2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/file.h>
13 #include <linux/sched.h>
14 #include <linux/namei.h>
15 #include <linux/slab.h>
17 #if BITS_PER_LONG >= 64
18 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
23 static inline u64 fuse_dentry_time(struct dentry *entry)
29 * On 32 bit archs store the high 32 bits of time in d_fsdata
31 static void fuse_dentry_settime(struct dentry *entry, u64 time)
34 entry->d_fsdata = (void *) (unsigned long) (time >> 32);
37 static u64 fuse_dentry_time(struct dentry *entry)
39 return (u64) entry->d_time +
40 ((u64) (unsigned long) entry->d_fsdata << 32);
45 * FUSE caches dentries and attributes with separate timeout. The
46 * time in jiffies until the dentry/attributes are valid is stored in
47 * dentry->d_time and fuse_inode->i_time respectively.
51 * Calculate the time in jiffies until a dentry/attributes are valid
53 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
56 struct timespec ts = {sec, nsec};
57 return get_jiffies_64() + timespec_to_jiffies(&ts);
63 * Set dentry and possibly attribute timeouts from the lookup/mk*
66 static void fuse_change_entry_timeout(struct dentry *entry,
67 struct fuse_entry_out *o)
69 fuse_dentry_settime(entry,
70 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
73 static u64 attr_timeout(struct fuse_attr_out *o)
75 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
78 static u64 entry_attr_timeout(struct fuse_entry_out *o)
80 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
84 * Mark the attributes as stale, so that at the next call to
85 * ->getattr() they will be fetched from userspace
87 void fuse_invalidate_attr(struct inode *inode)
89 get_fuse_inode(inode)->i_time = 0;
93 * Just mark the entry as stale, so that a next attempt to look it up
94 * will result in a new lookup call to userspace
96 * This is called when a dentry is about to become negative and the
97 * timeout is unknown (unlink, rmdir, rename and in some cases
100 void fuse_invalidate_entry_cache(struct dentry *entry)
102 fuse_dentry_settime(entry, 0);
106 * Same as fuse_invalidate_entry_cache(), but also try to remove the
107 * dentry from the hash
109 static void fuse_invalidate_entry(struct dentry *entry)
112 fuse_invalidate_entry_cache(entry);
115 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_req *req,
116 u64 nodeid, struct qstr *name,
117 struct fuse_entry_out *outarg)
119 memset(outarg, 0, sizeof(struct fuse_entry_out));
120 req->in.h.opcode = FUSE_LOOKUP;
121 req->in.h.nodeid = nodeid;
123 req->in.args[0].size = name->len + 1;
124 req->in.args[0].value = name->name;
125 req->out.numargs = 1;
127 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
129 req->out.args[0].size = sizeof(struct fuse_entry_out);
130 req->out.args[0].value = outarg;
133 u64 fuse_get_attr_version(struct fuse_conn *fc)
138 * The spin lock isn't actually needed on 64bit archs, but we
139 * don't yet care too much about such optimizations.
141 spin_lock(&fc->lock);
142 curr_version = fc->attr_version;
143 spin_unlock(&fc->lock);
149 * Check whether the dentry is still valid
151 * If the entry validity timeout has expired and the dentry is
152 * positive, try to redo the lookup. If the lookup results in a
153 * different inode, then let the VFS invalidate the dentry and redo
154 * the lookup once more. If the lookup results in the same inode,
155 * then refresh the attributes, timeouts and mark the dentry valid.
157 static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
161 if (nd && 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_forget_link *forget;
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 = 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 (!list_empty(&inode->i_dentry))
253 return ERR_PTR(-EBUSY);
257 return d_splice_alias(inode, entry);
260 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
261 struct fuse_entry_out *outarg, struct inode **inode)
263 struct fuse_conn *fc = get_fuse_conn_super(sb);
264 struct fuse_req *req;
265 struct fuse_forget_link *forget;
271 if (name->len > FUSE_NAME_MAX)
274 req = fuse_get_req(fc);
279 forget = fuse_alloc_forget();
282 fuse_put_request(fc, req);
286 attr_version = fuse_get_attr_version(fc);
288 fuse_lookup_init(fc, req, nodeid, name, outarg);
289 fuse_request_send(fc, req);
290 err = req->out.h.error;
291 fuse_put_request(fc, req);
292 /* Zero nodeid is same as -ENOENT, but with valid timeout */
293 if (err || !outarg->nodeid)
299 if (!fuse_valid_type(outarg->attr.mode))
302 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
303 &outarg->attr, entry_attr_timeout(outarg),
307 fuse_queue_forget(fc, forget, outarg->nodeid, 1);
318 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
319 struct nameidata *nd)
322 struct fuse_entry_out outarg;
324 struct dentry *newent;
325 struct fuse_conn *fc = get_fuse_conn(dir);
326 bool outarg_valid = true;
328 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
330 if (err == -ENOENT) {
331 outarg_valid = false;
338 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
341 if (inode && S_ISDIR(inode->i_mode)) {
342 mutex_lock(&fc->inst_mutex);
343 newent = fuse_d_add_directory(entry, inode);
344 mutex_unlock(&fc->inst_mutex);
345 err = PTR_ERR(newent);
349 newent = d_splice_alias(inode, entry);
352 entry = newent ? newent : entry;
354 fuse_change_entry_timeout(entry, &outarg);
356 fuse_invalidate_entry_cache(entry);
367 * Atomic create+open operation
369 * If the filesystem doesn't support this, then fall back to separate
370 * 'mknod' + 'open' requests.
372 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
373 struct nameidata *nd)
377 struct fuse_conn *fc = get_fuse_conn(dir);
378 struct fuse_req *req;
379 struct fuse_forget_link *forget;
380 struct fuse_create_in inarg;
381 struct fuse_open_out outopen;
382 struct fuse_entry_out outentry;
383 struct fuse_file *ff;
385 int flags = nd->intent.open.flags - 1;
390 if (flags & O_DIRECT)
393 forget = fuse_alloc_forget();
397 req = fuse_get_req(fc);
400 goto out_put_forget_req;
403 ff = fuse_file_alloc(fc);
405 goto out_put_request;
408 mode &= ~current_umask();
411 memset(&inarg, 0, sizeof(inarg));
412 memset(&outentry, 0, sizeof(outentry));
415 inarg.umask = current_umask();
416 req->in.h.opcode = FUSE_CREATE;
417 req->in.h.nodeid = get_node_id(dir);
419 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
421 req->in.args[0].value = &inarg;
422 req->in.args[1].size = entry->d_name.len + 1;
423 req->in.args[1].value = entry->d_name.name;
424 req->out.numargs = 2;
426 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
428 req->out.args[0].size = sizeof(outentry);
429 req->out.args[0].value = &outentry;
430 req->out.args[1].size = sizeof(outopen);
431 req->out.args[1].value = &outopen;
432 fuse_request_send(fc, req);
433 err = req->out.h.error;
441 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
444 fuse_put_request(fc, req);
446 ff->nodeid = outentry.nodeid;
447 ff->open_flags = outopen.open_flags;
448 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
449 &outentry.attr, entry_attr_timeout(&outentry), 0);
451 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
452 fuse_sync_release(ff, flags);
453 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
457 d_instantiate(entry, inode);
458 fuse_change_entry_timeout(entry, &outentry);
459 fuse_invalidate_attr(dir);
460 file = lookup_instantiate_filp(nd, entry, generic_file_open);
462 fuse_sync_release(ff, flags);
463 return PTR_ERR(file);
465 file->private_data = fuse_file_get(ff);
466 fuse_finish_open(inode, file);
472 fuse_put_request(fc, req);
479 * Code shared between mknod, mkdir, symlink and link
481 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
482 struct inode *dir, struct dentry *entry,
485 struct fuse_entry_out outarg;
488 struct fuse_forget_link *forget;
490 forget = fuse_alloc_forget();
492 fuse_put_request(fc, req);
496 memset(&outarg, 0, sizeof(outarg));
497 req->in.h.nodeid = get_node_id(dir);
498 req->out.numargs = 1;
500 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
502 req->out.args[0].size = sizeof(outarg);
503 req->out.args[0].value = &outarg;
504 fuse_request_send(fc, req);
505 err = req->out.h.error;
506 fuse_put_request(fc, req);
508 goto out_put_forget_req;
511 if (invalid_nodeid(outarg.nodeid))
512 goto out_put_forget_req;
514 if ((outarg.attr.mode ^ mode) & S_IFMT)
515 goto out_put_forget_req;
517 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
518 &outarg.attr, entry_attr_timeout(&outarg), 0);
520 fuse_queue_forget(fc, forget, outarg.nodeid, 1);
525 if (S_ISDIR(inode->i_mode)) {
526 struct dentry *alias;
527 mutex_lock(&fc->inst_mutex);
528 alias = d_find_alias(inode);
530 /* New directory must have moved since mkdir */
531 mutex_unlock(&fc->inst_mutex);
536 d_instantiate(entry, inode);
537 mutex_unlock(&fc->inst_mutex);
539 d_instantiate(entry, inode);
541 fuse_change_entry_timeout(entry, &outarg);
542 fuse_invalidate_attr(dir);
550 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
553 struct fuse_mknod_in inarg;
554 struct fuse_conn *fc = get_fuse_conn(dir);
555 struct fuse_req *req = fuse_get_req(fc);
560 mode &= ~current_umask();
562 memset(&inarg, 0, sizeof(inarg));
564 inarg.rdev = new_encode_dev(rdev);
565 inarg.umask = current_umask();
566 req->in.h.opcode = FUSE_MKNOD;
568 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
570 req->in.args[0].value = &inarg;
571 req->in.args[1].size = entry->d_name.len + 1;
572 req->in.args[1].value = entry->d_name.name;
573 return create_new_entry(fc, req, dir, entry, mode);
576 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
577 struct nameidata *nd)
579 if (nd && (nd->flags & LOOKUP_OPEN)) {
580 int err = fuse_create_open(dir, entry, mode, nd);
583 /* Fall back on mknod */
585 return fuse_mknod(dir, entry, mode, 0);
588 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
590 struct fuse_mkdir_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.umask = current_umask();
602 req->in.h.opcode = FUSE_MKDIR;
604 req->in.args[0].size = sizeof(inarg);
605 req->in.args[0].value = &inarg;
606 req->in.args[1].size = entry->d_name.len + 1;
607 req->in.args[1].value = entry->d_name.name;
608 return create_new_entry(fc, req, dir, entry, S_IFDIR);
611 static int fuse_symlink(struct inode *dir, struct dentry *entry,
614 struct fuse_conn *fc = get_fuse_conn(dir);
615 unsigned len = strlen(link) + 1;
616 struct fuse_req *req = fuse_get_req(fc);
620 req->in.h.opcode = FUSE_SYMLINK;
622 req->in.args[0].size = entry->d_name.len + 1;
623 req->in.args[0].value = entry->d_name.name;
624 req->in.args[1].size = len;
625 req->in.args[1].value = link;
626 return create_new_entry(fc, req, dir, entry, S_IFLNK);
629 static int fuse_unlink(struct inode *dir, struct dentry *entry)
632 struct fuse_conn *fc = get_fuse_conn(dir);
633 struct fuse_req *req = fuse_get_req(fc);
637 req->in.h.opcode = FUSE_UNLINK;
638 req->in.h.nodeid = get_node_id(dir);
640 req->in.args[0].size = entry->d_name.len + 1;
641 req->in.args[0].value = entry->d_name.name;
642 fuse_request_send(fc, req);
643 err = req->out.h.error;
644 fuse_put_request(fc, req);
646 struct inode *inode = entry->d_inode;
649 * Set nlink to zero so the inode can be cleared, if the inode
650 * does have more links this will be discovered at the next
654 fuse_invalidate_attr(inode);
655 fuse_invalidate_attr(dir);
656 fuse_invalidate_entry_cache(entry);
657 } else if (err == -EINTR)
658 fuse_invalidate_entry(entry);
662 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
665 struct fuse_conn *fc = get_fuse_conn(dir);
666 struct fuse_req *req = fuse_get_req(fc);
670 req->in.h.opcode = FUSE_RMDIR;
671 req->in.h.nodeid = get_node_id(dir);
673 req->in.args[0].size = entry->d_name.len + 1;
674 req->in.args[0].value = entry->d_name.name;
675 fuse_request_send(fc, req);
676 err = req->out.h.error;
677 fuse_put_request(fc, req);
679 clear_nlink(entry->d_inode);
680 fuse_invalidate_attr(dir);
681 fuse_invalidate_entry_cache(entry);
682 } else if (err == -EINTR)
683 fuse_invalidate_entry(entry);
687 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
688 struct inode *newdir, struct dentry *newent)
691 struct fuse_rename_in inarg;
692 struct fuse_conn *fc = get_fuse_conn(olddir);
693 struct fuse_req *req = fuse_get_req(fc);
697 memset(&inarg, 0, sizeof(inarg));
698 inarg.newdir = get_node_id(newdir);
699 req->in.h.opcode = FUSE_RENAME;
700 req->in.h.nodeid = get_node_id(olddir);
702 req->in.args[0].size = sizeof(inarg);
703 req->in.args[0].value = &inarg;
704 req->in.args[1].size = oldent->d_name.len + 1;
705 req->in.args[1].value = oldent->d_name.name;
706 req->in.args[2].size = newent->d_name.len + 1;
707 req->in.args[2].value = newent->d_name.name;
708 fuse_request_send(fc, req);
709 err = req->out.h.error;
710 fuse_put_request(fc, req);
713 fuse_invalidate_attr(oldent->d_inode);
715 fuse_invalidate_attr(olddir);
716 if (olddir != newdir)
717 fuse_invalidate_attr(newdir);
719 /* newent will end up negative */
720 if (newent->d_inode) {
721 fuse_invalidate_attr(newent->d_inode);
722 fuse_invalidate_entry_cache(newent);
724 } else if (err == -EINTR) {
725 /* If request was interrupted, DEITY only knows if the
726 rename actually took place. If the invalidation
727 fails (e.g. some process has CWD under the renamed
728 directory), then there can be inconsistency between
729 the dcache and the real filesystem. Tough luck. */
730 fuse_invalidate_entry(oldent);
732 fuse_invalidate_entry(newent);
738 static int fuse_link(struct dentry *entry, struct inode *newdir,
739 struct dentry *newent)
742 struct fuse_link_in inarg;
743 struct inode *inode = entry->d_inode;
744 struct fuse_conn *fc = get_fuse_conn(inode);
745 struct fuse_req *req = fuse_get_req(fc);
749 memset(&inarg, 0, sizeof(inarg));
750 inarg.oldnodeid = get_node_id(inode);
751 req->in.h.opcode = FUSE_LINK;
753 req->in.args[0].size = sizeof(inarg);
754 req->in.args[0].value = &inarg;
755 req->in.args[1].size = newent->d_name.len + 1;
756 req->in.args[1].value = newent->d_name.name;
757 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
758 /* Contrary to "normal" filesystems it can happen that link
759 makes two "logical" inodes point to the same "physical"
760 inode. We invalidate the attributes of the old one, so it
761 will reflect changes in the backing inode (link count,
764 if (!err || err == -EINTR)
765 fuse_invalidate_attr(inode);
769 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
772 stat->dev = inode->i_sb->s_dev;
773 stat->ino = attr->ino;
774 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
775 stat->nlink = attr->nlink;
776 stat->uid = attr->uid;
777 stat->gid = attr->gid;
778 stat->rdev = inode->i_rdev;
779 stat->atime.tv_sec = attr->atime;
780 stat->atime.tv_nsec = attr->atimensec;
781 stat->mtime.tv_sec = attr->mtime;
782 stat->mtime.tv_nsec = attr->mtimensec;
783 stat->ctime.tv_sec = attr->ctime;
784 stat->ctime.tv_nsec = attr->ctimensec;
785 stat->size = attr->size;
786 stat->blocks = attr->blocks;
787 stat->blksize = (1 << inode->i_blkbits);
790 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
794 struct fuse_getattr_in inarg;
795 struct fuse_attr_out outarg;
796 struct fuse_conn *fc = get_fuse_conn(inode);
797 struct fuse_req *req;
800 req = fuse_get_req(fc);
804 attr_version = fuse_get_attr_version(fc);
806 memset(&inarg, 0, sizeof(inarg));
807 memset(&outarg, 0, sizeof(outarg));
808 /* Directories have separate file-handle space */
809 if (file && S_ISREG(inode->i_mode)) {
810 struct fuse_file *ff = file->private_data;
812 inarg.getattr_flags |= FUSE_GETATTR_FH;
815 req->in.h.opcode = FUSE_GETATTR;
816 req->in.h.nodeid = get_node_id(inode);
818 req->in.args[0].size = sizeof(inarg);
819 req->in.args[0].value = &inarg;
820 req->out.numargs = 1;
822 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
824 req->out.args[0].size = sizeof(outarg);
825 req->out.args[0].value = &outarg;
826 fuse_request_send(fc, req);
827 err = req->out.h.error;
828 fuse_put_request(fc, req);
830 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
831 make_bad_inode(inode);
834 fuse_change_attributes(inode, &outarg.attr,
835 attr_timeout(&outarg),
838 fuse_fillattr(inode, &outarg.attr, stat);
844 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
845 struct file *file, bool *refreshed)
847 struct fuse_inode *fi = get_fuse_inode(inode);
851 if (fi->i_time < get_jiffies_64()) {
853 err = fuse_do_getattr(inode, stat, file);
858 generic_fillattr(inode, stat);
859 stat->mode = fi->orig_i_mode;
863 if (refreshed != NULL)
869 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
873 struct inode *parent;
875 struct dentry *entry;
877 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
881 mutex_lock(&parent->i_mutex);
882 if (!S_ISDIR(parent->i_mode))
886 dir = d_find_alias(parent);
890 entry = d_lookup(dir, name);
895 fuse_invalidate_attr(parent);
896 fuse_invalidate_entry(entry);
901 mutex_unlock(&parent->i_mutex);
907 * Calling into a user-controlled filesystem gives the filesystem
908 * daemon ptrace-like capabilities over the requester process. This
909 * means, that the filesystem daemon is able to record the exact
910 * filesystem operations performed, and can also control the behavior
911 * of the requester process in otherwise impossible ways. For example
912 * it can delay the operation for arbitrary length of time allowing
913 * DoS against the requester.
915 * For this reason only those processes can call into the filesystem,
916 * for which the owner of the mount has ptrace privilege. This
917 * excludes processes started by other users, suid or sgid processes.
919 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
921 const struct cred *cred;
924 if (fc->flags & FUSE_ALLOW_OTHER)
929 cred = __task_cred(task);
930 if (cred->euid == fc->user_id &&
931 cred->suid == fc->user_id &&
932 cred->uid == fc->user_id &&
933 cred->egid == fc->group_id &&
934 cred->sgid == fc->group_id &&
935 cred->gid == fc->group_id)
942 static int fuse_access(struct inode *inode, int mask)
944 struct fuse_conn *fc = get_fuse_conn(inode);
945 struct fuse_req *req;
946 struct fuse_access_in inarg;
952 req = fuse_get_req(fc);
956 memset(&inarg, 0, sizeof(inarg));
957 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
958 req->in.h.opcode = FUSE_ACCESS;
959 req->in.h.nodeid = get_node_id(inode);
961 req->in.args[0].size = sizeof(inarg);
962 req->in.args[0].value = &inarg;
963 fuse_request_send(fc, req);
964 err = req->out.h.error;
965 fuse_put_request(fc, req);
966 if (err == -ENOSYS) {
974 * Check permission. The two basic access models of FUSE are:
976 * 1) Local access checking ('default_permissions' mount option) based
977 * on file mode. This is the plain old disk filesystem permission
980 * 2) "Remote" access checking, where server is responsible for
981 * checking permission in each inode operation. An exception to this
982 * is if ->permission() was invoked from sys_access() in which case an
983 * access request is sent. Execute permission is still checked
984 * locally based on file mode.
986 static int fuse_permission(struct inode *inode, int mask, unsigned int flags)
988 struct fuse_conn *fc = get_fuse_conn(inode);
989 bool refreshed = false;
992 if (flags & IPERM_FLAG_RCU)
995 if (!fuse_allow_task(fc, current))
999 * If attributes are needed, refresh them before proceeding
1001 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1002 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1003 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
1008 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1009 err = generic_permission(inode, mask, flags, NULL);
1011 /* If permission is denied, try to refresh file
1012 attributes. This is also needed, because the root
1013 node will at first have no permissions */
1014 if (err == -EACCES && !refreshed) {
1015 err = fuse_do_getattr(inode, NULL, NULL);
1017 err = generic_permission(inode, mask,
1021 /* Note: the opposite of the above test does not
1022 exist. So if permissions are revoked this won't be
1023 noticed immediately, only after the attribute
1024 timeout has expired */
1025 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1026 err = fuse_access(inode, mask);
1027 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1028 if (!(inode->i_mode & S_IXUGO)) {
1032 err = fuse_do_getattr(inode, NULL, NULL);
1033 if (!err && !(inode->i_mode & S_IXUGO))
1040 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1041 void *dstbuf, filldir_t filldir)
1043 while (nbytes >= FUSE_NAME_OFFSET) {
1044 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1045 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1047 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1049 if (reclen > nbytes)
1052 over = filldir(dstbuf, dirent->name, dirent->namelen,
1053 file->f_pos, dirent->ino, dirent->type);
1059 file->f_pos = dirent->off;
1065 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1070 struct inode *inode = file->f_path.dentry->d_inode;
1071 struct fuse_conn *fc = get_fuse_conn(inode);
1072 struct fuse_req *req;
1074 if (is_bad_inode(inode))
1077 req = fuse_get_req(fc);
1079 return PTR_ERR(req);
1081 page = alloc_page(GFP_KERNEL);
1083 fuse_put_request(fc, req);
1086 req->out.argpages = 1;
1088 req->pages[0] = page;
1089 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1090 fuse_request_send(fc, req);
1091 nbytes = req->out.args[0].size;
1092 err = req->out.h.error;
1093 fuse_put_request(fc, req);
1095 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1099 fuse_invalidate_attr(inode); /* atime changed */
1103 static char *read_link(struct dentry *dentry)
1105 struct inode *inode = dentry->d_inode;
1106 struct fuse_conn *fc = get_fuse_conn(inode);
1107 struct fuse_req *req = fuse_get_req(fc);
1111 return ERR_CAST(req);
1113 link = (char *) __get_free_page(GFP_KERNEL);
1115 link = ERR_PTR(-ENOMEM);
1118 req->in.h.opcode = FUSE_READLINK;
1119 req->in.h.nodeid = get_node_id(inode);
1120 req->out.argvar = 1;
1121 req->out.numargs = 1;
1122 req->out.args[0].size = PAGE_SIZE - 1;
1123 req->out.args[0].value = link;
1124 fuse_request_send(fc, req);
1125 if (req->out.h.error) {
1126 free_page((unsigned long) link);
1127 link = ERR_PTR(req->out.h.error);
1129 link[req->out.args[0].size] = '\0';
1131 fuse_put_request(fc, req);
1132 fuse_invalidate_attr(inode); /* atime changed */
1136 static void free_link(char *link)
1139 free_page((unsigned long) link);
1142 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1144 nd_set_link(nd, read_link(dentry));
1148 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1150 free_link(nd_get_link(nd));
1153 static int fuse_dir_open(struct inode *inode, struct file *file)
1155 return fuse_open_common(inode, file, true);
1158 static int fuse_dir_release(struct inode *inode, struct file *file)
1160 fuse_release_common(file, FUSE_RELEASEDIR);
1165 static int fuse_dir_fsync(struct file *file, int datasync)
1167 return fuse_fsync_common(file, datasync, 1);
1170 static bool update_mtime(unsigned ivalid)
1172 /* Always update if mtime is explicitly set */
1173 if (ivalid & ATTR_MTIME_SET)
1176 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1177 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1180 /* In all other cases update */
1184 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1186 unsigned ivalid = iattr->ia_valid;
1188 if (ivalid & ATTR_MODE)
1189 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1190 if (ivalid & ATTR_UID)
1191 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1192 if (ivalid & ATTR_GID)
1193 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1194 if (ivalid & ATTR_SIZE)
1195 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1196 if (ivalid & ATTR_ATIME) {
1197 arg->valid |= FATTR_ATIME;
1198 arg->atime = iattr->ia_atime.tv_sec;
1199 arg->atimensec = iattr->ia_atime.tv_nsec;
1200 if (!(ivalid & ATTR_ATIME_SET))
1201 arg->valid |= FATTR_ATIME_NOW;
1203 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1204 arg->valid |= FATTR_MTIME;
1205 arg->mtime = iattr->ia_mtime.tv_sec;
1206 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1207 if (!(ivalid & ATTR_MTIME_SET))
1208 arg->valid |= FATTR_MTIME_NOW;
1213 * Prevent concurrent writepages on inode
1215 * This is done by adding a negative bias to the inode write counter
1216 * and waiting for all pending writes to finish.
1218 void fuse_set_nowrite(struct inode *inode)
1220 struct fuse_conn *fc = get_fuse_conn(inode);
1221 struct fuse_inode *fi = get_fuse_inode(inode);
1223 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1225 spin_lock(&fc->lock);
1226 BUG_ON(fi->writectr < 0);
1227 fi->writectr += FUSE_NOWRITE;
1228 spin_unlock(&fc->lock);
1229 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1233 * Allow writepages on inode
1235 * Remove the bias from the writecounter and send any queued
1238 static void __fuse_release_nowrite(struct inode *inode)
1240 struct fuse_inode *fi = get_fuse_inode(inode);
1242 BUG_ON(fi->writectr != FUSE_NOWRITE);
1244 fuse_flush_writepages(inode);
1247 void fuse_release_nowrite(struct inode *inode)
1249 struct fuse_conn *fc = get_fuse_conn(inode);
1251 spin_lock(&fc->lock);
1252 __fuse_release_nowrite(inode);
1253 spin_unlock(&fc->lock);
1257 * Set attributes, and at the same time refresh them.
1259 * Truncation is slightly complicated, because the 'truncate' request
1260 * may fail, in which case we don't want to touch the mapping.
1261 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1262 * and the actual truncation by hand.
1264 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1267 struct inode *inode = entry->d_inode;
1268 struct fuse_conn *fc = get_fuse_conn(inode);
1269 struct fuse_req *req;
1270 struct fuse_setattr_in inarg;
1271 struct fuse_attr_out outarg;
1272 bool is_truncate = false;
1276 if (!fuse_allow_task(fc, current))
1279 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1280 attr->ia_valid |= ATTR_FORCE;
1282 err = inode_change_ok(inode, attr);
1286 if (attr->ia_valid & ATTR_OPEN) {
1287 if (fc->atomic_o_trunc)
1292 if (attr->ia_valid & ATTR_SIZE)
1295 req = fuse_get_req(fc);
1297 return PTR_ERR(req);
1300 fuse_set_nowrite(inode);
1302 memset(&inarg, 0, sizeof(inarg));
1303 memset(&outarg, 0, sizeof(outarg));
1304 iattr_to_fattr(attr, &inarg);
1306 struct fuse_file *ff = file->private_data;
1307 inarg.valid |= FATTR_FH;
1310 if (attr->ia_valid & ATTR_SIZE) {
1311 /* For mandatory locking in truncate */
1312 inarg.valid |= FATTR_LOCKOWNER;
1313 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1315 req->in.h.opcode = FUSE_SETATTR;
1316 req->in.h.nodeid = get_node_id(inode);
1317 req->in.numargs = 1;
1318 req->in.args[0].size = sizeof(inarg);
1319 req->in.args[0].value = &inarg;
1320 req->out.numargs = 1;
1322 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1324 req->out.args[0].size = sizeof(outarg);
1325 req->out.args[0].value = &outarg;
1326 fuse_request_send(fc, req);
1327 err = req->out.h.error;
1328 fuse_put_request(fc, req);
1331 fuse_invalidate_attr(inode);
1335 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1336 make_bad_inode(inode);
1341 spin_lock(&fc->lock);
1342 fuse_change_attributes_common(inode, &outarg.attr,
1343 attr_timeout(&outarg));
1344 oldsize = inode->i_size;
1345 i_size_write(inode, outarg.attr.size);
1348 /* NOTE: this may release/reacquire fc->lock */
1349 __fuse_release_nowrite(inode);
1351 spin_unlock(&fc->lock);
1354 * Only call invalidate_inode_pages2() after removing
1355 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1357 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1358 truncate_pagecache(inode, oldsize, outarg.attr.size);
1359 invalidate_inode_pages2(inode->i_mapping);
1366 fuse_release_nowrite(inode);
1371 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1373 if (attr->ia_valid & ATTR_FILE)
1374 return fuse_do_setattr(entry, attr, attr->ia_file);
1376 return fuse_do_setattr(entry, attr, NULL);
1379 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1382 struct inode *inode = entry->d_inode;
1383 struct fuse_conn *fc = get_fuse_conn(inode);
1385 if (!fuse_allow_task(fc, current))
1388 return fuse_update_attributes(inode, stat, NULL, NULL);
1391 static int fuse_setxattr(struct dentry *entry, const char *name,
1392 const void *value, size_t size, int flags)
1394 struct inode *inode = entry->d_inode;
1395 struct fuse_conn *fc = get_fuse_conn(inode);
1396 struct fuse_req *req;
1397 struct fuse_setxattr_in inarg;
1400 if (fc->no_setxattr)
1403 req = fuse_get_req(fc);
1405 return PTR_ERR(req);
1407 memset(&inarg, 0, sizeof(inarg));
1409 inarg.flags = flags;
1410 req->in.h.opcode = FUSE_SETXATTR;
1411 req->in.h.nodeid = get_node_id(inode);
1412 req->in.numargs = 3;
1413 req->in.args[0].size = sizeof(inarg);
1414 req->in.args[0].value = &inarg;
1415 req->in.args[1].size = strlen(name) + 1;
1416 req->in.args[1].value = name;
1417 req->in.args[2].size = size;
1418 req->in.args[2].value = value;
1419 fuse_request_send(fc, req);
1420 err = req->out.h.error;
1421 fuse_put_request(fc, req);
1422 if (err == -ENOSYS) {
1423 fc->no_setxattr = 1;
1429 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1430 void *value, size_t size)
1432 struct inode *inode = entry->d_inode;
1433 struct fuse_conn *fc = get_fuse_conn(inode);
1434 struct fuse_req *req;
1435 struct fuse_getxattr_in inarg;
1436 struct fuse_getxattr_out outarg;
1439 if (fc->no_getxattr)
1442 req = fuse_get_req(fc);
1444 return PTR_ERR(req);
1446 memset(&inarg, 0, sizeof(inarg));
1448 req->in.h.opcode = FUSE_GETXATTR;
1449 req->in.h.nodeid = get_node_id(inode);
1450 req->in.numargs = 2;
1451 req->in.args[0].size = sizeof(inarg);
1452 req->in.args[0].value = &inarg;
1453 req->in.args[1].size = strlen(name) + 1;
1454 req->in.args[1].value = name;
1455 /* This is really two different operations rolled into one */
1456 req->out.numargs = 1;
1458 req->out.argvar = 1;
1459 req->out.args[0].size = size;
1460 req->out.args[0].value = value;
1462 req->out.args[0].size = sizeof(outarg);
1463 req->out.args[0].value = &outarg;
1465 fuse_request_send(fc, req);
1466 ret = req->out.h.error;
1468 ret = size ? req->out.args[0].size : outarg.size;
1470 if (ret == -ENOSYS) {
1471 fc->no_getxattr = 1;
1475 fuse_put_request(fc, req);
1479 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1481 struct inode *inode = entry->d_inode;
1482 struct fuse_conn *fc = get_fuse_conn(inode);
1483 struct fuse_req *req;
1484 struct fuse_getxattr_in inarg;
1485 struct fuse_getxattr_out outarg;
1488 if (!fuse_allow_task(fc, current))
1491 if (fc->no_listxattr)
1494 req = fuse_get_req(fc);
1496 return PTR_ERR(req);
1498 memset(&inarg, 0, sizeof(inarg));
1500 req->in.h.opcode = FUSE_LISTXATTR;
1501 req->in.h.nodeid = get_node_id(inode);
1502 req->in.numargs = 1;
1503 req->in.args[0].size = sizeof(inarg);
1504 req->in.args[0].value = &inarg;
1505 /* This is really two different operations rolled into one */
1506 req->out.numargs = 1;
1508 req->out.argvar = 1;
1509 req->out.args[0].size = size;
1510 req->out.args[0].value = list;
1512 req->out.args[0].size = sizeof(outarg);
1513 req->out.args[0].value = &outarg;
1515 fuse_request_send(fc, req);
1516 ret = req->out.h.error;
1518 ret = size ? req->out.args[0].size : outarg.size;
1520 if (ret == -ENOSYS) {
1521 fc->no_listxattr = 1;
1525 fuse_put_request(fc, req);
1529 static int fuse_removexattr(struct dentry *entry, const char *name)
1531 struct inode *inode = entry->d_inode;
1532 struct fuse_conn *fc = get_fuse_conn(inode);
1533 struct fuse_req *req;
1536 if (fc->no_removexattr)
1539 req = fuse_get_req(fc);
1541 return PTR_ERR(req);
1543 req->in.h.opcode = FUSE_REMOVEXATTR;
1544 req->in.h.nodeid = get_node_id(inode);
1545 req->in.numargs = 1;
1546 req->in.args[0].size = strlen(name) + 1;
1547 req->in.args[0].value = name;
1548 fuse_request_send(fc, req);
1549 err = req->out.h.error;
1550 fuse_put_request(fc, req);
1551 if (err == -ENOSYS) {
1552 fc->no_removexattr = 1;
1558 static const struct inode_operations fuse_dir_inode_operations = {
1559 .lookup = fuse_lookup,
1560 .mkdir = fuse_mkdir,
1561 .symlink = fuse_symlink,
1562 .unlink = fuse_unlink,
1563 .rmdir = fuse_rmdir,
1564 .rename = fuse_rename,
1566 .setattr = fuse_setattr,
1567 .create = fuse_create,
1568 .mknod = fuse_mknod,
1569 .permission = fuse_permission,
1570 .getattr = fuse_getattr,
1571 .setxattr = fuse_setxattr,
1572 .getxattr = fuse_getxattr,
1573 .listxattr = fuse_listxattr,
1574 .removexattr = fuse_removexattr,
1577 static const struct file_operations fuse_dir_operations = {
1578 .llseek = generic_file_llseek,
1579 .read = generic_read_dir,
1580 .readdir = fuse_readdir,
1581 .open = fuse_dir_open,
1582 .release = fuse_dir_release,
1583 .fsync = fuse_dir_fsync,
1586 static const struct inode_operations fuse_common_inode_operations = {
1587 .setattr = fuse_setattr,
1588 .permission = fuse_permission,
1589 .getattr = fuse_getattr,
1590 .setxattr = fuse_setxattr,
1591 .getxattr = fuse_getxattr,
1592 .listxattr = fuse_listxattr,
1593 .removexattr = fuse_removexattr,
1596 static const struct inode_operations fuse_symlink_inode_operations = {
1597 .setattr = fuse_setattr,
1598 .follow_link = fuse_follow_link,
1599 .put_link = fuse_put_link,
1600 .readlink = generic_readlink,
1601 .getattr = fuse_getattr,
1602 .setxattr = fuse_setxattr,
1603 .getxattr = fuse_getxattr,
1604 .listxattr = fuse_listxattr,
1605 .removexattr = fuse_removexattr,
1608 void fuse_init_common(struct inode *inode)
1610 inode->i_op = &fuse_common_inode_operations;
1613 void fuse_init_dir(struct inode *inode)
1615 inode->i_op = &fuse_dir_inode_operations;
1616 inode->i_fop = &fuse_dir_operations;
1619 void fuse_init_symlink(struct inode *inode)
1621 inode->i_op = &fuse_symlink_inode_operations;