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;
42 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
43 page = radix_tree_deref_slot(slot);
44 if (!page || radix_tree_exception(page)) {
45 if (radix_tree_deref_retry(page)) {
46 slot = radix_tree_iter_retry(&iter);
49 } else if (page_count(page) - page_mapcount(page) > 1) {
50 xa_lock_irq(&mapping->i_pages);
51 radix_tree_tag_set(&mapping->i_pages, iter.index,
53 xa_unlock_irq(&mapping->i_pages);
57 slot = radix_tree_iter_resume(slot, &iter);
65 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
66 * via get_user_pages(), drivers might have some pending I/O without any active
67 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
68 * and see whether it has an elevated ref-count. If so, we tag them and wait for
70 * The caller must guarantee that no new user will acquire writable references
71 * to those pages to avoid races.
73 static int memfd_wait_for_pins(struct address_space *mapping)
75 struct radix_tree_iter iter;
81 memfd_tag_pins(mapping);
84 for (scan = 0; scan <= LAST_SCAN; scan++) {
85 if (!radix_tree_tagged(&mapping->i_pages, MEMFD_TAG_PINNED))
90 else if (schedule_timeout_killable((HZ << scan) / 200))
95 radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter,
96 start, MEMFD_TAG_PINNED) {
98 page = radix_tree_deref_slot(slot);
99 if (radix_tree_exception(page)) {
100 if (radix_tree_deref_retry(page)) {
101 slot = radix_tree_iter_retry(&iter);
109 page_count(page) - page_mapcount(page) != 1) {
110 if (scan < LAST_SCAN)
111 goto continue_resched;
114 * On the last scan, we clean up all those tags
115 * we inserted; but make a note that we still
116 * found pages pinned.
121 xa_lock_irq(&mapping->i_pages);
122 radix_tree_tag_clear(&mapping->i_pages,
123 iter.index, MEMFD_TAG_PINNED);
124 xa_unlock_irq(&mapping->i_pages);
126 if (need_resched()) {
127 slot = radix_tree_iter_resume(slot, &iter);
137 static unsigned int *memfd_file_seals_ptr(struct file *file)
139 if (shmem_file(file))
140 return &SHMEM_I(file_inode(file))->seals;
142 #ifdef CONFIG_HUGETLBFS
143 if (is_file_hugepages(file))
144 return &HUGETLBFS_I(file_inode(file))->seals;
150 #define F_ALL_SEALS (F_SEAL_SEAL | \
155 static int memfd_add_seals(struct file *file, unsigned int seals)
157 struct inode *inode = file_inode(file);
158 unsigned int *file_seals;
163 * Sealing allows multiple parties to share a tmpfs or hugetlbfs file
164 * but restrict access to a specific subset of file operations. Seals
165 * can only be added, but never removed. This way, mutually untrusted
166 * parties can share common memory regions with a well-defined policy.
167 * A malicious peer can thus never perform unwanted operations on a
170 * Seals are only supported on special tmpfs or hugetlbfs files and
171 * always affect the whole underlying inode. Once a seal is set, it
172 * may prevent some kinds of access to the file. Currently, the
173 * following seals are defined:
174 * SEAL_SEAL: Prevent further seals from being set on this file
175 * SEAL_SHRINK: Prevent the file from shrinking
176 * SEAL_GROW: Prevent the file from growing
177 * SEAL_WRITE: Prevent write access to the file
179 * As we don't require any trust relationship between two parties, we
180 * must prevent seals from being removed. Therefore, sealing a file
181 * only adds a given set of seals to the file, it never touches
182 * existing seals. Furthermore, the "setting seals"-operation can be
183 * sealed itself, which basically prevents any further seal from being
186 * Semantics of sealing are only defined on volatile files. Only
187 * anonymous tmpfs and hugetlbfs files support sealing. More
188 * importantly, seals are never written to disk. Therefore, there's
189 * no plan to support it on other file types.
192 if (!(file->f_mode & FMODE_WRITE))
194 if (seals & ~(unsigned int)F_ALL_SEALS)
199 file_seals = memfd_file_seals_ptr(file);
205 if (*file_seals & F_SEAL_SEAL) {
210 if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) {
211 error = mapping_deny_writable(file->f_mapping);
215 error = memfd_wait_for_pins(file->f_mapping);
217 mapping_allow_writable(file->f_mapping);
222 *file_seals |= seals;
230 static int memfd_get_seals(struct file *file)
232 unsigned int *seals = memfd_file_seals_ptr(file);
234 return seals ? *seals : -EINVAL;
237 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
243 /* disallow upper 32bit */
247 error = memfd_add_seals(file, arg);
250 error = memfd_get_seals(file);
260 #define MFD_NAME_PREFIX "memfd:"
261 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
262 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
264 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
266 SYSCALL_DEFINE2(memfd_create,
267 const char __user *, uname,
270 unsigned int *file_seals;
276 if (!(flags & MFD_HUGETLB)) {
277 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
280 /* Allow huge page size encoding in flags. */
281 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
282 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
286 /* length includes terminating zero */
287 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
290 if (len > MFD_NAME_MAX_LEN + 1)
293 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
297 strcpy(name, MFD_NAME_PREFIX);
298 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
303 /* terminating-zero may have changed after strnlen_user() returned */
304 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
309 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
315 if (flags & MFD_HUGETLB) {
316 struct user_struct *user = NULL;
318 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
319 HUGETLB_ANONHUGE_INODE,
320 (flags >> MFD_HUGE_SHIFT) &
323 file = shmem_file_setup(name, 0, VM_NORESERVE);
325 error = PTR_ERR(file);
328 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
329 file->f_flags |= O_RDWR | O_LARGEFILE;
331 if (flags & MFD_ALLOW_SEALING) {
332 file_seals = memfd_file_seals_ptr(file);
333 *file_seals &= ~F_SEAL_SEAL;
336 fd_install(fd, file);