].size = sizeof(inarg);
1499 req->in.args[0].value = &inarg;
1500 /* This is really two different operations rolled into one */
1501 req->out.numargs = 1;
1503 req->out.argvar = 1;
1504 req->out.args[0].size = size;
1505 req->out.args[0].value = list;
1507 req->out.args[0].size = sizeof(outarg);
1508 req->out.args[0].value = &outarg;
1510 fuse_request_send(fc, req);
1511 ret = req->out.h.error;
1513 ret = size ? req->out.args[0].size : outarg.size;
1515 if (ret == -ENOSYS) {
1516 fc->no_listxattr = 1;
1520 fuse_put_request(fc, req);
1524 static int fuse_removexattr(struct dentry *entry, const char *name)
1526 struct inode *inode = entry->d_inode;
1527 struct fuse_conn *fc = get_fuse_conn(inode);
1528 struct fuse_req *req;
1531 if (fc->no_removexattr)
1534 req = fuse_get_req(fc);
1536 return PTR_ERR(req);
1538 req->in.h.opcode = FUSE_REMOVEXATTR;
1539 req->in.h.nodeid = get_node_id(inode);
1540 req->in.numargs = 1;
1541 req->in.args[0].size = strlen(name) + 1;
1542 req->in.args[0].value = name;
1543 fuse_request_send(fc, req);
1544 err = req->out.h.error;
1545 fuse_put_request(fc, req);
1546 if (err == -ENOSYS) {
1547 fc->no_removexattr = 1;
1553 static const struct inode_operations fuse_dir_inode_operations = {
1554 .lookup = fuse_lookup,
1555 .mkdir = fuse_mkdir,
1556 .symlink = fuse_symlink,
1557 .unlink = fuse_unlink,
1558 .rmdir = fuse_rmdir,
1559 .rename = fuse_rename,
1561 .setattr = fuse_setattr,
1562 .create = fuse_create,
1563 .mknod = fuse_mknod,
1564 .permission = fuse_permission,
1565 .getattr = fuse_getattr,
1566 .setxattr = fuse_setxattr,
1567 .getxattr = fuse_getxattr,
1568 .listxattr = fuse_listxattr,
1569 .removexattr = fuse_removexattr,
1572 static const struct file_operations fuse_dir_operations = {
1573 .llseek = generic_file_llseek,
1574 .read = generic_read_dir,
1575 .readdir = fuse_readdir,
1576 .open = fuse_dir_open,
1577 .release = fuse_dir_release,
1578 .fsync = fuse_dir_fsync,
1581 static const struct inode_operations fuse_common_inode_operations = {
1582 .setattr = fuse_setattr,
1583 .permission = fuse_permission,
1584 .getattr = fuse_getattr,
1585 .setxattr = fuse_setxattr,
1586 .getxattr = fuse_getxattr,
1587 .listxattr = fuse_listxattr,
1588 .removexattr = fuse_removexattr,
1591 static const struct inode_operations fuse_symlink_inode_operations = {
1592 .setattr = fuse_setattr,
1593 .follow_link = fuse_follow_link,
1594 .put_link = fuse_put_link,
1595 .readlink = generic_readlink,
1596 .getattr = fuse_getattr,
1597 .setxattr = fuse_setxattr,
1598 .getxattr = fuse_getxattr,
1599 .listxattr = fuse_listxattr,
1600 .removexattr = fuse_removexattr,
1603 void fuse_init_common(struct inode *inode)
1605 inode->i_op = &fuse_common_inode_operations;
1608 void fuse_init_dir(struct inode *inode)
1610 inode->i_op = &fuse_dir_inode_operations;
1611 inode->i_fop = &fuse_dir_operations;
1614 void fuse_init_symlink(struct inode *inode)
1616 inode->i_op = &fuse_symlink_inode_operations;