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 <linux/pid_namespace.h>
22 #include <uapi/linux/memfd.h>
26 * We need a tag: a new tag would expand every xa_node by 8 bytes,
27 * so reuse a tag which we firmly believe is never set or cleared on tmpfs
28 * or hugetlbfs because they are memory only filesystems.
30 #define MEMFD_TAG_PINNED PAGECACHE_TAG_TOWRITE
31 #define LAST_SCAN 4 /* about 150ms max */
33 static bool memfd_folio_has_extra_refs(struct folio *folio)
35 return folio_ref_count(folio) - folio_mapcount(folio) !=
36 folio_nr_pages(folio);
39 static void memfd_tag_pins(struct xa_state *xas)
47 xas_for_each(xas, folio, ULONG_MAX) {
48 if (!xa_is_value(folio) && memfd_folio_has_extra_refs(folio))
49 xas_set_mark(xas, MEMFD_TAG_PINNED);
51 if (++latency < XA_CHECK_SCHED)
64 * This is a helper function used by memfd_pin_user_pages() in GUP (gup.c).
65 * It is mainly called to allocate a folio in a memfd when the caller
66 * (memfd_pin_folios()) cannot find a folio in the page cache at a given
67 * index in the mapping.
69 struct folio *memfd_alloc_folio(struct file *memfd, pgoff_t idx)
71 #ifdef CONFIG_HUGETLB_PAGE
76 if (is_file_hugepages(memfd)) {
78 * The folio would most likely be accessed by a DMA driver,
79 * therefore, we have zone memory constraints where we can
80 * alloc from. Also, the folio will be pinned for an indefinite
81 * amount of time, so it is not expected to be migrated away.
83 struct hstate *h = hstate_file(memfd);
85 gfp_mask = htlb_alloc_mask(h);
86 gfp_mask &= ~(__GFP_HIGHMEM | __GFP_MOVABLE);
87 idx >>= huge_page_order(h);
89 folio = alloc_hugetlb_folio_reserve(h,
94 err = hugetlb_add_to_page_cache(folio,
104 return ERR_PTR(-ENOMEM);
107 return shmem_read_folio(memfd->f_mapping, idx);
111 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
112 * via get_user_pages(), drivers might have some pending I/O without any active
113 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all folios
114 * and see whether it has an elevated ref-count. If so, we tag them and wait for
115 * them to be dropped.
116 * The caller must guarantee that no new user will acquire writable references
117 * to those folios to avoid races.
119 static int memfd_wait_for_pins(struct address_space *mapping)
121 XA_STATE(xas, &mapping->i_pages, 0);
125 memfd_tag_pins(&xas);
128 for (scan = 0; scan <= LAST_SCAN; scan++) {
131 if (!xas_marked(&xas, MEMFD_TAG_PINNED))
136 else if (schedule_timeout_killable((HZ << scan) / 200))
141 xas_for_each_marked(&xas, folio, ULONG_MAX, MEMFD_TAG_PINNED) {
144 if (!xa_is_value(folio) &&
145 memfd_folio_has_extra_refs(folio)) {
147 * On the last scan, we clean up all those tags
148 * we inserted; but make a note that we still
149 * found folios pinned.
151 if (scan == LAST_SCAN)
157 xas_clear_mark(&xas, MEMFD_TAG_PINNED);
159 if (++latency < XA_CHECK_SCHED)
164 xas_unlock_irq(&xas);
168 xas_unlock_irq(&xas);
174 static unsigned int *memfd_file_seals_ptr(struct file *file)
176 if (shmem_file(file))
177 return &SHMEM_I(file_inode(file))->seals;
179 #ifdef CONFIG_HUGETLBFS
180 if (is_file_hugepages(file))
181 return &HUGETLBFS_I(file_inode(file))->seals;
187 #define F_ALL_SEALS (F_SEAL_SEAL | \
194 static int memfd_add_seals(struct file *file, unsigned int seals)
196 struct inode *inode = file_inode(file);
197 unsigned int *file_seals;
202 * Sealing allows multiple parties to share a tmpfs or hugetlbfs file
203 * but restrict access to a specific subset of file operations. Seals
204 * can only be added, but never removed. This way, mutually untrusted
205 * parties can share common memory regions with a well-defined policy.
206 * A malicious peer can thus never perform unwanted operations on a
209 * Seals are only supported on special tmpfs or hugetlbfs files and
210 * always affect the whole underlying inode. Once a seal is set, it
211 * may prevent some kinds of access to the file. Currently, the
212 * following seals are defined:
213 * SEAL_SEAL: Prevent further seals from being set on this file
214 * SEAL_SHRINK: Prevent the file from shrinking
215 * SEAL_GROW: Prevent the file from growing
216 * SEAL_WRITE: Prevent write access to the file
217 * SEAL_EXEC: Prevent modification of the exec bits in the file mode
219 * As we don't require any trust relationship between two parties, we
220 * must prevent seals from being removed. Therefore, sealing a file
221 * only adds a given set of seals to the file, it never touches
222 * existing seals. Furthermore, the "setting seals"-operation can be
223 * sealed itself, which basically prevents any further seal from being
226 * Semantics of sealing are only defined on volatile files. Only
227 * anonymous tmpfs and hugetlbfs files support sealing. More
228 * importantly, seals are never written to disk. Therefore, there's
229 * no plan to support it on other file types.
232 if (!(file->f_mode & FMODE_WRITE))
234 if (seals & ~(unsigned int)F_ALL_SEALS)
239 file_seals = memfd_file_seals_ptr(file);
245 if (*file_seals & F_SEAL_SEAL) {
250 if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) {
251 error = mapping_deny_writable(file->f_mapping);
255 error = memfd_wait_for_pins(file->f_mapping);
257 mapping_allow_writable(file->f_mapping);
263 * SEAL_EXEC implies SEAL_WRITE, making W^X from the start.
265 if (seals & F_SEAL_EXEC && inode->i_mode & 0111)
266 seals |= F_SEAL_SHRINK|F_SEAL_GROW|F_SEAL_WRITE|F_SEAL_FUTURE_WRITE;
268 *file_seals |= seals;
276 static int memfd_get_seals(struct file *file)
278 unsigned int *seals = memfd_file_seals_ptr(file);
280 return seals ? *seals : -EINVAL;
283 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned int arg)
289 error = memfd_add_seals(file, arg);
292 error = memfd_get_seals(file);
302 #define MFD_NAME_PREFIX "memfd:"
303 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
304 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
306 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB | MFD_NOEXEC_SEAL | MFD_EXEC)
308 static int check_sysctl_memfd_noexec(unsigned int *flags)
311 struct pid_namespace *ns = task_active_pid_ns(current);
312 int sysctl = pidns_memfd_noexec_scope(ns);
314 if (!(*flags & (MFD_EXEC | MFD_NOEXEC_SEAL))) {
315 if (sysctl >= MEMFD_NOEXEC_SCOPE_NOEXEC_SEAL)
316 *flags |= MFD_NOEXEC_SEAL;
321 if (!(*flags & MFD_NOEXEC_SEAL) && sysctl >= MEMFD_NOEXEC_SCOPE_NOEXEC_ENFORCED) {
323 "%s[%d]: memfd_create() requires MFD_NOEXEC_SEAL with vm.memfd_noexec=%d\n",
324 current->comm, task_pid_nr(current), sysctl);
331 static inline bool is_write_sealed(unsigned int seals)
333 return seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE);
336 static int check_write_seal(unsigned long *vm_flags_ptr)
338 unsigned long vm_flags = *vm_flags_ptr;
339 unsigned long mask = vm_flags & (VM_SHARED | VM_WRITE);
341 /* If a private mapping then writability is irrelevant. */
342 if (!(mask & VM_SHARED))
346 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
347 * write seals are active.
353 * This is a read-only mapping, disallow mprotect() from making a
354 * write-sealed mapping writable in future.
356 *vm_flags_ptr &= ~VM_MAYWRITE;
361 int memfd_check_seals_mmap(struct file *file, unsigned long *vm_flags_ptr)
364 unsigned int *seals_ptr = memfd_file_seals_ptr(file);
365 unsigned int seals = seals_ptr ? *seals_ptr : 0;
367 if (is_write_sealed(seals))
368 err = check_write_seal(vm_flags_ptr);
373 static int sanitize_flags(unsigned int *flags_ptr)
375 unsigned int flags = *flags_ptr;
377 if (!(flags & MFD_HUGETLB)) {
378 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
381 /* Allow huge page size encoding in flags. */
382 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
383 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
387 /* Invalid if both EXEC and NOEXEC_SEAL are set.*/
388 if ((flags & MFD_EXEC) && (flags & MFD_NOEXEC_SEAL))
391 return check_sysctl_memfd_noexec(flags_ptr);
394 static char *alloc_name(const char __user *uname)
400 name = kmalloc(NAME_MAX + 1, GFP_KERNEL);
402 return ERR_PTR(-ENOMEM);
404 strcpy(name, MFD_NAME_PREFIX);
405 /* returned length does not include terminating zero */
406 len = strncpy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, MFD_NAME_MAX_LEN + 1);
410 } else if (len > MFD_NAME_MAX_LEN) {
419 return ERR_PTR(error);
422 static struct file *alloc_file(const char *name, unsigned int flags)
424 unsigned int *file_seals;
427 if (flags & MFD_HUGETLB) {
428 file = hugetlb_file_setup(name, 0, VM_NORESERVE,
429 HUGETLB_ANONHUGE_INODE,
430 (flags >> MFD_HUGE_SHIFT) &
433 file = shmem_file_setup(name, 0, VM_NORESERVE);
437 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
438 file->f_flags |= O_LARGEFILE;
440 if (flags & MFD_NOEXEC_SEAL) {
441 struct inode *inode = file_inode(file);
443 inode->i_mode &= ~0111;
444 file_seals = memfd_file_seals_ptr(file);
446 *file_seals &= ~F_SEAL_SEAL;
447 *file_seals |= F_SEAL_EXEC;
449 } else if (flags & MFD_ALLOW_SEALING) {
450 /* MFD_EXEC and MFD_ALLOW_SEALING are set */
451 file_seals = memfd_file_seals_ptr(file);
453 *file_seals &= ~F_SEAL_SEAL;
459 SYSCALL_DEFINE2(memfd_create,
460 const char __user *, uname,
467 error = sanitize_flags(&flags);
471 name = alloc_name(uname);
473 return PTR_ERR(name);
475 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
481 file = alloc_file(name, flags);
483 error = PTR_ERR(file);
487 fd_install(fd, file);