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->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;
353 d_set_d_op(entry, &fuse_dentry_operations);
355 fuse_change_entry_timeout(entry, &outarg);
357 fuse_invalidate_entry_cache(entry);
368 * Atomic create+open operation
370 * If the filesystem doesn't support this, then fall back to separate
371 * 'mknod' + 'open' requests.
373 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
374 struct nameidata *nd)
378 struct fuse_conn *fc = get_fuse_conn(dir);
379 struct fuse_req *req;
380 struct fuse_forget_link *forget;
381 struct fuse_create_in inarg;
382 struct fuse_open_out outopen;
383 struct fuse_entry_out outentry;
384 struct fuse_file *ff;
386 int flags = nd->intent.open.flags - 1;
391 if (flags & O_DIRECT)
394 forget = fuse_alloc_forget();
398 req = fuse_get_req(fc);
401 goto out_put_forget_req;
404 ff = fuse_file_alloc(fc);
406 goto out_put_request;
409 mode &= ~current_umask();
412 memset(&inarg, 0, sizeof(inarg));
413 memset(&outentry, 0, sizeof(outentry));
416 inarg.umask = current_umask();
417 req->in.h.opcode = FUSE_CREATE;
418 req->in.h.nodeid = get_node_id(dir);
420 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
422 req->in.args[0].value = &inarg;
423 req->in.args[1].size = entry->d_name.len + 1;
424 req->in.args[1].value = entry->d_name.name;
425 req->out.numargs = 2;
427 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
429 req->out.args[0].size = sizeof(outentry);
430 req->out.args[0].value = &outentry;
431 req->out.args[1].size = sizeof(outopen);
432 req->out.args[1].value = &outopen;
433 fuse_request_send(fc, req);
434 err = req->out.h.error;
442 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
445 fuse_put_request(fc, req);
447 ff->nodeid = outentry.nodeid;
448 ff->open_flags = outopen.open_flags;
449 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
450 &outentry.attr, entry_attr_timeout(&outentry), 0);
452 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
453 fuse_sync_release(ff, flags);
454 fuse_queue_forget(fc, forget, outentry.nodeid, 1);
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);
463 fuse_sync_release(ff, flags);
464 return PTR_ERR(file);
466 file->private_data = fuse_file_get(ff);
467 fuse_finish_open(inode, file);
473 fuse_put_request(fc, 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_forget_link *forget;
491 forget = fuse_alloc_forget();
493 fuse_put_request(fc, 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_queue_forget(fc, forget, outarg.nodeid, 1);
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);
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);
561 mode &= ~current_umask();
563 memset(&inarg, 0, sizeof(inarg));
565 inarg.rdev = new_encode_dev(rdev);
566 inarg.umask = current_umask();
567 req->in.h.opcode = FUSE_MKNOD;
569 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
571 req->in.args[0].value = &inarg;
572 req->in.args[1].size = entry->d_name.len + 1;
573 req->in.args[1].value = entry->d_name.name;
574 return create_new_entry(fc, req, dir, entry, mode);
577 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
578 struct nameidata *nd)
580 if (nd && (nd->flags & LOOKUP_OPEN)) {
581 int err = fuse_create_open(dir, entry, mode, nd);
584 /* Fall back on mknod */
586 return fuse_mknod(dir, entry, mode, 0);
589 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
591 struct fuse_mkdir_in inarg;
592 struct fuse_conn *fc = get_fuse_conn(dir);
593 struct fuse_req *req = fuse_get_req(fc);
598 mode &= ~current_umask();
600 memset(&inarg, 0, sizeof(inarg));
602 inarg.umask = current_umask();
603 req->in.h.opcode = FUSE_MKDIR;
605 req->in.args[0].size = sizeof(inarg);
606 req->in.args[0].value = &inarg;
607 req->in.args[1].size = entry->d_name.len + 1;
608 req->in.args[1].value = entry->d_name.name;
609 return create_new_entry(fc, req, dir, entry, S_IFDIR);
612 static int fuse_symlink(struct inode *dir, struct dentry *entry,
615 struct fuse_conn *fc = get_fuse_conn(dir);
616 unsigned len = strlen(link) + 1;
617 struct fuse_req *req = fuse_get_req(fc);
621 req->in.h.opcode = FUSE_SYMLINK;
623 req->in.args[0].size = entry->d_name.len + 1;
624 req->in.args[0].value = entry->d_name.name;
625 req->in.args[1].size = len;
626 req->in.args[1].value = link;
627 return create_new_entry(fc, req, dir, entry, S_IFLNK);
630 static int fuse_unlink(struct inode *dir, struct dentry *entry)
633 struct fuse_conn *fc = get_fuse_conn(dir);
634 struct fuse_req *req = fuse_get_req(fc);
638 req->in.h.opcode = FUSE_UNLINK;
639 req->in.h.nodeid = get_node_id(dir);
641 req->in.args[0].size = entry->d_name.len + 1;
642 req->in.args[0].value = entry->d_name.name;
643 fuse_request_send(fc, req);
644 err = req->out.h.error;
645 fuse_put_request(fc, req);
647 struct inode *inode = entry->d_inode;
650 * Set nlink to zero so the inode can be cleared, if the inode
651 * does have more links this will be discovered at the next
655 fuse_invalidate_attr(inode);
656 fuse_invalidate_attr(dir);
657 fuse_invalidate_entry_cache(entry);
658 } else if (err == -EINTR)
659 fuse_invalidate_entry(entry);
663 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
666 struct fuse_conn *fc = get_fuse_conn(dir);
667 struct fuse_req *req = fuse_get_req(fc);
671 req->in.h.opcode = FUSE_RMDIR;
672 req->in.h.nodeid = get_node_id(dir);
674 req->in.args[0].size = entry->d_name.len + 1;
675 req->in.args[0].value = entry->d_name.name;
676 fuse_request_send(fc, req);
677 err = req->out.h.error;
678 fuse_put_request(fc, req);
680 clear_nlink(entry->d_inode);
681 fuse_invalidate_attr(dir);
682 fuse_invalidate_entry_cache(entry);
683 } else if (err == -EINTR)
684 fuse_invalidate_entry(entry);
688 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
689 struct inode *newdir, struct dentry *newent)
692 struct fuse_rename_in inarg;
693 struct fuse_conn *fc = get_fuse_conn(olddir);
694 struct fuse_req *req = fuse_get_req(fc);
698 memset(&inarg, 0, sizeof(inarg));
699 inarg.newdir = get_node_id(newdir);
700 req->in.h.opcode = FUSE_RENAME;
701 req->in.h.nodeid = get_node_id(olddir);
703 req->in.args[0].size = sizeof(inarg);
704 req->in.args[0].value = &inarg;
705 req->in.args[1].size = oldent->d_name.len + 1;
706 req->in.args[1].value = oldent->d_name.name;
707 req->in.args[2].size = newent->d_name.len + 1;
708 req->in.args[2].value = newent->d_name.name;
709 fuse_request_send(fc, req);
710 err = req->out.h.error;
711 fuse_put_request(fc, req);
714 fuse_invalidate_attr(oldent->d_inode);
716 fuse_invalidate_attr(olddir);
717 if (olddir != newdir)
718 fuse_invalidate_attr(newdir);
720 /* newent will end up negative */
721 if (newent->d_inode) {
722 fuse_invalidate_attr(newent->d_inode);
723 fuse_invalidate_entry_cache(newent);
725 } else if (err == -EINTR) {
726 /* If request was interrupted, DEITY only knows if the
727 rename actually took place. If the invalidation
728 fails (e.g. some process has CWD under the renamed
729 directory), then there can be inconsistency between
730 the dcache and the real filesystem. Tough luck. */
731 fuse_invalidate_entry(oldent);
733 fuse_invalidate_entry(newent);
739 static int fuse_link(struct dentry *entry, struct inode *newdir,
740 struct dentry *newent)
743 struct fuse_link_in inarg;
744 struct inode *inode = entry->d_inode;
745 struct fuse_conn *fc = get_fuse_conn(inode);
746 struct fuse_req *req = fuse_get_req(fc);
750 memset(&inarg, 0, sizeof(inarg));
751 inarg.oldnodeid = get_node_id(inode);
752 req->in.h.opcode = FUSE_LINK;
754 req->in.args[0].size = sizeof(inarg);
755 req->in.args[0].value = &inarg;
756 req->in.args[1].size = newent->d_name.len + 1;
757 req->in.args[1].value = newent->d_name.name;
758 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
759 /* Contrary to "normal" filesystems it can happen that link
760 makes two "logical" inodes point to the same "physical"
761 inode. We invalidate the attributes of the old one, so it
762 will reflect changes in the backing inode (link count,
765 if (!err || err == -EINTR)
766 fuse_invalidate_attr(inode);
770 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
773 stat->dev = inode->i_sb->s_dev;
774 stat->ino = attr->ino;
775 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
776 stat->nlink = attr->nlink;
777 stat->uid = attr->uid;
778 stat->gid = attr->gid;
779 stat->rdev = inode->i_rdev;
780 stat->atime.tv_sec = attr->atime;
781 stat->atime.tv_nsec = attr->atimensec;
782 stat->mtime.tv_sec = attr->mtime;
783 stat->mtime.tv_nsec = attr->mtimensec;
784 stat->ctime.tv_sec = attr->ctime;
785 stat->ctime.tv_nsec = attr->ctimensec;
786 stat->size = attr->size;
787 stat->blocks = attr->blocks;
788 stat->blksize = (1 << inode->i_blkbits);
791 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
795 struct fuse_getattr_in inarg;
796 struct fuse_attr_out outarg;
797 struct fuse_conn *fc = get_fuse_conn(inode);
798 struct fuse_req *req;
801 req = fuse_get_req(fc);
805 attr_version = fuse_get_attr_version(fc);
807 memset(&inarg, 0, sizeof(inarg));
808 memset(&outarg, 0, sizeof(outarg));
809 /* Directories have separate file-handle space */
810 if (file && S_ISREG(inode->i_mode)) {
811 struct fuse_file *ff = file->private_data;
813 inarg.getattr_flags |= FUSE_GETATTR_FH;
816 req->in.h.opcode = FUSE_GETATTR;
817 req->in.h.nodeid = get_node_id(inode);
819 req->in.args[0].size = sizeof(inarg);
820 req->in.args[0].value = &inarg;
821 req->out.numargs = 1;
823 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
825 req->out.args[0].size = sizeof(outarg);
826 req->out.args[0].value = &outarg;
827 fuse_request_send(fc, req);
828 err = req->out.h.error;
829 fuse_put_request(fc, req);
831 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
832 make_bad_inode(inode);
835 fuse_change_attributes(inode, &outarg.attr,
836 attr_timeout(&outarg),
839 fuse_fillattr(inode, &outarg.attr, stat);
845 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
846 struct file *file, bool *refreshed)
848 struct fuse_inode *fi = get_fuse_inode(inode);
852 if (fi->i_time < get_jiffies_64()) {
854 err = fuse_do_getattr(inode, stat, file);
859 generic_fillattr(inode, stat);
860 stat->mode = fi->orig_i_mode;
864 if (refreshed != NULL)
870 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
874 struct inode *parent;
876 struct dentry *entry;
878 parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
882 mutex_lock(&parent->i_mutex);
883 if (!S_ISDIR(parent->i_mode))
887 dir = d_find_alias(parent);
891 entry = d_lookup(dir, name);
896 fuse_invalidate_attr(parent);
897 fuse_invalidate_entry(entry);
902 mutex_unlock(&parent->i_mutex);
908 * Calling into a user-controlled filesystem gives the filesystem
909 * daemon ptrace-like capabilities over the requester process. This
910 * means, that the filesystem daemon is able to record the exact
911 * filesystem operations performed, and can also control the behavior
912 * of the requester process in otherwise impossible ways. For example
913 * it can delay the operation for arbitrary length of time allowing
914 * DoS against the requester.
916 * For this reason only those processes can call into the filesystem,
917 * for which the owner of the mount has ptrace privilege. This
918 * excludes processes started by other users, suid or sgid processes.
920 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
922 const struct cred *cred;
925 if (fc->flags & FUSE_ALLOW_OTHER)
930 cred = __task_cred(task);
931 if (cred->euid == fc->user_id &&
932 cred->suid == fc->user_id &&
933 cred->uid == fc->user_id &&
934 cred->egid == fc->group_id &&
935 cred->sgid == fc->group_id &&
936 cred->gid == fc->group_id)
943 static int fuse_access(struct inode *inode, int mask)
945 struct fuse_conn *fc = get_fuse_conn(inode);
946 struct fuse_req *req;
947 struct fuse_access_in inarg;
953 req = fuse_get_req(fc);
957 memset(&inarg, 0, sizeof(inarg));
958 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
959 req->in.h.opcode = FUSE_ACCESS;
960 req->in.h.nodeid = get_node_id(inode);
962 req->in.args[0].size = sizeof(inarg);
963 req->in.args[0].value = &inarg;
964 fuse_request_send(fc, req);
965 err = req->out.h.error;
966 fuse_put_request(fc, req);
967 if (err == -ENOSYS) {
975 * Check permission. The two basic access models of FUSE are:
977 * 1) Local access checking ('default_permissions' mount option) based
978 * on file mode. This is the plain old disk filesystem permission
981 * 2) "Remote" access checking, where server is responsible for
982 * checking permission in each inode operation. An exception to this
983 * is if ->permission() was invoked from sys_access() in which case an
984 * access request is sent. Execute permission is still checked
985 * locally based on file mode.
987 static int fuse_permission(struct inode *inode, int mask, unsigned int flags)
989 struct fuse_conn *fc = get_fuse_conn(inode);
990 bool refreshed = false;
993 if (flags & IPERM_FLAG_RCU)
996 if (!fuse_allow_task(fc, current))
1000 * If attributes are needed, refresh them before proceeding
1002 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1003 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1004 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
1009 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1010 err = generic_permission(inode, mask, flags, NULL);
1012 /* If permission is denied, try to refresh file
1013 attributes. This is also needed, because the root
1014 node will at first have no permissions */
1015 if (err == -EACCES && !refreshed) {
1016 err = fuse_do_getattr(inode, NULL, NULL);
1018 err = generic_permission(inode, mask,
1022 /* Note: the opposite of the above test does not
1023 exist. So if permissions are revoked this won't be
1024 noticed immediately, only after the attribute
1025 timeout has expired */
1026 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1027 err = fuse_access(inode, mask);
1028 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1029 if (!(inode->i_mode & S_IXUGO)) {
1033 err = fuse_do_getattr(inode, NULL, NULL);
1034 if (!err && !(inode->i_mode & S_IXUGO))
1041 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1042 void *dstbuf, filldir_t filldir)
1044 while (nbytes >= FUSE_NAME_OFFSET) {
1045 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1046 size_t reclen = FUSE_DIRENT_SIZE(dirent);
1048 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1050 if (reclen > nbytes)
1053 over = filldir(dstbuf, dirent->name, dirent->namelen,
1054 file->f_pos, dirent->ino, dirent->type);
1060 file->f_pos = dirent->off;
1066 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1071 struct inode *inode = file->f_path.dentry->d_inode;
1072 struct fuse_conn *fc = get_fuse_conn(inode);
1073 struct fuse_req *req;
1075 if (is_bad_inode(inode))
1078 req = fuse_get_req(fc);
1080 return PTR_ERR(req);
1082 page = alloc_page(GFP_KERNEL);
1084 fuse_put_request(fc, req);
1087 req->out.argpages = 1;
1089 req->pages[0] = page;
1090 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1091 fuse_request_send(fc, req);
1092 nbytes = req->out.args[0].size;
1093 err = req->out.h.error;
1094 fuse_put_request(fc, req);
1096 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1100 fuse_invalidate_attr(inode); /* atime changed */
1104 static char *read_link(struct dentry *dentry)
1106 struct inode *inode = dentry->d_inode;
1107 struct fuse_conn *fc = get_fuse_conn(inode);
1108 struct fuse_req *req = fuse_get_req(fc);
1112 return ERR_CAST(req);
1114 link = (char *) __get_free_page(GFP_KERNEL);
1116 link = ERR_PTR(-ENOMEM);
1119 req->in.h.opcode = FUSE_READLINK;
1120 req->in.h.nodeid = get_node_id(inode);
1121 req->out.argvar = 1;
1122 req->out.numargs = 1;
1123 req->out.args[0].size = PAGE_SIZE - 1;
1124 req->out.args[0].value = link;
1125 fuse_request_send(fc, req);
1126 if (req->out.h.error) {
1127 free_page((unsigned long) link);
1128 link = ERR_PTR(req->out.h.error);
1130 link[req->out.args[0].size] = '\0';
1132 fuse_put_request(fc, req);
1133 fuse_invalidate_attr(inode); /* atime changed */
1137 static void free_link(char *link)
1140 free_page((unsigned long) link);
1143 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1145 nd_set_link(nd, read_link(dentry));
1149 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1151 free_link(nd_get_link(nd));
1154 static int fuse_dir_open(struct inode *inode, struct file *file)
1156 return fuse_open_common(inode, file, true);
1159 static int fuse_dir_release(struct inode *inode, struct file *file)
1161 fuse_release_common(file, FUSE_RELEASEDIR);
1166 static int fuse_dir_fsync(struct file *file, int datasync)
1168 return fuse_fsync_common(file, datasync, 1);
1171 static bool update_mtime(unsigned ivalid)
1173 /* Always update if mtime is explicitly set */
1174 if (ivalid & ATTR_MTIME_SET)
1177 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1178 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1181 /* In all other cases update */
1185 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1187 unsigned ivalid = iattr->ia_valid;
1189 if (ivalid & ATTR_MODE)
1190 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1191 if (ivalid & ATTR_UID)
1192 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1193 if (ivalid & ATTR_GID)
1194 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1195 if (ivalid & ATTR_SIZE)
1196 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1197 if (ivalid & ATTR_ATIME) {
1198 arg->valid |= FATTR_ATIME;
1199 arg->atime = iattr->ia_atime.tv_sec;
1200 arg->atimensec = iattr->ia_atime.tv_nsec;
1201 if (!(ivalid & ATTR_ATIME_SET))
1202 arg->valid |= FATTR_ATIME_NOW;
1204 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1205 arg->valid |= FATTR_MTIME;
1206 arg->mtime = iattr->ia_mtime.tv_sec;
1207 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1208 if (!(ivalid & ATTR_MTIME_SET))
1209 arg->valid |= FATTR_MTIME_NOW;
1214 * Prevent concurrent writepages on inode
1216 * This is done by adding a negative bias to the inode write counter
1217 * and waiting for all pending writes to finish.
1219 void fuse_set_nowrite(struct inode *inode)
1221 struct fuse_conn *fc = get_fuse_conn(inode);
1222 struct fuse_inode *fi = get_fuse_inode(inode);
1224 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1226 spin_lock(&fc->lock);
1227 BUG_ON(fi->writectr < 0);
1228 fi->writectr += FUSE_NOWRITE;
1229 spin_unlock(&fc->lock);
1230 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1234 * Allow writepages on inode
1236 * Remove the bias from the writecounter and send any queued
1239 static void __fuse_release_nowrite(struct inode *inode)
1241 struct fuse_inode *fi = get_fuse_inode(inode);
1243 BUG_ON(fi->writectr != FUSE_NOWRITE);
1245 fuse_flush_writepages(inode);
1248 void fuse_release_nowrite(struct inode *inode)
1250 struct fuse_conn *fc = get_fuse_conn(inode);
1252 spin_lock(&fc->lock);
1253 __fuse_release_nowrite(inode);
1254 spin_unlock(&fc->lock);
1258 * Set attributes, and at the same time refresh them.
1260 * Truncation is slightly complicated, because the 'truncate' request
1261 * may fail, in which case we don't want to touch the mapping.
1262 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1263 * and the actual truncation by hand.
1265 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1268 struct inode *inode = entry->d_inode;
1269 struct fuse_conn *fc = get_fuse_conn(inode);
1270 struct fuse_req *req;
1271 struct fuse_setattr_in inarg;
1272 struct fuse_attr_out outarg;
1273 bool is_truncate = false;
1277 if (!fuse_allow_task(fc, current))
1280 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1281 attr->ia_valid |= ATTR_FORCE;
1283 err = inode_change_ok(inode, attr);
1287 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1290 if (attr->ia_valid & ATTR_SIZE)
1293 req = fuse_get_req(fc);
1295 return PTR_ERR(req);
1298 fuse_set_nowrite(inode);
1300 memset(&inarg, 0, sizeof(inarg));
1301 memset(&outarg, 0, sizeof(outarg));
1302 iattr_to_fattr(attr, &inarg);
1304 struct fuse_file *ff = file->private_data;
1305 inarg.valid |= FATTR_FH;
1308 if (attr->ia_valid & ATTR_SIZE) {
1309 /* For mandatory locking in truncate */
1310 inarg.valid |= FATTR_LOCKOWNER;
1311 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1313 req->in.h.opcode = FUSE_SETATTR;
1314 req->in.h.nodeid = get_node_id(inode);
1315 req->in.numargs = 1;
1316 req->in.args[0].size = sizeof(inarg);
1317 req->in.args[0].value = &inarg;
1318 req->out.numargs = 1;
1320 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1322 req->out.args[0].size = sizeof(outarg);
1323 req->out.args[0].value = &outarg;
1324 fuse_request_send(fc, req);
1325 err = req->out.h.error;
1326 fuse_put_request(fc, req);
1329 fuse_invalidate_attr(inode);
1333 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1334 make_bad_inode(inode);
1339 spin_lock(&fc->lock);
1340 fuse_change_attributes_common(inode, &outarg.attr,
1341 attr_timeout(&outarg));
1342 oldsize = inode->i_size;
1343 i_size_write(inode, outarg.attr.size);
1346 /* NOTE: this may release/reacquire fc->lock */
1347 __fuse_release_nowrite(inode);
1349 spin_unlock(&fc->lock);
1352 * Only call invalidate_inode_pages2() after removing
1353 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1355 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1356 truncate_pagecache(inode, oldsize, outarg.attr.size);
1357 invalidate_inode_pages2(inode->i_mapping);
1364 fuse_release_nowrite(inode);
1369 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1371 if (attr->ia_valid & ATTR_FILE)
1372 return fuse_do_setattr(entry, attr, attr->ia_file);
1374 return fuse_do_setattr(entry, attr, NULL);
1377 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1380 struct inode *inode = entry->d_inode;
1381 struct fuse_conn *fc = get_fuse_conn(inode);
1383 if (!fuse_allow_task(fc, current))
1386 return fuse_update_attributes(inode, stat, NULL, NULL);
1389 static int fuse_setxattr(struct dentry *entry, const char *name,
1390 const void *value, size_t size, int flags)
1392 struct inode *inode = entry->d_inode;
1393 struct fuse_conn *fc = get_fuse_conn(inode);
1394 struct fuse_req *req;
1395 struct fuse_setxattr_in inarg;
1398 if (fc->no_setxattr)
1401 req = fuse_get_req(fc);
1403 return PTR_ERR(req);
1405 memset(&inarg, 0, sizeof(inarg));
1407 inarg.flags = flags;
1408 req->in.h.opcode = FUSE_SETXATTR;
1409 req->in.h.nodeid = get_node_id(inode);
1410 req->in.numargs = 3;
1411 req->in.args[0].size = sizeof(inarg);
1412 req->in.args[0].value = &inarg;
1413 req->in.args[1].size = strlen(name) + 1;
1414 req->in.args[1].value = name;
1415 req->in.args[2].size = size;
1416 req->in.args[2].value = value;
1417 fuse_request_send(fc, req);
1418 err = req->out.h.error;
1419 fuse_put_request(fc, req);
1420 if (err == -ENOSYS) {
1421 fc->no_setxattr = 1;
1427 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1428 void *value, size_t size)
1430 struct inode *inode = entry->d_inode;
1431 struct fuse_conn *fc = get_fuse_conn(inode);
1432 struct fuse_req *req;
1433 struct fuse_getxattr_in inarg;
1434 struct fuse_getxattr_out outarg;
1437 if (fc->no_getxattr)
1440 req = fuse_get_req(fc);
1442 return PTR_ERR(req);
1444 memset(&inarg, 0, sizeof(inarg));
1446 req->in.h.opcode = FUSE_GETXATTR;
1447 req->in.h.nodeid = get_node_id(inode);
1448 req->in.numargs = 2;
1449 req->in.args[0].size = sizeof(inarg);
1450 req->in.args[0].value = &inarg;
1451 req->in.args[1].size = strlen(name) + 1;
1452 req->in.args[1].value = name;
1453 /* This is really two different operations rolled into one */
1454 req->out.numargs = 1;
1456 req->out.argvar = 1;
1457 req->out.args[0].size = size;
1458 req->out.args[0].value = value;
1460 req->out.args[0].size = sizeof(outarg);
1461 req->out.args[0].value = &outarg;
1463 fuse_request_send(fc, req);
1464 ret = req->out.h.error;
1466 ret = size ? req->out.args[0].size : outarg.size;
1468 if (ret == -ENOSYS) {
1469 fc->no_getxattr = 1;
1473 fuse_put_request(fc, req);
1477 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1479 struct inode *inode = entry->d_inode;
1480 struct fuse_conn *fc = get_fuse_conn(inode);
1481 struct fuse_req *req;
1482 struct fuse_getxattr_in inarg;
1483 struct fuse_getxattr_out outarg;
1486 if (!fuse_allow_task(fc, current))
1489 if (fc->no_listxattr)
1492 req = fuse_get_req(fc);
1494 return PTR_ERR(req);
1496 memset(&inarg, 0, sizeof(inarg));
1498 req->in.h.opcode = FUSE_LISTXATTR;
1499 req->in.h.nodeid = get_node_id(inode);
1500 req->in.numargs = 1;
1501 req->in.args[0].size = sizeof(inarg);
1502 req->in.args[0].value = &inarg;
1503 /* This is really two different operations rolled into one */
1504 req->out.numargs = 1;
1506 req->out.argvar = 1;
1507 req->out.args[0].size = size;
1508 req->out.args[0].value = list;
1510 req->out.args[0].size = sizeof(outarg);
1511 req->out.args[0].value = &outarg;
1513 fuse_request_send(fc, req);
1514 ret = req->out.h.error;
1516 ret = size ? req->out.args[0].size : outarg.size;
1518 if (ret == -ENOSYS) {
1519 fc->no_listxattr = 1;
1523 fuse_put_request(fc, req);
1527 static int fuse_removexattr(struct dentry *entry, const char *name)
1529 struct inode *inode = entry->d_inode;
1530 struct fuse_conn *fc = get_fuse_conn(inode);
1531 struct fuse_req *req;
1534 if (fc->no_removexattr)
1537 req = fuse_get_req(fc);
1539 return PTR_ERR(req);
1541 req->in.h.opcode = FUSE_REMOVEXATTR;
1542 req->in.h.nodeid = get_node_id(inode);
1543 req->in.numargs = 1;
1544 req->in.args[0].size = strlen(name) + 1;
1545 req->in.args[0].value = name;
1546 fuse_request_send(fc, req);
1547 err = req->out.h.error;
1548 fuse_put_request(fc, req);
1549 if (err == -ENOSYS) {
1550 fc->no_removexattr = 1;
1556 static const struct inode_operations fuse_dir_inode_operations = {
1557 .lookup = fuse_lookup,
1558 .mkdir = fuse_mkdir,
1559 .symlink = fuse_symlink,
1560 .unlink = fuse_unlink,
1561 .rmdir = fuse_rmdir,
1562 .rename = fuse_rename,
1564 .setattr = fuse_setattr,
1565 .create = fuse_create,
1566 .mknod = fuse_mknod,
1567 .permission = fuse_permission,
1568 .getattr = fuse_getattr,
1569 .setxattr = fuse_setxattr,
1570 .getxattr = fuse_getxattr,
1571 .listxattr = fuse_listxattr,
1572 .removexattr = fuse_removexattr,
1575 static const struct file_operations fuse_dir_operations = {
1576 .llseek = generic_file_llseek,
1577 .read = generic_read_dir,
1578 .readdir = fuse_readdir,
1579 .open = fuse_dir_open,
1580 .release = fuse_dir_release,
1581 .fsync = fuse_dir_fsync,
1584 static const struct inode_operations fuse_common_inode_operations = {
1585 .setattr = fuse_setattr,
1586 .permission = fuse_permission,
1587 .getattr = fuse_getattr,
1588 .setxattr = fuse_setxattr,
1589 .getxattr = fuse_getxattr,
1590 .listxattr = fuse_listxattr,
1591 .removexattr = fuse_removexattr,
1594 static const struct inode_operations fuse_symlink_inode_operations = {
1595 .setattr = fuse_setattr,
1596 .follow_link = fuse_follow_link,
1597 .put_link = fuse_put_link,
1598 .readlink = generic_readlink,
1599 .getattr = fuse_getattr,
1600 .setxattr = fuse_setxattr,
1601 .getxattr = fuse_getxattr,
1602 .listxattr = fuse_listxattr,
1603 .removexattr = fuse_removexattr,
1606 void fuse_init_common(struct inode *inode)
1608 inode->i_op = &fuse_common_inode_operations;
1611 void fuse_init_dir(struct inode *inode)
1613 inode->i_op = &fuse_dir_inode_operations;
1614 inode->i_fop = &fuse_dir_operations;
1617 void fuse_init_symlink(struct inode *inode)
1619 inode->i_op = &fuse_symlink_inode_operations;