2 * memfd_create system call and file sealing support
4 * Code was originally included in shmem.c, and broken out to facilitate
5 * use by hugetlbfs as well as tmpfs.
7 * This file is released under the GPL.
11 #include <linux/vfs.h>
12 #include <linux/pagemap.h>
13 #include <linux/file.h>
15 #include <linux/sched/signal.h>
16 #include <linux/khugepaged.h>
17 #include <linux/syscalls.h>
18 #include <linux/hugetlb.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/memfd.h>
21 #include <uapi/linux/memfd.h>
24 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
25 * so reuse a tag which we firmly believe is never set or cleared on tmpfs
26 * or hugetlbfs because they are memory only filesystems.
28 #define MEMFD_TAG_PINNED PAGECACHE_TAG_TOWRITE
29 #define LAST_SCAN 4 /* about 150ms max */
31 static void memfd_tag_pins(struct address_space *mapping)
33 struct radix_tree_iter iter;
37 unsigned int tagged = 0;
42 xa_lock_irq(&mapping->i_pages);
43 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
44 page = radix_tree_deref_slot_protected(slot, &mapping->i_pages.xa_lock);
45 if (!page || radix_tree_exception(page)) {
46 if (radix_tree_deref_retry(page)) {
47 slot = radix_tree_iter_retry(&iter);
50 } else if (page_count(page) - page_mapcount(page) > 1) {
51 radix_tree_tag_set(&mapping->i_pages, iter.index,
58 slot = radix_tree_iter_resume(slot, &iter);
59 xa_unlock_irq(&mapping->i_pages);
61 xa_lock_irq(&mapping->i_pages);
63 xa_unlock_irq(&mapping->i_pages);
67 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
68 * via get_user_pages(), drivers might have some pending I/O without any active
69 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
70 * and see whether it has an elevated ref-count. If so, we tag them and wait for
72 * The caller must guarantee that no new user will acquire writable references
73 * to those pages to avoid races.
75 static int memfd_wait_for_pins(struct address_space *mapping)
77 struct radix_tree_iter iter;
83 memfd_tag_pins(mapping);
86 for (scan = 0; scan <= LAST_SCAN; scan++) {
87 if (!radix_tree_tagged(&mapping->i_pages, MEMFD_TAG_PINNED))
92 else if (schedule_timeout_killable((HZ << scan) / 200))
97 radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter,
98 start, MEMFD_TAG_PINNED) {
100 page = radix_tree_deref_slot(slot);
101 if (radix_tree_exception(page)) {
102 if (radix_tree_deref_retry(page)) {
103 slot = radix_tree_iter_retry(&iter);
111 page_count(page) - page_mapcount(page) != 1) {
112 if (scan < LAST_SCAN)
113 goto continue_resched;
116 * On the last scan, we clean up all those tags
117 * we inserted; but make a note that we still
118 * found pages pinned.
123 xa_lock_irq(&mapping->i_pages);
124 radix_tree_tag_clear(&mapping->i_pages,
125 iter.index, MEMFD_TAG_PINNED);
126 xa_unlock_irq(&mapping->i_pages);
128 if (need_resched()) {
129 slot = radix_tree_iter_resume(slot, &iter);
139 static unsigned int *memfd_file_seals_ptr(struct file *file)
141 if (shmem_file(file))
142 return &SHMEM_I(file_inode(file))->seals;
144 #ifdef CONFIG_HUGETLBFS
145 if (is_file_hugepages(file))
146 return &HUGETLBFS_I(file_inode(file))->seals;
152 #define F_ALL_SEALS (F_SEAL_SEAL | \
157 static int memfd_add_seals(struct file *file, unsigned int seals)
159 struct inode *inode = file_inode(file);
160 unsigned int *file_seals;
165 * Sealing allows multiple parties to share a tmpfs or hugetlbfs file
166 * but restrict access to a specific subset of file operations. Seals
167 * can only be added, but never removed. This way, mutually untrusted
168 * parties can share common memory regions with a well-defined policy.
169 * A malicious peer can thus never perform unwanted operations on a
172 * Seals are only supported on special tmpfs or hugetlbfs files and
173 * always affect the whole underlying inode. Once a seal is set, it
174 * may prevent some kinds of access to the file. Currently, the
175 * following seals are defined:
176 * SEAL_SEAL: Prevent further seals from being set on this file
177 * SEAL_SHRINK: Prevent the file from shrinking
178 * SEAL_GROW: Prevent the file from growing
179 * SEAL_WRITE: Prevent write access to the file
181 * As we don't require any trust relationship between two parties, we
182 * must prevent seals from being removed. Therefore, sealing a file
183 * only adds a given set of seals to the file, it never touches
184 * existing seals. Furthermore, the "setting seals"-operation can be
185 * sealed itself, which basically prevents any further seal from being
188 * Semantics of sealing are only defined on volatile files. Only
189 * anonymous tmpfs and hugetlbfs files support sealing. More
190 * importantly, seals are never written to disk. Therefore, there's
191 * no plan to support it on other file types.
194 if (!(file->f_mode & FMODE_WRITE))
196 if (seals & ~(unsigned int)F_ALL_SEALS)
201 file_seals = memfd_file_seals_ptr(file);
207 if (*file_seals & F_SEAL_SEAL) {
212 if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) {
213 error = mapping_deny_writable(file->f_mapping);
217 error = memfd_wait_for_pins(file->f_mapping);
219 mapping_allow_writable(file->f_mapping);
224 *file_seals |= seals;
232 static int memfd_get_seals(struct file *file)
234 unsigned int *seals = memfd_file_seals_ptr(file);
236 return seals ? *seals : -EINVAL;
239 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
245 /* disallow upper 32bit */
249 error = memfd_add_seals(file, arg);
252 error = memfd_get_seals(file);
262 #define MFD_NAME_PREFIX "memfd:"
263 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
264 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
266 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
268 SYSCALL_DEFINE2(memfd_create,
269 const char __user *, uname,
272 unsigned int *file_seals;
278 if (!(flags & MFD_HUGETLB)) {
279 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
282 /* Allow huge page size encoding in flags. */
283 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
284 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
288 /* length includes terminating zero */
289 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
292 if (len > MFD_NAME_MAX_LEN + 1)
295 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
299 strcpy(name, MFD_NAME_PREFIX);
300 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
305 /* terminating-zero may have changed after strnlen_user() returned */
306 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
311 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
317 if (flags & MFD_HUGETLB) {
318 struct user_struct *user = NULL;
320 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
321 HUGETLB_ANONHUGE_INODE,
322 (flags >> MFD_HUGE_SHIFT) &
325 file = shmem_file_setup(name, 0, VM_NORESERVE);
327 error = PTR_ERR(file);
330 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
331 file->f_flags |= O_LARGEFILE;
333 if (flags & MFD_ALLOW_SEALING) {
334 file_seals = memfd_file_seals_ptr(file);
335 *file_seals &= ~F_SEAL_SEAL;
338 fd_install(fd, file);