1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Author: Andrei Vagin <avagin@openvz.org> 4 * Author: Dmitry Safonov <dima@arista.com> 5 */ 6 7 #include <linux/time_namespace.h> 8 #include <linux/user_namespace.h> 9 #include <linux/sched/signal.h> 10 #include <linux/sched/task.h> 11 #include <linux/clocksource.h> 12 #include <linux/seq_file.h> 13 #include <linux/proc_ns.h> 14 #include <linux/export.h> 15 #include <linux/time.h> 16 #include <linux/slab.h> 17 #include <linux/cred.h> 18 #include <linux/err.h> 19 #include <linux/mm.h> 20 21 #include <vdso/datapage.h> 22 23 ktime_t do_timens_ktime_to_host(clockid_t clockid, ktime_t tim, 24 struct timens_offsets *ns_offsets) 25 { 26 ktime_t offset; 27 28 switch (clockid) { 29 case CLOCK_MONOTONIC: 30 offset = timespec64_to_ktime(ns_offsets->monotonic); 31 break; 32 case CLOCK_BOOTTIME: 33 case CLOCK_BOOTTIME_ALARM: 34 offset = timespec64_to_ktime(ns_offsets->boottime); 35 break; 36 default: 37 return tim; 38 } 39 40 /* 41 * Check that @tim value is in [offset, KTIME_MAX + offset] 42 * and subtract offset. 43 */ 44 if (tim < offset) { 45 /* 46 * User can specify @tim *absolute* value - if it's lesser than 47 * the time namespace's offset - it's already expired. 48 */ 49 tim = 0; 50 } else { 51 tim = ktime_sub(tim, offset); 52 if (unlikely(tim > KTIME_MAX)) 53 tim = KTIME_MAX; 54 } 55 56 return tim; 57 } 58 59 static struct ucounts *inc_time_namespaces(struct user_namespace *ns) 60 { 61 return inc_ucount(ns, current_euid(), UCOUNT_TIME_NAMESPACES); 62 } 63 64 static void dec_time_namespaces(struct ucounts *ucounts) 65 { 66 dec_ucount(ucounts, UCOUNT_TIME_NAMESPACES); 67 } 68 69 /** 70 * clone_time_ns - Clone a time namespace 71 * @user_ns: User namespace which owns a new namespace. 72 * @old_ns: Namespace to clone 73 * 74 * Clone @old_ns and set the clone refcount to 1 75 * 76 * Return: The new namespace or ERR_PTR. 77 */ 78 static struct time_namespace *clone_time_ns(struct user_namespace *user_ns, 79 struct time_namespace *old_ns) 80 { 81 struct time_namespace *ns; 82 struct ucounts *ucounts; 83 int err; 84 85 err = -ENOSPC; 86 ucounts = inc_time_namespaces(user_ns); 87 if (!ucounts) 88 goto fail; 89 90 err = -ENOMEM; 91 ns = kmalloc(sizeof(*ns), GFP_KERNEL_ACCOUNT); 92 if (!ns) 93 goto fail_dec; 94 95 refcount_set(&ns->ns.count, 1); 96 97 ns->vvar_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); 98 if (!ns->vvar_page) 99 goto fail_free; 100 101 err = ns_alloc_inum(&ns->ns); 102 if (err) 103 goto fail_free_page; 104 105 ns->ucounts = ucounts; 106 ns->ns.ops = &timens_operations; 107 ns->user_ns = get_user_ns(user_ns); 108 ns->offsets = old_ns->offsets; 109 ns->frozen_offsets = false; 110 return ns; 111 112 fail_free_page: 113 __free_page(ns->vvar_page); 114 fail_free: 115 kfree(ns); 116 fail_dec: 117 dec_time_namespaces(ucounts); 118 fail: 119 return ERR_PTR(err); 120 } 121 122 /** 123 * copy_time_ns - Create timens_for_children from @old_ns 124 * @flags: Cloning flags 125 * @user_ns: User namespace which owns a new namespace. 126 * @old_ns: Namespace to clone 127 * 128 * If CLONE_NEWTIME specified in @flags, creates a new timens_for_children; 129 * adds a refcounter to @old_ns otherwise. 130 * 131 * Return: timens_for_children namespace or ERR_PTR. 132 */ 133 struct time_namespace *copy_time_ns(unsigned long flags, 134 struct user_namespace *user_ns, struct time_namespace *old_ns) 135 { 136 if (!(flags & CLONE_NEWTIME)) 137 return get_time_ns(old_ns); 138 139 return clone_time_ns(user_ns, old_ns); 140 } 141 142 static struct timens_offset offset_from_ts(struct timespec64 off) 143 { 144 struct timens_offset ret; 145 146 ret.sec = off.tv_sec; 147 ret.nsec = off.tv_nsec; 148 149 return ret; 150 } 151 152 /* 153 * A time namespace VVAR page has the same layout as the VVAR page which 154 * contains the system wide VDSO data. 155 * 156 * For a normal task the VVAR pages are installed in the normal ordering: 157 * VVAR 158 * PVCLOCK 159 * HVCLOCK 160 * TIMENS <- Not really required 161 * 162 * Now for a timens task the pages are installed in the following order: 163 * TIMENS 164 * PVCLOCK 165 * HVCLOCK 166 * VVAR 167 * 168 * The check for vdso_clock->clock_mode is in the unlikely path of 169 * the seq begin magic. So for the non-timens case most of the time 170 * 'seq' is even, so the branch is not taken. 171 * 172 * If 'seq' is odd, i.e. a concurrent update is in progress, the extra check 173 * for vdso_clock->clock_mode is a non-issue. The task is spin waiting for the 174 * update to finish and for 'seq' to become even anyway. 175 * 176 * Timens page has vdso_clock->clock_mode set to VDSO_CLOCKMODE_TIMENS which 177 * enforces the time namespace handling path. 178 */ 179 static void timens_setup_vdso_clock_data(struct vdso_clock *vc, 180 struct time_namespace *ns) 181 { 182 struct timens_offset *offset = vc->offset; 183 struct timens_offset monotonic = offset_from_ts(ns->offsets.monotonic); 184 struct timens_offset boottime = offset_from_ts(ns->offsets.boottime); 185 186 vc->seq = 1; 187 vc->clock_mode = VDSO_CLOCKMODE_TIMENS; 188 offset[CLOCK_MONOTONIC] = monotonic; 189 offset[CLOCK_MONOTONIC_RAW] = monotonic; 190 offset[CLOCK_MONOTONIC_COARSE] = monotonic; 191 offset[CLOCK_BOOTTIME] = boottime; 192 offset[CLOCK_BOOTTIME_ALARM] = boottime; 193 } 194 195 struct page *find_timens_vvar_page(struct vm_area_struct *vma) 196 { 197 if (likely(vma->vm_mm == current->mm)) 198 return current->nsproxy->time_ns->vvar_page; 199 200 /* 201 * VM_PFNMAP | VM_IO protect .fault() handler from being called 202 * through interfaces like /proc/$pid/mem or 203 * process_vm_{readv,writev}() as long as there's no .access() 204 * in special_mapping_vmops(). 205 * For more details check_vma_flags() and __access_remote_vm() 206 */ 207 208 WARN(1, "vvar_page accessed remotely"); 209 210 return NULL; 211 } 212 213 /* 214 * Protects possibly multiple offsets writers racing each other 215 * and tasks entering the namespace. 216 */ 217 static DEFINE_MUTEX(offset_lock); 218 219 static void timens_set_vvar_page(struct task_struct *task, 220 struct time_namespace *ns) 221 { 222 struct vdso_time_data *vdata; 223 struct vdso_clock *vc; 224 unsigned int i; 225 226 if (ns == &init_time_ns) 227 return; 228 229 /* Fast-path, taken by every task in namespace except the first. */ 230 if (likely(ns->frozen_offsets)) 231 return; 232 233 mutex_lock(&offset_lock); 234 /* Nothing to-do: vvar_page has been already initialized. */ 235 if (ns->frozen_offsets) 236 goto out; 237 238 ns->frozen_offsets = true; 239 vdata = page_address(ns->vvar_page); 240 vc = vdata->clock_data; 241 242 for (i = 0; i < CS_BASES; i++) 243 timens_setup_vdso_clock_data(&vc[i], ns); 244 245 out: 246 mutex_unlock(&offset_lock); 247 } 248 249 void free_time_ns(struct time_namespace *ns) 250 { 251 dec_time_namespaces(ns->ucounts); 252 put_user_ns(ns->user_ns); 253 ns_free_inum(&ns->ns); 254 __free_page(ns->vvar_page); 255 kfree(ns); 256 } 257 258 static struct time_namespace *to_time_ns(struct ns_common *ns) 259 { 260 return container_of(ns, struct time_namespace, ns); 261 } 262 263 static struct ns_common *timens_get(struct task_struct *task) 264 { 265 struct time_namespace *ns = NULL; 266 struct nsproxy *nsproxy; 267 268 task_lock(task); 269 nsproxy = task->nsproxy; 270 if (nsproxy) { 271 ns = nsproxy->time_ns; 272 get_time_ns(ns); 273 } 274 task_unlock(task); 275 276 return ns ? &ns->ns : NULL; 277 } 278 279 static struct ns_common *timens_for_children_get(struct task_struct *task) 280 { 281 struct time_namespace *ns = NULL; 282 struct nsproxy *nsproxy; 283 284 task_lock(task); 285 nsproxy = task->nsproxy; 286 if (nsproxy) { 287 ns = nsproxy->time_ns_for_children; 288 get_time_ns(ns); 289 } 290 task_unlock(task); 291 292 return ns ? &ns->ns : NULL; 293 } 294 295 static void timens_put(struct ns_common *ns) 296 { 297 put_time_ns(to_time_ns(ns)); 298 } 299 300 void timens_commit(struct task_struct *tsk, struct time_namespace *ns) 301 { 302 timens_set_vvar_page(tsk, ns); 303 vdso_join_timens(tsk, ns); 304 } 305 306 static int timens_install(struct nsset *nsset, struct ns_common *new) 307 { 308 struct nsproxy *nsproxy = nsset->nsproxy; 309 struct time_namespace *ns = to_time_ns(new); 310 311 if (!current_is_single_threaded()) 312 return -EUSERS; 313 314 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) || 315 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN)) 316 return -EPERM; 317 318 get_time_ns(ns); 319 put_time_ns(nsproxy->time_ns); 320 nsproxy->time_ns = ns; 321 322 get_time_ns(ns); 323 put_time_ns(nsproxy->time_ns_for_children); 324 nsproxy->time_ns_for_children = ns; 325 return 0; 326 } 327 328 void timens_on_fork(struct nsproxy *nsproxy, struct task_struct *tsk) 329 { 330 struct ns_common *nsc = &nsproxy->time_ns_for_children->ns; 331 struct time_namespace *ns = to_time_ns(nsc); 332 333 /* create_new_namespaces() already incremented the ref counter */ 334 if (nsproxy->time_ns == nsproxy->time_ns_for_children) 335 return; 336 337 get_time_ns(ns); 338 put_time_ns(nsproxy->time_ns); 339 nsproxy->time_ns = ns; 340 341 timens_commit(tsk, ns); 342 } 343 344 static struct user_namespace *timens_owner(struct ns_common *ns) 345 { 346 return to_time_ns(ns)->user_ns; 347 } 348 349 static void show_offset(struct seq_file *m, int clockid, struct timespec64 *ts) 350 { 351 char *clock; 352 353 switch (clockid) { 354 case CLOCK_BOOTTIME: 355 clock = "boottime"; 356 break; 357 case CLOCK_MONOTONIC: 358 clock = "monotonic"; 359 break; 360 default: 361 clock = "unknown"; 362 break; 363 } 364 seq_printf(m, "%-10s %10lld %9ld\n", clock, ts->tv_sec, ts->tv_nsec); 365 } 366 367 void proc_timens_show_offsets(struct task_struct *p, struct seq_file *m) 368 { 369 struct ns_common *ns; 370 struct time_namespace *time_ns; 371 372 ns = timens_for_children_get(p); 373 if (!ns) 374 return; 375 time_ns = to_time_ns(ns); 376 377 show_offset(m, CLOCK_MONOTONIC, &time_ns->offsets.monotonic); 378 show_offset(m, CLOCK_BOOTTIME, &time_ns->offsets.boottime); 379 put_time_ns(time_ns); 380 } 381 382 int proc_timens_set_offset(struct file *file, struct task_struct *p, 383 struct proc_timens_offset *offsets, int noffsets) 384 { 385 struct ns_common *ns; 386 struct time_namespace *time_ns; 387 struct timespec64 tp; 388 int i, err; 389 390 ns = timens_for_children_get(p); 391 if (!ns) 392 return -ESRCH; 393 time_ns = to_time_ns(ns); 394 395 if (!file_ns_capable(file, time_ns->user_ns, CAP_SYS_TIME)) { 396 put_time_ns(time_ns); 397 return -EPERM; 398 } 399 400 for (i = 0; i < noffsets; i++) { 401 struct proc_timens_offset *off = &offsets[i]; 402 403 switch (off->clockid) { 404 case CLOCK_MONOTONIC: 405 ktime_get_ts64(&tp); 406 break; 407 case CLOCK_BOOTTIME: 408 ktime_get_boottime_ts64(&tp); 409 break; 410 default: 411 err = -EINVAL; 412 goto out; 413 } 414 415 err = -ERANGE; 416 417 if (off->val.tv_sec > KTIME_SEC_MAX || 418 off->val.tv_sec < -KTIME_SEC_MAX) 419 goto out; 420 421 tp = timespec64_add(tp, off->val); 422 /* 423 * KTIME_SEC_MAX is divided by 2 to be sure that KTIME_MAX is 424 * still unreachable. 425 */ 426 if (tp.tv_sec < 0 || tp.tv_sec > KTIME_SEC_MAX / 2) 427 goto out; 428 } 429 430 mutex_lock(&offset_lock); 431 if (time_ns->frozen_offsets) { 432 err = -EACCES; 433 goto out_unlock; 434 } 435 436 err = 0; 437 /* Don't report errors after this line */ 438 for (i = 0; i < noffsets; i++) { 439 struct proc_timens_offset *off = &offsets[i]; 440 struct timespec64 *offset = NULL; 441 442 switch (off->clockid) { 443 case CLOCK_MONOTONIC: 444 offset = &time_ns->offsets.monotonic; 445 break; 446 case CLOCK_BOOTTIME: 447 offset = &time_ns->offsets.boottime; 448 break; 449 } 450 451 *offset = off->val; 452 } 453 454 out_unlock: 455 mutex_unlock(&offset_lock); 456 out: 457 put_time_ns(time_ns); 458 459 return err; 460 } 461 462 const struct proc_ns_operations timens_operations = { 463 .name = "time", 464 .type = CLONE_NEWTIME, 465 .get = timens_get, 466 .put = timens_put, 467 .install = timens_install, 468 .owner = timens_owner, 469 }; 470 471 const struct proc_ns_operations timens_for_children_operations = { 472 .name = "time_for_children", 473 .real_ns_name = "time", 474 .type = CLONE_NEWTIME, 475 .get = timens_for_children_get, 476 .put = timens_put, 477 .install = timens_install, 478 .owner = timens_owner, 479 }; 480 481 struct time_namespace init_time_ns = { 482 .ns.count = REFCOUNT_INIT(3), 483 .user_ns = &init_user_ns, 484 .ns.inum = PROC_TIME_INIT_INO, 485 .ns.ops = &timens_operations, 486 .frozen_offsets = true, 487 }; 488