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/fs_context.h>
14 #include <linux/sched.h>
15 #include <linux/namei.h>
16 #include <linux/slab.h>
17 #include <linux/xattr.h>
18 #include <linux/iversion.h>
19 #include <linux/posix_acl.h>
21 static void fuse_advise_use_readdirplus(struct inode *dir)
23 struct fuse_inode *fi = get_fuse_inode(dir);
25 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
28 #if BITS_PER_LONG >= 64
29 static inline void __fuse_dentry_settime(struct dentry *entry, u64 time)
31 entry->d_fsdata = (void *) time;
34 static inline u64 fuse_dentry_time(const struct dentry *entry)
36 return (u64)entry->d_fsdata;
45 static inline void __fuse_dentry_settime(struct dentry *dentry, u64 time)
47 ((union fuse_dentry *) dentry->d_fsdata)->time = time;
50 static inline u64 fuse_dentry_time(const struct dentry *entry)
52 return ((union fuse_dentry *) entry->d_fsdata)->time;
56 static void fuse_dentry_settime(struct dentry *dentry, u64 time)
58 struct fuse_conn *fc = get_fuse_conn_super(dentry->d_sb);
59 bool delete = !time && fc->delete_stale;
61 * Mess with DCACHE_OP_DELETE because dput() will be faster without it.
62 * Don't care about races, either way it's just an optimization
64 if ((!delete && (dentry->d_flags & DCACHE_OP_DELETE)) ||
65 (delete && !(dentry->d_flags & DCACHE_OP_DELETE))) {
66 spin_lock(&dentry->d_lock);
68 dentry->d_flags &= ~DCACHE_OP_DELETE;
70 dentry->d_flags |= DCACHE_OP_DELETE;
71 spin_unlock(&dentry->d_lock);
74 __fuse_dentry_settime(dentry, time);
78 * FUSE caches dentries and attributes with separate timeout. The
79 * time in jiffies until the dentry/attributes are valid is stored in
80 * dentry->d_fsdata and fuse_inode->i_time respectively.
84 * Calculate the time in jiffies until a dentry/attributes are valid
86 static u64 time_to_jiffies(u64 sec, u32 nsec)
89 struct timespec64 ts = {
91 min_t(u32, nsec, NSEC_PER_SEC - 1)
94 return get_jiffies_64() + timespec64_to_jiffies(&ts);
100 * Set dentry and possibly attribute timeouts from the lookup/mk*
103 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o)
105 fuse_dentry_settime(entry,
106 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
109 static u64 attr_timeout(struct fuse_attr_out *o)
111 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
114 u64 entry_attr_timeout(struct fuse_entry_out *o)
116 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
119 static void fuse_invalidate_attr_mask(struct inode *inode, u32 mask)
121 set_mask_bits(&get_fuse_inode(inode)->inval_mask, 0, mask);
125 * Mark the attributes as stale, so that at the next call to
126 * ->getattr() they will be fetched from userspace
128 void fuse_invalidate_attr(struct inode *inode)
130 fuse_invalidate_attr_mask(inode, STATX_BASIC_STATS);
133 static void fuse_dir_changed(struct inode *dir)
135 fuse_invalidate_attr(dir);
136 inode_maybe_inc_iversion(dir, false);
140 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
143 void fuse_invalidate_atime(struct inode *inode)
145 if (!IS_RDONLY(inode))
146 fuse_invalidate_attr_mask(inode, STATX_ATIME);
150 * Just mark the entry as stale, so that a next attempt to look it up
151 * will result in a new lookup call to userspace
153 * This is called when a dentry is about to become negative and the
154 * timeout is unknown (unlink, rmdir, rename and in some cases
157 void fuse_invalidate_entry_cache(struct dentry *entry)
159 fuse_dentry_settime(entry, 0);
163 * Same as fuse_invalidate_entry_cache(), but also try to remove the
164 * dentry from the hash
166 static void fuse_invalidate_entry(struct dentry *entry)
169 fuse_invalidate_entry_cache(entry);
172 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
173 u64 nodeid, const struct qstr *name,
174 struct fuse_entry_out *outarg)
176 memset(outarg, 0, sizeof(struct fuse_entry_out));
177 args->opcode = FUSE_LOOKUP;
178 args->nodeid = nodeid;
179 args->in_numargs = 1;
180 args->in_args[0].size = name->len + 1;
181 args->in_args[0].value = name->name;
182 args->out_numargs = 1;
183 args->out_args[0].size = sizeof(struct fuse_entry_out);
184 args->out_args[0].value = outarg;
188 * Check whether the dentry is still valid
190 * If the entry validity timeout has expired and the dentry is
191 * positive, try to redo the lookup. If the lookup results in a
192 * different inode, then let the VFS invalidate the dentry and redo
193 * the lookup once more. If the lookup results in the same inode,
194 * then refresh the attributes, timeouts and mark the dentry valid.
196 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
199 struct dentry *parent;
200 struct fuse_mount *fm;
201 struct fuse_inode *fi;
204 inode = d_inode_rcu(entry);
205 if (inode && fuse_is_bad(inode))
207 else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
208 (flags & (LOOKUP_EXCL | LOOKUP_REVAL))) {
209 struct fuse_entry_out outarg;
211 struct fuse_forget_link *forget;
214 /* For negative dentries, always do a fresh lookup */
219 if (flags & LOOKUP_RCU)
222 fm = get_fuse_mount(inode);
224 forget = fuse_alloc_forget();
229 attr_version = fuse_get_attr_version(fm->fc);
231 parent = dget_parent(entry);
232 fuse_lookup_init(fm->fc, &args, get_node_id(d_inode(parent)),
233 &entry->d_name, &outarg);
234 ret = fuse_simple_request(fm, &args);
236 /* Zero nodeid is same as -ENOENT */
237 if (!ret && !outarg.nodeid)
240 fi = get_fuse_inode(inode);
241 if (outarg.nodeid != get_node_id(inode) ||
242 (bool) IS_AUTOMOUNT(inode) != (bool) (outarg.attr.flags & FUSE_ATTR_SUBMOUNT)) {
243 fuse_queue_forget(fm->fc, forget,
247 spin_lock(&fi->lock);
249 spin_unlock(&fi->lock);
254 if (ret || fuse_invalid_attr(&outarg.attr) ||
255 (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
258 forget_all_cached_acls(inode);
259 fuse_change_attributes(inode, &outarg.attr,
260 entry_attr_timeout(&outarg),
262 fuse_change_entry_timeout(entry, &outarg);
264 fi = get_fuse_inode(inode);
265 if (flags & LOOKUP_RCU) {
266 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
268 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
269 parent = dget_parent(entry);
270 fuse_advise_use_readdirplus(d_inode(parent));
283 #if BITS_PER_LONG < 64
284 static int fuse_dentry_init(struct dentry *dentry)
286 dentry->d_fsdata = kzalloc(sizeof(union fuse_dentry),
287 GFP_KERNEL_ACCOUNT | __GFP_RECLAIMABLE);
289 return dentry->d_fsdata ? 0 : -ENOMEM;
291 static void fuse_dentry_release(struct dentry *dentry)
293 union fuse_dentry *fd = dentry->d_fsdata;
299 static int fuse_dentry_delete(const struct dentry *dentry)
301 return time_before64(fuse_dentry_time(dentry), get_jiffies_64());
305 * Create a fuse_mount object with a new superblock (with path->dentry
306 * as the root), and return that mount so it can be auto-mounted on
309 static struct vfsmount *fuse_dentry_automount(struct path *path)
311 struct fs_context *fsc;
312 struct fuse_mount *parent_fm = get_fuse_mount_super(path->mnt->mnt_sb);
313 struct fuse_conn *fc = parent_fm->fc;
314 struct fuse_mount *fm;
315 struct vfsmount *mnt;
316 struct fuse_inode *mp_fi = get_fuse_inode(d_inode(path->dentry));
317 struct super_block *sb;
320 fsc = fs_context_for_submount(path->mnt->mnt_sb->s_type, path->dentry);
327 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
332 sb = sget_fc(fsc, NULL, set_anon_super_fc);
338 fm->fc = fuse_conn_get(fc);
340 /* Initialize superblock, making @mp_fi its root */
341 err = fuse_fill_super_submount(sb, mp_fi);
345 sb->s_flags |= SB_ACTIVE;
346 fsc->root = dget(sb->s_root);
347 /* We are done configuring the superblock, so unlock it */
348 up_write(&sb->s_umount);
350 down_write(&fc->killsb);
351 list_add_tail(&fm->fc_entry, &fc->mounts);
352 up_write(&fc->killsb);
354 /* Create the submount */
355 mnt = vfs_create_mount(fsc);
366 * Only jump here when fsc->root is NULL and sb is still locked
367 * (otherwise put_fs_context() will put the superblock)
369 deactivate_locked_super(sb);
376 const struct dentry_operations fuse_dentry_operations = {
377 .d_revalidate = fuse_dentry_revalidate,
378 .d_delete = fuse_dentry_delete,
379 #if BITS_PER_LONG < 64
380 .d_init = fuse_dentry_init,
381 .d_release = fuse_dentry_release,
383 .d_automount = fuse_dentry_automount,
386 const struct dentry_operations fuse_root_dentry_operations = {
387 #if BITS_PER_LONG < 64
388 .d_init = fuse_dentry_init,
389 .d_release = fuse_dentry_release,
393 int fuse_valid_type(int m)
395 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
396 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
399 bool fuse_invalid_attr(struct fuse_attr *attr)
401 return !fuse_valid_type(attr->mode) ||
402 attr->size > LLONG_MAX;
405 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
406 struct fuse_entry_out *outarg, struct inode **inode)
408 struct fuse_mount *fm = get_fuse_mount_super(sb);
410 struct fuse_forget_link *forget;
416 if (name->len > FUSE_NAME_MAX)
420 forget = fuse_alloc_forget();
425 attr_version = fuse_get_attr_version(fm->fc);
427 fuse_lookup_init(fm->fc, &args, nodeid, name, outarg);
428 err = fuse_simple_request(fm, &args);
429 /* Zero nodeid is same as -ENOENT, but with valid timeout */
430 if (err || !outarg->nodeid)
436 if (fuse_invalid_attr(&outarg->attr))
439 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
440 &outarg->attr, entry_attr_timeout(outarg),
444 fuse_queue_forget(fm->fc, forget, outarg->nodeid, 1);
455 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
459 struct fuse_entry_out outarg;
461 struct dentry *newent;
462 bool outarg_valid = true;
465 if (fuse_is_bad(dir))
466 return ERR_PTR(-EIO);
468 locked = fuse_lock_inode(dir);
469 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
471 fuse_unlock_inode(dir, locked);
472 if (err == -ENOENT) {
473 outarg_valid = false;
480 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
483 newent = d_splice_alias(inode, entry);
484 err = PTR_ERR(newent);
488 entry = newent ? newent : entry;
490 fuse_change_entry_timeout(entry, &outarg);
492 fuse_invalidate_entry_cache(entry);
495 fuse_advise_use_readdirplus(dir);
505 * Atomic create+open operation
507 * If the filesystem doesn't support this, then fall back to separate
508 * 'mknod' + 'open' requests.
510 static int fuse_create_open(struct inode *dir, struct dentry *entry,
511 struct file *file, unsigned flags,
516 struct fuse_mount *fm = get_fuse_mount(dir);
518 struct fuse_forget_link *forget;
519 struct fuse_create_in inarg;
520 struct fuse_open_out outopen;
521 struct fuse_entry_out outentry;
522 struct fuse_inode *fi;
523 struct fuse_file *ff;
525 /* Userspace expects S_IFREG in create mode */
526 BUG_ON((mode & S_IFMT) != S_IFREG);
528 forget = fuse_alloc_forget();
534 ff = fuse_file_alloc(fm);
536 goto out_put_forget_req;
538 if (!fm->fc->dont_mask)
539 mode &= ~current_umask();
542 memset(&inarg, 0, sizeof(inarg));
543 memset(&outentry, 0, sizeof(outentry));
546 inarg.umask = current_umask();
548 if (fm->fc->handle_killpriv_v2 && (flags & O_TRUNC) &&
549 !(flags & O_EXCL) && !capable(CAP_FSETID)) {
550 inarg.open_flags |= FUSE_OPEN_KILL_SUIDGID;
553 args.opcode = FUSE_CREATE;
554 args.nodeid = get_node_id(dir);
556 args.in_args[0].size = sizeof(inarg);
557 args.in_args[0].value = &inarg;
558 args.in_args[1].size = entry->d_name.len + 1;
559 args.in_args[1].value = entry->d_name.name;
560 args.out_numargs = 2;
561 args.out_args[0].size = sizeof(outentry);
562 args.out_args[0].value = &outentry;
563 args.out_args[1].size = sizeof(outopen);
564 args.out_args[1].value = &outopen;
565 err = fuse_simple_request(fm, &args);
570 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid) ||
571 fuse_invalid_attr(&outentry.attr))
575 ff->nodeid = outentry.nodeid;
576 ff->open_flags = outopen.open_flags;
577 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
578 &outentry.attr, entry_attr_timeout(&outentry), 0);
580 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
581 fuse_sync_release(NULL, ff, flags);
582 fuse_queue_forget(fm->fc, forget, outentry.nodeid, 1);
587 d_instantiate(entry, inode);
588 fuse_change_entry_timeout(entry, &outentry);
589 fuse_dir_changed(dir);
590 err = finish_open(file, entry, generic_file_open);
592 fi = get_fuse_inode(inode);
593 fuse_sync_release(fi, ff, flags);
595 file->private_data = ff;
596 fuse_finish_open(inode, file);
608 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
609 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
610 struct file *file, unsigned flags,
614 struct fuse_conn *fc = get_fuse_conn(dir);
615 struct dentry *res = NULL;
617 if (fuse_is_bad(dir))
620 if (d_in_lookup(entry)) {
621 res = fuse_lookup(dir, entry, 0);
629 if (!(flags & O_CREAT) || d_really_is_positive(entry))
633 file->f_mode |= FMODE_CREATED;
638 err = fuse_create_open(dir, entry, file, flags, mode);
639 if (err == -ENOSYS) {
648 err = fuse_mknod(dir, entry, mode, 0);
652 return finish_no_open(file, res);
656 * Code shared between mknod, mkdir, symlink and link
658 static int create_new_entry(struct fuse_mount *fm, struct fuse_args *args,
659 struct inode *dir, struct dentry *entry,
662 struct fuse_entry_out outarg;
666 struct fuse_forget_link *forget;
668 if (fuse_is_bad(dir))
671 forget = fuse_alloc_forget();
675 memset(&outarg, 0, sizeof(outarg));
676 args->nodeid = get_node_id(dir);
677 args->out_numargs = 1;
678 args->out_args[0].size = sizeof(outarg);
679 args->out_args[0].value = &outarg;
680 err = fuse_simple_request(fm, args);
682 goto out_put_forget_req;
685 if (invalid_nodeid(outarg.nodeid) || fuse_invalid_attr(&outarg.attr))
686 goto out_put_forget_req;
688 if ((outarg.attr.mode ^ mode) & S_IFMT)
689 goto out_put_forget_req;
691 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
692 &outarg.attr, entry_attr_timeout(&outarg), 0);
694 fuse_queue_forget(fm->fc, forget, outarg.nodeid, 1);
700 d = d_splice_alias(inode, entry);
705 fuse_change_entry_timeout(d, &outarg);
708 fuse_change_entry_timeout(entry, &outarg);
710 fuse_dir_changed(dir);
718 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
721 struct fuse_mknod_in inarg;
722 struct fuse_mount *fm = get_fuse_mount(dir);
725 if (!fm->fc->dont_mask)
726 mode &= ~current_umask();
728 memset(&inarg, 0, sizeof(inarg));
730 inarg.rdev = new_encode_dev(rdev);
731 inarg.umask = current_umask();
732 args.opcode = FUSE_MKNOD;
734 args.in_args[0].size = sizeof(inarg);
735 args.in_args[0].value = &inarg;
736 args.in_args[1].size = entry->d_name.len + 1;
737 args.in_args[1].value = entry->d_name.name;
738 return create_new_entry(fm, &args, dir, entry, mode);
741 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
744 return fuse_mknod(dir, entry, mode, 0);
747 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
749 struct fuse_mkdir_in inarg;
750 struct fuse_mount *fm = get_fuse_mount(dir);
753 if (!fm->fc->dont_mask)
754 mode &= ~current_umask();
756 memset(&inarg, 0, sizeof(inarg));
758 inarg.umask = current_umask();
759 args.opcode = FUSE_MKDIR;
761 args.in_args[0].size = sizeof(inarg);
762 args.in_args[0].value = &inarg;
763 args.in_args[1].size = entry->d_name.len + 1;
764 args.in_args[1].value = entry->d_name.name;
765 return create_new_entry(fm, &args, dir, entry, S_IFDIR);
768 static int fuse_symlink(struct inode *dir, struct dentry *entry,
771 struct fuse_mount *fm = get_fuse_mount(dir);
772 unsigned len = strlen(link) + 1;
775 args.opcode = FUSE_SYMLINK;
777 args.in_args[0].size = entry->d_name.len + 1;
778 args.in_args[0].value = entry->d_name.name;
779 args.in_args[1].size = len;
780 args.in_args[1].value = link;
781 return create_new_entry(fm, &args, dir, entry, S_IFLNK);
784 void fuse_update_ctime(struct inode *inode)
786 if (!IS_NOCMTIME(inode)) {
787 inode->i_ctime = current_time(inode);
788 mark_inode_dirty_sync(inode);
792 static int fuse_unlink(struct inode *dir, struct dentry *entry)
795 struct fuse_mount *fm = get_fuse_mount(dir);
798 if (fuse_is_bad(dir))
801 args.opcode = FUSE_UNLINK;
802 args.nodeid = get_node_id(dir);
804 args.in_args[0].size = entry->d_name.len + 1;
805 args.in_args[0].value = entry->d_name.name;
806 err = fuse_simple_request(fm, &args);
808 struct inode *inode = d_inode(entry);
809 struct fuse_inode *fi = get_fuse_inode(inode);
811 spin_lock(&fi->lock);
812 fi->attr_version = atomic64_inc_return(&fm->fc->attr_version);
814 * If i_nlink == 0 then unlink doesn't make sense, yet this can
815 * happen if userspace filesystem is careless. It would be
816 * difficult to enforce correct nlink usage so just ignore this
819 if (inode->i_nlink > 0)
821 spin_unlock(&fi->lock);
822 fuse_invalidate_attr(inode);
823 fuse_dir_changed(dir);
824 fuse_invalidate_entry_cache(entry);
825 fuse_update_ctime(inode);
826 } else if (err == -EINTR)
827 fuse_invalidate_entry(entry);
831 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
834 struct fuse_mount *fm = get_fuse_mount(dir);
837 if (fuse_is_bad(dir))
840 args.opcode = FUSE_RMDIR;
841 args.nodeid = get_node_id(dir);
843 args.in_args[0].size = entry->d_name.len + 1;
844 args.in_args[0].value = entry->d_name.name;
845 err = fuse_simple_request(fm, &args);
847 clear_nlink(d_inode(entry));
848 fuse_dir_changed(dir);
849 fuse_invalidate_entry_cache(entry);
850 } else if (err == -EINTR)
851 fuse_invalidate_entry(entry);
855 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
856 struct inode *newdir, struct dentry *newent,
857 unsigned int flags, int opcode, size_t argsize)
860 struct fuse_rename2_in inarg;
861 struct fuse_mount *fm = get_fuse_mount(olddir);
864 memset(&inarg, 0, argsize);
865 inarg.newdir = get_node_id(newdir);
867 args.opcode = opcode;
868 args.nodeid = get_node_id(olddir);
870 args.in_args[0].size = argsize;
871 args.in_args[0].value = &inarg;
872 args.in_args[1].size = oldent->d_name.len + 1;
873 args.in_args[1].value = oldent->d_name.name;
874 args.in_args[2].size = newent->d_name.len + 1;
875 args.in_args[2].value = newent->d_name.name;
876 err = fuse_simple_request(fm, &args);
879 fuse_invalidate_attr(d_inode(oldent));
880 fuse_update_ctime(d_inode(oldent));
882 if (flags & RENAME_EXCHANGE) {
883 fuse_invalidate_attr(d_inode(newent));
884 fuse_update_ctime(d_inode(newent));
887 fuse_dir_changed(olddir);
888 if (olddir != newdir)
889 fuse_dir_changed(newdir);
891 /* newent will end up negative */
892 if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
893 fuse_invalidate_attr(d_inode(newent));
894 fuse_invalidate_entry_cache(newent);
895 fuse_update_ctime(d_inode(newent));
897 } else if (err == -EINTR) {
898 /* If request was interrupted, DEITY only knows if the
899 rename actually took place. If the invalidation
900 fails (e.g. some process has CWD under the renamed
901 directory), then there can be inconsistency between
902 the dcache and the real filesystem. Tough luck. */
903 fuse_invalidate_entry(oldent);
904 if (d_really_is_positive(newent))
905 fuse_invalidate_entry(newent);
911 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
912 struct inode *newdir, struct dentry *newent,
915 struct fuse_conn *fc = get_fuse_conn(olddir);
918 if (fuse_is_bad(olddir))
921 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
925 if (fc->no_rename2 || fc->minor < 23)
928 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
930 sizeof(struct fuse_rename2_in));
931 if (err == -ENOSYS) {
936 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
938 sizeof(struct fuse_rename_in));
944 static int fuse_link(struct dentry *entry, struct inode *newdir,
945 struct dentry *newent)
948 struct fuse_link_in inarg;
949 struct inode *inode = d_inode(entry);
950 struct fuse_mount *fm = get_fuse_mount(inode);
953 memset(&inarg, 0, sizeof(inarg));
954 inarg.oldnodeid = get_node_id(inode);
955 args.opcode = FUSE_LINK;
957 args.in_args[0].size = sizeof(inarg);
958 args.in_args[0].value = &inarg;
959 args.in_args[1].size = newent->d_name.len + 1;
960 args.in_args[1].value = newent->d_name.name;
961 err = create_new_entry(fm, &args, newdir, newent, inode->i_mode);
962 /* Contrary to "normal" filesystems it can happen that link
963 makes two "logical" inodes point to the same "physical"
964 inode. We invalidate the attributes of the old one, so it
965 will reflect changes in the backing inode (link count,
969 struct fuse_inode *fi = get_fuse_inode(inode);
971 spin_lock(&fi->lock);
972 fi->attr_version = atomic64_inc_return(&fm->fc->attr_version);
973 if (likely(inode->i_nlink < UINT_MAX))
975 spin_unlock(&fi->lock);
976 fuse_invalidate_attr(inode);
977 fuse_update_ctime(inode);
978 } else if (err == -EINTR) {
979 fuse_invalidate_attr(inode);
984 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
987 unsigned int blkbits;
988 struct fuse_conn *fc = get_fuse_conn(inode);
990 /* see the comment in fuse_change_attributes() */
991 if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
992 attr->size = i_size_read(inode);
993 attr->mtime = inode->i_mtime.tv_sec;
994 attr->mtimensec = inode->i_mtime.tv_nsec;
995 attr->ctime = inode->i_ctime.tv_sec;
996 attr->ctimensec = inode->i_ctime.tv_nsec;
999 stat->dev = inode->i_sb->s_dev;
1000 stat->ino = attr->ino;
1001 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
1002 stat->nlink = attr->nlink;
1003 stat->uid = make_kuid(fc->user_ns, attr->uid);
1004 stat->gid = make_kgid(fc->user_ns, attr->gid);
1005 stat->rdev = inode->i_rdev;
1006 stat->atime.tv_sec = attr->atime;
1007 stat->atime.tv_nsec = attr->atimensec;
1008 stat->mtime.tv_sec = attr->mtime;
1009 stat->mtime.tv_nsec = attr->mtimensec;
1010 stat->ctime.tv_sec = attr->ctime;
1011 stat->ctime.tv_nsec = attr->ctimensec;
1012 stat->size = attr->size;
1013 stat->blocks = attr->blocks;
1015 if (attr->blksize != 0)
1016 blkbits = ilog2(attr->blksize);
1018 blkbits = inode->i_sb->s_blocksize_bits;
1020 stat->blksize = 1 << blkbits;
1023 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
1027 struct fuse_getattr_in inarg;
1028 struct fuse_attr_out outarg;
1029 struct fuse_mount *fm = get_fuse_mount(inode);
1033 attr_version = fuse_get_attr_version(fm->fc);
1035 memset(&inarg, 0, sizeof(inarg));
1036 memset(&outarg, 0, sizeof(outarg));
1037 /* Directories have separate file-handle space */
1038 if (file && S_ISREG(inode->i_mode)) {
1039 struct fuse_file *ff = file->private_data;
1041 inarg.getattr_flags |= FUSE_GETATTR_FH;
1044 args.opcode = FUSE_GETATTR;
1045 args.nodeid = get_node_id(inode);
1046 args.in_numargs = 1;
1047 args.in_args[0].size = sizeof(inarg);
1048 args.in_args[0].value = &inarg;
1049 args.out_numargs = 1;
1050 args.out_args[0].size = sizeof(outarg);
1051 args.out_args[0].value = &outarg;
1052 err = fuse_simple_request(fm, &args);
1054 if (fuse_invalid_attr(&outarg.attr) ||
1055 (inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1056 fuse_make_bad(inode);
1059 fuse_change_attributes(inode, &outarg.attr,
1060 attr_timeout(&outarg),
1063 fuse_fillattr(inode, &outarg.attr, stat);
1069 static int fuse_update_get_attr(struct inode *inode, struct file *file,
1070 struct kstat *stat, u32 request_mask,
1073 struct fuse_inode *fi = get_fuse_inode(inode);
1077 if (flags & AT_STATX_FORCE_SYNC)
1079 else if (flags & AT_STATX_DONT_SYNC)
1081 else if (request_mask & READ_ONCE(fi->inval_mask))
1084 sync = time_before64(fi->i_time, get_jiffies_64());
1087 forget_all_cached_acls(inode);
1088 err = fuse_do_getattr(inode, stat, file);
1090 generic_fillattr(&init_user_ns, inode, stat);
1091 stat->mode = fi->orig_i_mode;
1092 stat->ino = fi->orig_ino;
1098 int fuse_update_attributes(struct inode *inode, struct file *file)
1100 /* Do *not* need to get atime for internal purposes */
1101 return fuse_update_get_attr(inode, file, NULL,
1102 STATX_BASIC_STATS & ~STATX_ATIME, 0);
1105 int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
1106 u64 child_nodeid, struct qstr *name)
1109 struct inode *parent;
1111 struct dentry *entry;
1113 parent = fuse_ilookup(fc, parent_nodeid, NULL);
1118 if (!S_ISDIR(parent->i_mode))
1122 dir = d_find_alias(parent);
1126 name->hash = full_name_hash(dir, name->name, name->len);
1127 entry = d_lookup(dir, name);
1132 fuse_dir_changed(parent);
1133 fuse_invalidate_entry(entry);
1135 if (child_nodeid != 0 && d_really_is_positive(entry)) {
1136 inode_lock(d_inode(entry));
1137 if (get_node_id(d_inode(entry)) != child_nodeid) {
1141 if (d_mountpoint(entry)) {
1145 if (d_is_dir(entry)) {
1146 shrink_dcache_parent(entry);
1147 if (!simple_empty(entry)) {
1151 d_inode(entry)->i_flags |= S_DEAD;
1154 clear_nlink(d_inode(entry));
1157 inode_unlock(d_inode(entry));
1166 inode_unlock(parent);
1172 * Calling into a user-controlled filesystem gives the filesystem
1173 * daemon ptrace-like capabilities over the current process. This
1174 * means, that the filesystem daemon is able to record the exact
1175 * filesystem operations performed, and can also control the behavior
1176 * of the requester process in otherwise impossible ways. For example
1177 * it can delay the operation for arbitrary length of time allowing
1178 * DoS against the requester.
1180 * For this reason only those processes can call into the filesystem,
1181 * for which the owner of the mount has ptrace privilege. This
1182 * excludes processes started by other users, suid or sgid processes.
1184 int fuse_allow_current_process(struct fuse_conn *fc)
1186 const struct cred *cred;
1188 if (fc->allow_other)
1189 return current_in_userns(fc->user_ns);
1191 cred = current_cred();
1192 if (uid_eq(cred->euid, fc->user_id) &&
1193 uid_eq(cred->suid, fc->user_id) &&
1194 uid_eq(cred->uid, fc->user_id) &&
1195 gid_eq(cred->egid, fc->group_id) &&
1196 gid_eq(cred->sgid, fc->group_id) &&
1197 gid_eq(cred->gid, fc->group_id))
1203 static int fuse_access(struct inode *inode, int mask)
1205 struct fuse_mount *fm = get_fuse_mount(inode);
1207 struct fuse_access_in inarg;
1210 BUG_ON(mask & MAY_NOT_BLOCK);
1212 if (fm->fc->no_access)
1215 memset(&inarg, 0, sizeof(inarg));
1216 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1217 args.opcode = FUSE_ACCESS;
1218 args.nodeid = get_node_id(inode);
1219 args.in_numargs = 1;
1220 args.in_args[0].size = sizeof(inarg);
1221 args.in_args[0].value = &inarg;
1222 err = fuse_simple_request(fm, &args);
1223 if (err == -ENOSYS) {
1224 fm->fc->no_access = 1;
1230 static int fuse_perm_getattr(struct inode *inode, int mask)
1232 if (mask & MAY_NOT_BLOCK)
1235 forget_all_cached_acls(inode);
1236 return fuse_do_getattr(inode, NULL, NULL);
1240 * Check permission. The two basic access models of FUSE are:
1242 * 1) Local access checking ('default_permissions' mount option) based
1243 * on file mode. This is the plain old disk filesystem permission
1246 * 2) "Remote" access checking, where server is responsible for
1247 * checking permission in each inode operation. An exception to this
1248 * is if ->permission() was invoked from sys_access() in which case an
1249 * access request is sent. Execute permission is still checked
1250 * locally based on file mode.
1252 static int fuse_permission(struct inode *inode, int mask)
1254 struct fuse_conn *fc = get_fuse_conn(inode);
1255 bool refreshed = false;
1258 if (fuse_is_bad(inode))
1261 if (!fuse_allow_current_process(fc))
1265 * If attributes are needed, refresh them before proceeding
1267 if (fc->default_permissions ||
1268 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1269 struct fuse_inode *fi = get_fuse_inode(inode);
1270 u32 perm_mask = STATX_MODE | STATX_UID | STATX_GID;
1272 if (perm_mask & READ_ONCE(fi->inval_mask) ||
1273 time_before64(fi->i_time, get_jiffies_64())) {
1276 err = fuse_perm_getattr(inode, mask);
1282 if (fc->default_permissions) {
1283 err = generic_permission(&init_user_ns, inode, mask);
1285 /* If permission is denied, try to refresh file
1286 attributes. This is also needed, because the root
1287 node will at first have no permissions */
1288 if (err == -EACCES && !refreshed) {
1289 err = fuse_perm_getattr(inode, mask);
1291 err = generic_permission(&init_user_ns,
1295 /* Note: the opposite of the above test does not
1296 exist. So if permissions are revoked this won't be
1297 noticed immediately, only after the attribute
1298 timeout has expired */
1299 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1300 err = fuse_access(inode, mask);
1301 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1302 if (!(inode->i_mode & S_IXUGO)) {
1306 err = fuse_perm_getattr(inode, mask);
1307 if (!err && !(inode->i_mode & S_IXUGO))
1314 static int fuse_readlink_page(struct inode *inode, struct page *page)
1316 struct fuse_mount *fm = get_fuse_mount(inode);
1317 struct fuse_page_desc desc = { .length = PAGE_SIZE - 1 };
1318 struct fuse_args_pages ap = {
1326 ap.args.opcode = FUSE_READLINK;
1327 ap.args.nodeid = get_node_id(inode);
1328 ap.args.out_pages = true;
1329 ap.args.out_argvar = true;
1330 ap.args.page_zeroing = true;
1331 ap.args.out_numargs = 1;
1332 ap.args.out_args[0].size = desc.length;
1333 res = fuse_simple_request(fm, &ap.args);
1335 fuse_invalidate_atime(inode);
1340 if (WARN_ON(res >= PAGE_SIZE))
1343 link = page_address(page);
1349 static const char *fuse_get_link(struct dentry *dentry, struct inode *inode,
1350 struct delayed_call *callback)
1352 struct fuse_conn *fc = get_fuse_conn(inode);
1357 if (fuse_is_bad(inode))
1360 if (fc->cache_symlinks)
1361 return page_get_link(dentry, inode, callback);
1367 page = alloc_page(GFP_KERNEL);
1372 err = fuse_readlink_page(inode, page);
1378 set_delayed_call(callback, page_put_link, page);
1380 return page_address(page);
1383 return ERR_PTR(err);
1386 static int fuse_dir_open(struct inode *inode, struct file *file)
1388 return fuse_open_common(inode, file, true);
1391 static int fuse_dir_release(struct inode *inode, struct file *file)
1393 fuse_release_common(file, true);
1398 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1401 struct inode *inode = file->f_mapping->host;
1402 struct fuse_conn *fc = get_fuse_conn(inode);
1405 if (fuse_is_bad(inode))
1408 if (fc->no_fsyncdir)
1412 err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNCDIR);
1413 if (err == -ENOSYS) {
1414 fc->no_fsyncdir = 1;
1417 inode_unlock(inode);
1422 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1425 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1427 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1431 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1434 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1437 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1442 return fuse_ioctl_common(file, cmd, arg,
1443 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1446 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1448 /* Always update if mtime is explicitly set */
1449 if (ivalid & ATTR_MTIME_SET)
1452 /* Or if kernel i_mtime is the official one */
1453 if (trust_local_mtime)
1456 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1457 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1460 /* In all other cases update */
1464 static void iattr_to_fattr(struct fuse_conn *fc, struct iattr *iattr,
1465 struct fuse_setattr_in *arg, bool trust_local_cmtime)
1467 unsigned ivalid = iattr->ia_valid;
1469 if (ivalid & ATTR_MODE)
1470 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1471 if (ivalid & ATTR_UID)
1472 arg->valid |= FATTR_UID, arg->uid = from_kuid(fc->user_ns, iattr->ia_uid);
1473 if (ivalid & ATTR_GID)
1474 arg->valid |= FATTR_GID, arg->gid = from_kgid(fc->user_ns, iattr->ia_gid);
1475 if (ivalid & ATTR_SIZE)
1476 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1477 if (ivalid & ATTR_ATIME) {
1478 arg->valid |= FATTR_ATIME;
1479 arg->atime = iattr->ia_atime.tv_sec;
1480 arg->atimensec = iattr->ia_atime.tv_nsec;
1481 if (!(ivalid & ATTR_ATIME_SET))
1482 arg->valid |= FATTR_ATIME_NOW;
1484 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1485 arg->valid |= FATTR_MTIME;
1486 arg->mtime = iattr->ia_mtime.tv_sec;
1487 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1488 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1489 arg->valid |= FATTR_MTIME_NOW;
1491 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1492 arg->valid |= FATTR_CTIME;
1493 arg->ctime = iattr->ia_ctime.tv_sec;
1494 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1499 * Prevent concurrent writepages on inode
1501 * This is done by adding a negative bias to the inode write counter
1502 * and waiting for all pending writes to finish.
1504 void fuse_set_nowrite(struct inode *inode)
1506 struct fuse_inode *fi = get_fuse_inode(inode);
1508 BUG_ON(!inode_is_locked(inode));
1510 spin_lock(&fi->lock);
1511 BUG_ON(fi->writectr < 0);
1512 fi->writectr += FUSE_NOWRITE;
1513 spin_unlock(&fi->lock);
1514 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1518 * Allow writepages on inode
1520 * Remove the bias from the writecounter and send any queued
1523 static void __fuse_release_nowrite(struct inode *inode)
1525 struct fuse_inode *fi = get_fuse_inode(inode);
1527 BUG_ON(fi->writectr != FUSE_NOWRITE);
1529 fuse_flush_writepages(inode);
1532 void fuse_release_nowrite(struct inode *inode)
1534 struct fuse_inode *fi = get_fuse_inode(inode);
1536 spin_lock(&fi->lock);
1537 __fuse_release_nowrite(inode);
1538 spin_unlock(&fi->lock);
1541 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1542 struct inode *inode,
1543 struct fuse_setattr_in *inarg_p,
1544 struct fuse_attr_out *outarg_p)
1546 args->opcode = FUSE_SETATTR;
1547 args->nodeid = get_node_id(inode);
1548 args->in_numargs = 1;
1549 args->in_args[0].size = sizeof(*inarg_p);
1550 args->in_args[0].value = inarg_p;
1551 args->out_numargs = 1;
1552 args->out_args[0].size = sizeof(*outarg_p);
1553 args->out_args[0].value = outarg_p;
1557 * Flush inode->i_mtime to the server
1559 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1561 struct fuse_mount *fm = get_fuse_mount(inode);
1563 struct fuse_setattr_in inarg;
1564 struct fuse_attr_out outarg;
1566 memset(&inarg, 0, sizeof(inarg));
1567 memset(&outarg, 0, sizeof(outarg));
1569 inarg.valid = FATTR_MTIME;
1570 inarg.mtime = inode->i_mtime.tv_sec;
1571 inarg.mtimensec = inode->i_mtime.tv_nsec;
1572 if (fm->fc->minor >= 23) {
1573 inarg.valid |= FATTR_CTIME;
1574 inarg.ctime = inode->i_ctime.tv_sec;
1575 inarg.ctimensec = inode->i_ctime.tv_nsec;
1578 inarg.valid |= FATTR_FH;
1581 fuse_setattr_fill(fm->fc, &args, inode, &inarg, &outarg);
1583 return fuse_simple_request(fm, &args);
1587 * Set attributes, and at the same time refresh them.
1589 * Truncation is slightly complicated, because the 'truncate' request
1590 * may fail, in which case we don't want to touch the mapping.
1591 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1592 * and the actual truncation by hand.
1594 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1597 struct inode *inode = d_inode(dentry);
1598 struct fuse_mount *fm = get_fuse_mount(inode);
1599 struct fuse_conn *fc = fm->fc;
1600 struct fuse_inode *fi = get_fuse_inode(inode);
1602 struct fuse_setattr_in inarg;
1603 struct fuse_attr_out outarg;
1604 bool is_truncate = false;
1605 bool is_wb = fc->writeback_cache;
1608 bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1609 bool fault_blocked = false;
1611 if (!fc->default_permissions)
1612 attr->ia_valid |= ATTR_FORCE;
1614 err = setattr_prepare(&init_user_ns, dentry, attr);
1618 if (attr->ia_valid & ATTR_SIZE) {
1619 if (WARN_ON(!S_ISREG(inode->i_mode)))
1624 if (FUSE_IS_DAX(inode) && is_truncate) {
1625 down_write(&fi->i_mmap_sem);
1626 fault_blocked = true;
1627 err = fuse_dax_break_layouts(inode, 0, 0);
1629 up_write(&fi->i_mmap_sem);
1634 if (attr->ia_valid & ATTR_OPEN) {
1635 /* This is coming from open(..., ... | O_TRUNC); */
1636 WARN_ON(!(attr->ia_valid & ATTR_SIZE));
1637 WARN_ON(attr->ia_size != 0);
1638 if (fc->atomic_o_trunc) {
1640 * No need to send request to userspace, since actual
1641 * truncation has already been done by OPEN. But still
1642 * need to truncate page cache.
1644 i_size_write(inode, 0);
1645 truncate_pagecache(inode, 0);
1651 /* Flush dirty data/metadata before non-truncate SETATTR */
1652 if (is_wb && S_ISREG(inode->i_mode) &&
1654 (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_MTIME_SET |
1656 err = write_inode_now(inode, true);
1660 fuse_set_nowrite(inode);
1661 fuse_release_nowrite(inode);
1665 fuse_set_nowrite(inode);
1666 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1667 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1668 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1671 memset(&inarg, 0, sizeof(inarg));
1672 memset(&outarg, 0, sizeof(outarg));
1673 iattr_to_fattr(fc, attr, &inarg, trust_local_cmtime);
1675 struct fuse_file *ff = file->private_data;
1676 inarg.valid |= FATTR_FH;
1680 /* Kill suid/sgid for non-directory chown unconditionally */
1681 if (fc->handle_killpriv_v2 && !S_ISDIR(inode->i_mode) &&
1682 attr->ia_valid & (ATTR_UID | ATTR_GID))
1683 inarg.valid |= FATTR_KILL_SUIDGID;
1685 if (attr->ia_valid & ATTR_SIZE) {
1686 /* For mandatory locking in truncate */
1687 inarg.valid |= FATTR_LOCKOWNER;
1688 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1690 /* Kill suid/sgid for truncate only if no CAP_FSETID */
1691 if (fc->handle_killpriv_v2 && !capable(CAP_FSETID))
1692 inarg.valid |= FATTR_KILL_SUIDGID;
1694 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1695 err = fuse_simple_request(fm, &args);
1698 fuse_invalidate_attr(inode);
1702 if (fuse_invalid_attr(&outarg.attr) ||
1703 (inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1704 fuse_make_bad(inode);
1709 spin_lock(&fi->lock);
1710 /* the kernel maintains i_mtime locally */
1711 if (trust_local_cmtime) {
1712 if (attr->ia_valid & ATTR_MTIME)
1713 inode->i_mtime = attr->ia_mtime;
1714 if (attr->ia_valid & ATTR_CTIME)
1715 inode->i_ctime = attr->ia_ctime;
1716 /* FIXME: clear I_DIRTY_SYNC? */
1719 fuse_change_attributes_common(inode, &outarg.attr,
1720 attr_timeout(&outarg));
1721 oldsize = inode->i_size;
1722 /* see the comment in fuse_change_attributes() */
1723 if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1724 i_size_write(inode, outarg.attr.size);
1727 /* NOTE: this may release/reacquire fi->lock */
1728 __fuse_release_nowrite(inode);
1730 spin_unlock(&fi->lock);
1733 * Only call invalidate_inode_pages2() after removing
1734 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1736 if ((is_truncate || !is_wb) &&
1737 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1738 truncate_pagecache(inode, outarg.attr.size);
1739 invalidate_inode_pages2(inode->i_mapping);
1742 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1745 up_write(&fi->i_mmap_sem);
1751 fuse_release_nowrite(inode);
1753 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1756 up_write(&fi->i_mmap_sem);
1760 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1762 struct inode *inode = d_inode(entry);
1763 struct fuse_conn *fc = get_fuse_conn(inode);
1764 struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1767 if (fuse_is_bad(inode))
1770 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1773 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1774 attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1778 * The only sane way to reliably kill suid/sgid is to do it in
1779 * the userspace filesystem
1781 * This should be done on write(), truncate() and chown().
1783 if (!fc->handle_killpriv && !fc->handle_killpriv_v2) {
1785 * ia_mode calculation may have used stale i_mode.
1786 * Refresh and recalculate.
1788 ret = fuse_do_getattr(inode, NULL, file);
1792 attr->ia_mode = inode->i_mode;
1793 if (inode->i_mode & S_ISUID) {
1794 attr->ia_valid |= ATTR_MODE;
1795 attr->ia_mode &= ~S_ISUID;
1797 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1798 attr->ia_valid |= ATTR_MODE;
1799 attr->ia_mode &= ~S_ISGID;
1803 if (!attr->ia_valid)
1806 ret = fuse_do_setattr(entry, attr, file);
1809 * If filesystem supports acls it may have updated acl xattrs in
1810 * the filesystem, so forget cached acls for the inode.
1813 forget_all_cached_acls(inode);
1815 /* Directory mode changed, may need to revalidate access */
1816 if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1817 fuse_invalidate_entry_cache(entry);
1822 static int fuse_getattr(const struct path *path, struct kstat *stat,
1823 u32 request_mask, unsigned int flags)
1825 struct inode *inode = d_inode(path->dentry);
1826 struct fuse_conn *fc = get_fuse_conn(inode);
1828 if (fuse_is_bad(inode))
1831 if (!fuse_allow_current_process(fc)) {
1832 if (!request_mask) {
1834 * If user explicitly requested *nothing* then don't
1835 * error out, but return st_dev only.
1837 stat->result_mask = 0;
1838 stat->dev = inode->i_sb->s_dev;
1844 return fuse_update_get_attr(inode, NULL, stat, request_mask, flags);
1847 static const struct inode_operations fuse_dir_inode_operations = {
1848 .lookup = fuse_lookup,
1849 .mkdir = fuse_mkdir,
1850 .symlink = fuse_symlink,
1851 .unlink = fuse_unlink,
1852 .rmdir = fuse_rmdir,
1853 .rename = fuse_rename2,
1855 .setattr = fuse_setattr,
1856 .create = fuse_create,
1857 .atomic_open = fuse_atomic_open,
1858 .mknod = fuse_mknod,
1859 .permission = fuse_permission,
1860 .getattr = fuse_getattr,
1861 .listxattr = fuse_listxattr,
1862 .get_acl = fuse_get_acl,
1863 .set_acl = fuse_set_acl,
1866 static const struct file_operations fuse_dir_operations = {
1867 .llseek = generic_file_llseek,
1868 .read = generic_read_dir,
1869 .iterate_shared = fuse_readdir,
1870 .open = fuse_dir_open,
1871 .release = fuse_dir_release,
1872 .fsync = fuse_dir_fsync,
1873 .unlocked_ioctl = fuse_dir_ioctl,
1874 .compat_ioctl = fuse_dir_compat_ioctl,
1877 static const struct inode_operations fuse_common_inode_operations = {
1878 .setattr = fuse_setattr,
1879 .permission = fuse_permission,
1880 .getattr = fuse_getattr,
1881 .listxattr = fuse_listxattr,
1882 .get_acl = fuse_get_acl,
1883 .set_acl = fuse_set_acl,
1886 static const struct inode_operations fuse_symlink_inode_operations = {
1887 .setattr = fuse_setattr,
1888 .get_link = fuse_get_link,
1889 .getattr = fuse_getattr,
1890 .listxattr = fuse_listxattr,
1893 void fuse_init_common(struct inode *inode)
1895 inode->i_op = &fuse_common_inode_operations;
1898 void fuse_init_dir(struct inode *inode)
1900 struct fuse_inode *fi = get_fuse_inode(inode);
1902 inode->i_op = &fuse_dir_inode_operations;
1903 inode->i_fop = &fuse_dir_operations;
1905 spin_lock_init(&fi->rdc.lock);
1906 fi->rdc.cached = false;
1909 fi->rdc.version = 0;
1912 static int fuse_symlink_readpage(struct file *null, struct page *page)
1914 int err = fuse_readlink_page(page->mapping->host, page);
1917 SetPageUptodate(page);
1924 static const struct address_space_operations fuse_symlink_aops = {
1925 .readpage = fuse_symlink_readpage,
1928 void fuse_init_symlink(struct inode *inode)
1930 inode->i_op = &fuse_symlink_inode_operations;
1931 inode->i_data.a_ops = &fuse_symlink_aops;
1932 inode_nohighmem(inode);