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 xa_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 31 static void memfd_tag_pins(struct xa_state *xas) 32 { 33 struct page *page; 34 int latency = 0; 35 int cache_count; 36 37 lru_add_drain(); 38 39 xas_lock_irq(xas); 40 xas_for_each(xas, page, ULONG_MAX) { 41 cache_count = 1; 42 if (!xa_is_value(page) && 43 PageTransHuge(page) && !PageHuge(page)) 44 cache_count = HPAGE_PMD_NR; 45 46 if (!xa_is_value(page) && 47 page_count(page) - total_mapcount(page) != cache_count) 48 xas_set_mark(xas, MEMFD_TAG_PINNED); 49 if (cache_count != 1) 50 xas_set(xas, page->index + cache_count); 51 52 latency += cache_count; 53 if (latency < XA_CHECK_SCHED) 54 continue; 55 latency = 0; 56 57 xas_pause(xas); 58 xas_unlock_irq(xas); 59 cond_resched(); 60 xas_lock_irq(xas); 61 } 62 xas_unlock_irq(xas); 63 } 64 65 /* 66 * Setting SEAL_WRITE requires us to verify there's no pending writer. However, 67 * via get_user_pages(), drivers might have some pending I/O without any active 68 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages 69 * and see whether it has an elevated ref-count. If so, we tag them and wait for 70 * them to be dropped. 71 * The caller must guarantee that no new user will acquire writable references 72 * to those pages to avoid races. 73 */ 74 static int memfd_wait_for_pins(struct address_space *mapping) 75 { 76 XA_STATE(xas, &mapping->i_pages, 0); 77 struct page *page; 78 int error, scan; 79 80 memfd_tag_pins(&xas); 81 82 error = 0; 83 for (scan = 0; scan <= LAST_SCAN; scan++) { 84 int latency = 0; 85 int cache_count; 86 87 if (!xas_marked(&xas, 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 xas_set(&xas, 0); 96 xas_lock_irq(&xas); 97 xas_for_each_marked(&xas, page, ULONG_MAX, MEMFD_TAG_PINNED) { 98 bool clear = true; 99 100 cache_count = 1; 101 if (!xa_is_value(page) && 102 PageTransHuge(page) && !PageHuge(page)) 103 cache_count = HPAGE_PMD_NR; 104 105 if (!xa_is_value(page) && cache_count != 106 page_count(page) - total_mapcount(page)) { 107 /* 108 * On the last scan, we clean up all those tags 109 * we inserted; but make a note that we still 110 * found pages pinned. 111 */ 112 if (scan == LAST_SCAN) 113 error = -EBUSY; 114 else 115 clear = false; 116 } 117 if (clear) 118 xas_clear_mark(&xas, MEMFD_TAG_PINNED); 119 120 latency += cache_count; 121 if (latency < XA_CHECK_SCHED) 122 continue; 123 latency = 0; 124 125 xas_pause(&xas); 126 xas_unlock_irq(&xas); 127 cond_resched(); 128 xas_lock_irq(&xas); 129 } 130 xas_unlock_irq(&xas); 131 } 132 133 return error; 134 } 135 136 static unsigned int *memfd_file_seals_ptr(struct file *file) 137 { 138 if (shmem_file(file)) 139 return &SHMEM_I(file_inode(file))->seals; 140 141 #ifdef CONFIG_HUGETLBFS 142 if (is_file_hugepages(file)) 143 return &HUGETLBFS_I(file_inode(file))->seals; 144 #endif 145 146 return NULL; 147 } 148 149 #define F_ALL_SEALS (F_SEAL_SEAL | \ 150 F_SEAL_EXEC | \ 151 F_SEAL_SHRINK | \ 152 F_SEAL_GROW | \ 153 F_SEAL_WRITE | \ 154 F_SEAL_FUTURE_WRITE) 155 156 static int memfd_add_seals(struct file *file, unsigned int seals) 157 { 158 struct inode *inode = file_inode(file); 159 unsigned int *file_seals; 160 int error; 161 162 /* 163 * SEALING 164 * Sealing allows multiple parties to share a tmpfs or hugetlbfs file 165 * but restrict access to a specific subset of file operations. Seals 166 * can only be added, but never removed. This way, mutually untrusted 167 * parties can share common memory regions with a well-defined policy. 168 * A malicious peer can thus never perform unwanted operations on a 169 * shared object. 170 * 171 * Seals are only supported on special tmpfs or hugetlbfs files and 172 * always affect the whole underlying inode. Once a seal is set, it 173 * may prevent some kinds of access to the file. Currently, the 174 * following seals are defined: 175 * SEAL_SEAL: Prevent further seals from being set on this file 176 * SEAL_SHRINK: Prevent the file from shrinking 177 * SEAL_GROW: Prevent the file from growing 178 * SEAL_WRITE: Prevent write access to the file 179 * SEAL_EXEC: Prevent modification of the exec bits in the file mode 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 } 231 232 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 } 238 239 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) 267 268 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 file = hugetlb_file_setup(name, 0, VM_NORESERVE, 319 HUGETLB_ANONHUGE_INODE, 320 (flags >> MFD_HUGE_SHIFT) & 321 MFD_HUGE_MASK); 322 } else 323 file = shmem_file_setup(name, 0, VM_NORESERVE); 324 if (IS_ERR(file)) { 325 error = PTR_ERR(file); 326 goto err_fd; 327 } 328 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; 329 file->f_flags |= O_LARGEFILE; 330 331 if (flags & MFD_ALLOW_SEALING) { 332 file_seals = memfd_file_seals_ptr(file); 333 *file_seals &= ~F_SEAL_SEAL; 334 } 335 336 fd_install(fd, file); 337 kfree(name); 338 return fd; 339 340 err_fd: 341 put_unused_fd(fd); 342 err_name: 343 kfree(name); 344 return error; 345 } 346