2 FUSE: Filesystem in Userspace
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/moduleparam.h>
15 #include <linux/sched.h>
16 #include <linux/namei.h>
17 #include <linux/slab.h>
18 #include <linux/xattr.h>
19 #include <linux/iversion.h>
20 #include <linux/posix_acl.h>
21 #include <linux/security.h>
22 #include <linux/types.h>
23 #include <linux/kernel.h>
25 static bool __read_mostly allow_sys_admin_access;
26 module_param(allow_sys_admin_access, bool, 0644);
27 MODULE_PARM_DESC(allow_sys_admin_access,
28 "Allow users with CAP_SYS_ADMIN in initial userns to bypass allow_other access check");
30 static void fuse_advise_use_readdirplus(struct inode *dir)
32 struct fuse_inode *fi = get_fuse_inode(dir);
34 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
37 #if BITS_PER_LONG >= 64
38 static inline void __fuse_dentry_settime(struct dentry *entry, u64 time)
40 entry->d_fsdata = (void *) time;
43 static inline u64 fuse_dentry_time(const struct dentry *entry)
45 return (u64)entry->d_fsdata;
54 static inline void __fuse_dentry_settime(struct dentry *dentry, u64 time)
56 ((union fuse_dentry *) dentry->d_fsdata)->time = time;
59 static inline u64 fuse_dentry_time(const struct dentry *entry)
61 return ((union fuse_dentry *) entry->d_fsdata)->time;
65 static void fuse_dentry_settime(struct dentry *dentry, u64 time)
67 struct fuse_conn *fc = get_fuse_conn_super(dentry->d_sb);
68 bool delete = !time && fc->delete_stale;
70 * Mess with DCACHE_OP_DELETE because dput() will be faster without it.
71 * Don't care about races, either way it's just an optimization
73 if ((!delete && (dentry->d_flags & DCACHE_OP_DELETE)) ||
74 (delete && !(dentry->d_flags & DCACHE_OP_DELETE))) {
75 spin_lock(&dentry->d_lock);
77 dentry->d_flags &= ~DCACHE_OP_DELETE;
79 dentry->d_flags |= DCACHE_OP_DELETE;
80 spin_unlock(&dentry->d_lock);
83 __fuse_dentry_settime(dentry, time);
87 * FUSE caches dentries and attributes with separate timeout. The
88 * time in jiffies until the dentry/attributes are valid is stored in
89 * dentry->d_fsdata and fuse_inode->i_time respectively.
93 * Calculate the time in jiffies until a dentry/attributes are valid
95 static u64 time_to_jiffies(u64 sec, u32 nsec)
98 struct timespec64 ts = {
100 min_t(u32, nsec, NSEC_PER_SEC - 1)
103 return get_jiffies_64() + timespec64_to_jiffies(&ts);
109 * Set dentry and possibly attribute timeouts from the lookup/mk*
112 void fuse_change_entry_timeout(struct dentry *entry, struct fuse_entry_out *o)
114 fuse_dentry_settime(entry,
115 time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
118 static u64 attr_timeout(struct fuse_attr_out *o)
120 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
123 u64 entry_attr_timeout(struct fuse_entry_out *o)
125 return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
128 void fuse_invalidate_attr_mask(struct inode *inode, u32 mask)
130 set_mask_bits(&get_fuse_inode(inode)->inval_mask, 0, mask);
134 * Mark the attributes as stale, so that at the next call to
135 * ->getattr() they will be fetched from userspace
137 void fuse_invalidate_attr(struct inode *inode)
139 fuse_invalidate_attr_mask(inode, STATX_BASIC_STATS);
142 static void fuse_dir_changed(struct inode *dir)
144 fuse_invalidate_attr(dir);
145 inode_maybe_inc_iversion(dir, false);
149 * Mark the attributes as stale due to an atime change. Avoid the invalidate if
152 void fuse_invalidate_atime(struct inode *inode)
154 if (!IS_RDONLY(inode))
155 fuse_invalidate_attr_mask(inode, STATX_ATIME);
159 * Just mark the entry as stale, so that a next attempt to look it up
160 * will result in a new lookup call to userspace
162 * This is called when a dentry is about to become negative and the
163 * timeout is unknown (unlink, rmdir, rename and in some cases
166 void fuse_invalidate_entry_cache(struct dentry *entry)
168 fuse_dentry_settime(entry, 0);
172 * Same as fuse_invalidate_entry_cache(), but also try to remove the
173 * dentry from the hash
175 static void fuse_invalidate_entry(struct dentry *entry)
178 fuse_invalidate_entry_cache(entry);
181 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
182 u64 nodeid, const struct qstr *name,
183 struct fuse_entry_out *outarg)
185 memset(outarg, 0, sizeof(struct fuse_entry_out));
186 args->opcode = FUSE_LOOKUP;
187 args->nodeid = nodeid;
188 args->in_numargs = 1;
189 args->in_args[0].size = name->len + 1;
190 args->in_args[0].value = name->name;
191 args->out_numargs = 1;
192 args->out_args[0].size = sizeof(struct fuse_entry_out);
193 args->out_args[0].value = outarg;
197 * Check whether the dentry is still valid
199 * If the entry validity timeout has expired and the dentry is
200 * positive, try to redo the lookup. If the lookup results in a
201 * different inode, then let the VFS invalidate the dentry and redo
202 * the lookup once more. If the lookup results in the same inode,
203 * then refresh the attributes, timeouts and mark the dentry valid.
205 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
208 struct dentry *parent;
209 struct fuse_mount *fm;
210 struct fuse_inode *fi;
213 inode = d_inode_rcu(entry);
214 if (inode && fuse_is_bad(inode))
216 else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
217 (flags & (LOOKUP_EXCL | LOOKUP_REVAL))) {
218 struct fuse_entry_out outarg;
220 struct fuse_forget_link *forget;
223 /* For negative dentries, always do a fresh lookup */
228 if (flags & LOOKUP_RCU)
231 fm = get_fuse_mount(inode);
233 forget = fuse_alloc_forget();
238 attr_version = fuse_get_attr_version(fm->fc);
240 parent = dget_parent(entry);
241 fuse_lookup_init(fm->fc, &args, get_node_id(d_inode(parent)),
242 &entry->d_name, &outarg);
243 ret = fuse_simple_request(fm, &args);
245 /* Zero nodeid is same as -ENOENT */
246 if (!ret && !outarg.nodeid)
249 fi = get_fuse_inode(inode);
250 if (outarg.nodeid != get_node_id(inode) ||
251 (bool) IS_AUTOMOUNT(inode) != (bool) (outarg.attr.flags & FUSE_ATTR_SUBMOUNT)) {
252 fuse_queue_forget(fm->fc, forget,
256 spin_lock(&fi->lock);
258 spin_unlock(&fi->lock);
263 if (ret || fuse_invalid_attr(&outarg.attr) ||
264 fuse_stale_inode(inode, outarg.generation, &outarg.attr))
267 forget_all_cached_acls(inode);
268 fuse_change_attributes(inode, &outarg.attr,
269 entry_attr_timeout(&outarg),
271 fuse_change_entry_timeout(entry, &outarg);
273 fi = get_fuse_inode(inode);
274 if (flags & LOOKUP_RCU) {
275 if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
277 } else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
278 parent = dget_parent(entry);
279 fuse_advise_use_readdirplus(d_inode(parent));
292 #if BITS_PER_LONG < 64
293 static int fuse_dentry_init(struct dentry *dentry)
295 dentry->d_fsdata = kzalloc(sizeof(union fuse_dentry),
296 GFP_KERNEL_ACCOUNT | __GFP_RECLAIMABLE);
298 return dentry->d_fsdata ? 0 : -ENOMEM;
300 static void fuse_dentry_release(struct dentry *dentry)
302 union fuse_dentry *fd = dentry->d_fsdata;
308 static int fuse_dentry_delete(const struct dentry *dentry)
310 return time_before64(fuse_dentry_time(dentry), get_jiffies_64());
314 * Create a fuse_mount object with a new superblock (with path->dentry
315 * as the root), and return that mount so it can be auto-mounted on
318 static struct vfsmount *fuse_dentry_automount(struct path *path)
320 struct fs_context *fsc;
321 struct vfsmount *mnt;
322 struct fuse_inode *mp_fi = get_fuse_inode(d_inode(path->dentry));
324 fsc = fs_context_for_submount(path->mnt->mnt_sb->s_type, path->dentry);
326 return ERR_CAST(fsc);
328 /* Pass the FUSE inode of the mount for fuse_get_tree_submount() */
329 fsc->fs_private = mp_fi;
331 /* Create the submount */
340 const struct dentry_operations fuse_dentry_operations = {
341 .d_revalidate = fuse_dentry_revalidate,
342 .d_delete = fuse_dentry_delete,
343 #if BITS_PER_LONG < 64
344 .d_init = fuse_dentry_init,
345 .d_release = fuse_dentry_release,
347 .d_automount = fuse_dentry_automount,
350 const struct dentry_operations fuse_root_dentry_operations = {
351 #if BITS_PER_LONG < 64
352 .d_init = fuse_dentry_init,
353 .d_release = fuse_dentry_release,
357 int fuse_valid_type(int m)
359 return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
360 S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
363 bool fuse_invalid_attr(struct fuse_attr *attr)
365 return !fuse_valid_type(attr->mode) ||
366 attr->size > LLONG_MAX;
369 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
370 struct fuse_entry_out *outarg, struct inode **inode)
372 struct fuse_mount *fm = get_fuse_mount_super(sb);
374 struct fuse_forget_link *forget;
380 if (name->len > FUSE_NAME_MAX)
384 forget = fuse_alloc_forget();
389 attr_version = fuse_get_attr_version(fm->fc);
391 fuse_lookup_init(fm->fc, &args, nodeid, name, outarg);
392 err = fuse_simple_request(fm, &args);
393 /* Zero nodeid is same as -ENOENT, but with valid timeout */
394 if (err || !outarg->nodeid)
400 if (fuse_invalid_attr(&outarg->attr))
403 *inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
404 &outarg->attr, entry_attr_timeout(outarg),
408 fuse_queue_forget(fm->fc, forget, outarg->nodeid, 1);
419 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
423 struct fuse_entry_out outarg;
425 struct dentry *newent;
426 bool outarg_valid = true;
429 if (fuse_is_bad(dir))
430 return ERR_PTR(-EIO);
432 locked = fuse_lock_inode(dir);
433 err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
435 fuse_unlock_inode(dir, locked);
436 if (err == -ENOENT) {
437 outarg_valid = false;
444 if (inode && get_node_id(inode) == FUSE_ROOT_ID)
447 newent = d_splice_alias(inode, entry);
448 err = PTR_ERR(newent);
452 entry = newent ? newent : entry;
454 fuse_change_entry_timeout(entry, &outarg);
456 fuse_invalidate_entry_cache(entry);
459 fuse_advise_use_readdirplus(dir);
468 static int get_security_context(struct dentry *entry, umode_t mode,
469 void **security_ctx, u32 *security_ctxlen)
471 struct fuse_secctx *fctx;
472 struct fuse_secctx_header *header;
473 void *ctx = NULL, *ptr;
474 u32 ctxlen, total_len = sizeof(*header);
479 err = security_dentry_init_security(entry, mode, &entry->d_name,
480 &name, &ctx, &ctxlen);
482 if (err != -EOPNOTSUPP)
484 /* No LSM is supporting this security hook. Ignore error */
491 namelen = strlen(name) + 1;
493 if (WARN_ON(namelen > XATTR_NAME_MAX + 1 || ctxlen > S32_MAX))
495 total_len += FUSE_REC_ALIGN(sizeof(*fctx) + namelen + ctxlen);
499 header = ptr = kzalloc(total_len, GFP_KERNEL);
503 header->nr_secctx = nr_ctx;
504 header->size = total_len;
505 ptr += sizeof(*header);
509 ptr += sizeof(*fctx);
514 memcpy(ptr, ctx, ctxlen);
516 *security_ctxlen = total_len;
517 *security_ctx = header;
525 * Atomic create+open operation
527 * If the filesystem doesn't support this, then fall back to separate
528 * 'mknod' + 'open' requests.
530 static int fuse_create_open(struct inode *dir, struct dentry *entry,
531 struct file *file, unsigned int flags,
536 struct fuse_mount *fm = get_fuse_mount(dir);
538 struct fuse_forget_link *forget;
539 struct fuse_create_in inarg;
540 struct fuse_open_out outopen;
541 struct fuse_entry_out outentry;
542 struct fuse_inode *fi;
543 struct fuse_file *ff;
544 void *security_ctx = NULL;
546 bool trunc = flags & O_TRUNC;
548 /* Userspace expects S_IFREG in create mode */
549 BUG_ON((mode & S_IFMT) != S_IFREG);
551 forget = fuse_alloc_forget();
557 ff = fuse_file_alloc(fm);
559 goto out_put_forget_req;
561 if (!fm->fc->dont_mask)
562 mode &= ~current_umask();
565 memset(&inarg, 0, sizeof(inarg));
566 memset(&outentry, 0, sizeof(outentry));
569 inarg.umask = current_umask();
571 if (fm->fc->handle_killpriv_v2 && trunc &&
572 !(flags & O_EXCL) && !capable(CAP_FSETID)) {
573 inarg.open_flags |= FUSE_OPEN_KILL_SUIDGID;
576 args.opcode = FUSE_CREATE;
577 args.nodeid = get_node_id(dir);
579 args.in_args[0].size = sizeof(inarg);
580 args.in_args[0].value = &inarg;
581 args.in_args[1].size = entry->d_name.len + 1;
582 args.in_args[1].value = entry->d_name.name;
583 args.out_numargs = 2;
584 args.out_args[0].size = sizeof(outentry);
585 args.out_args[0].value = &outentry;
586 args.out_args[1].size = sizeof(outopen);
587 args.out_args[1].value = &outopen;
589 if (fm->fc->init_security) {
590 err = get_security_context(entry, mode, &security_ctx,
593 goto out_put_forget_req;
596 args.in_args[2].size = security_ctxlen;
597 args.in_args[2].value = security_ctx;
600 err = fuse_simple_request(fm, &args);
606 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid) ||
607 fuse_invalid_attr(&outentry.attr))
611 ff->nodeid = outentry.nodeid;
612 ff->open_flags = outopen.open_flags;
613 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
614 &outentry.attr, entry_attr_timeout(&outentry), 0);
616 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
617 fuse_sync_release(NULL, ff, flags);
618 fuse_queue_forget(fm->fc, forget, outentry.nodeid, 1);
623 d_instantiate(entry, inode);
624 fuse_change_entry_timeout(entry, &outentry);
625 fuse_dir_changed(dir);
626 err = finish_open(file, entry, generic_file_open);
628 fi = get_fuse_inode(inode);
629 fuse_sync_release(fi, ff, flags);
631 file->private_data = ff;
632 fuse_finish_open(inode, file);
633 if (fm->fc->atomic_o_trunc && trunc)
634 truncate_pagecache(inode, 0);
635 else if (!(ff->open_flags & FOPEN_KEEP_CACHE))
636 invalidate_inode_pages2(inode->i_mapping);
648 static int fuse_mknod(struct user_namespace *, struct inode *, struct dentry *,
650 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
651 struct file *file, unsigned flags,
655 struct fuse_conn *fc = get_fuse_conn(dir);
656 struct dentry *res = NULL;
658 if (fuse_is_bad(dir))
661 if (d_in_lookup(entry)) {
662 res = fuse_lookup(dir, entry, 0);
670 if (!(flags & O_CREAT) || d_really_is_positive(entry))
674 file->f_mode |= FMODE_CREATED;
679 err = fuse_create_open(dir, entry, file, flags, mode);
680 if (err == -ENOSYS) {
689 err = fuse_mknod(&init_user_ns, dir, entry, mode, 0);
693 return finish_no_open(file, res);
697 * Code shared between mknod, mkdir, symlink and link
699 static int create_new_entry(struct fuse_mount *fm, struct fuse_args *args,
700 struct inode *dir, struct dentry *entry,
703 struct fuse_entry_out outarg;
707 struct fuse_forget_link *forget;
708 void *security_ctx = NULL;
711 if (fuse_is_bad(dir))
714 forget = fuse_alloc_forget();
718 memset(&outarg, 0, sizeof(outarg));
719 args->nodeid = get_node_id(dir);
720 args->out_numargs = 1;
721 args->out_args[0].size = sizeof(outarg);
722 args->out_args[0].value = &outarg;
724 if (fm->fc->init_security && args->opcode != FUSE_LINK) {
725 err = get_security_context(entry, mode, &security_ctx,
728 goto out_put_forget_req;
730 BUG_ON(args->in_numargs != 2);
732 args->in_numargs = 3;
733 args->in_args[2].size = security_ctxlen;
734 args->in_args[2].value = security_ctx;
737 err = fuse_simple_request(fm, args);
740 goto out_put_forget_req;
743 if (invalid_nodeid(outarg.nodeid) || fuse_invalid_attr(&outarg.attr))
744 goto out_put_forget_req;
746 if ((outarg.attr.mode ^ mode) & S_IFMT)
747 goto out_put_forget_req;
749 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
750 &outarg.attr, entry_attr_timeout(&outarg), 0);
752 fuse_queue_forget(fm->fc, forget, outarg.nodeid, 1);
758 d = d_splice_alias(inode, entry);
763 fuse_change_entry_timeout(d, &outarg);
766 fuse_change_entry_timeout(entry, &outarg);
768 fuse_dir_changed(dir);
776 static int fuse_mknod(struct user_namespace *mnt_userns, struct inode *dir,
777 struct dentry *entry, umode_t mode, dev_t rdev)
779 struct fuse_mknod_in inarg;
780 struct fuse_mount *fm = get_fuse_mount(dir);
783 if (!fm->fc->dont_mask)
784 mode &= ~current_umask();
786 memset(&inarg, 0, sizeof(inarg));
788 inarg.rdev = new_encode_dev(rdev);
789 inarg.umask = current_umask();
790 args.opcode = FUSE_MKNOD;
792 args.in_args[0].size = sizeof(inarg);
793 args.in_args[0].value = &inarg;
794 args.in_args[1].size = entry->d_name.len + 1;
795 args.in_args[1].value = entry->d_name.name;
796 return create_new_entry(fm, &args, dir, entry, mode);
799 static int fuse_create(struct user_namespace *mnt_userns, struct inode *dir,
800 struct dentry *entry, umode_t mode, bool excl)
802 return fuse_mknod(&init_user_ns, dir, entry, mode, 0);
805 static int fuse_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
806 struct dentry *entry, umode_t mode)
808 struct fuse_mkdir_in inarg;
809 struct fuse_mount *fm = get_fuse_mount(dir);
812 if (!fm->fc->dont_mask)
813 mode &= ~current_umask();
815 memset(&inarg, 0, sizeof(inarg));
817 inarg.umask = current_umask();
818 args.opcode = FUSE_MKDIR;
820 args.in_args[0].size = sizeof(inarg);
821 args.in_args[0].value = &inarg;
822 args.in_args[1].size = entry->d_name.len + 1;
823 args.in_args[1].value = entry->d_name.name;
824 return create_new_entry(fm, &args, dir, entry, S_IFDIR);
827 static int fuse_symlink(struct user_namespace *mnt_userns, struct inode *dir,
828 struct dentry *entry, const char *link)
830 struct fuse_mount *fm = get_fuse_mount(dir);
831 unsigned len = strlen(link) + 1;
834 args.opcode = FUSE_SYMLINK;
836 args.in_args[0].size = entry->d_name.len + 1;
837 args.in_args[0].value = entry->d_name.name;
838 args.in_args[1].size = len;
839 args.in_args[1].value = link;
840 return create_new_entry(fm, &args, dir, entry, S_IFLNK);
843 void fuse_flush_time_update(struct inode *inode)
845 int err = sync_inode_metadata(inode, 1);
847 mapping_set_error(inode->i_mapping, err);
850 static void fuse_update_ctime_in_cache(struct inode *inode)
852 if (!IS_NOCMTIME(inode)) {
853 inode->i_ctime = current_time(inode);
854 mark_inode_dirty_sync(inode);
855 fuse_flush_time_update(inode);
859 void fuse_update_ctime(struct inode *inode)
861 fuse_invalidate_attr_mask(inode, STATX_CTIME);
862 fuse_update_ctime_in_cache(inode);
865 static void fuse_entry_unlinked(struct dentry *entry)
867 struct inode *inode = d_inode(entry);
868 struct fuse_conn *fc = get_fuse_conn(inode);
869 struct fuse_inode *fi = get_fuse_inode(inode);
871 spin_lock(&fi->lock);
872 fi->attr_version = atomic64_inc_return(&fc->attr_version);
874 * If i_nlink == 0 then unlink doesn't make sense, yet this can
875 * happen if userspace filesystem is careless. It would be
876 * difficult to enforce correct nlink usage so just ignore this
879 if (S_ISDIR(inode->i_mode))
881 else if (inode->i_nlink > 0)
883 spin_unlock(&fi->lock);
884 fuse_invalidate_entry_cache(entry);
885 fuse_update_ctime(inode);
888 static int fuse_unlink(struct inode *dir, struct dentry *entry)
891 struct fuse_mount *fm = get_fuse_mount(dir);
894 if (fuse_is_bad(dir))
897 args.opcode = FUSE_UNLINK;
898 args.nodeid = get_node_id(dir);
900 args.in_args[0].size = entry->d_name.len + 1;
901 args.in_args[0].value = entry->d_name.name;
902 err = fuse_simple_request(fm, &args);
904 fuse_dir_changed(dir);
905 fuse_entry_unlinked(entry);
906 } else if (err == -EINTR)
907 fuse_invalidate_entry(entry);
911 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
914 struct fuse_mount *fm = get_fuse_mount(dir);
917 if (fuse_is_bad(dir))
920 args.opcode = FUSE_RMDIR;
921 args.nodeid = get_node_id(dir);
923 args.in_args[0].size = entry->d_name.len + 1;
924 args.in_args[0].value = entry->d_name.name;
925 err = fuse_simple_request(fm, &args);
927 fuse_dir_changed(dir);
928 fuse_entry_unlinked(entry);
929 } else if (err == -EINTR)
930 fuse_invalidate_entry(entry);
934 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
935 struct inode *newdir, struct dentry *newent,
936 unsigned int flags, int opcode, size_t argsize)
939 struct fuse_rename2_in inarg;
940 struct fuse_mount *fm = get_fuse_mount(olddir);
943 memset(&inarg, 0, argsize);
944 inarg.newdir = get_node_id(newdir);
946 args.opcode = opcode;
947 args.nodeid = get_node_id(olddir);
949 args.in_args[0].size = argsize;
950 args.in_args[0].value = &inarg;
951 args.in_args[1].size = oldent->d_name.len + 1;
952 args.in_args[1].value = oldent->d_name.name;
953 args.in_args[2].size = newent->d_name.len + 1;
954 args.in_args[2].value = newent->d_name.name;
955 err = fuse_simple_request(fm, &args);
958 fuse_update_ctime(d_inode(oldent));
960 if (flags & RENAME_EXCHANGE)
961 fuse_update_ctime(d_inode(newent));
963 fuse_dir_changed(olddir);
964 if (olddir != newdir)
965 fuse_dir_changed(newdir);
967 /* newent will end up negative */
968 if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent))
969 fuse_entry_unlinked(newent);
970 } else if (err == -EINTR) {
971 /* If request was interrupted, DEITY only knows if the
972 rename actually took place. If the invalidation
973 fails (e.g. some process has CWD under the renamed
974 directory), then there can be inconsistency between
975 the dcache and the real filesystem. Tough luck. */
976 fuse_invalidate_entry(oldent);
977 if (d_really_is_positive(newent))
978 fuse_invalidate_entry(newent);
984 static int fuse_rename2(struct user_namespace *mnt_userns, struct inode *olddir,
985 struct dentry *oldent, struct inode *newdir,
986 struct dentry *newent, unsigned int flags)
988 struct fuse_conn *fc = get_fuse_conn(olddir);
991 if (fuse_is_bad(olddir))
994 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
998 if (fc->no_rename2 || fc->minor < 23)
1001 err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
1003 sizeof(struct fuse_rename2_in));
1004 if (err == -ENOSYS) {
1009 err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
1011 sizeof(struct fuse_rename_in));
1017 static int fuse_link(struct dentry *entry, struct inode *newdir,
1018 struct dentry *newent)
1021 struct fuse_link_in inarg;
1022 struct inode *inode = d_inode(entry);
1023 struct fuse_mount *fm = get_fuse_mount(inode);
1026 memset(&inarg, 0, sizeof(inarg));
1027 inarg.oldnodeid = get_node_id(inode);
1028 args.opcode = FUSE_LINK;
1029 args.in_numargs = 2;
1030 args.in_args[0].size = sizeof(inarg);
1031 args.in_args[0].value = &inarg;
1032 args.in_args[1].size = newent->d_name.len + 1;
1033 args.in_args[1].value = newent->d_name.name;
1034 err = create_new_entry(fm, &args, newdir, newent, inode->i_mode);
1036 fuse_update_ctime_in_cache(inode);
1037 else if (err == -EINTR)
1038 fuse_invalidate_attr(inode);
1043 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
1046 unsigned int blkbits;
1047 struct fuse_conn *fc = get_fuse_conn(inode);
1049 stat->dev = inode->i_sb->s_dev;
1050 stat->ino = attr->ino;
1051 stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
1052 stat->nlink = attr->nlink;
1053 stat->uid = make_kuid(fc->user_ns, attr->uid);
1054 stat->gid = make_kgid(fc->user_ns, attr->gid);
1055 stat->rdev = inode->i_rdev;
1056 stat->atime.tv_sec = attr->atime;
1057 stat->atime.tv_nsec = attr->atimensec;
1058 stat->mtime.tv_sec = attr->mtime;
1059 stat->mtime.tv_nsec = attr->mtimensec;
1060 stat->ctime.tv_sec = attr->ctime;
1061 stat->ctime.tv_nsec = attr->ctimensec;
1062 stat->size = attr->size;
1063 stat->blocks = attr->blocks;
1065 if (attr->blksize != 0)
1066 blkbits = ilog2(attr->blksize);
1068 blkbits = inode->i_sb->s_blocksize_bits;
1070 stat->blksize = 1 << blkbits;
1073 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
1077 struct fuse_getattr_in inarg;
1078 struct fuse_attr_out outarg;
1079 struct fuse_mount *fm = get_fuse_mount(inode);
1083 attr_version = fuse_get_attr_version(fm->fc);
1085 memset(&inarg, 0, sizeof(inarg));
1086 memset(&outarg, 0, sizeof(outarg));
1087 /* Directories have separate file-handle space */
1088 if (file && S_ISREG(inode->i_mode)) {
1089 struct fuse_file *ff = file->private_data;
1091 inarg.getattr_flags |= FUSE_GETATTR_FH;
1094 args.opcode = FUSE_GETATTR;
1095 args.nodeid = get_node_id(inode);
1096 args.in_numargs = 1;
1097 args.in_args[0].size = sizeof(inarg);
1098 args.in_args[0].value = &inarg;
1099 args.out_numargs = 1;
1100 args.out_args[0].size = sizeof(outarg);
1101 args.out_args[0].value = &outarg;
1102 err = fuse_simple_request(fm, &args);
1104 if (fuse_invalid_attr(&outarg.attr) ||
1105 inode_wrong_type(inode, outarg.attr.mode)) {
1106 fuse_make_bad(inode);
1109 fuse_change_attributes(inode, &outarg.attr,
1110 attr_timeout(&outarg),
1113 fuse_fillattr(inode, &outarg.attr, stat);
1119 static int fuse_update_get_attr(struct inode *inode, struct file *file,
1120 struct kstat *stat, u32 request_mask,
1123 struct fuse_inode *fi = get_fuse_inode(inode);
1126 u32 inval_mask = READ_ONCE(fi->inval_mask);
1127 u32 cache_mask = fuse_get_cache_mask(inode);
1129 if (flags & AT_STATX_FORCE_SYNC)
1131 else if (flags & AT_STATX_DONT_SYNC)
1133 else if (request_mask & inval_mask & ~cache_mask)
1136 sync = time_before64(fi->i_time, get_jiffies_64());
1139 forget_all_cached_acls(inode);
1140 err = fuse_do_getattr(inode, stat, file);
1142 generic_fillattr(&init_user_ns, inode, stat);
1143 stat->mode = fi->orig_i_mode;
1144 stat->ino = fi->orig_ino;
1150 int fuse_update_attributes(struct inode *inode, struct file *file, u32 mask)
1152 return fuse_update_get_attr(inode, file, NULL, mask, 0);
1155 int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
1156 u64 child_nodeid, struct qstr *name)
1159 struct inode *parent;
1161 struct dentry *entry;
1163 parent = fuse_ilookup(fc, parent_nodeid, NULL);
1167 inode_lock_nested(parent, I_MUTEX_PARENT);
1168 if (!S_ISDIR(parent->i_mode))
1172 dir = d_find_alias(parent);
1176 name->hash = full_name_hash(dir, name->name, name->len);
1177 entry = d_lookup(dir, name);
1182 fuse_dir_changed(parent);
1183 fuse_invalidate_entry(entry);
1185 if (child_nodeid != 0 && d_really_is_positive(entry)) {
1186 inode_lock(d_inode(entry));
1187 if (get_node_id(d_inode(entry)) != child_nodeid) {
1191 if (d_mountpoint(entry)) {
1195 if (d_is_dir(entry)) {
1196 shrink_dcache_parent(entry);
1197 if (!simple_empty(entry)) {
1201 d_inode(entry)->i_flags |= S_DEAD;
1204 clear_nlink(d_inode(entry));
1207 inode_unlock(d_inode(entry));
1216 inode_unlock(parent);
1222 * Calling into a user-controlled filesystem gives the filesystem
1223 * daemon ptrace-like capabilities over the current process. This
1224 * means, that the filesystem daemon is able to record the exact
1225 * filesystem operations performed, and can also control the behavior
1226 * of the requester process in otherwise impossible ways. For example
1227 * it can delay the operation for arbitrary length of time allowing
1228 * DoS against the requester.
1230 * For this reason only those processes can call into the filesystem,
1231 * for which the owner of the mount has ptrace privilege. This
1232 * excludes processes started by other users, suid or sgid processes.
1234 int fuse_allow_current_process(struct fuse_conn *fc)
1236 const struct cred *cred;
1238 if (allow_sys_admin_access && capable(CAP_SYS_ADMIN))
1241 if (fc->allow_other)
1242 return current_in_userns(fc->user_ns);
1244 cred = current_cred();
1245 if (uid_eq(cred->euid, fc->user_id) &&
1246 uid_eq(cred->suid, fc->user_id) &&
1247 uid_eq(cred->uid, fc->user_id) &&
1248 gid_eq(cred->egid, fc->group_id) &&
1249 gid_eq(cred->sgid, fc->group_id) &&
1250 gid_eq(cred->gid, fc->group_id))
1256 static int fuse_access(struct inode *inode, int mask)
1258 struct fuse_mount *fm = get_fuse_mount(inode);
1260 struct fuse_access_in inarg;
1263 BUG_ON(mask & MAY_NOT_BLOCK);
1265 if (fm->fc->no_access)
1268 memset(&inarg, 0, sizeof(inarg));
1269 inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1270 args.opcode = FUSE_ACCESS;
1271 args.nodeid = get_node_id(inode);
1272 args.in_numargs = 1;
1273 args.in_args[0].size = sizeof(inarg);
1274 args.in_args[0].value = &inarg;
1275 err = fuse_simple_request(fm, &args);
1276 if (err == -ENOSYS) {
1277 fm->fc->no_access = 1;
1283 static int fuse_perm_getattr(struct inode *inode, int mask)
1285 if (mask & MAY_NOT_BLOCK)
1288 forget_all_cached_acls(inode);
1289 return fuse_do_getattr(inode, NULL, NULL);
1293 * Check permission. The two basic access models of FUSE are:
1295 * 1) Local access checking ('default_permissions' mount option) based
1296 * on file mode. This is the plain old disk filesystem permission
1299 * 2) "Remote" access checking, where server is responsible for
1300 * checking permission in each inode operation. An exception to this
1301 * is if ->permission() was invoked from sys_access() in which case an
1302 * access request is sent. Execute permission is still checked
1303 * locally based on file mode.
1305 static int fuse_permission(struct user_namespace *mnt_userns,
1306 struct inode *inode, int mask)
1308 struct fuse_conn *fc = get_fuse_conn(inode);
1309 bool refreshed = false;
1312 if (fuse_is_bad(inode))
1315 if (!fuse_allow_current_process(fc))
1319 * If attributes are needed, refresh them before proceeding
1321 if (fc->default_permissions ||
1322 ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1323 struct fuse_inode *fi = get_fuse_inode(inode);
1324 u32 perm_mask = STATX_MODE | STATX_UID | STATX_GID;
1326 if (perm_mask & READ_ONCE(fi->inval_mask) ||
1327 time_before64(fi->i_time, get_jiffies_64())) {
1330 err = fuse_perm_getattr(inode, mask);
1336 if (fc->default_permissions) {
1337 err = generic_permission(&init_user_ns, inode, mask);
1339 /* If permission is denied, try to refresh file
1340 attributes. This is also needed, because the root
1341 node will at first have no permissions */
1342 if (err == -EACCES && !refreshed) {
1343 err = fuse_perm_getattr(inode, mask);
1345 err = generic_permission(&init_user_ns,
1349 /* Note: the opposite of the above test does not
1350 exist. So if permissions are revoked this won't be
1351 noticed immediately, only after the attribute
1352 timeout has expired */
1353 } else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1354 err = fuse_access(inode, mask);
1355 } else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1356 if (!(inode->i_mode & S_IXUGO)) {
1360 err = fuse_perm_getattr(inode, mask);
1361 if (!err && !(inode->i_mode & S_IXUGO))
1368 static int fuse_readlink_page(struct inode *inode, struct page *page)
1370 struct fuse_mount *fm = get_fuse_mount(inode);
1371 struct fuse_page_desc desc = { .length = PAGE_SIZE - 1 };
1372 struct fuse_args_pages ap = {
1380 ap.args.opcode = FUSE_READLINK;
1381 ap.args.nodeid = get_node_id(inode);
1382 ap.args.out_pages = true;
1383 ap.args.out_argvar = true;
1384 ap.args.page_zeroing = true;
1385 ap.args.out_numargs = 1;
1386 ap.args.out_args[0].size = desc.length;
1387 res = fuse_simple_request(fm, &ap.args);
1389 fuse_invalidate_atime(inode);
1394 if (WARN_ON(res >= PAGE_SIZE))
1397 link = page_address(page);
1403 static const char *fuse_get_link(struct dentry *dentry, struct inode *inode,
1404 struct delayed_call *callback)
1406 struct fuse_conn *fc = get_fuse_conn(inode);
1411 if (fuse_is_bad(inode))
1414 if (fc->cache_symlinks)
1415 return page_get_link(dentry, inode, callback);
1421 page = alloc_page(GFP_KERNEL);
1426 err = fuse_readlink_page(inode, page);
1432 set_delayed_call(callback, page_put_link, page);
1434 return page_address(page);
1437 return ERR_PTR(err);
1440 static int fuse_dir_open(struct inode *inode, struct file *file)
1442 return fuse_open_common(inode, file, true);
1445 static int fuse_dir_release(struct inode *inode, struct file *file)
1447 fuse_release_common(file, true);
1452 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1455 struct inode *inode = file->f_mapping->host;
1456 struct fuse_conn *fc = get_fuse_conn(inode);
1459 if (fuse_is_bad(inode))
1462 if (fc->no_fsyncdir)
1466 err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNCDIR);
1467 if (err == -ENOSYS) {
1468 fc->no_fsyncdir = 1;
1471 inode_unlock(inode);
1476 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1479 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1481 /* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1485 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1488 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1491 struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1496 return fuse_ioctl_common(file, cmd, arg,
1497 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1500 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1502 /* Always update if mtime is explicitly set */
1503 if (ivalid & ATTR_MTIME_SET)
1506 /* Or if kernel i_mtime is the official one */
1507 if (trust_local_mtime)
1510 /* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1511 if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1514 /* In all other cases update */
1518 static void iattr_to_fattr(struct fuse_conn *fc, struct iattr *iattr,
1519 struct fuse_setattr_in *arg, bool trust_local_cmtime)
1521 unsigned ivalid = iattr->ia_valid;
1523 if (ivalid & ATTR_MODE)
1524 arg->valid |= FATTR_MODE, arg->mode = iattr->ia_mode;
1525 if (ivalid & ATTR_UID)
1526 arg->valid |= FATTR_UID, arg->uid = from_kuid(fc->user_ns, iattr->ia_uid);
1527 if (ivalid & ATTR_GID)
1528 arg->valid |= FATTR_GID, arg->gid = from_kgid(fc->user_ns, iattr->ia_gid);
1529 if (ivalid & ATTR_SIZE)
1530 arg->valid |= FATTR_SIZE, arg->size = iattr->ia_size;
1531 if (ivalid & ATTR_ATIME) {
1532 arg->valid |= FATTR_ATIME;
1533 arg->atime = iattr->ia_atime.tv_sec;
1534 arg->atimensec = iattr->ia_atime.tv_nsec;
1535 if (!(ivalid & ATTR_ATIME_SET))
1536 arg->valid |= FATTR_ATIME_NOW;
1538 if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1539 arg->valid |= FATTR_MTIME;
1540 arg->mtime = iattr->ia_mtime.tv_sec;
1541 arg->mtimensec = iattr->ia_mtime.tv_nsec;
1542 if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1543 arg->valid |= FATTR_MTIME_NOW;
1545 if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1546 arg->valid |= FATTR_CTIME;
1547 arg->ctime = iattr->ia_ctime.tv_sec;
1548 arg->ctimensec = iattr->ia_ctime.tv_nsec;
1553 * Prevent concurrent writepages on inode
1555 * This is done by adding a negative bias to the inode write counter
1556 * and waiting for all pending writes to finish.
1558 void fuse_set_nowrite(struct inode *inode)
1560 struct fuse_inode *fi = get_fuse_inode(inode);
1562 BUG_ON(!inode_is_locked(inode));
1564 spin_lock(&fi->lock);
1565 BUG_ON(fi->writectr < 0);
1566 fi->writectr += FUSE_NOWRITE;
1567 spin_unlock(&fi->lock);
1568 wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1572 * Allow writepages on inode
1574 * Remove the bias from the writecounter and send any queued
1577 static void __fuse_release_nowrite(struct inode *inode)
1579 struct fuse_inode *fi = get_fuse_inode(inode);
1581 BUG_ON(fi->writectr != FUSE_NOWRITE);
1583 fuse_flush_writepages(inode);
1586 void fuse_release_nowrite(struct inode *inode)
1588 struct fuse_inode *fi = get_fuse_inode(inode);
1590 spin_lock(&fi->lock);
1591 __fuse_release_nowrite(inode);
1592 spin_unlock(&fi->lock);
1595 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1596 struct inode *inode,
1597 struct fuse_setattr_in *inarg_p,
1598 struct fuse_attr_out *outarg_p)
1600 args->opcode = FUSE_SETATTR;
1601 args->nodeid = get_node_id(inode);
1602 args->in_numargs = 1;
1603 args->in_args[0].size = sizeof(*inarg_p);
1604 args->in_args[0].value = inarg_p;
1605 args->out_numargs = 1;
1606 args->out_args[0].size = sizeof(*outarg_p);
1607 args->out_args[0].value = outarg_p;
1611 * Flush inode->i_mtime to the server
1613 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1615 struct fuse_mount *fm = get_fuse_mount(inode);
1617 struct fuse_setattr_in inarg;
1618 struct fuse_attr_out outarg;
1620 memset(&inarg, 0, sizeof(inarg));
1621 memset(&outarg, 0, sizeof(outarg));
1623 inarg.valid = FATTR_MTIME;
1624 inarg.mtime = inode->i_mtime.tv_sec;
1625 inarg.mtimensec = inode->i_mtime.tv_nsec;
1626 if (fm->fc->minor >= 23) {
1627 inarg.valid |= FATTR_CTIME;
1628 inarg.ctime = inode->i_ctime.tv_sec;
1629 inarg.ctimensec = inode->i_ctime.tv_nsec;
1632 inarg.valid |= FATTR_FH;
1635 fuse_setattr_fill(fm->fc, &args, inode, &inarg, &outarg);
1637 return fuse_simple_request(fm, &args);
1641 * Set attributes, and at the same time refresh them.
1643 * Truncation is slightly complicated, because the 'truncate' request
1644 * may fail, in which case we don't want to touch the mapping.
1645 * vmtruncate() doesn't allow for this case, so do the rlimit checking
1646 * and the actual truncation by hand.
1648 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1651 struct inode *inode = d_inode(dentry);
1652 struct fuse_mount *fm = get_fuse_mount(inode);
1653 struct fuse_conn *fc = fm->fc;
1654 struct fuse_inode *fi = get_fuse_inode(inode);
1655 struct address_space *mapping = inode->i_mapping;
1657 struct fuse_setattr_in inarg;
1658 struct fuse_attr_out outarg;
1659 bool is_truncate = false;
1660 bool is_wb = fc->writeback_cache && S_ISREG(inode->i_mode);
1663 bool trust_local_cmtime = is_wb;
1664 bool fault_blocked = false;
1666 if (!fc->default_permissions)
1667 attr->ia_valid |= ATTR_FORCE;
1669 err = setattr_prepare(&init_user_ns, dentry, attr);
1673 if (attr->ia_valid & ATTR_SIZE) {
1674 if (WARN_ON(!S_ISREG(inode->i_mode)))
1679 if (FUSE_IS_DAX(inode) && is_truncate) {
1680 filemap_invalidate_lock(mapping);
1681 fault_blocked = true;
1682 err = fuse_dax_break_layouts(inode, 0, 0);
1684 filemap_invalidate_unlock(mapping);
1689 if (attr->ia_valid & ATTR_OPEN) {
1690 /* This is coming from open(..., ... | O_TRUNC); */
1691 WARN_ON(!(attr->ia_valid & ATTR_SIZE));
1692 WARN_ON(attr->ia_size != 0);
1693 if (fc->atomic_o_trunc) {
1695 * No need to send request to userspace, since actual
1696 * truncation has already been done by OPEN. But still
1697 * need to truncate page cache.
1699 i_size_write(inode, 0);
1700 truncate_pagecache(inode, 0);
1706 /* Flush dirty data/metadata before non-truncate SETATTR */
1709 (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_MTIME_SET |
1711 err = write_inode_now(inode, true);
1715 fuse_set_nowrite(inode);
1716 fuse_release_nowrite(inode);
1720 fuse_set_nowrite(inode);
1721 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1722 if (trust_local_cmtime && attr->ia_size != inode->i_size)
1723 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1726 memset(&inarg, 0, sizeof(inarg));
1727 memset(&outarg, 0, sizeof(outarg));
1728 iattr_to_fattr(fc, attr, &inarg, trust_local_cmtime);
1730 struct fuse_file *ff = file->private_data;
1731 inarg.valid |= FATTR_FH;
1735 /* Kill suid/sgid for non-directory chown unconditionally */
1736 if (fc->handle_killpriv_v2 && !S_ISDIR(inode->i_mode) &&
1737 attr->ia_valid & (ATTR_UID | ATTR_GID))
1738 inarg.valid |= FATTR_KILL_SUIDGID;
1740 if (attr->ia_valid & ATTR_SIZE) {
1741 /* For mandatory locking in truncate */
1742 inarg.valid |= FATTR_LOCKOWNER;
1743 inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1745 /* Kill suid/sgid for truncate only if no CAP_FSETID */
1746 if (fc->handle_killpriv_v2 && !capable(CAP_FSETID))
1747 inarg.valid |= FATTR_KILL_SUIDGID;
1749 fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1750 err = fuse_simple_request(fm, &args);
1753 fuse_invalidate_attr(inode);
1757 if (fuse_invalid_attr(&outarg.attr) ||
1758 inode_wrong_type(inode, outarg.attr.mode)) {
1759 fuse_make_bad(inode);
1764 spin_lock(&fi->lock);
1765 /* the kernel maintains i_mtime locally */
1766 if (trust_local_cmtime) {
1767 if (attr->ia_valid & ATTR_MTIME)
1768 inode->i_mtime = attr->ia_mtime;
1769 if (attr->ia_valid & ATTR_CTIME)
1770 inode->i_ctime = attr->ia_ctime;
1771 /* FIXME: clear I_DIRTY_SYNC? */
1774 fuse_change_attributes_common(inode, &outarg.attr,
1775 attr_timeout(&outarg),
1776 fuse_get_cache_mask(inode));
1777 oldsize = inode->i_size;
1778 /* see the comment in fuse_change_attributes() */
1779 if (!is_wb || is_truncate)
1780 i_size_write(inode, outarg.attr.size);
1783 /* NOTE: this may release/reacquire fi->lock */
1784 __fuse_release_nowrite(inode);
1786 spin_unlock(&fi->lock);
1789 * Only call invalidate_inode_pages2() after removing
1790 * FUSE_NOWRITE, otherwise fuse_launder_folio() would deadlock.
1792 if ((is_truncate || !is_wb) &&
1793 S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1794 truncate_pagecache(inode, outarg.attr.size);
1795 invalidate_inode_pages2(mapping);
1798 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1801 filemap_invalidate_unlock(mapping);
1807 fuse_release_nowrite(inode);
1809 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1812 filemap_invalidate_unlock(mapping);
1816 static int fuse_setattr(struct user_namespace *mnt_userns, struct dentry *entry,
1819 struct inode *inode = d_inode(entry);
1820 struct fuse_conn *fc = get_fuse_conn(inode);
1821 struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1824 if (fuse_is_bad(inode))
1827 if (!fuse_allow_current_process(get_fuse_conn(inode)))
1830 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1831 attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1835 * The only sane way to reliably kill suid/sgid is to do it in
1836 * the userspace filesystem
1838 * This should be done on write(), truncate() and chown().
1840 if (!fc->handle_killpriv && !fc->handle_killpriv_v2) {
1842 * ia_mode calculation may have used stale i_mode.
1843 * Refresh and recalculate.
1845 ret = fuse_do_getattr(inode, NULL, file);
1849 attr->ia_mode = inode->i_mode;
1850 if (inode->i_mode & S_ISUID) {
1851 attr->ia_valid |= ATTR_MODE;
1852 attr->ia_mode &= ~S_ISUID;
1854 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1855 attr->ia_valid |= ATTR_MODE;
1856 attr->ia_mode &= ~S_ISGID;
1860 if (!attr->ia_valid)
1863 ret = fuse_do_setattr(entry, attr, file);
1866 * If filesystem supports acls it may have updated acl xattrs in
1867 * the filesystem, so forget cached acls for the inode.
1870 forget_all_cached_acls(inode);
1872 /* Directory mode changed, may need to revalidate access */
1873 if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1874 fuse_invalidate_entry_cache(entry);
1879 static int fuse_getattr(struct user_namespace *mnt_userns,
1880 const struct path *path, struct kstat *stat,
1881 u32 request_mask, unsigned int flags)
1883 struct inode *inode = d_inode(path->dentry);
1884 struct fuse_conn *fc = get_fuse_conn(inode);
1886 if (fuse_is_bad(inode))
1889 if (!fuse_allow_current_process(fc)) {
1890 if (!request_mask) {
1892 * If user explicitly requested *nothing* then don't
1893 * error out, but return st_dev only.
1895 stat->result_mask = 0;
1896 stat->dev = inode->i_sb->s_dev;
1902 return fuse_update_get_attr(inode, NULL, stat, request_mask, flags);
1905 static const struct inode_operations fuse_dir_inode_operations = {
1906 .lookup = fuse_lookup,
1907 .mkdir = fuse_mkdir,
1908 .symlink = fuse_symlink,
1909 .unlink = fuse_unlink,
1910 .rmdir = fuse_rmdir,
1911 .rename = fuse_rename2,
1913 .setattr = fuse_setattr,
1914 .create = fuse_create,
1915 .atomic_open = fuse_atomic_open,
1916 .mknod = fuse_mknod,
1917 .permission = fuse_permission,
1918 .getattr = fuse_getattr,
1919 .listxattr = fuse_listxattr,
1920 .get_acl = fuse_get_acl,
1921 .set_acl = fuse_set_acl,
1922 .fileattr_get = fuse_fileattr_get,
1923 .fileattr_set = fuse_fileattr_set,
1926 static const struct file_operations fuse_dir_operations = {
1927 .llseek = generic_file_llseek,
1928 .read = generic_read_dir,
1929 .iterate_shared = fuse_readdir,
1930 .open = fuse_dir_open,
1931 .release = fuse_dir_release,
1932 .fsync = fuse_dir_fsync,
1933 .unlocked_ioctl = fuse_dir_ioctl,
1934 .compat_ioctl = fuse_dir_compat_ioctl,
1937 static const struct inode_operations fuse_common_inode_operations = {
1938 .setattr = fuse_setattr,
1939 .permission = fuse_permission,
1940 .getattr = fuse_getattr,
1941 .listxattr = fuse_listxattr,
1942 .get_acl = fuse_get_acl,
1943 .set_acl = fuse_set_acl,
1944 .fileattr_get = fuse_fileattr_get,
1945 .fileattr_set = fuse_fileattr_set,
1948 static const struct inode_operations fuse_symlink_inode_operations = {
1949 .setattr = fuse_setattr,
1950 .get_link = fuse_get_link,
1951 .getattr = fuse_getattr,
1952 .listxattr = fuse_listxattr,
1955 void fuse_init_common(struct inode *inode)
1957 inode->i_op = &fuse_common_inode_operations;
1960 void fuse_init_dir(struct inode *inode)
1962 struct fuse_inode *fi = get_fuse_inode(inode);
1964 inode->i_op = &fuse_dir_inode_operations;
1965 inode->i_fop = &fuse_dir_operations;
1967 spin_lock_init(&fi->rdc.lock);
1968 fi->rdc.cached = false;
1971 fi->rdc.version = 0;
1974 static int fuse_symlink_read_folio(struct file *null, struct folio *folio)
1976 int err = fuse_readlink_page(folio->mapping->host, &folio->page);
1979 folio_mark_uptodate(folio);
1981 folio_unlock(folio);
1986 static const struct address_space_operations fuse_symlink_aops = {
1987 .read_folio = fuse_symlink_read_folio,
1990 void fuse_init_symlink(struct inode *inode)
1992 inode->i_op = &fuse_symlink_inode_operations;
1993 inode->i_data.a_ops = &fuse_symlink_aops;
1994 inode_nohighmem(inode);