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 * Atomic create+open operation
367 * If the filesystem doesn't support this, then fall back to separate
368 * 'mknod' + 'open' requests.
370 static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
371 struct nameidata *nd)
375 struct fuse_conn *fc = get_fuse_conn(dir);
376 struct fuse_req *req;
377 struct fuse_req *forget_req;
378 struct fuse_create_in inarg;
379 struct fuse_open_out outopen;
380 struct fuse_entry_out outentry;
381 struct fuse_file *ff;
383 int flags = nd->intent.open.flags - 1;
388 forget_req = fuse_get_req(fc);
389 if (IS_ERR(forget_req))
390 return PTR_ERR(forget_req);
392 req = fuse_get_req(fc);
395 goto out_put_forget_req;
398 ff = fuse_file_alloc(fc);
400 goto out_put_request;
403 mode &= ~current_umask();
406 memset(&inarg, 0, sizeof(inarg));
407 memset(&outentry, 0, sizeof(outentry));
410 inarg.umask = current_umask();
411 req->in.h.opcode = FUSE_CREATE;
412 req->in.h.nodeid = get_node_id(dir);
414 req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
416 req->in.args[0].value = &inarg;
417 req->in.args[1].size = entry->d_name.len + 1;
418 req->in.args[1].value = entry->d_name.name;
419 req->out.numargs = 2;
421 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
423 req->out.args[0].size = sizeof(outentry);
424 req->out.args[0].value = &outentry;
425 req->out.args[1].size = sizeof(outopen);
426 req->out.args[1].value = &outopen;
427 fuse_request_send(fc, req);
428 err = req->out.h.error;
436 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
439 fuse_put_request(fc, req);
441 ff->nodeid = outentry.nodeid;
442 ff->open_flags = outopen.open_flags;
443 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
444 &outentry.attr, entry_attr_timeout(&outentry), 0);
446 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
447 fuse_sync_release(ff, flags);
448 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
451 fuse_put_request(fc, forget_req);
452 d_instantiate(entry, inode);
453 fuse_change_entry_timeout(entry, &outentry);
454 fuse_invalidate_attr(dir);
455 file = lookup_instantiate_filp(nd, entry, generic_file_open);
457 fuse_sync_release(ff, flags);
458 return PTR_ERR(file);
460 file->private_data = fuse_file_get(ff);
461 fuse_finish_open(inode, file);
467 fuse_put_request(fc, req);
469 fuse_put_request(fc, forget_req);
474 * Code shared between mknod, mkdir, symlink and link
476 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
477 struct inode *dir, struct dentry *entry,
480 struct fuse_entry_out outarg;
483 struct fuse_req *forget_req;
485 forget_req = fuse_get_req(fc);
486 if (IS_ERR(forget_req)) {
487 fuse_put_request(fc, req);
488 return PTR_ERR(forget_req);
491 memset(&outarg, 0, sizeof(outarg));
492 req->in.h.nodeid = get_node_id(dir);
493 req->out.numargs = 1;
495 req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
497 req->out.args[0].size = sizeof(outarg);
498 req->out.args[0].value = &outarg;
499 fuse_request_send(fc, req);
500 err = req->out.h.error;
501 fuse_put_request(fc, req);
503 goto out_put_forget_req;
506 if (invalid_nodeid(outarg.nodeid))
507 goto out_put_forget_req;
509 if ((outarg.attr.mode ^ mode) & S_IFMT)
510 goto out_put_forget_req;
512 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
513 &outarg.attr, entry_attr_timeout(&outarg), 0);
515 fuse_send_forget(fc, forget_req, outarg.nodeid, 1);
518 fuse_put_request(fc, forget_req);
520 if (S_ISDIR(inode->i_mode)) {
521 struct dentry *alias;
522 mutex_lock(&fc->inst_mutex);
523 alias = d_find_alias(inode);
525 /* New directory must have moved since mkdir */
526 mutex_unlock(&fc->inst_mutex);
531 d_instantiate(entry, inode);
532 mutex_unlock(&fc->inst_mutex);
534 d_instantiate(entry, inode);
536 fuse_change_entry_timeout(entry, &outarg);
537 fuse_invalidate_attr(dir);
541 fuse_put_request(fc, forget_req);
545 static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
548 struct fuse_mknod_in inarg;
549 struct fuse_conn *fc = get_fuse_conn(dir);
550 struct fuse_req *req = fuse_get_req(fc);
555 mode &= ~current_umask();
557 memset(&inarg, 0, sizeof(inarg));
559 inarg.rdev = new_encode_dev(rdev);
560 inarg.umask = current_umask();
561 req->in.h.opcode = FUSE_MKNOD;
563 req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
565 req->in.args[0].value = &inarg;
566 req->in.args[1].size = entry->d_name.len + 1;
567 req->in.args[1].value = entry->d_name.name;
568 return create_new_entry(fc, req, dir, entry, mode);
571 static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
572 struct nameidata *nd)
574 if (nd && (nd->flags & LOOKUP_OPEN)) {
575 int err = fuse_create_open(dir, entry, mode, nd);
578 /* Fall back on mknod */
580 return fuse_mknod(dir, entry, mode, 0);
583 static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
585 struct fuse_mkdir_in inarg;
586 struct fuse_conn *fc = get_fuse_conn(dir);
587 struct fuse_req *req = fuse_get_req(fc);
592 mode &= ~current_umask();
594 memset(&inarg, 0, sizeof(inarg));
596 inarg.umask = current_umask();
597 req->in.h.opcode = FUSE_MKDIR;
599 req->in.args[0].size = sizeof(inarg);
600 req->in.args[0].value = &inarg;
601 req->in.args[1].size = entry->d_name.len + 1;
602 req->in.args[1].value = entry->d_name.name;
603 return create_new_entry(fc, req, dir, entry, S_IFDIR);
606 static int fuse_symlink(struct inode *dir, struct dentry *entry,
609 struct fuse_conn *fc = get_fuse_conn(dir);
610 unsigned len = strlen(link) + 1;
611 struct fuse_req *req = fuse_get_req(fc);
615 req->in.h.opcode = FUSE_SYMLINK;
617 req->in.args[0].size = entry->d_name.len + 1;
618 req->in.args[0].value = entry->d_name.name;
619 req->in.args[1].size = len;
620 req->in.args[1].value = link;
621 return create_new_entry(fc, req, dir, entry, S_IFLNK);
624 static int fuse_unlink(struct inode *dir, struct dentry *entry)
627 struct fuse_conn *fc = get_fuse_conn(dir);
628 struct fuse_req *req = fuse_get_req(fc);
632 req->in.h.opcode = FUSE_UNLINK;
633 req->in.h.nodeid = get_node_id(dir);
635 req->in.args[0].size = entry->d_name.len + 1;
636 req->in.args[0].value = entry->d_name.name;
637 fuse_request_send(fc, req);
638 err = req->out.h.error;
639 fuse_put_request(fc, req);
641 struct inode *inode = entry->d_inode;
644 * Set nlink to zero so the inode can be cleared, if the inode
645 * does have more links this will be discovered at the next
649 fuse_invalidate_attr(inode);
650 fuse_invalidate_attr(dir);
651 fuse_invalidate_entry_cache(entry);
652 } else if (err == -EINTR)
653 fuse_invalidate_entry(entry);
657 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
660 struct fuse_conn *fc = get_fuse_conn(dir);
661 struct fuse_req *req = fuse_get_req(fc);
665 req->in.h.opcode = FUSE_RMDIR;
666 req->in.h.nodeid = get_node_id(dir);
668 req->in.args[0].size = entry->d_name.len + 1;
669 req->in.args[0].value = entry->d_name.name;
670 fuse_request_send(fc, req);
671 err = req->out.h.error;
672 fuse_put_request(fc, req);
674 clear_nlink(entry->d_inode);
675 fuse_invalidate_attr(dir);
676 fuse_invalidate_entry_cache(entry);
677 } else if (err == -EINTR)
678 fuse_invalidate_entry(entry);
682 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
683 struct inode *newdir, struct dentry *newent)
686 struct fuse_rename_in inarg;
687 struct fuse_conn *fc = get_fuse_conn(olddir);
688 struct fuse_req *req = fuse_get_req(fc);
692 memset(&inarg, 0, sizeof(inarg));
693 inarg.newdir = get_node_id(newdir);
694 req->in.h.opcode = FUSE_RENAME;
695 req->in.h.nodeid = get_node_id(olddir);
697 req->in.args[0].size = sizeof(inarg);
698 req->in.args[0].value = &inarg;
699 req->in.args[1].size = oldent->d_name.len + 1;
700 req->in.args[1].value = oldent->d_name.name;
701 req->in.args[2].size = newent->d_name.len + 1;
702 req->in.args[2].value = newent->d_name.name;
703 fuse_request_send(fc, req);
704 err = req->out.h.error;
705 fuse_put_request(fc, req);
708 fuse_invalidate_attr(oldent->d_inode);
710 fuse_invalidate_attr(olddir);
711 if (olddir != newdir)
712 fuse_invalidate_attr(newdir);
714 /* newent will end up negative */
716 fuse_invalidate_entry_cache(newent);
717 } else if (err == -EINTR) {
718 /* If request was interrupted, DEITY only knows if the
719 rename actually took place. If the invalidation
720 fails (e.g. some process has CWD under the renamed
721 directory), then there can be inconsistency between
722 the dcache and the real filesystem. Tough luck. */
723 fuse_invalidate_entry(oldent);
725 fuse_invalidate_entry(newent);
731 static int fuse_link(struct dentry *entry, struct inode *newdir,
732 struct dentry *newent)
735 struct fuse_link_in inarg;
736 struct inode *inode = entry->d_inode;
737 struct fuse_conn *fc = get_fuse_conn(inode);
738 struct fuse_req *req = fuse_get_req(fc);
742 memset(&inarg, 0, sizeof(inarg));
743 inarg.oldnodeid = get_node_id(inode);
744 req->in.h.opcode = FUSE_LINK;
746 req->in.args[0].size = sizeof(inarg);
747 req->in.args[0].value = &inarg;
748 req->in.args[1].size = newent->d_name.len + 1;
749 req->in.args[1].value = newent->d_name.name;
750 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
751 /* Contrary to "normal" filesystems it can happen that link
752 makes two "logical" inodes point to the same "physical"
753 inode. We invalidate the attributes of the old one, so it
754 will reflect changes in the backing inode (link count,
757 if (!err || err == -EINTR)
758 fuse_invalidate_attr(inode);
762 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
765 stat->dev = inode->i_sb->s_dev;
766 stat->ino = attr->ino;
767 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
768 stat->nlink = attr->nlink;
769 stat->uid = attr->uid;
770 stat->gid = attr->gid;
771 stat->rdev = inode->i_rdev;
772 stat->atime.tv_sec = attr->atime;
773 stat->atime.tv_nsec = attr->atimensec;
774 stat->mtime.tv_sec = attr->mtime;
775 stat->mtime.tv_nsec = attr->mtimensec;
776 stat->ctime.tv_sec = attr->ctime;
777 stat->ctime.tv_nsec = attr->ctimensec;
778 stat->size = attr->size;
779 stat->blocks = attr->blocks;
780 stat->blksize = (1 << inode->i_blkbits);
783 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
787 struct fuse_getattr_in inarg;
788 struct fuse_attr_out outarg;
789 struct fuse_conn *fc = get_fuse_conn(inode);
790 struct fuse_req *req;
793 req = fuse_get_req(fc);
797 attr_version = fuse_get_attr_version(fc);
799 memset(&inarg, 0, sizeof(inarg));
800 memset(&outarg, 0, sizeof(outarg));
801 /* Directories have separate file-handle space */
802 if (file && S_ISREG(inode->i_mode)) {
803 struct fuse_file *ff = file->private_data;
805 inarg.getattr_flags |= FUSE_GETATTR_FH;
808 req->in.h.opcode = FUSE_GETATTR;
809 req->in.h.nodeid = get_node_id(inode);
811 req->in.args[0].size = sizeof(inarg);
812 req->in.args[0].value = &inarg;
813 req->out.numargs = 1;
815 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
817 req->out.args[0].size = sizeof(outarg);
818 req->out.args[0].value = &outarg;
819 fuse_request_send(fc, req);
820 err = req->out.h.error;
821 fuse_put_request(fc, req);
823 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
824 make_bad_inode(inode);
827 fuse_change_attributes(inode, &outarg.attr,
828 attr_timeout(&outarg),
831 fuse_fillattr(inode, &outarg.attr, stat);
837 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
838 struct file *file, bool *refreshed)
840 struct fuse_inode *fi = get_fuse_inode(inode);
844 if (fi->i_time < get_jiffies_64()) {
846 err = fuse_do_getattr(inode, stat, file);
851 generic_fillattr(inode, stat);
852 stat->mode = fi->orig_i_mode;
856 if (refreshed != NULL)
863 * Calling into a user-controlled filesystem gives the filesystem
864 * daemon ptrace-like capabilities over the requester process. This
865 * means, that the filesystem daemon is able to record the exact
866 * filesystem operations performed, and can also control the behavior
867 * of the requester process in otherwise impossible ways. For example
868 * it can delay the operation for arbitrary length of time allowing
869 * DoS against the requester.
871 * For this reason only those processes can call into the filesystem,
872 * for which the owner of the mount has ptrace privilege. This
873 * excludes processes started by other users, suid or sgid processes.
875 int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
877 const struct cred *cred;
880 if (fc->flags & FUSE_ALLOW_OTHER)
885 cred = __task_cred(task);
886 if (cred->euid == fc->user_id &&
887 cred->suid == fc->user_id &&
888 cred->uid == fc->user_id &&
889 cred->egid == fc->group_id &&
890 cred->sgid == fc->group_id &&
891 cred->gid == fc->group_id)
898 static int fuse_access(struct inode *inode, int mask)
900 struct fuse_conn *fc = get_fuse_conn(inode);
901 struct fuse_req *req;
902 struct fuse_access_in inarg;
908 req = fuse_get_req(fc);
912 memset(&inarg, 0, sizeof(inarg));
913 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
914 req->in.h.opcode = FUSE_ACCESS;
915 req->in.h.nodeid = get_node_id(inode);
917 req->in.args[0].size = sizeof(inarg);
918 req->in.args[0].value = &inarg;
919 fuse_request_send(fc, req);
920 err = req->out.h.error;
921 fuse_put_request(fc, req);
922 if (err == -ENOSYS) {
930 * Check permission. The two basic access models of FUSE are:
932 * 1) Local access checking ('default_permissions' mount option) based
933 * on file mode. This is the plain old disk filesystem permission
936 * 2) "Remote" access checking, where server is responsible for
937 * checking permission in each inode operation. An exception to this
938 * is if ->permission() was invoked from sys_access() in which case an
939 * access request is sent. Execute permission is still checked
940 * locally based on file mode.
942 static int fuse_permission(struct inode *inode, int mask)
944 struct fuse_conn *fc = get_fuse_conn(inode);
945 bool refreshed = false;
948 if (!fuse_allow_task(fc, current))
952 * If attributes are needed, refresh them before proceeding
954 if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
955 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
956 err = fuse_update_attributes(inode, NULL, NULL, &refreshed);
961 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
962 err = generic_permission(inode, mask, NULL);
964 /* If permission is denied, try to refresh file
965 attributes. This is also needed, because the root
966 node will at first have no permissions */
967 if (err == -EACCES && !refreshed) {
968 err = fuse_do_getattr(inode, NULL, NULL);
970 err = generic_permission(inode, mask, NULL);
973 /* Note: the opposite of the above test does not
974 exist. So if permissions are revoked this won't be
975 noticed immediately, only after the attribute
976 timeout has expired */
977 } else if (mask & MAY_ACCESS) {
978 err = fuse_access(inode, mask);
979 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
980 if (!(inode->i_mode & S_IXUGO)) {
984 err = fuse_do_getattr(inode, NULL, NULL);
985 if (!err && !(inode->i_mode & S_IXUGO))
992 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
993 void *dstbuf, filldir_t filldir)
995 while (nbytes >= FUSE_NAME_OFFSET) {
996 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
997 size_t reclen = FUSE_DIRENT_SIZE(dirent);
999 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1001 if (reclen > nbytes)
1004 over = filldir(dstbuf, dirent->name, dirent->namelen,
1005 file->f_pos, dirent->ino, dirent->type);
1011 file->f_pos = dirent->off;
1017 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1022 struct inode *inode = file->f_path.dentry->d_inode;
1023 struct fuse_conn *fc = get_fuse_conn(inode);
1024 struct fuse_req *req;
1026 if (is_bad_inode(inode))
1029 req = fuse_get_req(fc);
1031 return PTR_ERR(req);
1033 page = alloc_page(GFP_KERNEL);
1035 fuse_put_request(fc, req);
1038 req->out.argpages = 1;
1040 req->pages[0] = page;
1041 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
1042 fuse_request_send(fc, req);
1043 nbytes = req->out.args[0].size;
1044 err = req->out.h.error;
1045 fuse_put_request(fc, req);
1047 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
1051 fuse_invalidate_attr(inode); /* atime changed */
1055 static char *read_link(struct dentry *dentry)
1057 struct inode *inode = dentry->d_inode;
1058 struct fuse_conn *fc = get_fuse_conn(inode);
1059 struct fuse_req *req = fuse_get_req(fc);
1063 return ERR_CAST(req);
1065 link = (char *) __get_free_page(GFP_KERNEL);
1067 link = ERR_PTR(-ENOMEM);
1070 req->in.h.opcode = FUSE_READLINK;
1071 req->in.h.nodeid = get_node_id(inode);
1072 req->out.argvar = 1;
1073 req->out.numargs = 1;
1074 req->out.args[0].size = PAGE_SIZE - 1;
1075 req->out.args[0].value = link;
1076 fuse_request_send(fc, req);
1077 if (req->out.h.error) {
1078 free_page((unsigned long) link);
1079 link = ERR_PTR(req->out.h.error);
1081 link[req->out.args[0].size] = '\0';
1083 fuse_put_request(fc, req);
1084 fuse_invalidate_attr(inode); /* atime changed */
1088 static void free_link(char *link)
1091 free_page((unsigned long) link);
1094 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1096 nd_set_link(nd, read_link(dentry));
1100 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1102 free_link(nd_get_link(nd));
1105 static int fuse_dir_open(struct inode *inode, struct file *file)
1107 return fuse_open_common(inode, file, true);
1110 static int fuse_dir_release(struct inode *inode, struct file *file)
1112 fuse_release_common(file, FUSE_RELEASEDIR);
1117 static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
1119 /* nfsd can call this with no file */
1120 return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
1123 static bool update_mtime(unsigned ivalid)
1125 /* Always update if mtime is explicitly set */
1126 if (ivalid & ATTR_MTIME_SET)
1129 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1130 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1133 /* In all other cases update */
1137 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1139 unsigned ivalid = iattr->ia_valid;
1141 if (ivalid & ATTR_MODE)
1142 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1143 if (ivalid & ATTR_UID)
1144 arg->valid |= FATTR_UID, arg->uid = iattr->ia_uid;
1145 if (ivalid & ATTR_GID)
1146 arg->valid |= FATTR_GID, arg->gid = iattr->ia_gid;
1147 if (ivalid & ATTR_SIZE)
1148 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1149 if (ivalid & ATTR_ATIME) {
1150 arg->valid |= FATTR_ATIME;
1151 arg->atime = iattr->ia_atime.tv_sec;
1152 arg->atimensec = iattr->ia_atime.tv_nsec;
1153 if (!(ivalid & ATTR_ATIME_SET))
1154 arg->valid |= FATTR_ATIME_NOW;
1156 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1157 arg->valid |= FATTR_MTIME;
1158 arg->mtime = iattr->ia_mtime.tv_sec;
1159 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1160 if (!(ivalid & ATTR_MTIME_SET))
1161 arg->valid |= FATTR_MTIME_NOW;
1166 * Prevent concurrent writepages on inode
1168 * This is done by adding a negative bias to the inode write counter
1169 * and waiting for all pending writes to finish.
1171 void fuse_set_nowrite(struct inode *inode)
1173 struct fuse_conn *fc = get_fuse_conn(inode);
1174 struct fuse_inode *fi = get_fuse_inode(inode);
1176 BUG_ON(!mutex_is_locked(&inode->i_mutex));
1178 spin_lock(&fc->lock);
1179 BUG_ON(fi->writectr < 0);
1180 fi->writectr += FUSE_NOWRITE;
1181 spin_unlock(&fc->lock);
1182 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1186 * Allow writepages on inode
1188 * Remove the bias from the writecounter and send any queued
1191 static void __fuse_release_nowrite(struct inode *inode)
1193 struct fuse_inode *fi = get_fuse_inode(inode);
1195 BUG_ON(fi->writectr != FUSE_NOWRITE);
1197 fuse_flush_writepages(inode);
1200 void fuse_release_nowrite(struct inode *inode)
1202 struct fuse_conn *fc = get_fuse_conn(inode);
1204 spin_lock(&fc->lock);
1205 __fuse_release_nowrite(inode);
1206 spin_unlock(&fc->lock);
1210 * Set attributes, and at the same time refresh them.
1212 * Truncation is slightly complicated, because the 'truncate' request
1213 * may fail, in which case we don't want to touch the mapping.
1214 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1215 * and the actual truncation by hand.
1217 static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1220 struct inode *inode = entry->d_inode;
1221 struct fuse_conn *fc = get_fuse_conn(inode);
1222 struct fuse_req *req;
1223 struct fuse_setattr_in inarg;
1224 struct fuse_attr_out outarg;
1225 bool is_truncate = false;
1229 if (!fuse_allow_task(fc, current))
1232 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1233 err = inode_change_ok(inode, attr);
1238 if ((attr->ia_valid & ATTR_OPEN) && fc->atomic_o_trunc)
1241 if (attr->ia_valid & ATTR_SIZE) {
1242 unsigned long limit;
1243 if (IS_SWAPFILE(inode))
1245 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1246 if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
1247 send_sig(SIGXFSZ, current, 0);
1253 req = fuse_get_req(fc);
1255 return PTR_ERR(req);
1258 fuse_set_nowrite(inode);
1260 memset(&inarg, 0, sizeof(inarg));
1261 memset(&outarg, 0, sizeof(outarg));
1262 iattr_to_fattr(attr, &inarg);
1264 struct fuse_file *ff = file->private_data;
1265 inarg.valid |= FATTR_FH;
1268 if (attr->ia_valid & ATTR_SIZE) {
1269 /* For mandatory locking in truncate */
1270 inarg.valid |= FATTR_LOCKOWNER;
1271 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1273 req->in.h.opcode = FUSE_SETATTR;
1274 req->in.h.nodeid = get_node_id(inode);
1275 req->in.numargs = 1;
1276 req->in.args[0].size = sizeof(inarg);
1277 req->in.args[0].value = &inarg;
1278 req->out.numargs = 1;
1280 req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1282 req->out.args[0].size = sizeof(outarg);
1283 req->out.args[0].value = &outarg;
1284 fuse_request_send(fc, req);
1285 err = req->out.h.error;
1286 fuse_put_request(fc, req);
1289 fuse_invalidate_attr(inode);
1293 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1294 make_bad_inode(inode);
1299 spin_lock(&fc->lock);
1300 fuse_change_attributes_common(inode, &outarg.attr,
1301 attr_timeout(&outarg));
1302 oldsize = inode->i_size;
1303 i_size_write(inode, outarg.attr.size);
1306 /* NOTE: this may release/reacquire fc->lock */
1307 __fuse_release_nowrite(inode);
1309 spin_unlock(&fc->lock);
1312 * Only call invalidate_inode_pages2() after removing
1313 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1315 if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1316 if (outarg.attr.size < oldsize)
1317 fuse_truncate(inode->i_mapping, outarg.attr.size);
1318 invalidate_inode_pages2(inode->i_mapping);
1325 fuse_release_nowrite(inode);
1330 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1332 if (attr->ia_valid & ATTR_FILE)
1333 return fuse_do_setattr(entry, attr, attr->ia_file);
1335 return fuse_do_setattr(entry, attr, NULL);
1338 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1341 struct inode *inode = entry->d_inode;
1342 struct fuse_conn *fc = get_fuse_conn(inode);
1344 if (!fuse_allow_task(fc, current))
1347 return fuse_update_attributes(inode, stat, NULL, NULL);
1350 static int fuse_setxattr(struct dentry *entry, const char *name,
1351 const void *value, size_t size, int flags)
1353 struct inode *inode = entry->d_inode;
1354 struct fuse_conn *fc = get_fuse_conn(inode);
1355 struct fuse_req *req;
1356 struct fuse_setxattr_in inarg;
1359 if (fc->no_setxattr)
1362 req = fuse_get_req(fc);
1364 return PTR_ERR(req);
1366 memset(&inarg, 0, sizeof(inarg));
1368 inarg.flags = flags;
1369 req->in.h.opcode = FUSE_SETXATTR;
1370 req->in.h.nodeid = get_node_id(inode);
1371 req->in.numargs = 3;
1372 req->in.args[0].size = sizeof(inarg);
1373 req->in.args[0].value = &inarg;
1374 req->in.args[1].size = strlen(name) + 1;
1375 req->in.args[1].value = name;
1376 req->in.args[2].size = size;
1377 req->in.args[2].value = value;
1378 fuse_request_send(fc, req);
1379 err = req->out.h.error;
1380 fuse_put_request(fc, req);
1381 if (err == -ENOSYS) {
1382 fc->no_setxattr = 1;
1388 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1389 void *value, size_t size)
1391 struct inode *inode = entry->d_inode;
1392 struct fuse_conn *fc = get_fuse_conn(inode);
1393 struct fuse_req *req;
1394 struct fuse_getxattr_in inarg;
1395 struct fuse_getxattr_out outarg;
1398 if (fc->no_getxattr)
1401 req = fuse_get_req(fc);
1403 return PTR_ERR(req);
1405 memset(&inarg, 0, sizeof(inarg));
1407 req->in.h.opcode = FUSE_GETXATTR;
1408 req->in.h.nodeid = get_node_id(inode);
1409 req->in.numargs = 2;
1410 req->in.args[0].size = sizeof(inarg);
1411 req->in.args[0].value = &inarg;
1412 req->in.args[1].size = strlen(name) + 1;
1413 req->in.args[1].value = name;
1414 /* This is really two different operations rolled into one */
1415 req->out.numargs = 1;
1417 req->out.argvar = 1;
1418 req->out.args[0].size = size;
1419 req->out.args[0].value = value;
1421 req->out.args[0].size = sizeof(outarg);
1422 req->out.args[0].value = &outarg;
1424 fuse_request_send(fc, req);
1425 ret = req->out.h.error;
1427 ret = size ? req->out.args[0].size : outarg.size;
1429 if (ret == -ENOSYS) {
1430 fc->no_getxattr = 1;
1434 fuse_put_request(fc, req);
1438 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1440 struct inode *inode = entry->d_inode;
1441 struct fuse_conn *fc = get_fuse_conn(inode);
1442 struct fuse_req *req;
1443 struct fuse_getxattr_in inarg;
1444 struct fuse_getxattr_out outarg;
1447 if (!fuse_allow_task(fc, current))
1450 if (fc->no_listxattr)
1453 req = fuse_get_req(fc);
1455 return PTR_ERR(req);
1457 memset(&inarg, 0, sizeof(inarg));
1459 req->in.h.opcode = FUSE_LISTXATTR;
1460 req->in.h.nodeid = get_node_id(inode);
1461 req->in.numargs = 1;
1462 req->in.args[0].size = sizeof(inarg);
1463 req->in.args[0].value = &inarg;
1464 /* This is really two different operations rolled into one */
1465 req->out.numargs = 1;
1467 req->out.argvar = 1;
1468 req->out.args[0].size = size;
1469 req->out.args[0].value = list;
1471 req->out.args[0].size = sizeof(outarg);
1472 req->out.args[0].value = &outarg;
1474 fuse_request_send(fc, req);
1475 ret = req->out.h.error;
1477 ret = size ? req->out.args[0].size : outarg.size;
1479 if (ret == -ENOSYS) {
1480 fc->no_listxattr = 1;
1484 fuse_put_request(fc, req);
1488 static int fuse_removexattr(struct dentry *entry, const char *name)
1490 struct inode *inode = entry->d_inode;
1491 struct fuse_conn *fc = get_fuse_conn(inode);
1492 struct fuse_req *req;
1495 if (fc->no_removexattr)
1498 req = fuse_get_req(fc);
1500 return PTR_ERR(req);
1502 req->in.h.opcode = FUSE_REMOVEXATTR;
1503 req->in.h.nodeid = get_node_id(inode);
1504 req->in.numargs = 1;
1505 req->in.args[0].size = strlen(name) + 1;
1506 req->in.args[0].value = name;
1507 fuse_request_send(fc, req);
1508 err = req->out.h.error;
1509 fuse_put_request(fc, req);
1510 if (err == -ENOSYS) {
1511 fc->no_removexattr = 1;
1517 static const struct inode_operations fuse_dir_inode_operations = {
1518 .lookup = fuse_lookup,
1519 .mkdir = fuse_mkdir,
1520 .symlink = fuse_symlink,
1521 .unlink = fuse_unlink,
1522 .rmdir = fuse_rmdir,
1523 .rename = fuse_rename,
1525 .setattr = fuse_setattr,
1526 .create = fuse_create,
1527 .mknod = fuse_mknod,
1528 .permission = fuse_permission,
1529 .getattr = fuse_getattr,
1530 .setxattr = fuse_setxattr,
1531 .getxattr = fuse_getxattr,
1532 .listxattr = fuse_listxattr,
1533 .removexattr = fuse_removexattr,
1536 static const struct file_operations fuse_dir_operations = {
1537 .llseek = generic_file_llseek,
1538 .read = generic_read_dir,
1539 .readdir = fuse_readdir,
1540 .open = fuse_dir_open,
1541 .release = fuse_dir_release,
1542 .fsync = fuse_dir_fsync,
1545 static const struct inode_operations fuse_common_inode_operations = {
1546 .setattr = fuse_setattr,
1547 .permission = fuse_permission,
1548 .getattr = fuse_getattr,
1549 .setxattr = fuse_setxattr,
1550 .getxattr = fuse_getxattr,
1551 .listxattr = fuse_listxattr,
1552 .removexattr = fuse_removexattr,
1555 static const struct inode_operations fuse_symlink_inode_operations = {
1556 .setattr = fuse_setattr,
1557 .follow_link = fuse_follow_link,
1558 .put_link = fuse_put_link,
1559 .readlink = generic_readlink,
1560 .getattr = fuse_getattr,
1561 .setxattr = fuse_setxattr,
1562 .getxattr = fuse_getxattr,
1563 .listxattr = fuse_listxattr,
1564 .removexattr = fuse_removexattr,
1567 void fuse_init_common(struct inode *inode)
1569 inode->i_op = &fuse_common_inode_operations;
1572 void fuse_init_dir(struct inode *inode)
1574 inode->i_op = &fuse_dir_inode_operations;
1575 inode->i_fop = &fuse_dir_operations;
1578 void fuse_init_symlink(struct inode *inode)
1580 inode->i_op = &fuse_symlink_inode_operations;