1 /*
2 * memfd_create system call and file sealing support
3 *
4 * Code was originally included in shmem.c, and broken out to facilitate
5 * use by hugetlbfs as well as tmpfs.
6 *
7 * This file is released under the GPL.
8 */
9
10 #include <linux/fs.h>
11 #include <linux/vfs.h>
12 #include <linux/pagemap.h>
13 #include <linux/file.h>
14 #include <linux/mm.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>
22
23 /*
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.
27 */
28 #define MEMFD_TAG_PINNED PAGECACHE_TAG_TOWRITE
29 #define LAST_SCAN 4 /* about 150ms max */
30
memfd_tag_pins(struct address_space * mapping)31 static void memfd_tag_pins(struct address_space *mapping)
32 {
33 struct radix_tree_iter iter;
34 void __rcu **slot;
35 pgoff_t start;
36 struct page *page;
37 unsigned int tagged = 0;
38
39 lru_add_drain();
40 start = 0;
41
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);
48 continue;
49 }
50 } else if (page_count(page) - page_mapcount(page) > 1) {
51 radix_tree_tag_set(&mapping->i_pages, iter.index,
52 MEMFD_TAG_PINNED);
53 }
54
55 if (++tagged % 1024)
56 continue;
57
58 slot = radix_tree_iter_resume(slot, &iter);
59 xa_unlock_irq(&mapping->i_pages);
60 cond_resched();
61 xa_lock_irq(&mapping->i_pages);
62 }
63 xa_unlock_irq(&mapping->i_pages);
64 }
65
66 /*
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
71 * them to be dropped.
72 * The caller must guarantee that no new user will acquire writable references
73 * to those pages to avoid races.
74 */
memfd_wait_for_pins(struct address_space * mapping)75 static int memfd_wait_for_pins(struct address_space *mapping)
76 {
77 struct radix_tree_iter iter;
78 void __rcu **slot;
79 pgoff_t start;
80 struct page *page;
81 int error, scan;
82
83 memfd_tag_pins(mapping);
84
85 error = 0;
86 for (scan = 0; scan <= LAST_SCAN; scan++) {
87 if (!radix_tree_tagged(&mapping->i_pages, MEMFD_TAG_PINNED))
88 break;
89
90 if (!scan)
91 lru_add_drain_all();
92 else if (schedule_timeout_killable((HZ << scan) / 200))
93 scan = LAST_SCAN;
94
95 start = 0;
96 rcu_read_lock();
97 radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter,
98 start, MEMFD_TAG_PINNED) {
99
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);
104 continue;
105 }
106
107 page = NULL;
108 }
109
110 if (page &&
111 page_count(page) - page_mapcount(page) != 1) {
112 if (scan < LAST_SCAN)
113 goto continue_resched;
114
115 /*
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.
119 */
120 error = -EBUSY;
121 }
122
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);
127 continue_resched:
128 if (need_resched()) {
129 slot = radix_tree_iter_resume(slot, &iter);
130 cond_resched_rcu();
131 }
132 }
133 rcu_read_unlock();
134 }
135
136 return error;
137 }
138
memfd_file_seals_ptr(struct file * file)139 static unsigned int *memfd_file_seals_ptr(struct file *file)
140 {
141 if (shmem_file(file))
142 return &SHMEM_I(file_inode(file))->seals;
143
144 #ifdef CONFIG_HUGETLBFS
145 if (is_file_hugepages(file))
146 return &HUGETLBFS_I(file_inode(file))->seals;
147 #endif
148
149 return NULL;
150 }
151
152 #define F_ALL_SEALS (F_SEAL_SEAL | \
153 F_SEAL_SHRINK | \
154 F_SEAL_GROW | \
155 F_SEAL_WRITE)
156
memfd_add_seals(struct file * file,unsigned int seals)157 static int memfd_add_seals(struct file *file, unsigned int seals)
158 {
159 struct inode *inode = file_inode(file);
160 unsigned int *file_seals;
161 int error;
162
163 /*
164 * SEALING
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
170 * shared object.
171 *
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
180 *
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
186 * added.
187 *
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.
192 */
193
194 if (!(file->f_mode & FMODE_WRITE))
195 return -EPERM;
196 if (seals & ~(unsigned int)F_ALL_SEALS)
197 return -EINVAL;
198
199 inode_lock(inode);
200
201 file_seals = memfd_file_seals_ptr(file);
202 if (!file_seals) {
203 error = -EINVAL;
204 goto unlock;
205 }
206
207 if (*file_seals & F_SEAL_SEAL) {
208 error = -EPERM;
209 goto unlock;
210 }
211
212 if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) {
213 error = mapping_deny_writable(file->f_mapping);
214 if (error)
215 goto unlock;
216
217 error = memfd_wait_for_pins(file->f_mapping);
218 if (error) {
219 mapping_allow_writable(file->f_mapping);
220 goto unlock;
221 }
222 }
223
224 *file_seals |= seals;
225 error = 0;
226
227 unlock:
228 inode_unlock(inode);
229 return error;
230 }
231
memfd_get_seals(struct file * file)232 static int memfd_get_seals(struct file *file)
233 {
234 unsigned int *seals = memfd_file_seals_ptr(file);
235
236 return seals ? *seals : -EINVAL;
237 }
238
memfd_fcntl(struct file * file,unsigned int cmd,unsigned long arg)239 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
240 {
241 long error;
242
243 switch (cmd) {
244 case F_ADD_SEALS:
245 /* disallow upper 32bit */
246 if (arg > UINT_MAX)
247 return -EINVAL;
248
249 error = memfd_add_seals(file, arg);
250 break;
251 case F_GET_SEALS:
252 error = memfd_get_seals(file);
253 break;
254 default:
255 error = -EINVAL;
256 break;
257 }
258
259 return error;
260 }
261
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)
265
266 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
267
SYSCALL_DEFINE2(memfd_create,const char __user *,uname,unsigned int,flags)268 SYSCALL_DEFINE2(memfd_create,
269 const char __user *, uname,
270 unsigned int, flags)
271 {
272 unsigned int *file_seals;
273 struct file *file;
274 int fd, error;
275 char *name;
276 long len;
277
278 if (!(flags & MFD_HUGETLB)) {
279 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
280 return -EINVAL;
281 } else {
282 /* Allow huge page size encoding in flags. */
283 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
284 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
285 return -EINVAL;
286 }
287
288 /* length includes terminating zero */
289 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
290 if (len <= 0)
291 return -EFAULT;
292 if (len > MFD_NAME_MAX_LEN + 1)
293 return -EINVAL;
294
295 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
296 if (!name)
297 return -ENOMEM;
298
299 strcpy(name, MFD_NAME_PREFIX);
300 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
301 error = -EFAULT;
302 goto err_name;
303 }
304
305 /* terminating-zero may have changed after strnlen_user() returned */
306 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
307 error = -EFAULT;
308 goto err_name;
309 }
310
311 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
312 if (fd < 0) {
313 error = fd;
314 goto err_name;
315 }
316
317 if (flags & MFD_HUGETLB) {
318 struct user_struct *user = NULL;
319
320 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
321 HUGETLB_ANONHUGE_INODE,
322 (flags >> MFD_HUGE_SHIFT) &
323 MFD_HUGE_MASK);
324 } else
325 file = shmem_file_setup(name, 0, VM_NORESERVE);
326 if (IS_ERR(file)) {
327 error = PTR_ERR(file);
328 goto err_fd;
329 }
330 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
331 file->f_flags |= O_LARGEFILE;
332
333 if (flags & MFD_ALLOW_SEALING) {
334 file_seals = memfd_file_seals_ptr(file);
335 *file_seals &= ~F_SEAL_SEAL;
336 }
337
338 fd_install(fd, file);
339 kfree(name);
340 return fd;
341
342 err_fd:
343 put_unused_fd(fd);
344 err_name:
345 kfree(name);
346 return error;
347 }
348