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/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)
159 struct inode *inode = entry->d_inode;
161 if (inode && is_bad_inode(inode))
163 else if (fuse_dentry_time(entry) < get_jiffies_64()) {
165 struct fuse_entry_out outarg;
166 struct fuse_conn *fc;
167 struct fuse_req *req;
168 struct fuse_req *forget_req;
169 struct dentry *parent;
172 /* For negative dentries, always do a fresh lookup */
176 fc = get_fuse_conn(inode);
177 req = fuse_get_req(fc);
181 forget_req = fuse_get_req(fc);
182 if (IS_ERR(forget_req)) {
183 fuse_put_request(fc, req);
187 attr_version = fuse_get_attr_version(fc);
189 parent = dget_parent(entry);
190 fuse_lookup_init(fc, req, get_node_id(parent->d_inode),
191 &entry->d_name, &outarg);
192 fuse_request_send(fc, req);
194 err = req->out.h.error;
195 fuse_put_request(fc, req);
196 /* Zero nodeid is same as -ENOENT */
197 if (!err && !outarg.nodeid)
200 struct fuse_inode *fi = get_fuse_inode(inode);
201 if (outarg.nodeid != get_node_id(inode)) {
202 fuse_send_forget(fc, forget_req,
206 spin_lock(&fc->lock);
208 spin_unlock(&fc->lock);
210 fuse_put_request(fc, forget_req);
211 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
214 fuse_change_attributes(inode, &outarg.attr,
215 entry_attr_timeout(&outarg),
217 fuse_change_entry_timeout(entry, &outarg);
222 static int invalid_nodeid(u64 nodeid)
224 return !nodeid || nodeid == FUSE_ROOT_ID;
227 const struct dentry_operations fuse_dentry_operations = {
228 .d_revalidate = fuse_dentry_revalidate,
231 int fuse_valid_type(int m)
233 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
234 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
238 * Add a directory inode to a dentry, ensuring that no other dentry
239 * refers to this inode. Called with fc->inst_mutex.
241 static struct dentry *fuse_d_add_directory(struct dentry *entry,
244 struct dentry *alias = d_find_alias(inode);
245 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
246 /* This tries to shrink the subtree below alias */
247 fuse_invalidate_entry(alias);
249 if (!list_empty(&inode->i_dentry))
250 return ERR_PTR(-EBUSY);
254 return d_splice_alias(inode, entry);
257 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
258 struct fuse_entry_out *outarg, struct inode **inode)
260 struct fuse_conn *fc = get_fuse_conn_super(sb);
261 struct fuse_req *req;
262 struct fuse_req *forget_req;
268 if (name->len > FUSE_NAME_MAX)
271 req = fuse_get_req(fc);
276 forget_req = fuse_get_req(fc);
277 err = PTR_ERR(forget_req);
278 if (IS_ERR(forget_req)) {
279 fuse_put_request(fc, req);
283 attr_version = fuse_get_attr_version(fc);
285 fuse_lookup_init(fc, req, nodeid, name, outarg);
286 fuse_request_send(fc, req);
287 err = req->out.h.error;
288 fuse_put_request(fc, req);
289 /* Zero nodeid is same as -ENOENT, but with valid timeout */
290 if (err || !outarg->nodeid)
296 if (!fuse_valid_type(outarg->attr.mode))
299 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
300 &outarg->attr, entry_attr_timeout(outarg),
304 fuse_send_forget(fc, forget_req, outarg->nodeid, 1);
310 fuse_put_request(fc, forget_req);
315 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
316 struct nameidata *nd)
319 struct fuse_entry_out outarg;
321 struct dentry *newent;
322 struct fuse_conn *fc = get_fuse_conn(dir);
323 bool outarg_valid = true;
325 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
327 if (err == -ENOENT) {
328 outarg_valid = false;
335 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
338 if (inode && S_ISDIR(inode->i_mode)) {
339 mutex_lock(&fc->inst_mutex);
340 newent = fuse_d_add_directory(entry, inode);
341 mutex_unlock(&fc->inst_mutex);
342 err = PTR_ERR(newent);
346 newent = d_splice_alias(inode, entry);
349 entry = newent ? newent : entry;
350 entry->d_op = &fuse_dentry_operations;
352 fuse_change_entry_timeout(entry, &outarg);
354 fuse_invalidate_entry_cache(entry);
365 * Synchronous release for the case when something goes wrong in CREATE_OPEN
367 static void fuse_sync_release(struct fuse_conn *fc, struct fuse_file *ff,
368 u64 nodeid, int flags)
370 fuse_release_fill(ff, nodeid, flags, FUSE_RELEASE);
371 ff->reserved_req->force = 1;
372 fuse_request_send(fc, ff->reserved_req);
373 fuse_put_request(fc, ff->reserved_req);
378 * Atomic create+open operation
380 * If the filesystem doesn't support this, then fall back to separate
381 * 'mknod' + 'open' requests.
383 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
384 struct nameidata *nd)
388 struct fuse_conn *fc = get_fuse_conn(dir);
389 struct fuse_req *req;
390 struct fuse_req *forget_req;
391 struct fuse_open_in inarg;
392 struct fuse_open_out outopen;
393 struct fuse_entry_out outentry;
394 struct fuse_file *ff;
396 int flags = nd->intent.open.flags - 1;
401 forget_req = fuse_get_req(fc);
402 if (IS_ERR(forget_req))
403 return PTR_ERR(forget_req);
405 req = fuse_get_req(fc);
408 goto out_put_forget_req;
411 ff = fuse_file_alloc(fc);
413 goto out_put_request;
416 memset(&inarg, 0, sizeof(inarg));
417 memset(&outentry, 0, sizeof(outentry));
420 req->in.h.opcode = FUSE_CREATE;
421 req->in.h.nodeid = get_node_id(dir);
423 req->in.args[0].size = sizeof(inarg);
424 req->in.args[0].value = &inarg;
425 req->in.args[1].size = entry->d_name.len + 1;
426 req->in.args[1].value = entry->d_name.name;
427 req->out.numargs = 2;
429 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
431 req->out.args[0].size = sizeof(outentry);
432 req->out.args[0].value = &outentry;
433 req->out.args[1].size = sizeof(outopen);
434 req->out.args[1].value = &outopen;
435 fuse_request_send(fc, req);
436 err = req->out.h.error;
444 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
447 fuse_put_request(fc, req);
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);
453 fuse_sync_release(fc, ff, outentry.nodeid, flags);
454 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
457 fuse_put_request(fc, forget_req);
458 d_instantiate(entry, inode);
459 fuse_change_entry_timeout(entry, &outentry);
460 fuse_invalidate_attr(dir);
461 file = lookup_instantiate_filp(nd, entry, generic_file_open);
464 fuse_sync_release(fc, ff, outentry.nodeid, flags);
465 return PTR_ERR(file);
467 fuse_finish_open(inode, file, ff, &outopen);
473 fuse_put_request(fc, req);
475 fuse_put_request(fc, forget_req);
480 * Code shared between mknod, mkdir, symlink and link
482 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
483 struct inode *dir, struct dentry *entry,
486 struct fuse_entry_out outarg;
489 struct fuse_req *forget_req;
491 forget_req = fuse_get_req(fc);
492 if (IS_ERR(forget_req)) {
493 fuse_put_request(fc, req);
494 return PTR_ERR(forget_req);
497 memset(&outarg, 0, sizeof(outarg));
498 req->in.h.nodeid = get_node_id(dir);
499 req->out.numargs = 1;
501 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
503 req->out.args[0].size = sizeof(outarg);
504 req->out.args[0].value = &outarg;
505 fuse_request_send(fc, req);
506 err = req->out.h.error;
507 fuse_put_request(fc, req);
509 goto out_put_forget_req;
512 if (invalid_nodeid(outarg.nodeid))
513 goto out_put_forget_req;
515 if ((outarg.attr.mode ^ mode) & S_IFMT)
516 goto out_put_forget_req;
518 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
519 &outarg.attr, entry_attr_timeout(&outarg), 0);
521 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
524 fuse_put_request(fc, forget_req);
526 if (S_ISDIR(inode->i_mode)) {
527 struct dentry *alias;
528 mutex_lock(&fc->inst_mutex);
529 alias = d_find_alias(inode);
531 /* New directory must have moved since mkdir */
532 mutex_unlock(&fc->inst_mutex);
537 d_instantiate(entry, inode);
538 mutex_unlock(&fc->inst_mutex);
540 d_instantiate(entry, inode);
542 fuse_change_entry_timeout(entry, &outarg);
543 fuse_invalidate_attr(dir);
547 fuse_put_request(fc, forget_req);
551 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
554 struct fuse_mknod_in inarg;
555 struct fuse_conn *fc = get_fuse_conn(dir);
556 struct fuse_req *req = fuse_get_req(fc);
560 memset(&inarg, 0, sizeof(inarg));
562 inarg.rdev = new_encode_dev(rdev);
563 req->in.h.opcode = FUSE_MKNOD;
565 req->in.args[0].size = sizeof(inarg);
566 req->in.args[0].value = &inarg;
567 req->in.args[1].size = entry->d_name.len + 1;
568 req->in.args[1].value = entry->d_name.name;
569 return create_new_entry(fc, req, dir, entry, mode);
572 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
573 struct nameidata *nd)
575 if (nd && (nd->flags & LOOKUP_OPEN)) {
576 int err = fuse_create_open(dir, entry, mode, nd);
579 /* Fall back on mknod */
581 return fuse_mknod(dir, entry, mode, 0);
584 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
586 struct fuse_mkdir_in inarg;
587 struct fuse_conn *fc = get_fuse_conn(dir);
588 struct fuse_req *req = fuse_get_req(fc);
592 memset(&inarg, 0, sizeof(inarg));
594 req->in.h.opcode = FUSE_MKDIR;
596 req->in.args[0].size = sizeof(inarg);
597 req->in.args[0].value = &inarg;
598 req->in.args[1].size = entry->d_name.len + 1;
599 req->in.args[1].value = entry->d_name.name;
600 return create_new_entry(fc, req, dir, entry, S_IFDIR);
603 static int fuse_symlink(struct inode *dir, struct dentry *entry,
606 struct fuse_conn *fc = get_fuse_conn(dir);
607 unsigned len = strlen(link) + 1;
608 struct fuse_req *req = fuse_get_req(fc);
612 req->in.h.opcode = FUSE_SYMLINK;
614 req->in.args[0].size = entry->d_name.len + 1;
615 req->in.args[0].value = entry->d_name.name;
616 req->in.args[1].size = len;
617 req->in.args[1].value = link;
618 return create_new_entry(fc, req, dir, entry, S_IFLNK);
621 static int fuse_unlink(struct inode *dir, struct dentry *entry)
624 struct fuse_conn *fc = get_fuse_conn(dir);
625 struct fuse_req *req = fuse_get_req(fc);
629 req->in.h.opcode = FUSE_UNLINK;
630 req->in.h.nodeid = get_node_id(dir);
632 req->in.args[0].size = entry->d_name.len + 1;
633 req->in.args[0].value = entry->d_name.name;
634 fuse_request_send(fc, req);
635 err = req->out.h.error;
636 fuse_put_request(fc, req);
638 struct inode *inode = entry->d_inode;
641 * Set nlink to zero so the inode can be cleared, if the inode
642 * does have more links this will be discovered at the next
646 fuse_invalidate_attr(inode);
647 fuse_invalidate_attr(dir);
648 fuse_invalidate_entry_cache(entry);
649 } else if (err == -EINTR)
650 fuse_invalidate_entry(entry);
654 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
657 struct fuse_conn *fc = get_fuse_conn(dir);
658 struct fuse_req *req = fuse_get_req(fc);
662 req->in.h.opcode = FUSE_RMDIR;
663 req->in.h.nodeid = get_node_id(dir);
665 req->in.args[0].size = entry->d_name.len + 1;
666 req->in.args[0].value = entry->d_name.name;
667 fuse_request_send(fc, req);
668 err = req->out.h.error;
669 fuse_put_request(fc, req);
671 clear_nlink(entry->d_inode);
672 fuse_invalidate_attr(dir);
673 fuse_invalidate_entry_cache(entry);
674 } else if (err == -EINTR)
675 fuse_invalidate_entry(entry);
679 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
680 struct inode *newdir, struct dentry *newent)
683 struct fuse_rename_in inarg;
684 struct fuse_conn *fc = get_fuse_conn(olddir);
685 struct fuse_req *req = fuse_get_req(fc);
689 memset(&inarg, 0, sizeof(inarg));
690 inarg.newdir = get_node_id(newdir);
691 req->in.h.opcode = FUSE_RENAME;
692 req->in.h.nodeid = get_node_id(olddir);
694 req->in.args[0].size = sizeof(inarg);
695 req->in.args[0].value = &inarg;
696 req->in.args[1].size = oldent->d_name.len + 1;
697 req->in.args[1].value = oldent->d_name.name;
698 req->in.args[2].size = newent->d_name.len + 1;
699 req->in.args[2].value = newent->d_name.name;
700 fuse_request_send(fc, req);
701 err = req->out.h.error;
702 fuse_put_request(fc, req);
705 fuse_invalidate_attr(oldent->d_inode);
707 fuse_invalidate_attr(olddir);
708 if (olddir != newdir)
709 fuse_invalidate_attr(newdir);
711 /* newent will end up negative */
713 fuse_invalidate_entry_cache(newent);
714 } else if (err == -EINTR) {
715 /* If request was interrupted, DEITY only knows if the
716 rename actually took place. If the invalidation
717 fails (e.g. some process has CWD under the renamed
718 directory), then there can be inconsistency between
719 the dcache and the real filesystem. Tough luck. */
720 fuse_invalidate_entry(oldent);
722 fuse_invalidate_entry(newent);
728 static int fuse_link(struct dentry *entry, struct inode *newdir,
729 struct dentry *newent)
732 struct fuse_link_in inarg;
733 struct inode *inode = entry->d_inode;
734 struct fuse_conn *fc = get_fuse_conn(inode);
735 struct fuse_req *req = fuse_get_req(fc);
739 memset(&inarg, 0, sizeof(inarg));
740 inarg.oldnodeid = get_node_id(inode);
741 req->in.h.opcode = FUSE_LINK;
743 req->in.args[0].size = sizeof(inarg);
744 req->in.args[0].value = &inarg;
745 req->in.args[1].size = newent->d_name.len + 1;
746 req->in.args[1].value = newent->d_name.name;
747 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
748 /* Contrary to "normal" filesystems it can happen that link
749 makes two "logical" inodes point to the same "physical"
750 inode. We invalidate the attributes of the old one, so it
751 will reflect changes in the backing inode (link count,
754 if (!err || err == -EINTR)
755 fuse_invalidate_attr(inode);
759 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
762 stat->dev = inode->i_sb->s_dev;
763 stat->ino = attr->ino;
764 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
765 stat->nlink = attr->nlink;
766 stat->uid = attr->uid;
767 stat->gid = attr->gid;
768 stat->rdev = inode->i_rdev;
769 stat->atime.tv_sec = attr->atime;
770 stat->atime.tv_nsec = attr->atimensec;
771 stat->mtime.tv_sec = attr->mtime;
772 stat->mtime.tv_nsec = attr->mtimensec;
773 stat->ctime.tv_sec = attr->ctime;
774 stat->ctime.tv_nsec = attr->ctimensec;
775 stat->size = attr->size;
776 stat->blocks = attr->blocks;
777 stat->blksize = (1 << inode->i_blkbits);
780 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
784 struct fuse_getattr_in inarg;
785 struct fuse_attr_out outarg;
786 struct fuse_conn *fc = get_fuse_conn(inode);
787 struct fuse_req *req;
790 req = fuse_get_req(fc);
794 attr_version = fuse_get_attr_version(fc);
796 memset(&inarg, 0, sizeof(inarg));
797 memset(&outarg, 0, sizeof(outarg));
798 /* Directories have separate file-handle space */
799 if (file && S_ISREG(inode->i_mode)) {
800 struct fuse_file *ff = file->private_data;
802 inarg.getattr_flags |= FUSE_GETATTR_FH;
805 req->in.h.opcode = FUSE_GETATTR;
806 req->in.h.nodeid = get_node_id(inode);
808 req->in.args[0].size = sizeof(inarg);
809 req->in.args[0].value = &inarg;
810 req->out.numargs = 1;
812 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
814 req->out.args[0].size = sizeof(outarg);
815 req->out.args[0].value = &outarg;
816 fuse_request_send(fc, req);
817 err = req->out.h.error;
818 fuse_put_request(fc, req);
820 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
821 make_bad_inode(inode);
824 fuse_change_attributes(inode, &outarg.attr,
825 attr_timeout(&outarg),
828 fuse_fillattr(inode, &outarg.attr, stat);
834 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
835 struct file *file, bool *refreshed)
837 struct fuse_inode *fi = get_fuse_inode(inode);
841 if (fi->i_time < get_jiffies_64()) {
843 err = fuse_do_getattr(inode, stat, file);
848 generic_fillattr(inode, stat);
849 stat->mode = fi->orig_i_mode;
853 if (refreshed != NULL)
860 * Calling into a user-controlled filesystem gives the filesystem
861 * daemon ptrace-like capabilities over the requester process. This
862 * means, that the filesystem daemon is able to record the exact
863 * filesystem operations performed, and can also control the behavior
864 * of the requester process in otherwise impossible ways. For example
865 * it can delay the operation for arbitrary length of time allowing
866 * DoS against the requester.
868 * For this reason only those processes can call into the filesystem,
869 * for which the owner of the mount has ptrace privilege. This
870 * excludes processes started by other users, suid or sgid processes.
872 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
874 const struct cred *cred;
877 if (fc->flags & FUSE_ALLOW_OTHER)
882 cred = __task_cred(task);
883 if (cred->euid == fc->user_id &&
884 cred->suid == fc->user_id &&
885 cred->uid == fc->user_id &&
886 cred->egid == fc->group_id &&
887 cred->sgid == fc->group_id &&
888 cred->gid == fc->group_id)
895 static int fuse_access(struct inode *inode, int mask)
897 struct fuse_conn *fc = get_fuse_conn(inode);
898 struct fuse_req *req;
899 struct fuse_access_in inarg;
905 req = fuse_get_req(fc);
909 memset(&inarg, 0, sizeof(inarg));
910 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
911 req->in.h.opcode = FUSE_ACCESS;
912 req->in.h.nodeid = get_node_id(inode);
914 req->in.args[0].size = sizeof(inarg);
915 req->in.args[0].value = &inarg;
916 fuse_request_send(fc, req);
917 err = req->out.h.error;
918 fuse_put_request(fc, req);
919 if (err == -ENOSYS) {
927 * Check permission. The two basic access models of FUSE are:
929 * 1) Local access checking ('default_permissions' mount option) based
930 * on file mode. This is the plain old disk filesystem permission
933 * 2) "Remote" access checking, where server is responsible for
934 * checking permission in each inode operation. An exception to this
935 * is if ->permission() was invoked from sys_access() in which case an
936 * access request is sent. Execute permission is still checked
937 * locally based on file mode.
939 static int fuse_permission(struct inode *inode, int mask)
941 struct fuse_conn *fc = get_fuse_conn(inode);
942 bool refreshed = false;
945 if (!fuse_allow_task(fc, current))
949 * If attributes are needed, refresh them before proceeding
951 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
952 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
953 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
958 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
959 err = generic_permission(inode, mask, NULL);
961 /* If permission is denied, try to refresh file
962 attributes. This is also needed, because the root
963 node will at first have no permissions */
964 if (err == -EACCES && !refreshed) {
965 err = fuse_do_getattr(inode, NULL, NULL);
967 err = generic_permission(inode, mask, NULL);
970 /* Note: the opposite of the above test does not
971 exist. So if permissions are revoked this won't be
972 noticed immediately, only after the attribute
973 timeout has expired */
974 } else if (mask & MAY_ACCESS) {
975 err = fuse_access(inode, mask);
976 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
977 if (!(inode->i_mode & S_IXUGO)) {
981 err = fuse_do_getattr(inode, NULL, NULL);
982 if (!err && !(inode->i_mode & S_IXUGO))
989 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
990 void *dstbuf, filldir_t filldir)
992 while (nbytes >= FUSE_NAME_OFFSET) {
993 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
994 size_t reclen = FUSE_DIRENT_SIZE(dirent);
996 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1001 over = filldir(dstbuf, dirent->name, dirent->namelen,
1002 file->f_pos, dirent->ino, dirent->type);
1008 file->f_pos = dirent->off;
1014 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1019 struct inode *inode = file->f_path.dentry->d_inode;
1020 struct fuse_conn *fc = get_fuse_conn(inode);
1021 struct fuse_req *req;
1023 if (is_bad_inode(inode))
1026 req = fuse_get_req(fc);
1028 return PTR_ERR(req);
1030 page = alloc_page(GFP_KERNEL);
1032 fuse_put_request(fc, req);
1036 req->pages[0] = page;
1037 fuse_read_fill(req, file, inode, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1038 fuse_request_send(fc, req);
1039 nbytes = req->out.args[0].size;
1040 err = req->out.h.error;
1041 fuse_put_request(fc, req);
1043 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1047 fuse_invalidate_attr(inode); /* atime changed */
1051 static char *read_link(struct dentry *dentry)
1053 struct inode *inode = dentry->d_inode;
1054 struct fuse_conn *fc = get_fuse_conn(inode);
1055 struct fuse_req *req = fuse_get_req(fc);
1059 return ERR_CAST(req);
1061 link = (char *) __get_free_page(GFP_KERNEL);
1063 link = ERR_PTR(-ENOMEM);
1066 req->in.h.opcode = FUSE_READLINK;
1067 req->in.h.nodeid = get_node_id(inode);
1068 req->out.argvar = 1;
1069 req->out.numargs = 1;
1070 req->out.args[0].size = PAGE_SIZE - 1;
1071 req->out.args[0].value = link;
1072 fuse_request_send(fc, req);
1073 if (req->out.h.error) {
1074 free_page((unsigned long) link);
1075 link = ERR_PTR(req->out.h.error);
1077 link[req->out.args[0].size] = '\0';
1079 fuse_put_request(fc, req);
1080 fuse_invalidate_attr(inode); /* atime changed */
1084 static void free_link(char *link)
1087 free_page((unsigned long) link);
1090 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1092 nd_set_link(nd, read_link(dentry));
1096 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1098 free_link(nd_get_link(nd));
1101 static int fuse_dir_open(struct inode *inode, struct file *file)
1103 return fuse_open_common(inode, file, 1);
1106 static int fuse_dir_release(struct inode *inode, struct file *file)
1108 return fuse_release_common(inode, file, 1);
1111 static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
1113 /* nfsd can call this with no file */
1114 return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
1117 static bool update_mtime(unsigned ivalid)
1119 /* Always update if mtime is explicitly set */
1120 if (ivalid & ATTR_MTIME_SET)
1123 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1124 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1127 /* In all other cases update */
1131 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1133 unsigned ivalid = iattr->ia_valid;
1135 if (ivalid & ATTR_MODE)
1136 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1137 if (ivalid & ATTR_UID)
1138 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1139 if (ivalid & ATTR_GID)
1140 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1141 if (ivalid & ATTR_SIZE)
1142 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1143 if (ivalid & ATTR_ATIME) {
1144 arg->valid |= FATTR_ATIME;
1145 arg->atime = iattr->ia_atime.tv_sec;
1146 arg->atimensec = iattr->ia_atime.tv_nsec;
1147 if (!(ivalid & ATTR_ATIME_SET))
1148 arg->valid |= FATTR_ATIME_NOW;
1150 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1151 arg->valid |= FATTR_MTIME;
1152 arg->mtime = iattr->ia_mtime.tv_sec;
1153 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1154 if (!(ivalid & ATTR_MTIME_SET))
1155 arg->valid |= FATTR_MTIME_NOW;
1160 * Prevent concurrent writepages on inode
1162 * This is done by adding a negative bias to the inode write counter
1163 * and waiting for all pending writes to finish.
1165 void fuse_set_nowrite(struct inode *inode)
1167 struct fuse_conn *fc = get_fuse_conn(inode);
1168 struct fuse_inode *fi = get_fuse_inode(inode);
1170 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1172 spin_lock(&fc->lock);
1173 BUG_ON(fi->writectr < 0);
1174 fi->writectr += FUSE_NOWRITE;
1175 spin_unlock(&fc->lock);
1176 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1180 * Allow writepages on inode
1182 * Remove the bias from the writecounter and send any queued
1185 static void __fuse_release_nowrite(struct inode *inode)
1187 struct fuse_inode *fi = get_fuse_inode(inode);
1189 BUG_ON(fi->writectr != FUSE_NOWRITE);
1191 fuse_flush_writepages(inode);
1194 void fuse_release_nowrite(struct inode *inode)
1196 struct fuse_conn *fc = get_fuse_conn(inode);
1198 spin_lock(&fc->lock);
1199 __fuse_release_nowrite(inode);
1200 spin_unlock(&fc->lock);
1204 * Set attributes, and at the same time refresh them.
1206 * Truncation is slightly complicated, because the 'truncate' request
1207 * may fail, in which case we don't want to touch the mapping.
1208 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1209 * and the actual truncation by hand.
1211 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1214 struct inode *inode = entry->d_inode;
1215 struct fuse_conn *fc = get_fuse_conn(inode);
1216 struct fuse_req *req;
1217 struct fuse_setattr_in inarg;
1218 struct fuse_attr_out outarg;
1219 bool is_truncate = false;
1223 if (!fuse_allow_task(fc, current))
1226 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1227 err = inode_change_ok(inode, attr);
1232 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1235 if (attr->ia_valid & ATTR_SIZE) {
1236 unsigned long limit;
1237 if (IS_SWAPFILE(inode))
1239 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1240 if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
1241 send_sig(SIGXFSZ, current, 0);
1247 req = fuse_get_req(fc);
1249 return PTR_ERR(req);
1252 fuse_set_nowrite(inode);
1254 memset(&inarg, 0, sizeof(inarg));
1255 memset(&outarg, 0, sizeof(outarg));
1256 iattr_to_fattr(attr, &inarg);
1258 struct fuse_file *ff = file->private_data;
1259 inarg.valid |= FATTR_FH;
1262 if (attr->ia_valid & ATTR_SIZE) {
1263 /* For mandatory locking in truncate */
1264 inarg.valid |= FATTR_LOCKOWNER;
1265 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1267 req->in.h.opcode = FUSE_SETATTR;
1268 req->in.h.nodeid = get_node_id(inode);
1269 req->in.numargs = 1;
1270 req->in.args[0].size = sizeof(inarg);
1271 req->in.args[0].value = &inarg;
1272 req->out.numargs = 1;
1274 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1276 req->out.args[0].size = sizeof(outarg);
1277 req->out.args[0].value = &outarg;
1278 fuse_request_send(fc, req);
1279 err = req->out.h.error;
1280 fuse_put_request(fc, req);
1283 fuse_invalidate_attr(inode);
1287 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1288 make_bad_inode(inode);
1293 spin_lock(&fc->lock);
1294 fuse_change_attributes_common(inode, &outarg.attr,
1295 attr_timeout(&outarg));
1296 oldsize = inode->i_size;
1297 i_size_write(inode, outarg.attr.size);
1300 /* NOTE: this may release/reacquire fc->lock */
1301 __fuse_release_nowrite(inode);
1303 spin_unlock(&fc->lock);
1306 * Only call invalidate_inode_pages2() after removing
1307 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1309 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1310 if (outarg.attr.size < oldsize)
1311 fuse_truncate(inode->i_mapping, outarg.attr.size);
1312 invalidate_inode_pages2(inode->i_mapping);
1319 fuse_release_nowrite(inode);
1324 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1326 if (attr->ia_valid & ATTR_FILE)
1327 return fuse_do_setattr(entry, attr, attr->ia_file);
1329 return fuse_do_setattr(entry, attr, NULL);
1332 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1335 struct inode *inode = entry->d_inode;
1336 struct fuse_conn *fc = get_fuse_conn(inode);
1338 if (!fuse_allow_task(fc, current))
1341 return fuse_update_attributes(inode, stat, NULL, NULL);
1344 static int fuse_setxattr(struct dentry *entry, const char *name,
1345 const void *value, size_t size, int flags)
1347 struct inode *inode = entry->d_inode;
1348 struct fuse_conn *fc = get_fuse_conn(inode);
1349 struct fuse_req *req;
1350 struct fuse_setxattr_in inarg;
1353 if (fc->no_setxattr)
1356 req = fuse_get_req(fc);
1358 return PTR_ERR(req);
1360 memset(&inarg, 0, sizeof(inarg));
1362 inarg.flags = flags;
1363 req->in.h.opcode = FUSE_SETXATTR;
1364 req->in.h.nodeid = get_node_id(inode);
1365 req->in.numargs = 3;
1366 req->in.args[0].size = sizeof(inarg);
1367 req->in.args[0].value = &inarg;
1368 req->in.args[1].size = strlen(name) + 1;
1369 req->in.args[1].value = name;
1370 req->in.args[2].size = size;
1371 req->in.args[2].value = value;
1372 fuse_request_send(fc, req);
1373 err = req->out.h.error;
1374 fuse_put_request(fc, req);
1375 if (err == -ENOSYS) {
1376 fc->no_setxattr = 1;
1382 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1383 void *value, size_t size)
1385 struct inode *inode = entry->d_inode;
1386 struct fuse_conn *fc = get_fuse_conn(inode);
1387 struct fuse_req *req;
1388 struct fuse_getxattr_in inarg;
1389 struct fuse_getxattr_out outarg;
1392 if (fc->no_getxattr)
1395 req = fuse_get_req(fc);
1397 return PTR_ERR(req);
1399 memset(&inarg, 0, sizeof(inarg));
1401 req->in.h.opcode = FUSE_GETXATTR;
1402 req->in.h.nodeid = get_node_id(inode);
1403 req->in.numargs = 2;
1404 req->in.args[0].size = sizeof(inarg);
1405 req->in.args[0].value = &inarg;
1406 req->in.args[1].size = strlen(name) + 1;
1407 req->in.args[1].value = name;
1408 /* This is really two different operations rolled into one */
1409 req->out.numargs = 1;
1411 req->out.argvar = 1;
1412 req->out.args[0].size = size;
1413 req->out.args[0].value = value;
1415 req->out.args[0].size = sizeof(outarg);
1416 req->out.args[0].value = &outarg;
1418 fuse_request_send(fc, req);
1419 ret = req->out.h.error;
1421 ret = size ? req->out.args[0].size : outarg.size;
1423 if (ret == -ENOSYS) {
1424 fc->no_getxattr = 1;
1428 fuse_put_request(fc, req);
1432 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1434 struct inode *inode = entry->d_inode;
1435 struct fuse_conn *fc = get_fuse_conn(inode);
1436 struct fuse_req *req;
1437 struct fuse_getxattr_in inarg;
1438 struct fuse_getxattr_out outarg;
1441 if (!fuse_allow_task(fc, current))
1444 if (fc->no_listxattr)
1447 req = fuse_get_req(fc);
1449 return PTR_ERR(req);
1451 memset(&inarg, 0, sizeof(inarg));
1453 req->in.h.opcode = FUSE_LISTXATTR;
1454 req->in.h.nodeid = get_node_id(inode);
1455 req->in.numargs = 1;
1456 req->in.args[0].size = sizeof(inarg);
1457 req->in.args[0].value = &inarg;
1458 /* This is really two different operations rolled into one */
1459 req->out.numargs = 1;
1461 req->out.argvar = 1;
1462 req->out.args[0].size = size;
1463 req->out.args[0].value = list;
1465 req->out.args[0].size = sizeof(outarg);
1466 req->out.args[0].value = &outarg;
1468 fuse_request_send(fc, req);
1469 ret = req->out.h.error;
1471 ret = size ? req->out.args[0].size : outarg.size;
1473 if (ret == -ENOSYS) {
1474 fc->no_listxattr = 1;
1478 fuse_put_request(fc, req);
1482 static int fuse_removexattr(struct dentry *entry, const char *name)
1484 struct inode *inode = entry->d_inode;
1485 struct fuse_conn *fc = get_fuse_conn(inode);
1486 struct fuse_req *req;
1489 if (fc->no_removexattr)
1492 req = fuse_get_req(fc);
1494 return PTR_ERR(req);
1496 req->in.h.opcode = FUSE_REMOVEXATTR;
1497 req->in.h.nodeid = get_node_id(inode);
1498 req->in.numargs = 1;
1499 req->in.args[0].size = strlen(name) + 1;
1500 req->in.args[0].value = name;
1501 fuse_request_send(fc, req);
1502 err = req->out.h.error;
1503 fuse_put_request(fc, req);
1504 if (err == -ENOSYS) {
1505 fc->no_removexattr = 1;
1511 static const struct inode_operations fuse_dir_inode_operations = {
1512 .lookup = fuse_lookup,
1513 .mkdir = fuse_mkdir,
1514 .symlink = fuse_symlink,
1515 .unlink = fuse_unlink,
1516 .rmdir = fuse_rmdir,
1517 .rename = fuse_rename,
1519 .setattr = fuse_setattr,
1520 .create = fuse_create,
1521 .mknod = fuse_mknod,
1522 .permission = fuse_permission,
1523 .getattr = fuse_getattr,
1524 .setxattr = fuse_setxattr,
1525 .getxattr = fuse_getxattr,
1526 .listxattr = fuse_listxattr,
1527 .removexattr = fuse_removexattr,
1530 static const struct file_operations fuse_dir_operations = {
1531 .llseek = generic_file_llseek,
1532 .read = generic_read_dir,
1533 .readdir = fuse_readdir,
1534 .open = fuse_dir_open,
1535 .release = fuse_dir_release,
1536 .fsync = fuse_dir_fsync,
1539 static const struct inode_operations fuse_common_inode_operations = {
1540 .setattr = fuse_setattr,
1541 .permission = fuse_permission,
1542 .getattr = fuse_getattr,
1543 .setxattr = fuse_setxattr,
1544 .getxattr = fuse_getxattr,
1545 .listxattr = fuse_listxattr,
1546 .removexattr = fuse_removexattr,
1549 static const struct inode_operations fuse_symlink_inode_operations = {
1550 .setattr = fuse_setattr,
1551 .follow_link = fuse_follow_link,
1552 .put_link = fuse_put_link,
1553 .readlink = generic_readlink,
1554 .getattr = fuse_getattr,
1555 .setxattr = fuse_setxattr,
1556 .getxattr = fuse_getxattr,
1557 .listxattr = fuse_listxattr,
1558 .removexattr = fuse_removexattr,
1561 void fuse_init_common(struct inode *inode)
1563 inode->i_op = &fuse_common_inode_operations;
1566 void fuse_init_dir(struct inode *inode)
1568 inode->i_op = &fuse_dir_inode_operations;
1569 inode->i_fop = &fuse_dir_operations;
1572 void fuse_init_symlink(struct inode *inode)
1574 inode->i_op = &fuse_symlink_inode_operations;