1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/kernel/signal.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson 8 * 9 * 2003-06-02 Jim Houston - Concurrent Computer Corp. 10 * Changes to use preallocated sigqueue structures 11 * to allow signals to be sent reliably. 12 */ 13 14 #include <linux/slab.h> 15 #include <linux/export.h> 16 #include <linux/init.h> 17 #include <linux/sched/mm.h> 18 #include <linux/sched/user.h> 19 #include <linux/sched/debug.h> 20 #include <linux/sched/task.h> 21 #include <linux/sched/task_stack.h> 22 #include <linux/sched/cputime.h> 23 #include <linux/file.h> 24 #include <linux/fs.h> 25 #include <linux/proc_fs.h> 26 #include <linux/tty.h> 27 #include <linux/binfmts.h> 28 #include <linux/coredump.h> 29 #include <linux/security.h> 30 #include <linux/syscalls.h> 31 #include <linux/ptrace.h> 32 #include <linux/signal.h> 33 #include <linux/signalfd.h> 34 #include <linux/ratelimit.h> 35 #include <linux/tracehook.h> 36 #include <linux/capability.h> 37 #include <linux/freezer.h> 38 #include <linux/pid_namespace.h> 39 #include <linux/nsproxy.h> 40 #include <linux/user_namespace.h> 41 #include <linux/uprobes.h> 42 #include <linux/compat.h> 43 #include <linux/cn_proc.h> 44 #include <linux/compiler.h> 45 #include <linux/posix-timers.h> 46 #include <linux/livepatch.h> 47 #include <linux/cgroup.h> 48 #include <linux/audit.h> 49 50 #define CREATE_TRACE_POINTS 51 #include <trace/events/signal.h> 52 53 #include <asm/param.h> 54 #include <linux/uaccess.h> 55 #include <asm/unistd.h> 56 #include <asm/siginfo.h> 57 #include <asm/cacheflush.h> 58 59 /* 60 * SLAB caches for signal bits. 61 */ 62 63 static struct kmem_cache *sigqueue_cachep; 64 65 int print_fatal_signals __read_mostly; 66 67 static void __user *sig_handler(struct task_struct *t, int sig) 68 { 69 return t->sighand->action[sig - 1].sa.sa_handler; 70 } 71 72 static inline bool sig_handler_ignored(void __user *handler, int sig) 73 { 74 /* Is it explicitly or implicitly ignored? */ 75 return handler == SIG_IGN || 76 (handler == SIG_DFL && sig_kernel_ignore(sig)); 77 } 78 79 static bool sig_task_ignored(struct task_struct *t, int sig, bool force) 80 { 81 void __user *handler; 82 83 handler = sig_handler(t, sig); 84 85 /* SIGKILL and SIGSTOP may not be sent to the global init */ 86 if (unlikely(is_global_init(t) && sig_kernel_only(sig))) 87 return true; 88 89 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) && 90 handler == SIG_DFL && !(force && sig_kernel_only(sig))) 91 return true; 92 93 /* Only allow kernel generated signals to this kthread */ 94 if (unlikely((t->flags & PF_KTHREAD) && 95 (handler == SIG_KTHREAD_KERNEL) && !force)) 96 return true; 97 98 return sig_handler_ignored(handler, sig); 99 } 100 101 static bool sig_ignored(struct task_struct *t, int sig, bool force) 102 { 103 /* 104 * Blocked signals are never ignored, since the 105 * signal handler may change by the time it is 106 * unblocked. 107 */ 108 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig)) 109 return false; 110 111 /* 112 * Tracers may want to know about even ignored signal unless it 113 * is SIGKILL which can't be reported anyway but can be ignored 114 * by SIGNAL_UNKILLABLE task. 115 */ 116 if (t->ptrace && sig != SIGKILL) 117 return false; 118 119 return sig_task_ignored(t, sig, force); 120 } 121 122 /* 123 * Re-calculate pending state from the set of locally pending 124 * signals, globally pending signals, and blocked signals. 125 */ 126 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked) 127 { 128 unsigned long ready; 129 long i; 130 131 switch (_NSIG_WORDS) { 132 default: 133 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;) 134 ready |= signal->sig[i] &~ blocked->sig[i]; 135 break; 136 137 case 4: ready = signal->sig[3] &~ blocked->sig[3]; 138 ready |= signal->sig[2] &~ blocked->sig[2]; 139 ready |= signal->sig[1] &~ blocked->sig[1]; 140 ready |= signal->sig[0] &~ blocked->sig[0]; 141 break; 142 143 case 2: ready = signal->sig[1] &~ blocked->sig[1]; 144 ready |= signal->sig[0] &~ blocked->sig[0]; 145 break; 146 147 case 1: ready = signal->sig[0] &~ blocked->sig[0]; 148 } 149 return ready != 0; 150 } 151 152 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b)) 153 154 static bool recalc_sigpending_tsk(struct task_struct *t) 155 { 156 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) || 157 PENDING(&t->pending, &t->blocked) || 158 PENDING(&t->signal->shared_pending, &t->blocked) || 159 cgroup_task_frozen(t)) { 160 set_tsk_thread_flag(t, TIF_SIGPENDING); 161 return true; 162 } 163 164 /* 165 * We must never clear the flag in another thread, or in current 166 * when it's possible the current syscall is returning -ERESTART*. 167 * So we don't clear it here, and only callers who know they should do. 168 */ 169 return false; 170 } 171 172 /* 173 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up. 174 * This is superfluous when called on current, the wakeup is a harmless no-op. 175 */ 176 void recalc_sigpending_and_wake(struct task_struct *t) 177 { 178 if (recalc_sigpending_tsk(t)) 179 signal_wake_up(t, 0); 180 } 181 182 void recalc_sigpending(void) 183 { 184 if (!recalc_sigpending_tsk(current) && !freezing(current) && 185 !klp_patch_pending(current)) 186 clear_thread_flag(TIF_SIGPENDING); 187 188 } 189 EXPORT_SYMBOL(recalc_sigpending); 190 191 void calculate_sigpending(void) 192 { 193 /* Have any signals or users of TIF_SIGPENDING been delayed 194 * until after fork? 195 */ 196 spin_lock_irq(¤t->sighand->siglock); 197 set_tsk_thread_flag(current, TIF_SIGPENDING); 198 recalc_sigpending(); 199 spin_unlock_irq(¤t->sighand->siglock); 200 } 201 202 /* Given the mask, find the first available signal that should be serviced. */ 203 204 #define SYNCHRONOUS_MASK \ 205 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \ 206 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS)) 207 208 int next_signal(struct sigpending *pending, sigset_t *mask) 209 { 210 unsigned long i, *s, *m, x; 211 int sig = 0; 212 213 s = pending->signal.sig; 214 m = mask->sig; 215 216 /* 217 * Handle the first word specially: it contains the 218 * synchronous signals that need to be dequeued first. 219 */ 220 x = *s &~ *m; 221 if (x) { 222 if (x & SYNCHRONOUS_MASK) 223 x &= SYNCHRONOUS_MASK; 224 sig = ffz(~x) + 1; 225 return sig; 226 } 227 228 switch (_NSIG_WORDS) { 229 default: 230 for (i = 1; i < _NSIG_WORDS; ++i) { 231 x = *++s &~ *++m; 232 if (!x) 233 continue; 234 sig = ffz(~x) + i*_NSIG_BPW + 1; 235 break; 236 } 237 break; 238 239 case 2: 240 x = s[1] &~ m[1]; 241 if (!x) 242 break; 243 sig = ffz(~x) + _NSIG_BPW + 1; 244 break; 245 246 case 1: 247 /* Nothing to do */ 248 break; 249 } 250 251 return sig; 252 } 253 254 static inline void print_dropped_signal(int sig) 255 { 256 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10); 257 258 if (!print_fatal_signals) 259 return; 260 261 if (!__ratelimit(&ratelimit_state)) 262 return; 263 264 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n", 265 current->comm, current->pid, sig); 266 } 267 268 /** 269 * task_set_jobctl_pending - set jobctl pending bits 270 * @task: target task 271 * @mask: pending bits to set 272 * 273 * Clear @mask from @task->jobctl. @mask must be subset of 274 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK | 275 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is 276 * cleared. If @task is already being killed or exiting, this function 277 * becomes noop. 278 * 279 * CONTEXT: 280 * Must be called with @task->sighand->siglock held. 281 * 282 * RETURNS: 283 * %true if @mask is set, %false if made noop because @task was dying. 284 */ 285 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask) 286 { 287 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME | 288 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING)); 289 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK)); 290 291 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING))) 292 return false; 293 294 if (mask & JOBCTL_STOP_SIGMASK) 295 task->jobctl &= ~JOBCTL_STOP_SIGMASK; 296 297 task->jobctl |= mask; 298 return true; 299 } 300 301 /** 302 * task_clear_jobctl_trapping - clear jobctl trapping bit 303 * @task: target task 304 * 305 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED. 306 * Clear it and wake up the ptracer. Note that we don't need any further 307 * locking. @task->siglock guarantees that @task->parent points to the 308 * ptracer. 309 * 310 * CONTEXT: 311 * Must be called with @task->sighand->siglock held. 312 */ 313 void task_clear_jobctl_trapping(struct task_struct *task) 314 { 315 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) { 316 task->jobctl &= ~JOBCTL_TRAPPING; 317 smp_mb(); /* advised by wake_up_bit() */ 318 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT); 319 } 320 } 321 322 /** 323 * task_clear_jobctl_pending - clear jobctl pending bits 324 * @task: target task 325 * @mask: pending bits to clear 326 * 327 * Clear @mask from @task->jobctl. @mask must be subset of 328 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other 329 * STOP bits are cleared together. 330 * 331 * If clearing of @mask leaves no stop or trap pending, this function calls 332 * task_clear_jobctl_trapping(). 333 * 334 * CONTEXT: 335 * Must be called with @task->sighand->siglock held. 336 */ 337 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask) 338 { 339 BUG_ON(mask & ~JOBCTL_PENDING_MASK); 340 341 if (mask & JOBCTL_STOP_PENDING) 342 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED; 343 344 task->jobctl &= ~mask; 345 346 if (!(task->jobctl & JOBCTL_PENDING_MASK)) 347 task_clear_jobctl_trapping(task); 348 } 349 350 /** 351 * task_participate_group_stop - participate in a group stop 352 * @task: task participating in a group stop 353 * 354 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop. 355 * Group stop states are cleared and the group stop count is consumed if 356 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group 357 * stop, the appropriate `SIGNAL_*` flags are set. 358 * 359 * CONTEXT: 360 * Must be called with @task->sighand->siglock held. 361 * 362 * RETURNS: 363 * %true if group stop completion should be notified to the parent, %false 364 * otherwise. 365 */ 366 static bool task_participate_group_stop(struct task_struct *task) 367 { 368 struct signal_struct *sig = task->signal; 369 bool consume = task->jobctl & JOBCTL_STOP_CONSUME; 370 371 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING)); 372 373 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING); 374 375 if (!consume) 376 return false; 377 378 if (!WARN_ON_ONCE(sig->group_stop_count == 0)) 379 sig->group_stop_count--; 380 381 /* 382 * Tell the caller to notify completion iff we are entering into a 383 * fresh group stop. Read comment in do_signal_stop() for details. 384 */ 385 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) { 386 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED); 387 return true; 388 } 389 return false; 390 } 391 392 void task_join_group_stop(struct task_struct *task) 393 { 394 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK; 395 struct signal_struct *sig = current->signal; 396 397 if (sig->group_stop_count) { 398 sig->group_stop_count++; 399 mask |= JOBCTL_STOP_CONSUME; 400 } else if (!(sig->flags & SIGNAL_STOP_STOPPED)) 401 return; 402 403 /* Have the new thread join an on-going signal group stop */ 404 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING); 405 } 406 407 /* 408 * allocate a new signal queue record 409 * - this may be called without locks if and only if t == current, otherwise an 410 * appropriate lock must be held to stop the target task from exiting 411 */ 412 static struct sigqueue * 413 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit) 414 { 415 struct sigqueue *q = NULL; 416 struct user_struct *user; 417 int sigpending; 418 419 /* 420 * Protect access to @t credentials. This can go away when all 421 * callers hold rcu read lock. 422 * 423 * NOTE! A pending signal will hold on to the user refcount, 424 * and we get/put the refcount only when the sigpending count 425 * changes from/to zero. 426 */ 427 rcu_read_lock(); 428 user = __task_cred(t)->user; 429 sigpending = atomic_inc_return(&user->sigpending); 430 if (sigpending == 1) 431 get_uid(user); 432 rcu_read_unlock(); 433 434 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) { 435 q = kmem_cache_alloc(sigqueue_cachep, flags); 436 } else { 437 print_dropped_signal(sig); 438 } 439 440 if (unlikely(q == NULL)) { 441 if (atomic_dec_and_test(&user->sigpending)) 442 free_uid(user); 443 } else { 444 INIT_LIST_HEAD(&q->list); 445 q->flags = 0; 446 q->user = user; 447 } 448 449 return q; 450 } 451 452 static void __sigqueue_free(struct sigqueue *q) 453 { 454 if (q->flags & SIGQUEUE_PREALLOC) 455 return; 456 if (atomic_dec_and_test(&q->user->sigpending)) 457 free_uid(q->user); 458 kmem_cache_free(sigqueue_cachep, q); 459 } 460 461 void flush_sigqueue(struct sigpending *queue) 462 { 463 struct sigqueue *q; 464 465 sigemptyset(&queue->signal); 466 while (!list_empty(&queue->list)) { 467 q = list_entry(queue->list.next, struct sigqueue , list); 468 list_del_init(&q->list); 469 __sigqueue_free(q); 470 } 471 } 472 473 /* 474 * Flush all pending signals for this kthread. 475 */ 476 void flush_signals(struct task_struct *t) 477 { 478 unsigned long flags; 479 480 spin_lock_irqsave(&t->sighand->siglock, flags); 481 clear_tsk_thread_flag(t, TIF_SIGPENDING); 482 flush_sigqueue(&t->pending); 483 flush_sigqueue(&t->signal->shared_pending); 484 spin_unlock_irqrestore(&t->sighand->siglock, flags); 485 } 486 EXPORT_SYMBOL(flush_signals); 487 488 #ifdef CONFIG_POSIX_TIMERS 489 static void __flush_itimer_signals(struct sigpending *pending) 490 { 491 sigset_t signal, retain; 492 struct sigqueue *q, *n; 493 494 signal = pending->signal; 495 sigemptyset(&retain); 496 497 list_for_each_entry_safe(q, n, &pending->list, list) { 498 int sig = q->info.si_signo; 499 500 if (likely(q->info.si_code != SI_TIMER)) { 501 sigaddset(&retain, sig); 502 } else { 503 sigdelset(&signal, sig); 504 list_del_init(&q->list); 505 __sigqueue_free(q); 506 } 507 } 508 509 sigorsets(&pending->signal, &signal, &retain); 510 } 511 512 void flush_itimer_signals(void) 513 { 514 struct task_struct *tsk = current; 515 unsigned long flags; 516 517 spin_lock_irqsave(&tsk->sighand->siglock, flags); 518 __flush_itimer_signals(&tsk->pending); 519 __flush_itimer_signals(&tsk->signal->shared_pending); 520 spin_unlock_irqrestore(&tsk->sighand->siglock, flags); 521 } 522 #endif 523 524 void ignore_signals(struct task_struct *t) 525 { 526 int i; 527 528 for (i = 0; i < _NSIG; ++i) 529 t->sighand->action[i].sa.sa_handler = SIG_IGN; 530 531 flush_signals(t); 532 } 533 534 /* 535 * Flush all handlers for a task. 536 */ 537 538 void 539 flush_signal_handlers(struct task_struct *t, int force_default) 540 { 541 int i; 542 struct k_sigaction *ka = &t->sighand->action[0]; 543 for (i = _NSIG ; i != 0 ; i--) { 544 if (force_default || ka->sa.sa_handler != SIG_IGN) 545 ka->sa.sa_handler = SIG_DFL; 546 ka->sa.sa_flags = 0; 547 #ifdef __ARCH_HAS_SA_RESTORER 548 ka->sa.sa_restorer = NULL; 549 #endif 550 sigemptyset(&ka->sa.sa_mask); 551 ka++; 552 } 553 } 554 555 bool unhandled_signal(struct task_struct *tsk, int sig) 556 { 557 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler; 558 if (is_global_init(tsk)) 559 return true; 560 561 if (handler != SIG_IGN && handler != SIG_DFL) 562 return false; 563 564 /* if ptraced, let the tracer determine */ 565 return !tsk->ptrace; 566 } 567 568 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info, 569 bool *resched_timer) 570 { 571 struct sigqueue *q, *first = NULL; 572 573 /* 574 * Collect the siginfo appropriate to this signal. Check if 575 * there is another siginfo for the same signal. 576 */ 577 list_for_each_entry(q, &list->list, list) { 578 if (q->info.si_signo == sig) { 579 if (first) 580 goto still_pending; 581 first = q; 582 } 583 } 584 585 sigdelset(&list->signal, sig); 586 587 if (first) { 588 still_pending: 589 list_del_init(&first->list); 590 copy_siginfo(info, &first->info); 591 592 *resched_timer = 593 (first->flags & SIGQUEUE_PREALLOC) && 594 (info->si_code == SI_TIMER) && 595 (info->si_sys_private); 596 597 __sigqueue_free(first); 598 } else { 599 /* 600 * Ok, it wasn't in the queue. This must be 601 * a fast-pathed signal or we must have been 602 * out of queue space. So zero out the info. 603 */ 604 clear_siginfo(info); 605 info->si_signo = sig; 606 info->si_errno = 0; 607 info->si_code = SI_USER; 608 info->si_pid = 0; 609 info->si_uid = 0; 610 } 611 } 612 613 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask, 614 kernel_siginfo_t *info, bool *resched_timer) 615 { 616 int sig = next_signal(pending, mask); 617 618 if (sig) 619 collect_signal(sig, pending, info, resched_timer); 620 return sig; 621 } 622 623 /* 624 * Dequeue a signal and return the element to the caller, which is 625 * expected to free it. 626 * 627 * All callers have to hold the siglock. 628 */ 629 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info) 630 { 631 bool resched_timer = false; 632 int signr; 633 634 /* We only dequeue private signals from ourselves, we don't let 635 * signalfd steal them 636 */ 637 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer); 638 if (!signr) { 639 signr = __dequeue_signal(&tsk->signal->shared_pending, 640 mask, info, &resched_timer); 641 #ifdef CONFIG_POSIX_TIMERS 642 /* 643 * itimer signal ? 644 * 645 * itimers are process shared and we restart periodic 646 * itimers in the signal delivery path to prevent DoS 647 * attacks in the high resolution timer case. This is 648 * compliant with the old way of self-restarting 649 * itimers, as the SIGALRM is a legacy signal and only 650 * queued once. Changing the restart behaviour to 651 * restart the timer in the signal dequeue path is 652 * reducing the timer noise on heavy loaded !highres 653 * systems too. 654 */ 655 if (unlikely(signr == SIGALRM)) { 656 struct hrtimer *tmr = &tsk->signal->real_timer; 657 658 if (!hrtimer_is_queued(tmr) && 659 tsk->signal->it_real_incr != 0) { 660 hrtimer_forward(tmr, tmr->base->get_time(), 661 tsk->signal->it_real_incr); 662 hrtimer_restart(tmr); 663 } 664 } 665 #endif 666 } 667 668 recalc_sigpending(); 669 if (!signr) 670 return 0; 671 672 if (unlikely(sig_kernel_stop(signr))) { 673 /* 674 * Set a marker that we have dequeued a stop signal. Our 675 * caller might release the siglock and then the pending 676 * stop signal it is about to process is no longer in the 677 * pending bitmasks, but must still be cleared by a SIGCONT 678 * (and overruled by a SIGKILL). So those cases clear this 679 * shared flag after we've set it. Note that this flag may 680 * remain set after the signal we return is ignored or 681 * handled. That doesn't matter because its only purpose 682 * is to alert stop-signal processing code when another 683 * processor has come along and cleared the flag. 684 */ 685 current->jobctl |= JOBCTL_STOP_DEQUEUED; 686 } 687 #ifdef CONFIG_POSIX_TIMERS 688 if (resched_timer) { 689 /* 690 * Release the siglock to ensure proper locking order 691 * of timer locks outside of siglocks. Note, we leave 692 * irqs disabled here, since the posix-timers code is 693 * about to disable them again anyway. 694 */ 695 spin_unlock(&tsk->sighand->siglock); 696 posixtimer_rearm(info); 697 spin_lock(&tsk->sighand->siglock); 698 699 /* Don't expose the si_sys_private value to userspace */ 700 info->si_sys_private = 0; 701 } 702 #endif 703 return signr; 704 } 705 EXPORT_SYMBOL_GPL(dequeue_signal); 706 707 static int dequeue_synchronous_signal(kernel_siginfo_t *info) 708 { 709 struct task_struct *tsk = current; 710 struct sigpending *pending = &tsk->pending; 711 struct sigqueue *q, *sync = NULL; 712 713 /* 714 * Might a synchronous signal be in the queue? 715 */ 716 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK)) 717 return 0; 718 719 /* 720 * Return the first synchronous signal in the queue. 721 */ 722 list_for_each_entry(q, &pending->list, list) { 723 /* Synchronous signals have a positive si_code */ 724 if ((q->info.si_code > SI_USER) && 725 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) { 726 sync = q; 727 goto next; 728 } 729 } 730 return 0; 731 next: 732 /* 733 * Check if there is another siginfo for the same signal. 734 */ 735 list_for_each_entry_continue(q, &pending->list, list) { 736 if (q->info.si_signo == sync->info.si_signo) 737 goto still_pending; 738 } 739 740 sigdelset(&pending->signal, sync->info.si_signo); 741 recalc_sigpending(); 742 still_pending: 743 list_del_init(&sync->list); 744 copy_siginfo(info, &sync->info); 745 __sigqueue_free(sync); 746 return info->si_signo; 747 } 748 749 /* 750 * Tell a process that it has a new active signal.. 751 * 752 * NOTE! we rely on the previous spin_lock to 753 * lock interrupts for us! We can only be called with 754 * "siglock" held, and the local interrupt must 755 * have been disabled when that got acquired! 756 * 757 * No need to set need_resched since signal event passing 758 * goes through ->blocked 759 */ 760 void signal_wake_up_state(struct task_struct *t, unsigned int state) 761 { 762 set_tsk_thread_flag(t, TIF_SIGPENDING); 763 /* 764 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable 765 * case. We don't check t->state here because there is a race with it 766 * executing another processor and just now entering stopped state. 767 * By using wake_up_state, we ensure the process will wake up and 768 * handle its death signal. 769 */ 770 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE)) 771 kick_process(t); 772 } 773 774 /* 775 * Remove signals in mask from the pending set and queue. 776 * Returns 1 if any signals were found. 777 * 778 * All callers must be holding the siglock. 779 */ 780 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s) 781 { 782 struct sigqueue *q, *n; 783 sigset_t m; 784 785 sigandsets(&m, mask, &s->signal); 786 if (sigisemptyset(&m)) 787 return; 788 789 sigandnsets(&s->signal, &s->signal, mask); 790 list_for_each_entry_safe(q, n, &s->list, list) { 791 if (sigismember(mask, q->info.si_signo)) { 792 list_del_init(&q->list); 793 __sigqueue_free(q); 794 } 795 } 796 } 797 798 static inline int is_si_special(const struct kernel_siginfo *info) 799 { 800 return info <= SEND_SIG_PRIV; 801 } 802 803 static inline bool si_fromuser(const struct kernel_siginfo *info) 804 { 805 return info == SEND_SIG_NOINFO || 806 (!is_si_special(info) && SI_FROMUSER(info)); 807 } 808 809 /* 810 * called with RCU read lock from check_kill_permission() 811 */ 812 static bool kill_ok_by_cred(struct task_struct *t) 813 { 814 const struct cred *cred = current_cred(); 815 const struct cred *tcred = __task_cred(t); 816 817 return uid_eq(cred->euid, tcred->suid) || 818 uid_eq(cred->euid, tcred->uid) || 819 uid_eq(cred->uid, tcred->suid) || 820 uid_eq(cred->uid, tcred->uid) || 821 ns_capable(tcred->user_ns, CAP_KILL); 822 } 823 824 /* 825 * Bad permissions for sending the signal 826 * - the caller must hold the RCU read lock 827 */ 828 static int check_kill_permission(int sig, struct kernel_siginfo *info, 829 struct task_struct *t) 830 { 831 struct pid *sid; 832 int error; 833 834 if (!valid_signal(sig)) 835 return -EINVAL; 836 837 if (!si_fromuser(info)) 838 return 0; 839 840 error = audit_signal_info(sig, t); /* Let audit system see the signal */ 841 if (error) 842 return error; 843 844 if (!same_thread_group(current, t) && 845 !kill_ok_by_cred(t)) { 846 switch (sig) { 847 case SIGCONT: 848 sid = task_session(t); 849 /* 850 * We don't return the error if sid == NULL. The 851 * task was unhashed, the caller must notice this. 852 */ 853 if (!sid || sid == task_session(current)) 854 break; 855 fallthrough; 856 default: 857 return -EPERM; 858 } 859 } 860 861 return security_task_kill(t, info, sig, NULL); 862 } 863 864 /** 865 * ptrace_trap_notify - schedule trap to notify ptracer 866 * @t: tracee wanting to notify tracer 867 * 868 * This function schedules sticky ptrace trap which is cleared on the next 869 * TRAP_STOP to notify ptracer of an event. @t must have been seized by 870 * ptracer. 871 * 872 * If @t is running, STOP trap will be taken. If trapped for STOP and 873 * ptracer is listening for events, tracee is woken up so that it can 874 * re-trap for the new event. If trapped otherwise, STOP trap will be 875 * eventually taken without returning to userland after the existing traps 876 * are finished by PTRACE_CONT. 877 * 878 * CONTEXT: 879 * Must be called with @task->sighand->siglock held. 880 */ 881 static void ptrace_trap_notify(struct task_struct *t) 882 { 883 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED)); 884 assert_spin_locked(&t->sighand->siglock); 885 886 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY); 887 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING); 888 } 889 890 /* 891 * Handle magic process-wide effects of stop/continue signals. Unlike 892 * the signal actions, these happen immediately at signal-generation 893 * time regardless of blocking, ignoring, or handling. This does the 894 * actual continuing for SIGCONT, but not the actual stopping for stop 895 * signals. The process stop is done as a signal action for SIG_DFL. 896 * 897 * Returns true if the signal should be actually delivered, otherwise 898 * it should be dropped. 899 */ 900 static bool prepare_signal(int sig, struct task_struct *p, bool force) 901 { 902 struct signal_struct *signal = p->signal; 903 struct task_struct *t; 904 sigset_t flush; 905 906 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) { 907 if (!(signal->flags & SIGNAL_GROUP_EXIT)) 908 return sig == SIGKILL; 909 /* 910 * The process is in the middle of dying, nothing to do. 911 */ 912 } else if (sig_kernel_stop(sig)) { 913 /* 914 * This is a stop signal. Remove SIGCONT from all queues. 915 */ 916 siginitset(&flush, sigmask(SIGCONT)); 917 flush_sigqueue_mask(&flush, &signal->shared_pending); 918 for_each_thread(p, t) 919 flush_sigqueue_mask(&flush, &t->pending); 920 } else if (sig == SIGCONT) { 921 unsigned int why; 922 /* 923 * Remove all stop signals from all queues, wake all threads. 924 */ 925 siginitset(&flush, SIG_KERNEL_STOP_MASK); 926 flush_sigqueue_mask(&flush, &signal->shared_pending); 927 for_each_thread(p, t) { 928 flush_sigqueue_mask(&flush, &t->pending); 929 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING); 930 if (likely(!(t->ptrace & PT_SEIZED))) 931 wake_up_state(t, __TASK_STOPPED); 932 else 933 ptrace_trap_notify(t); 934 } 935 936 /* 937 * Notify the parent with CLD_CONTINUED if we were stopped. 938 * 939 * If we were in the middle of a group stop, we pretend it 940 * was already finished, and then continued. Since SIGCHLD 941 * doesn't queue we report only CLD_STOPPED, as if the next 942 * CLD_CONTINUED was dropped. 943 */ 944 why = 0; 945 if (signal->flags & SIGNAL_STOP_STOPPED) 946 why |= SIGNAL_CLD_CONTINUED; 947 else if (signal->group_stop_count) 948 why |= SIGNAL_CLD_STOPPED; 949 950 if (why) { 951 /* 952 * The first thread which returns from do_signal_stop() 953 * will take ->siglock, notice SIGNAL_CLD_MASK, and 954 * notify its parent. See get_signal(). 955 */ 956 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED); 957 signal->group_stop_count = 0; 958 signal->group_exit_code = 0; 959 } 960 } 961 962 return !sig_ignored(p, sig, force); 963 } 964 965 /* 966 * Test if P wants to take SIG. After we've checked all threads with this, 967 * it's equivalent to finding no threads not blocking SIG. Any threads not 968 * blocking SIG were ruled out because they are not running and already 969 * have pending signals. Such threads will dequeue from the shared queue 970 * as soon as they're available, so putting the signal on the shared queue 971 * will be equivalent to sending it to one such thread. 972 */ 973 static inline bool wants_signal(int sig, struct task_struct *p) 974 { 975 if (sigismember(&p->blocked, sig)) 976 return false; 977 978 if (p->flags & PF_EXITING) 979 return false; 980 981 if (sig == SIGKILL) 982 return true; 983 984 if (task_is_stopped_or_traced(p)) 985 return false; 986 987 return task_curr(p) || !task_sigpending(p); 988 } 989 990 static void complete_signal(int sig, struct task_struct *p, enum pid_type type) 991 { 992 struct signal_struct *signal = p->signal; 993 struct task_struct *t; 994 995 /* 996 * Now find a thread we can wake up to take the signal off the queue. 997 * 998 * If the main thread wants the signal, it gets first crack. 999 * Probably the least surprising to the average bear. 1000 */ 1001 if (wants_signal(sig, p)) 1002 t = p; 1003 else if ((type == PIDTYPE_PID) || thread_group_empty(p)) 1004 /* 1005 * There is just one thread and it does not need to be woken. 1006 * It will dequeue unblocked signals before it runs again. 1007 */ 1008 return; 1009 else { 1010 /* 1011 * Otherwise try to find a suitable thread. 1012 */ 1013 t = signal->curr_target; 1014 while (!wants_signal(sig, t)) { 1015 t = next_thread(t); 1016 if (t == signal->curr_target) 1017 /* 1018 * No thread needs to be woken. 1019 * Any eligible threads will see 1020 * the signal in the queue soon. 1021 */ 1022 return; 1023 } 1024 signal->curr_target = t; 1025 } 1026 1027 /* 1028 * Found a killable thread. If the signal will be fatal, 1029 * then start taking the whole group down immediately. 1030 */ 1031 if (sig_fatal(p, sig) && 1032 !(signal->flags & SIGNAL_GROUP_EXIT) && 1033 !sigismember(&t->real_blocked, sig) && 1034 (sig == SIGKILL || !p->ptrace)) { 1035 /* 1036 * This signal will be fatal to the whole group. 1037 */ 1038 if (!sig_kernel_coredump(sig)) { 1039 /* 1040 * Start a group exit and wake everybody up. 1041 * This way we don't have other threads 1042 * running and doing things after a slower 1043 * thread has the fatal signal pending. 1044 */ 1045 signal->flags = SIGNAL_GROUP_EXIT; 1046 signal->group_exit_code = sig; 1047 signal->group_stop_count = 0; 1048 t = p; 1049 do { 1050 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); 1051 sigaddset(&t->pending.signal, SIGKILL); 1052 signal_wake_up(t, 1); 1053 } while_each_thread(p, t); 1054 return; 1055 } 1056 } 1057 1058 /* 1059 * The signal is already in the shared-pending queue. 1060 * Tell the chosen thread to wake up and dequeue it. 1061 */ 1062 signal_wake_up(t, sig == SIGKILL); 1063 return; 1064 } 1065 1066 static inline bool legacy_queue(struct sigpending *signals, int sig) 1067 { 1068 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig); 1069 } 1070 1071 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t, 1072 enum pid_type type, bool force) 1073 { 1074 struct sigpending *pending; 1075 struct sigqueue *q; 1076 int override_rlimit; 1077 int ret = 0, result; 1078 1079 assert_spin_locked(&t->sighand->siglock); 1080 1081 result = TRACE_SIGNAL_IGNORED; 1082 if (!prepare_signal(sig, t, force)) 1083 goto ret; 1084 1085 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending; 1086 /* 1087 * Short-circuit ignored signals and support queuing 1088 * exactly one non-rt signal, so that we can get more 1089 * detailed information about the cause of the signal. 1090 */ 1091 result = TRACE_SIGNAL_ALREADY_PENDING; 1092 if (legacy_queue(pending, sig)) 1093 goto ret; 1094 1095 result = TRACE_SIGNAL_DELIVERED; 1096 /* 1097 * Skip useless siginfo allocation for SIGKILL and kernel threads. 1098 */ 1099 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD)) 1100 goto out_set; 1101 1102 /* 1103 * Real-time signals must be queued if sent by sigqueue, or 1104 * some other real-time mechanism. It is implementation 1105 * defined whether kill() does so. We attempt to do so, on 1106 * the principle of least surprise, but since kill is not 1107 * allowed to fail with EAGAIN when low on memory we just 1108 * make sure at least one signal gets delivered and don't 1109 * pass on the info struct. 1110 */ 1111 if (sig < SIGRTMIN) 1112 override_rlimit = (is_si_special(info) || info->si_code >= 0); 1113 else 1114 override_rlimit = 0; 1115 1116 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit); 1117 if (q) { 1118 list_add_tail(&q->list, &pending->list); 1119 switch ((unsigned long) info) { 1120 case (unsigned long) SEND_SIG_NOINFO: 1121 clear_siginfo(&q->info); 1122 q->info.si_signo = sig; 1123 q->info.si_errno = 0; 1124 q->info.si_code = SI_USER; 1125 q->info.si_pid = task_tgid_nr_ns(current, 1126 task_active_pid_ns(t)); 1127 rcu_read_lock(); 1128 q->info.si_uid = 1129 from_kuid_munged(task_cred_xxx(t, user_ns), 1130 current_uid()); 1131 rcu_read_unlock(); 1132 break; 1133 case (unsigned long) SEND_SIG_PRIV: 1134 clear_siginfo(&q->info); 1135 q->info.si_signo = sig; 1136 q->info.si_errno = 0; 1137 q->info.si_code = SI_KERNEL; 1138 q->info.si_pid = 0; 1139 q->info.si_uid = 0; 1140 break; 1141 default: 1142 copy_siginfo(&q->info, info); 1143 break; 1144 } 1145 } else if (!is_si_special(info) && 1146 sig >= SIGRTMIN && info->si_code != SI_USER) { 1147 /* 1148 * Queue overflow, abort. We may abort if the 1149 * signal was rt and sent by user using something 1150 * other than kill(). 1151 */ 1152 result = TRACE_SIGNAL_OVERFLOW_FAIL; 1153 ret = -EAGAIN; 1154 goto ret; 1155 } else { 1156 /* 1157 * This is a silent loss of information. We still 1158 * send the signal, but the *info bits are lost. 1159 */ 1160 result = TRACE_SIGNAL_LOSE_INFO; 1161 } 1162 1163 out_set: 1164 signalfd_notify(t, sig); 1165 sigaddset(&pending->signal, sig); 1166 1167 /* Let multiprocess signals appear after on-going forks */ 1168 if (type > PIDTYPE_TGID) { 1169 struct multiprocess_signals *delayed; 1170 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) { 1171 sigset_t *signal = &delayed->signal; 1172 /* Can't queue both a stop and a continue signal */ 1173 if (sig == SIGCONT) 1174 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK); 1175 else if (sig_kernel_stop(sig)) 1176 sigdelset(signal, SIGCONT); 1177 sigaddset(signal, sig); 1178 } 1179 } 1180 1181 complete_signal(sig, t, type); 1182 ret: 1183 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result); 1184 return ret; 1185 } 1186 1187 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info) 1188 { 1189 bool ret = false; 1190 switch (siginfo_layout(info->si_signo, info->si_code)) { 1191 case SIL_KILL: 1192 case SIL_CHLD: 1193 case SIL_RT: 1194 ret = true; 1195 break; 1196 case SIL_TIMER: 1197 case SIL_POLL: 1198 case SIL_FAULT: 1199 case SIL_FAULT_MCEERR: 1200 case SIL_FAULT_BNDERR: 1201 case SIL_FAULT_PKUERR: 1202 case SIL_SYS: 1203 ret = false; 1204 break; 1205 } 1206 return ret; 1207 } 1208 1209 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t, 1210 enum pid_type type) 1211 { 1212 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */ 1213 bool force = false; 1214 1215 if (info == SEND_SIG_NOINFO) { 1216 /* Force if sent from an ancestor pid namespace */ 1217 force = !task_pid_nr_ns(current, task_active_pid_ns(t)); 1218 } else if (info == SEND_SIG_PRIV) { 1219 /* Don't ignore kernel generated signals */ 1220 force = true; 1221 } else if (has_si_pid_and_uid(info)) { 1222 /* SIGKILL and SIGSTOP is special or has ids */ 1223 struct user_namespace *t_user_ns; 1224 1225 rcu_read_lock(); 1226 t_user_ns = task_cred_xxx(t, user_ns); 1227 if (current_user_ns() != t_user_ns) { 1228 kuid_t uid = make_kuid(current_user_ns(), info->si_uid); 1229 info->si_uid = from_kuid_munged(t_user_ns, uid); 1230 } 1231 rcu_read_unlock(); 1232 1233 /* A kernel generated signal? */ 1234 force = (info->si_code == SI_KERNEL); 1235 1236 /* From an ancestor pid namespace? */ 1237 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) { 1238 info->si_pid = 0; 1239 force = true; 1240 } 1241 } 1242 return __send_signal(sig, info, t, type, force); 1243 } 1244 1245 static void print_fatal_signal(int signr) 1246 { 1247 struct pt_regs *regs = signal_pt_regs(); 1248 pr_info("potentially unexpected fatal signal %d.\n", signr); 1249 1250 #if defined(__i386__) && !defined(__arch_um__) 1251 pr_info("code at %08lx: ", regs->ip); 1252 { 1253 int i; 1254 for (i = 0; i < 16; i++) { 1255 unsigned char insn; 1256 1257 if (get_user(insn, (unsigned char *)(regs->ip + i))) 1258 break; 1259 pr_cont("%02x ", insn); 1260 } 1261 } 1262 pr_cont("\n"); 1263 #endif 1264 preempt_disable(); 1265 show_regs(regs); 1266 preempt_enable(); 1267 } 1268 1269 static int __init setup_print_fatal_signals(char *str) 1270 { 1271 get_option (&str, &print_fatal_signals); 1272 1273 return 1; 1274 } 1275 1276 __setup("print-fatal-signals=", setup_print_fatal_signals); 1277 1278 int 1279 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p) 1280 { 1281 return send_signal(sig, info, p, PIDTYPE_TGID); 1282 } 1283 1284 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p, 1285 enum pid_type type) 1286 { 1287 unsigned long flags; 1288 int ret = -ESRCH; 1289 1290 if (lock_task_sighand(p, &flags)) { 1291 ret = send_signal(sig, info, p, type); 1292 unlock_task_sighand(p, &flags); 1293 } 1294 1295 return ret; 1296 } 1297 1298 /* 1299 * Force a signal that the process can't ignore: if necessary 1300 * we unblock the signal and change any SIG_IGN to SIG_DFL. 1301 * 1302 * Note: If we unblock the signal, we always reset it to SIG_DFL, 1303 * since we do not want to have a signal handler that was blocked 1304 * be invoked when user space had explicitly blocked it. 1305 * 1306 * We don't want to have recursive SIGSEGV's etc, for example, 1307 * that is why we also clear SIGNAL_UNKILLABLE. 1308 */ 1309 static int 1310 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t) 1311 { 1312 unsigned long int flags; 1313 int ret, blocked, ignored; 1314 struct k_sigaction *action; 1315 int sig = info->si_signo; 1316 1317 spin_lock_irqsave(&t->sighand->siglock, flags); 1318 action = &t->sighand->action[sig-1]; 1319 ignored = action->sa.sa_handler == SIG_IGN; 1320 blocked = sigismember(&t->blocked, sig); 1321 if (blocked || ignored) { 1322 action->sa.sa_handler = SIG_DFL; 1323 if (blocked) { 1324 sigdelset(&t->blocked, sig); 1325 recalc_sigpending_and_wake(t); 1326 } 1327 } 1328 /* 1329 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect 1330 * debugging to leave init killable. 1331 */ 1332 if (action->sa.sa_handler == SIG_DFL && !t->ptrace) 1333 t->signal->flags &= ~SIGNAL_UNKILLABLE; 1334 ret = send_signal(sig, info, t, PIDTYPE_PID); 1335 spin_unlock_irqrestore(&t->sighand->siglock, flags); 1336 1337 return ret; 1338 } 1339 1340 int force_sig_info(struct kernel_siginfo *info) 1341 { 1342 return force_sig_info_to_task(info, current); 1343 } 1344 1345 /* 1346 * Nuke all other threads in the group. 1347 */ 1348 int zap_other_threads(struct task_struct *p) 1349 { 1350 struct task_struct *t = p; 1351 int count = 0; 1352 1353 p->signal->group_stop_count = 0; 1354 1355 while_each_thread(p, t) { 1356 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); 1357 count++; 1358 1359 /* Don't bother with already dead threads */ 1360 if (t->exit_state) 1361 continue; 1362 sigaddset(&t->pending.signal, SIGKILL); 1363 signal_wake_up(t, 1); 1364 } 1365 1366 return count; 1367 } 1368 1369 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk, 1370 unsigned long *flags) 1371 { 1372 struct sighand_struct *sighand; 1373 1374 rcu_read_lock(); 1375 for (;;) { 1376 sighand = rcu_dereference(tsk->sighand); 1377 if (unlikely(sighand == NULL)) 1378 break; 1379 1380 /* 1381 * This sighand can be already freed and even reused, but 1382 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which 1383 * initializes ->siglock: this slab can't go away, it has 1384 * the same object type, ->siglock can't be reinitialized. 1385 * 1386 * We need to ensure that tsk->sighand is still the same 1387 * after we take the lock, we can race with de_thread() or 1388 * __exit_signal(). In the latter case the next iteration 1389 * must see ->sighand == NULL. 1390 */ 1391 spin_lock_irqsave(&sighand->siglock, *flags); 1392 if (likely(sighand == rcu_access_pointer(tsk->sighand))) 1393 break; 1394 spin_unlock_irqrestore(&sighand->siglock, *flags); 1395 } 1396 rcu_read_unlock(); 1397 1398 return sighand; 1399 } 1400 1401 /* 1402 * send signal info to all the members of a group 1403 */ 1404 int group_send_sig_info(int sig, struct kernel_siginfo *info, 1405 struct task_struct *p, enum pid_type type) 1406 { 1407 int ret; 1408 1409 rcu_read_lock(); 1410 ret = check_kill_permission(sig, info, p); 1411 rcu_read_unlock(); 1412 1413 if (!ret && sig) 1414 ret = do_send_sig_info(sig, info, p, type); 1415 1416 return ret; 1417 } 1418 1419 /* 1420 * __kill_pgrp_info() sends a signal to a process group: this is what the tty 1421 * control characters do (^C, ^Z etc) 1422 * - the caller must hold at least a readlock on tasklist_lock 1423 */ 1424 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp) 1425 { 1426 struct task_struct *p = NULL; 1427 int retval, success; 1428 1429 success = 0; 1430 retval = -ESRCH; 1431 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 1432 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID); 1433 success |= !err; 1434 retval = err; 1435 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 1436 return success ? 0 : retval; 1437 } 1438 1439 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid) 1440 { 1441 int error = -ESRCH; 1442 struct task_struct *p; 1443 1444 for (;;) { 1445 rcu_read_lock(); 1446 p = pid_task(pid, PIDTYPE_PID); 1447 if (p) 1448 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID); 1449 rcu_read_unlock(); 1450 if (likely(!p || error != -ESRCH)) 1451 return error; 1452 1453 /* 1454 * The task was unhashed in between, try again. If it 1455 * is dead, pid_task() will return NULL, if we race with 1456 * de_thread() it will find the new leader. 1457 */ 1458 } 1459 } 1460 1461 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid) 1462 { 1463 int error; 1464 rcu_read_lock(); 1465 error = kill_pid_info(sig, info, find_vpid(pid)); 1466 rcu_read_unlock(); 1467 return error; 1468 } 1469 1470 static inline bool kill_as_cred_perm(const struct cred *cred, 1471 struct task_struct *target) 1472 { 1473 const struct cred *pcred = __task_cred(target); 1474 1475 return uid_eq(cred->euid, pcred->suid) || 1476 uid_eq(cred->euid, pcred->uid) || 1477 uid_eq(cred->uid, pcred->suid) || 1478 uid_eq(cred->uid, pcred->uid); 1479 } 1480 1481 /* 1482 * The usb asyncio usage of siginfo is wrong. The glibc support 1483 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT. 1484 * AKA after the generic fields: 1485 * kernel_pid_t si_pid; 1486 * kernel_uid32_t si_uid; 1487 * sigval_t si_value; 1488 * 1489 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout 1490 * after the generic fields is: 1491 * void __user *si_addr; 1492 * 1493 * This is a practical problem when there is a 64bit big endian kernel 1494 * and a 32bit userspace. As the 32bit address will encoded in the low 1495 * 32bits of the pointer. Those low 32bits will be stored at higher 1496 * address than appear in a 32 bit pointer. So userspace will not 1497 * see the address it was expecting for it's completions. 1498 * 1499 * There is nothing in the encoding that can allow 1500 * copy_siginfo_to_user32 to detect this confusion of formats, so 1501 * handle this by requiring the caller of kill_pid_usb_asyncio to 1502 * notice when this situration takes place and to store the 32bit 1503 * pointer in sival_int, instead of sival_addr of the sigval_t addr 1504 * parameter. 1505 */ 1506 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr, 1507 struct pid *pid, const struct cred *cred) 1508 { 1509 struct kernel_siginfo info; 1510 struct task_struct *p; 1511 unsigned long flags; 1512 int ret = -EINVAL; 1513 1514 if (!valid_signal(sig)) 1515 return ret; 1516 1517 clear_siginfo(&info); 1518 info.si_signo = sig; 1519 info.si_errno = errno; 1520 info.si_code = SI_ASYNCIO; 1521 *((sigval_t *)&info.si_pid) = addr; 1522 1523 rcu_read_lock(); 1524 p = pid_task(pid, PIDTYPE_PID); 1525 if (!p) { 1526 ret = -ESRCH; 1527 goto out_unlock; 1528 } 1529 if (!kill_as_cred_perm(cred, p)) { 1530 ret = -EPERM; 1531 goto out_unlock; 1532 } 1533 ret = security_task_kill(p, &info, sig, cred); 1534 if (ret) 1535 goto out_unlock; 1536 1537 if (sig) { 1538 if (lock_task_sighand(p, &flags)) { 1539 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false); 1540 unlock_task_sighand(p, &flags); 1541 } else 1542 ret = -ESRCH; 1543 } 1544 out_unlock: 1545 rcu_read_unlock(); 1546 return ret; 1547 } 1548 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio); 1549 1550 /* 1551 * kill_something_info() interprets pid in interesting ways just like kill(2). 1552 * 1553 * POSIX specifies that kill(-1,sig) is unspecified, but what we have 1554 * is probably wrong. Should make it like BSD or SYSV. 1555 */ 1556 1557 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid) 1558 { 1559 int ret; 1560 1561 if (pid > 0) 1562 return kill_proc_info(sig, info, pid); 1563 1564 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */ 1565 if (pid == INT_MIN) 1566 return -ESRCH; 1567 1568 read_lock(&tasklist_lock); 1569 if (pid != -1) { 1570 ret = __kill_pgrp_info(sig, info, 1571 pid ? find_vpid(-pid) : task_pgrp(current)); 1572 } else { 1573 int retval = 0, count = 0; 1574 struct task_struct * p; 1575 1576 for_each_process(p) { 1577 if (task_pid_vnr(p) > 1 && 1578 !same_thread_group(p, current)) { 1579 int err = group_send_sig_info(sig, info, p, 1580 PIDTYPE_MAX); 1581 ++count; 1582 if (err != -EPERM) 1583 retval = err; 1584 } 1585 } 1586 ret = count ? retval : -ESRCH; 1587 } 1588 read_unlock(&tasklist_lock); 1589 1590 return ret; 1591 } 1592 1593 /* 1594 * These are for backward compatibility with the rest of the kernel source. 1595 */ 1596 1597 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p) 1598 { 1599 /* 1600 * Make sure legacy kernel users don't send in bad values 1601 * (normal paths check this in check_kill_permission). 1602 */ 1603 if (!valid_signal(sig)) 1604 return -EINVAL; 1605 1606 return do_send_sig_info(sig, info, p, PIDTYPE_PID); 1607 } 1608 EXPORT_SYMBOL(send_sig_info); 1609 1610 #define __si_special(priv) \ 1611 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO) 1612 1613 int 1614 send_sig(int sig, struct task_struct *p, int priv) 1615 { 1616 return send_sig_info(sig, __si_special(priv), p); 1617 } 1618 EXPORT_SYMBOL(send_sig); 1619 1620 void force_sig(int sig) 1621 { 1622 struct kernel_siginfo info; 1623 1624 clear_siginfo(&info); 1625 info.si_signo = sig; 1626 info.si_errno = 0; 1627 info.si_code = SI_KERNEL; 1628 info.si_pid = 0; 1629 info.si_uid = 0; 1630 force_sig_info(&info); 1631 } 1632 EXPORT_SYMBOL(force_sig); 1633 1634 /* 1635 * When things go south during signal handling, we 1636 * will force a SIGSEGV. And if the signal that caused 1637 * the problem was already a SIGSEGV, we'll want to 1638 * make sure we don't even try to deliver the signal.. 1639 */ 1640 void force_sigsegv(int sig) 1641 { 1642 struct task_struct *p = current; 1643 1644 if (sig == SIGSEGV) { 1645 unsigned long flags; 1646 spin_lock_irqsave(&p->sighand->siglock, flags); 1647 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL; 1648 spin_unlock_irqrestore(&p->sighand->siglock, flags); 1649 } 1650 force_sig(SIGSEGV); 1651 } 1652 1653 int force_sig_fault_to_task(int sig, int code, void __user *addr 1654 ___ARCH_SI_TRAPNO(int trapno) 1655 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr) 1656 , struct task_struct *t) 1657 { 1658 struct kernel_siginfo info; 1659 1660 clear_siginfo(&info); 1661 info.si_signo = sig; 1662 info.si_errno = 0; 1663 info.si_code = code; 1664 info.si_addr = addr; 1665 #ifdef __ARCH_SI_TRAPNO 1666 info.si_trapno = trapno; 1667 #endif 1668 #ifdef __ia64__ 1669 info.si_imm = imm; 1670 info.si_flags = flags; 1671 info.si_isr = isr; 1672 #endif 1673 return force_sig_info_to_task(&info, t); 1674 } 1675 1676 int force_sig_fault(int sig, int code, void __user *addr 1677 ___ARCH_SI_TRAPNO(int trapno) 1678 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)) 1679 { 1680 return force_sig_fault_to_task(sig, code, addr 1681 ___ARCH_SI_TRAPNO(trapno) 1682 ___ARCH_SI_IA64(imm, flags, isr), current); 1683 } 1684 1685 int send_sig_fault(int sig, int code, void __user *addr 1686 ___ARCH_SI_TRAPNO(int trapno) 1687 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr) 1688 , struct task_struct *t) 1689 { 1690 struct kernel_siginfo info; 1691 1692 clear_siginfo(&info); 1693 info.si_signo = sig; 1694 info.si_errno = 0; 1695 info.si_code = code; 1696 info.si_addr = addr; 1697 #ifdef __ARCH_SI_TRAPNO 1698 info.si_trapno = trapno; 1699 #endif 1700 #ifdef __ia64__ 1701 info.si_imm = imm; 1702 info.si_flags = flags; 1703 info.si_isr = isr; 1704 #endif 1705 return send_sig_info(info.si_signo, &info, t); 1706 } 1707 1708 int force_sig_mceerr(int code, void __user *addr, short lsb) 1709 { 1710 struct kernel_siginfo info; 1711 1712 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR)); 1713 clear_siginfo(&info); 1714 info.si_signo = SIGBUS; 1715 info.si_errno = 0; 1716 info.si_code = code; 1717 info.si_addr = addr; 1718 info.si_addr_lsb = lsb; 1719 return force_sig_info(&info); 1720 } 1721 1722 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t) 1723 { 1724 struct kernel_siginfo info; 1725 1726 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR)); 1727 clear_siginfo(&info); 1728 info.si_signo = SIGBUS; 1729 info.si_errno = 0; 1730 info.si_code = code; 1731 info.si_addr = addr; 1732 info.si_addr_lsb = lsb; 1733 return send_sig_info(info.si_signo, &info, t); 1734 } 1735 EXPORT_SYMBOL(send_sig_mceerr); 1736 1737 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper) 1738 { 1739 struct kernel_siginfo info; 1740 1741 clear_siginfo(&info); 1742 info.si_signo = SIGSEGV; 1743 info.si_errno = 0; 1744 info.si_code = SEGV_BNDERR; 1745 info.si_addr = addr; 1746 info.si_lower = lower; 1747 info.si_upper = upper; 1748 return force_sig_info(&info); 1749 } 1750 1751 #ifdef SEGV_PKUERR 1752 int force_sig_pkuerr(void __user *addr, u32 pkey) 1753 { 1754 struct kernel_siginfo info; 1755 1756 clear_siginfo(&info); 1757 info.si_signo = SIGSEGV; 1758 info.si_errno = 0; 1759 info.si_code = SEGV_PKUERR; 1760 info.si_addr = addr; 1761 info.si_pkey = pkey; 1762 return force_sig_info(&info); 1763 } 1764 #endif 1765 1766 /* For the crazy architectures that include trap information in 1767 * the errno field, instead of an actual errno value. 1768 */ 1769 int force_sig_ptrace_errno_trap(int errno, void __user *addr) 1770 { 1771 struct kernel_siginfo info; 1772 1773 clear_siginfo(&info); 1774 info.si_signo = SIGTRAP; 1775 info.si_errno = errno; 1776 info.si_code = TRAP_HWBKPT; 1777 info.si_addr = addr; 1778 return force_sig_info(&info); 1779 } 1780 1781 int kill_pgrp(struct pid *pid, int sig, int priv) 1782 { 1783 int ret; 1784 1785 read_lock(&tasklist_lock); 1786 ret = __kill_pgrp_info(sig, __si_special(priv), pid); 1787 read_unlock(&tasklist_lock); 1788 1789 return ret; 1790 } 1791 EXPORT_SYMBOL(kill_pgrp); 1792 1793 int kill_pid(struct pid *pid, int sig, int priv) 1794 { 1795 return kill_pid_info(sig, __si_special(priv), pid); 1796 } 1797 EXPORT_SYMBOL(kill_pid); 1798 1799 /* 1800 * These functions support sending signals using preallocated sigqueue 1801 * structures. This is needed "because realtime applications cannot 1802 * afford to lose notifications of asynchronous events, like timer 1803 * expirations or I/O completions". In the case of POSIX Timers 1804 * we allocate the sigqueue structure from the timer_create. If this 1805 * allocation fails we are able to report the failure to the application 1806 * with an EAGAIN error. 1807 */ 1808 struct sigqueue *sigqueue_alloc(void) 1809 { 1810 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0); 1811 1812 if (q) 1813 q->flags |= SIGQUEUE_PREALLOC; 1814 1815 return q; 1816 } 1817 1818 void sigqueue_free(struct sigqueue *q) 1819 { 1820 unsigned long flags; 1821 spinlock_t *lock = ¤t->sighand->siglock; 1822 1823 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC)); 1824 /* 1825 * We must hold ->siglock while testing q->list 1826 * to serialize with collect_signal() or with 1827 * __exit_signal()->flush_sigqueue(). 1828 */ 1829 spin_lock_irqsave(lock, flags); 1830 q->flags &= ~SIGQUEUE_PREALLOC; 1831 /* 1832 * If it is queued it will be freed when dequeued, 1833 * like the "regular" sigqueue. 1834 */ 1835 if (!list_empty(&q->list)) 1836 q = NULL; 1837 spin_unlock_irqrestore(lock, flags); 1838 1839 if (q) 1840 __sigqueue_free(q); 1841 } 1842 1843 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type) 1844 { 1845 int sig = q->info.si_signo; 1846 struct sigpending *pending; 1847 struct task_struct *t; 1848 unsigned long flags; 1849 int ret, result; 1850 1851 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC)); 1852 1853 ret = -1; 1854 rcu_read_lock(); 1855 t = pid_task(pid, type); 1856 if (!t || !likely(lock_task_sighand(t, &flags))) 1857 goto ret; 1858 1859 ret = 1; /* the signal is ignored */ 1860 result = TRACE_SIGNAL_IGNORED; 1861 if (!prepare_signal(sig, t, false)) 1862 goto out; 1863 1864 ret = 0; 1865 if (unlikely(!list_empty(&q->list))) { 1866 /* 1867 * If an SI_TIMER entry is already queue just increment 1868 * the overrun count. 1869 */ 1870 BUG_ON(q->info.si_code != SI_TIMER); 1871 q->info.si_overrun++; 1872 result = TRACE_SIGNAL_ALREADY_PENDING; 1873 goto out; 1874 } 1875 q->info.si_overrun = 0; 1876 1877 signalfd_notify(t, sig); 1878 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending; 1879 list_add_tail(&q->list, &pending->list); 1880 sigaddset(&pending->signal, sig); 1881 complete_signal(sig, t, type); 1882 result = TRACE_SIGNAL_DELIVERED; 1883 out: 1884 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result); 1885 unlock_task_sighand(t, &flags); 1886 ret: 1887 rcu_read_unlock(); 1888 return ret; 1889 } 1890 1891 static void do_notify_pidfd(struct task_struct *task) 1892 { 1893 struct pid *pid; 1894 1895 WARN_ON(task->exit_state == 0); 1896 pid = task_pid(task); 1897 wake_up_all(&pid->wait_pidfd); 1898 } 1899 1900 /* 1901 * Let a parent know about the death of a child. 1902 * For a stopped/continued status change, use do_notify_parent_cldstop instead. 1903 * 1904 * Returns true if our parent ignored us and so we've switched to 1905 * self-reaping. 1906 */ 1907 bool do_notify_parent(struct task_struct *tsk, int sig) 1908 { 1909 struct kernel_siginfo info; 1910 unsigned long flags; 1911 struct sighand_struct *psig; 1912 bool autoreap = false; 1913 u64 utime, stime; 1914 1915 BUG_ON(sig == -1); 1916 1917 /* do_notify_parent_cldstop should have been called instead. */ 1918 BUG_ON(task_is_stopped_or_traced(tsk)); 1919 1920 BUG_ON(!tsk->ptrace && 1921 (tsk->group_leader != tsk || !thread_group_empty(tsk))); 1922 1923 /* Wake up all pidfd waiters */ 1924 do_notify_pidfd(tsk); 1925 1926 if (sig != SIGCHLD) { 1927 /* 1928 * This is only possible if parent == real_parent. 1929 * Check if it has changed security domain. 1930 */ 1931 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id)) 1932 sig = SIGCHLD; 1933 } 1934 1935 clear_siginfo(&info); 1936 info.si_signo = sig; 1937 info.si_errno = 0; 1938 /* 1939 * We are under tasklist_lock here so our parent is tied to 1940 * us and cannot change. 1941 * 1942 * task_active_pid_ns will always return the same pid namespace 1943 * until a task passes through release_task. 1944 * 1945 * write_lock() currently calls preempt_disable() which is the 1946 * same as rcu_read_lock(), but according to Oleg, this is not 1947 * correct to rely on this 1948 */ 1949 rcu_read_lock(); 1950 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent)); 1951 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns), 1952 task_uid(tsk)); 1953 rcu_read_unlock(); 1954 1955 task_cputime(tsk, &utime, &stime); 1956 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime); 1957 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime); 1958 1959 info.si_status = tsk->exit_code & 0x7f; 1960 if (tsk->exit_code & 0x80) 1961 info.si_code = CLD_DUMPED; 1962 else if (tsk->exit_code & 0x7f) 1963 info.si_code = CLD_KILLED; 1964 else { 1965 info.si_code = CLD_EXITED; 1966 info.si_status = tsk->exit_code >> 8; 1967 } 1968 1969 psig = tsk->parent->sighand; 1970 spin_lock_irqsave(&psig->siglock, flags); 1971 if (!tsk->ptrace && sig == SIGCHLD && 1972 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN || 1973 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) { 1974 /* 1975 * We are exiting and our parent doesn't care. POSIX.1 1976 * defines special semantics for setting SIGCHLD to SIG_IGN 1977 * or setting the SA_NOCLDWAIT flag: we should be reaped 1978 * automatically and not left for our parent's wait4 call. 1979 * Rather than having the parent do it as a magic kind of 1980 * signal handler, we just set this to tell do_exit that we 1981 * can be cleaned up without becoming a zombie. Note that 1982 * we still call __wake_up_parent in this case, because a 1983 * blocked sys_wait4 might now return -ECHILD. 1984 * 1985 * Whether we send SIGCHLD or not for SA_NOCLDWAIT 1986 * is implementation-defined: we do (if you don't want 1987 * it, just use SIG_IGN instead). 1988 */ 1989 autoreap = true; 1990 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) 1991 sig = 0; 1992 } 1993 /* 1994 * Send with __send_signal as si_pid and si_uid are in the 1995 * parent's namespaces. 1996 */ 1997 if (valid_signal(sig) && sig) 1998 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false); 1999 __wake_up_parent(tsk, tsk->parent); 2000 spin_unlock_irqrestore(&psig->siglock, flags); 2001 2002 return autoreap; 2003 } 2004 2005 /** 2006 * do_notify_parent_cldstop - notify parent of stopped/continued state change 2007 * @tsk: task reporting the state change 2008 * @for_ptracer: the notification is for ptracer 2009 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report 2010 * 2011 * Notify @tsk's parent that the stopped/continued state has changed. If 2012 * @for_ptracer is %false, @tsk's group leader notifies to its real parent. 2013 * If %true, @tsk reports to @tsk->parent which should be the ptracer. 2014 * 2015 * CONTEXT: 2016 * Must be called with tasklist_lock at least read locked. 2017 */ 2018 static void do_notify_parent_cldstop(struct task_struct *tsk, 2019 bool for_ptracer, int why) 2020 { 2021 struct kernel_siginfo info; 2022 unsigned long flags; 2023 struct task_struct *parent; 2024 struct sighand_struct *sighand; 2025 u64 utime, stime; 2026 2027 if (for_ptracer) { 2028 parent = tsk->parent; 2029 } else { 2030 tsk = tsk->group_leader; 2031 parent = tsk->real_parent; 2032 } 2033 2034 clear_siginfo(&info); 2035 info.si_signo = SIGCHLD; 2036 info.si_errno = 0; 2037 /* 2038 * see comment in do_notify_parent() about the following 4 lines 2039 */ 2040 rcu_read_lock(); 2041 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent)); 2042 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk)); 2043 rcu_read_unlock(); 2044 2045 task_cputime(tsk, &utime, &stime); 2046 info.si_utime = nsec_to_clock_t(utime); 2047 info.si_stime = nsec_to_clock_t(stime); 2048 2049 info.si_code = why; 2050 switch (why) { 2051 case CLD_CONTINUED: 2052 info.si_status = SIGCONT; 2053 break; 2054 case CLD_STOPPED: 2055 info.si_status = tsk->signal->group_exit_code & 0x7f; 2056 break; 2057 case CLD_TRAPPED: 2058 info.si_status = tsk->exit_code & 0x7f; 2059 break; 2060 default: 2061 BUG(); 2062 } 2063 2064 sighand = parent->sighand; 2065 spin_lock_irqsave(&sighand->siglock, flags); 2066 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN && 2067 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP)) 2068 __group_send_sig_info(SIGCHLD, &info, parent); 2069 /* 2070 * Even if SIGCHLD is not generated, we must wake up wait4 calls. 2071 */ 2072 __wake_up_parent(tsk, parent); 2073 spin_unlock_irqrestore(&sighand->siglock, flags); 2074 } 2075 2076 static inline bool may_ptrace_stop(void) 2077 { 2078 if (!likely(current->ptrace)) 2079 return false; 2080 /* 2081 * Are we in the middle of do_coredump? 2082 * If so and our tracer is also part of the coredump stopping 2083 * is a deadlock situation, and pointless because our tracer 2084 * is dead so don't allow us to stop. 2085 * If SIGKILL was already sent before the caller unlocked 2086 * ->siglock we must see ->core_state != NULL. Otherwise it 2087 * is safe to enter schedule(). 2088 * 2089 * This is almost outdated, a task with the pending SIGKILL can't 2090 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported 2091 * after SIGKILL was already dequeued. 2092 */ 2093 if (unlikely(current->mm->core_state) && 2094 unlikely(current->mm == current->parent->mm)) 2095 return false; 2096 2097 return true; 2098 } 2099 2100 /* 2101 * Return non-zero if there is a SIGKILL that should be waking us up. 2102 * Called with the siglock held. 2103 */ 2104 static bool sigkill_pending(struct task_struct *tsk) 2105 { 2106 return sigismember(&tsk->pending.signal, SIGKILL) || 2107 sigismember(&tsk->signal->shared_pending.signal, SIGKILL); 2108 } 2109 2110 /* 2111 * This must be called with current->sighand->siglock held. 2112 * 2113 * This should be the path for all ptrace stops. 2114 * We always set current->last_siginfo while stopped here. 2115 * That makes it a way to test a stopped process for 2116 * being ptrace-stopped vs being job-control-stopped. 2117 * 2118 * If we actually decide not to stop at all because the tracer 2119 * is gone, we keep current->exit_code unless clear_code. 2120 */ 2121 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info) 2122 __releases(¤t->sighand->siglock) 2123 __acquires(¤t->sighand->siglock) 2124 { 2125 bool gstop_done = false; 2126 2127 if (arch_ptrace_stop_needed(exit_code, info)) { 2128 /* 2129 * The arch code has something special to do before a 2130 * ptrace stop. This is allowed to block, e.g. for faults 2131 * on user stack pages. We can't keep the siglock while 2132 * calling arch_ptrace_stop, so we must release it now. 2133 * To preserve proper semantics, we must do this before 2134 * any signal bookkeeping like checking group_stop_count. 2135 * Meanwhile, a SIGKILL could come in before we retake the 2136 * siglock. That must prevent us from sleeping in TASK_TRACED. 2137 * So after regaining the lock, we must check for SIGKILL. 2138 */ 2139 spin_unlock_irq(¤t->sighand->siglock); 2140 arch_ptrace_stop(exit_code, info); 2141 spin_lock_irq(¤t->sighand->siglock); 2142 if (sigkill_pending(current)) 2143 return; 2144 } 2145 2146 set_special_state(TASK_TRACED); 2147 2148 /* 2149 * We're committing to trapping. TRACED should be visible before 2150 * TRAPPING is cleared; otherwise, the tracer might fail do_wait(). 2151 * Also, transition to TRACED and updates to ->jobctl should be 2152 * atomic with respect to siglock and should be done after the arch 2153 * hook as siglock is released and regrabbed across it. 2154 * 2155 * TRACER TRACEE 2156 * 2157 * ptrace_attach() 2158 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED) 2159 * do_wait() 2160 * set_current_state() smp_wmb(); 2161 * ptrace_do_wait() 2162 * wait_task_stopped() 2163 * task_stopped_code() 2164 * [L] task_is_traced() [S] task_clear_jobctl_trapping(); 2165 */ 2166 smp_wmb(); 2167 2168 current->last_siginfo = info; 2169 current->exit_code = exit_code; 2170 2171 /* 2172 * If @why is CLD_STOPPED, we're trapping to participate in a group 2173 * stop. Do the bookkeeping. Note that if SIGCONT was delievered 2174 * across siglock relocks since INTERRUPT was scheduled, PENDING 2175 * could be clear now. We act as if SIGCONT is received after 2176 * TASK_TRACED is entered - ignore it. 2177 */ 2178 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING)) 2179 gstop_done = task_participate_group_stop(current); 2180 2181 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */ 2182 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP); 2183 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP) 2184 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY); 2185 2186 /* entering a trap, clear TRAPPING */ 2187 task_clear_jobctl_trapping(current); 2188 2189 spin_unlock_irq(¤t->sighand->siglock); 2190 read_lock(&tasklist_lock); 2191 if (may_ptrace_stop()) { 2192 /* 2193 * Notify parents of the stop. 2194 * 2195 * While ptraced, there are two parents - the ptracer and 2196 * the real_parent of the group_leader. The ptracer should 2197 * know about every stop while the real parent is only 2198 * interested in the completion of group stop. The states 2199 * for the two don't interact with each other. Notify 2200 * separately unless they're gonna be duplicates. 2201 */ 2202 do_notify_parent_cldstop(current, true, why); 2203 if (gstop_done && ptrace_reparented(current)) 2204 do_notify_parent_cldstop(current, false, why); 2205 2206 /* 2207 * Don't want to allow preemption here, because 2208 * sys_ptrace() needs this task to be inactive. 2209 * 2210 * XXX: implement read_unlock_no_resched(). 2211 */ 2212 preempt_disable(); 2213 read_unlock(&tasklist_lock); 2214 cgroup_enter_frozen(); 2215 preempt_enable_no_resched(); 2216 freezable_schedule(); 2217 cgroup_leave_frozen(true); 2218 } else { 2219 /* 2220 * By the time we got the lock, our tracer went away. 2221 * Don't drop the lock yet, another tracer may come. 2222 * 2223 * If @gstop_done, the ptracer went away between group stop 2224 * completion and here. During detach, it would have set 2225 * JOBCTL_STOP_PENDING on us and we'll re-enter 2226 * TASK_STOPPED in do_signal_stop() on return, so notifying 2227 * the real parent of the group stop completion is enough. 2228 */ 2229 if (gstop_done) 2230 do_notify_parent_cldstop(current, false, why); 2231 2232 /* tasklist protects us from ptrace_freeze_traced() */ 2233 __set_current_state(TASK_RUNNING); 2234 if (clear_code) 2235 current->exit_code = 0; 2236 read_unlock(&tasklist_lock); 2237 } 2238 2239 /* 2240 * We are back. Now reacquire the siglock before touching 2241 * last_siginfo, so that we are sure to have synchronized with 2242 * any signal-sending on another CPU that wants to examine it. 2243 */ 2244 spin_lock_irq(¤t->sighand->siglock); 2245 current->last_siginfo = NULL; 2246 2247 /* LISTENING can be set only during STOP traps, clear it */ 2248 current->jobctl &= ~JOBCTL_LISTENING; 2249 2250 /* 2251 * Queued signals ignored us while we were stopped for tracing. 2252 * So check for any that we should take before resuming user mode. 2253 * This sets TIF_SIGPENDING, but never clears it. 2254 */ 2255 recalc_sigpending_tsk(current); 2256 } 2257 2258 static void ptrace_do_notify(int signr, int exit_code, int why) 2259 { 2260 kernel_siginfo_t info; 2261 2262 clear_siginfo(&info); 2263 info.si_signo = signr; 2264 info.si_code = exit_code; 2265 info.si_pid = task_pid_vnr(current); 2266 info.si_uid = from_kuid_munged(current_user_ns(), current_uid()); 2267 2268 /* Let the debugger run. */ 2269 ptrace_stop(exit_code, why, 1, &info); 2270 } 2271 2272 void ptrace_notify(int exit_code) 2273 { 2274 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP); 2275 if (unlikely(current->task_works)) 2276 task_work_run(); 2277 2278 spin_lock_irq(¤t->sighand->siglock); 2279 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED); 2280 spin_unlock_irq(¤t->sighand->siglock); 2281 } 2282 2283 /** 2284 * do_signal_stop - handle group stop for SIGSTOP and other stop signals 2285 * @signr: signr causing group stop if initiating 2286 * 2287 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr 2288 * and participate in it. If already set, participate in the existing 2289 * group stop. If participated in a group stop (and thus slept), %true is 2290 * returned with siglock released. 2291 * 2292 * If ptraced, this function doesn't handle stop itself. Instead, 2293 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock 2294 * untouched. The caller must ensure that INTERRUPT trap handling takes 2295 * places afterwards. 2296 * 2297 * CONTEXT: 2298 * Must be called with @current->sighand->siglock held, which is released 2299 * on %true return. 2300 * 2301 * RETURNS: 2302 * %false if group stop is already cancelled or ptrace trap is scheduled. 2303 * %true if participated in group stop. 2304 */ 2305 static bool do_signal_stop(int signr) 2306 __releases(¤t->sighand->siglock) 2307 { 2308 struct signal_struct *sig = current->signal; 2309 2310 if (!(current->jobctl & JOBCTL_STOP_PENDING)) { 2311 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME; 2312 struct task_struct *t; 2313 2314 /* signr will be recorded in task->jobctl for retries */ 2315 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK); 2316 2317 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) || 2318 unlikely(signal_group_exit(sig))) 2319 return false; 2320 /* 2321 * There is no group stop already in progress. We must 2322 * initiate one now. 2323 * 2324 * While ptraced, a task may be resumed while group stop is 2325 * still in effect and then receive a stop signal and 2326 * initiate another group stop. This deviates from the 2327 * usual behavior as two consecutive stop signals can't 2328 * cause two group stops when !ptraced. That is why we 2329 * also check !task_is_stopped(t) below. 2330 * 2331 * The condition can be distinguished by testing whether 2332 * SIGNAL_STOP_STOPPED is already set. Don't generate 2333 * group_exit_code in such case. 2334 * 2335 * This is not necessary for SIGNAL_STOP_CONTINUED because 2336 * an intervening stop signal is required to cause two 2337 * continued events regardless of ptrace. 2338 */ 2339 if (!(sig->flags & SIGNAL_STOP_STOPPED)) 2340 sig->group_exit_code = signr; 2341 2342 sig->group_stop_count = 0; 2343 2344 if (task_set_jobctl_pending(current, signr | gstop)) 2345 sig->group_stop_count++; 2346 2347 t = current; 2348 while_each_thread(current, t) { 2349 /* 2350 * Setting state to TASK_STOPPED for a group 2351 * stop is always done with the siglock held, 2352 * so this check has no races. 2353 */ 2354 if (!task_is_stopped(t) && 2355 task_set_jobctl_pending(t, signr | gstop)) { 2356 sig->group_stop_count++; 2357 if (likely(!(t->ptrace & PT_SEIZED))) 2358 signal_wake_up(t, 0); 2359 else 2360 ptrace_trap_notify(t); 2361 } 2362 } 2363 } 2364 2365 if (likely(!current->ptrace)) { 2366 int notify = 0; 2367 2368 /* 2369 * If there are no other threads in the group, or if there 2370 * is a group stop in progress and we are the last to stop, 2371 * report to the parent. 2372 */ 2373 if (task_participate_group_stop(current)) 2374 notify = CLD_STOPPED; 2375 2376 set_special_state(TASK_STOPPED); 2377 spin_unlock_irq(¤t->sighand->siglock); 2378 2379 /* 2380 * Notify the parent of the group stop completion. Because 2381 * we're not holding either the siglock or tasklist_lock 2382 * here, ptracer may attach inbetween; however, this is for 2383 * group stop and should always be delivered to the real 2384 * parent of the group leader. The new ptracer will get 2385 * its notification when this task transitions into 2386 * TASK_TRACED. 2387 */ 2388 if (notify) { 2389 read_lock(&tasklist_lock); 2390 do_notify_parent_cldstop(current, false, notify); 2391 read_unlock(&tasklist_lock); 2392 } 2393 2394 /* Now we don't run again until woken by SIGCONT or SIGKILL */ 2395 cgroup_enter_frozen(); 2396 freezable_schedule(); 2397 return true; 2398 } else { 2399 /* 2400 * While ptraced, group stop is handled by STOP trap. 2401 * Schedule it and let the caller deal with it. 2402 */ 2403 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP); 2404 return false; 2405 } 2406 } 2407 2408 /** 2409 * do_jobctl_trap - take care of ptrace jobctl traps 2410 * 2411 * When PT_SEIZED, it's used for both group stop and explicit 2412 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with 2413 * accompanying siginfo. If stopped, lower eight bits of exit_code contain 2414 * the stop signal; otherwise, %SIGTRAP. 2415 * 2416 * When !PT_SEIZED, it's used only for group stop trap with stop signal 2417 * number as exit_code and no siginfo. 2418 * 2419 * CONTEXT: 2420 * Must be called with @current->sighand->siglock held, which may be 2421 * released and re-acquired before returning with intervening sleep. 2422 */ 2423 static void do_jobctl_trap(void) 2424 { 2425 struct signal_struct *signal = current->signal; 2426 int signr = current->jobctl & JOBCTL_STOP_SIGMASK; 2427 2428 if (current->ptrace & PT_SEIZED) { 2429 if (!signal->group_stop_count && 2430 !(signal->flags & SIGNAL_STOP_STOPPED)) 2431 signr = SIGTRAP; 2432 WARN_ON_ONCE(!signr); 2433 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8), 2434 CLD_STOPPED); 2435 } else { 2436 WARN_ON_ONCE(!signr); 2437 ptrace_stop(signr, CLD_STOPPED, 0, NULL); 2438 current->exit_code = 0; 2439 } 2440 } 2441 2442 /** 2443 * do_freezer_trap - handle the freezer jobctl trap 2444 * 2445 * Puts the task into frozen state, if only the task is not about to quit. 2446 * In this case it drops JOBCTL_TRAP_FREEZE. 2447 * 2448 * CONTEXT: 2449 * Must be called with @current->sighand->siglock held, 2450 * which is always released before returning. 2451 */ 2452 static void do_freezer_trap(void) 2453 __releases(¤t->sighand->siglock) 2454 { 2455 /* 2456 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE, 2457 * let's make another loop to give it a chance to be handled. 2458 * In any case, we'll return back. 2459 */ 2460 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) != 2461 JOBCTL_TRAP_FREEZE) { 2462 spin_unlock_irq(¤t->sighand->siglock); 2463 return; 2464 } 2465 2466 /* 2467 * Now we're sure that there is no pending fatal signal and no 2468 * pending traps. Clear TIF_SIGPENDING to not get out of schedule() 2469 * immediately (if there is a non-fatal signal pending), and 2470 * put the task into sleep. 2471 */ 2472 __set_current_state(TASK_INTERRUPTIBLE); 2473 clear_thread_flag(TIF_SIGPENDING); 2474 spin_unlock_irq(¤t->sighand->siglock); 2475 cgroup_enter_frozen(); 2476 freezable_schedule(); 2477 } 2478 2479 static int ptrace_signal(int signr, kernel_siginfo_t *info) 2480 { 2481 /* 2482 * We do not check sig_kernel_stop(signr) but set this marker 2483 * unconditionally because we do not know whether debugger will 2484 * change signr. This flag has no meaning unless we are going 2485 * to stop after return from ptrace_stop(). In this case it will 2486 * be checked in do_signal_stop(), we should only stop if it was 2487 * not cleared by SIGCONT while we were sleeping. See also the 2488 * comment in dequeue_signal(). 2489 */ 2490 current->jobctl |= JOBCTL_STOP_DEQUEUED; 2491 ptrace_stop(signr, CLD_TRAPPED, 0, info); 2492 2493 /* We're back. Did the debugger cancel the sig? */ 2494 signr = current->exit_code; 2495 if (signr == 0) 2496 return signr; 2497 2498 current->exit_code = 0; 2499 2500 /* 2501 * Update the siginfo structure if the signal has 2502 * changed. If the debugger wanted something 2503 * specific in the siginfo structure then it should 2504 * have updated *info via PTRACE_SETSIGINFO. 2505 */ 2506 if (signr != info->si_signo) { 2507 clear_siginfo(info); 2508 info->si_signo = signr; 2509 info->si_errno = 0; 2510 info->si_code = SI_USER; 2511 rcu_read_lock(); 2512 info->si_pid = task_pid_vnr(current->parent); 2513 info->si_uid = from_kuid_munged(current_user_ns(), 2514 task_uid(current->parent)); 2515 rcu_read_unlock(); 2516 } 2517 2518 /* If the (new) signal is now blocked, requeue it. */ 2519 if (sigismember(¤t->blocked, signr)) { 2520 send_signal(signr, info, current, PIDTYPE_PID); 2521 signr = 0; 2522 } 2523 2524 return signr; 2525 } 2526 2527 static void hide_si_addr_tag_bits(struct ksignal *ksig) 2528 { 2529 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) { 2530 case SIL_FAULT: 2531 case SIL_FAULT_MCEERR: 2532 case SIL_FAULT_BNDERR: 2533 case SIL_FAULT_PKUERR: 2534 ksig->info.si_addr = arch_untagged_si_addr( 2535 ksig->info.si_addr, ksig->sig, ksig->info.si_code); 2536 break; 2537 case SIL_KILL: 2538 case SIL_TIMER: 2539 case SIL_POLL: 2540 case SIL_CHLD: 2541 case SIL_RT: 2542 case SIL_SYS: 2543 break; 2544 } 2545 } 2546 2547 bool get_signal(struct ksignal *ksig) 2548 { 2549 struct sighand_struct *sighand = current->sighand; 2550 struct signal_struct *signal = current->signal; 2551 int signr; 2552 2553 /* 2554 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so 2555 * that the arch handlers don't all have to do it. If we get here 2556 * without TIF_SIGPENDING, just exit after running signal work. 2557 */ 2558 #ifdef TIF_NOTIFY_SIGNAL 2559 if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) { 2560 if (test_thread_flag(TIF_NOTIFY_SIGNAL)) 2561 tracehook_notify_signal(); 2562 if (!task_sigpending(current)) 2563 return false; 2564 } 2565 #endif 2566 2567 if (unlikely(uprobe_deny_signal())) 2568 return false; 2569 2570 /* 2571 * Do this once, we can't return to user-mode if freezing() == T. 2572 * do_signal_stop() and ptrace_stop() do freezable_schedule() and 2573 * thus do not need another check after return. 2574 */ 2575 try_to_freeze(); 2576 2577 relock: 2578 spin_lock_irq(&sighand->siglock); 2579 /* 2580 * Make sure we can safely read ->jobctl() in task_work add. As Oleg 2581 * states: 2582 * 2583 * It pairs with mb (implied by cmpxchg) before READ_ONCE. So we 2584 * roughly have 2585 * 2586 * task_work_add: get_signal: 2587 * STORE(task->task_works, new_work); STORE(task->jobctl); 2588 * mb(); mb(); 2589 * LOAD(task->jobctl); LOAD(task->task_works); 2590 * 2591 * and we can rely on STORE-MB-LOAD [ in task_work_add]. 2592 */ 2593 smp_store_mb(current->jobctl, current->jobctl & ~JOBCTL_TASK_WORK); 2594 if (unlikely(current->task_works)) { 2595 spin_unlock_irq(&sighand->siglock); 2596 task_work_run(); 2597 goto relock; 2598 } 2599 2600 /* 2601 * Every stopped thread goes here after wakeup. Check to see if 2602 * we should notify the parent, prepare_signal(SIGCONT) encodes 2603 * the CLD_ si_code into SIGNAL_CLD_MASK bits. 2604 */ 2605 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) { 2606 int why; 2607 2608 if (signal->flags & SIGNAL_CLD_CONTINUED) 2609 why = CLD_CONTINUED; 2610 else 2611 why = CLD_STOPPED; 2612 2613 signal->flags &= ~SIGNAL_CLD_MASK; 2614 2615 spin_unlock_irq(&sighand->siglock); 2616 2617 /* 2618 * Notify the parent that we're continuing. This event is 2619 * always per-process and doesn't make whole lot of sense 2620 * for ptracers, who shouldn't consume the state via 2621 * wait(2) either, but, for backward compatibility, notify 2622 * the ptracer of the group leader too unless it's gonna be 2623 * a duplicate. 2624 */ 2625 read_lock(&tasklist_lock); 2626 do_notify_parent_cldstop(current, false, why); 2627 2628 if (ptrace_reparented(current->group_leader)) 2629 do_notify_parent_cldstop(current->group_leader, 2630 true, why); 2631 read_unlock(&tasklist_lock); 2632 2633 goto relock; 2634 } 2635 2636 /* Has this task already been marked for death? */ 2637 if (signal_group_exit(signal)) { 2638 ksig->info.si_signo = signr = SIGKILL; 2639 sigdelset(¤t->pending.signal, SIGKILL); 2640 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO, 2641 &sighand->action[SIGKILL - 1]); 2642 recalc_sigpending(); 2643 goto fatal; 2644 } 2645 2646 for (;;) { 2647 struct k_sigaction *ka; 2648 2649 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) && 2650 do_signal_stop(0)) 2651 goto relock; 2652 2653 if (unlikely(current->jobctl & 2654 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) { 2655 if (current->jobctl & JOBCTL_TRAP_MASK) { 2656 do_jobctl_trap(); 2657 spin_unlock_irq(&sighand->siglock); 2658 } else if (current->jobctl & JOBCTL_TRAP_FREEZE) 2659 do_freezer_trap(); 2660 2661 goto relock; 2662 } 2663 2664 /* 2665 * If the task is leaving the frozen state, let's update 2666 * cgroup counters and reset the frozen bit. 2667 */ 2668 if (unlikely(cgroup_task_frozen(current))) { 2669 spin_unlock_irq(&sighand->siglock); 2670 cgroup_leave_frozen(false); 2671 goto relock; 2672 } 2673 2674 /* 2675 * Signals generated by the execution of an instruction 2676 * need to be delivered before any other pending signals 2677 * so that the instruction pointer in the signal stack 2678 * frame points to the faulting instruction. 2679 */ 2680 signr = dequeue_synchronous_signal(&ksig->info); 2681 if (!signr) 2682 signr = dequeue_signal(current, ¤t->blocked, &ksig->info); 2683 2684 if (!signr) 2685 break; /* will return 0 */ 2686 2687 if (unlikely(current->ptrace) && signr != SIGKILL) { 2688 signr = ptrace_signal(signr, &ksig->info); 2689 if (!signr) 2690 continue; 2691 } 2692 2693 ka = &sighand->action[signr-1]; 2694 2695 /* Trace actually delivered signals. */ 2696 trace_signal_deliver(signr, &ksig->info, ka); 2697 2698 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */ 2699 continue; 2700 if (ka->sa.sa_handler != SIG_DFL) { 2701 /* Run the handler. */ 2702 ksig->ka = *ka; 2703 2704 if (ka->sa.sa_flags & SA_ONESHOT) 2705 ka->sa.sa_handler = SIG_DFL; 2706 2707 break; /* will return non-zero "signr" value */ 2708 } 2709 2710 /* 2711 * Now we are doing the default action for this signal. 2712 */ 2713 if (sig_kernel_ignore(signr)) /* Default is nothing. */ 2714 continue; 2715 2716 /* 2717 * Global init gets no signals it doesn't want. 2718 * Container-init gets no signals it doesn't want from same 2719 * container. 2720 * 2721 * Note that if global/container-init sees a sig_kernel_only() 2722 * signal here, the signal must have been generated internally 2723 * or must have come from an ancestor namespace. In either 2724 * case, the signal cannot be dropped. 2725 */ 2726 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) && 2727 !sig_kernel_only(signr)) 2728 continue; 2729 2730 if (sig_kernel_stop(signr)) { 2731 /* 2732 * The default action is to stop all threads in 2733 * the thread group. The job control signals 2734 * do nothing in an orphaned pgrp, but SIGSTOP 2735 * always works. Note that siglock needs to be 2736 * dropped during the call to is_orphaned_pgrp() 2737 * because of lock ordering with tasklist_lock. 2738 * This allows an intervening SIGCONT to be posted. 2739 * We need to check for that and bail out if necessary. 2740 */ 2741 if (signr != SIGSTOP) { 2742 spin_unlock_irq(&sighand->siglock); 2743 2744 /* signals can be posted during this window */ 2745 2746 if (is_current_pgrp_orphaned()) 2747 goto relock; 2748 2749 spin_lock_irq(&sighand->siglock); 2750 } 2751 2752 if (likely(do_signal_stop(ksig->info.si_signo))) { 2753 /* It released the siglock. */ 2754 goto relock; 2755 } 2756 2757 /* 2758 * We didn't actually stop, due to a race 2759 * with SIGCONT or something like that. 2760 */ 2761 continue; 2762 } 2763 2764 fatal: 2765 spin_unlock_irq(&sighand->siglock); 2766 if (unlikely(cgroup_task_frozen(current))) 2767 cgroup_leave_frozen(true); 2768 2769 /* 2770 * Anything else is fatal, maybe with a core dump. 2771 */ 2772 current->flags |= PF_SIGNALED; 2773 2774 if (sig_kernel_coredump(signr)) { 2775 if (print_fatal_signals) 2776 print_fatal_signal(ksig->info.si_signo); 2777 proc_coredump_connector(current); 2778 /* 2779 * If it was able to dump core, this kills all 2780 * other threads in the group and synchronizes with 2781 * their demise. If we lost the race with another 2782 * thread getting here, it set group_exit_code 2783 * first and our do_group_exit call below will use 2784 * that value and ignore the one we pass it. 2785 */ 2786 do_coredump(&ksig->info); 2787 } 2788 2789 /* 2790 * Death signals, no core dump. 2791 */ 2792 do_group_exit(ksig->info.si_signo); 2793 /* NOTREACHED */ 2794 } 2795 spin_unlock_irq(&sighand->siglock); 2796 2797 ksig->sig = signr; 2798 2799 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS)) 2800 hide_si_addr_tag_bits(ksig); 2801 2802 return ksig->sig > 0; 2803 } 2804 2805 /** 2806 * signal_delivered - 2807 * @ksig: kernel signal struct 2808 * @stepping: nonzero if debugger single-step or block-step in use 2809 * 2810 * This function should be called when a signal has successfully been 2811 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask 2812 * is always blocked, and the signal itself is blocked unless %SA_NODEFER 2813 * is set in @ksig->ka.sa.sa_flags. Tracing is notified. 2814 */ 2815 static void signal_delivered(struct ksignal *ksig, int stepping) 2816 { 2817 sigset_t blocked; 2818 2819 /* A signal was successfully delivered, and the 2820 saved sigmask was stored on the signal frame, 2821 and will be restored by sigreturn. So we can 2822 simply clear the restore sigmask flag. */ 2823 clear_restore_sigmask(); 2824 2825 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask); 2826 if (!(ksig->ka.sa.sa_flags & SA_NODEFER)) 2827 sigaddset(&blocked, ksig->sig); 2828 set_current_blocked(&blocked); 2829 tracehook_signal_handler(stepping); 2830 } 2831 2832 void signal_setup_done(int failed, struct ksignal *ksig, int stepping) 2833 { 2834 if (failed) 2835 force_sigsegv(ksig->sig); 2836 else 2837 signal_delivered(ksig, stepping); 2838 } 2839 2840 /* 2841 * It could be that complete_signal() picked us to notify about the 2842 * group-wide signal. Other threads should be notified now to take 2843 * the shared signals in @which since we will not. 2844 */ 2845 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which) 2846 { 2847 sigset_t retarget; 2848 struct task_struct *t; 2849 2850 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which); 2851 if (sigisemptyset(&retarget)) 2852 return; 2853 2854 t = tsk; 2855 while_each_thread(tsk, t) { 2856 if (t->flags & PF_EXITING) 2857 continue; 2858 2859 if (!has_pending_signals(&retarget, &t->blocked)) 2860 continue; 2861 /* Remove the signals this thread can handle. */ 2862 sigandsets(&retarget, &retarget, &t->blocked); 2863 2864 if (!task_sigpending(t)) 2865 signal_wake_up(t, 0); 2866 2867 if (sigisemptyset(&retarget)) 2868 break; 2869 } 2870 } 2871 2872 void exit_signals(struct task_struct *tsk) 2873 { 2874 int group_stop = 0; 2875 sigset_t unblocked; 2876 2877 /* 2878 * @tsk is about to have PF_EXITING set - lock out users which 2879 * expect stable threadgroup. 2880 */ 2881 cgroup_threadgroup_change_begin(tsk); 2882 2883 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) { 2884 tsk->flags |= PF_EXITING; 2885 cgroup_threadgroup_change_end(tsk); 2886 return; 2887 } 2888 2889 spin_lock_irq(&tsk->sighand->siglock); 2890 /* 2891 * From now this task is not visible for group-wide signals, 2892 * see wants_signal(), do_signal_stop(). 2893 */ 2894 tsk->flags |= PF_EXITING; 2895 2896 cgroup_threadgroup_change_end(tsk); 2897 2898 if (!task_sigpending(tsk)) 2899 goto out; 2900 2901 unblocked = tsk->blocked; 2902 signotset(&unblocked); 2903 retarget_shared_pending(tsk, &unblocked); 2904 2905 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) && 2906 task_participate_group_stop(tsk)) 2907 group_stop = CLD_STOPPED; 2908 out: 2909 spin_unlock_irq(&tsk->sighand->siglock); 2910 2911 /* 2912 * If group stop has completed, deliver the notification. This 2913 * should always go to the real parent of the group leader. 2914 */ 2915 if (unlikely(group_stop)) { 2916 read_lock(&tasklist_lock); 2917 do_notify_parent_cldstop(tsk, false, group_stop); 2918 read_unlock(&tasklist_lock); 2919 } 2920 } 2921 2922 /* 2923 * System call entry points. 2924 */ 2925 2926 /** 2927 * sys_restart_syscall - restart a system call 2928 */ 2929 SYSCALL_DEFINE0(restart_syscall) 2930 { 2931 struct restart_block *restart = ¤t->restart_block; 2932 return restart->fn(restart); 2933 } 2934 2935 long do_no_restart_syscall(struct restart_block *param) 2936 { 2937 return -EINTR; 2938 } 2939 2940 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset) 2941 { 2942 if (task_sigpending(tsk) && !thread_group_empty(tsk)) { 2943 sigset_t newblocked; 2944 /* A set of now blocked but previously unblocked signals. */ 2945 sigandnsets(&newblocked, newset, ¤t->blocked); 2946 retarget_shared_pending(tsk, &newblocked); 2947 } 2948 tsk->blocked = *newset; 2949 recalc_sigpending(); 2950 } 2951 2952 /** 2953 * set_current_blocked - change current->blocked mask 2954 * @newset: new mask 2955 * 2956 * It is wrong to change ->blocked directly, this helper should be used 2957 * to ensure the process can't miss a shared signal we are going to block. 2958 */ 2959 void set_current_blocked(sigset_t *newset) 2960 { 2961 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP)); 2962 __set_current_blocked(newset); 2963 } 2964 2965 void __set_current_blocked(const sigset_t *newset) 2966 { 2967 struct task_struct *tsk = current; 2968 2969 /* 2970 * In case the signal mask hasn't changed, there is nothing we need 2971 * to do. The current->blocked shouldn't be modified by other task. 2972 */ 2973 if (sigequalsets(&tsk->blocked, newset)) 2974 return; 2975 2976 spin_lock_irq(&tsk->sighand->siglock); 2977 __set_task_blocked(tsk, newset); 2978 spin_unlock_irq(&tsk->sighand->siglock); 2979 } 2980 2981 /* 2982 * This is also useful for kernel threads that want to temporarily 2983 * (or permanently) block certain signals. 2984 * 2985 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel 2986 * interface happily blocks "unblockable" signals like SIGKILL 2987 * and friends. 2988 */ 2989 int sigprocmask(int how, sigset_t *set, sigset_t *oldset) 2990 { 2991 struct task_struct *tsk = current; 2992 sigset_t newset; 2993 2994 /* Lockless, only current can change ->blocked, never from irq */ 2995 if (oldset) 2996 *oldset = tsk->blocked; 2997 2998 switch (how) { 2999 case SIG_BLOCK: 3000 sigorsets(&newset, &tsk->blocked, set); 3001 break; 3002 case SIG_UNBLOCK: 3003 sigandnsets(&newset, &tsk->blocked, set); 3004 break; 3005 case SIG_SETMASK: 3006 newset = *set; 3007 break; 3008 default: 3009 return -EINVAL; 3010 } 3011 3012 __set_current_blocked(&newset); 3013 return 0; 3014 } 3015 EXPORT_SYMBOL(sigprocmask); 3016 3017 /* 3018 * The api helps set app-provided sigmasks. 3019 * 3020 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and 3021 * epoll_pwait where a new sigmask is passed from userland for the syscalls. 3022 * 3023 * Note that it does set_restore_sigmask() in advance, so it must be always 3024 * paired with restore_saved_sigmask_unless() before return from syscall. 3025 */ 3026 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize) 3027 { 3028 sigset_t kmask; 3029 3030 if (!umask) 3031 return 0; 3032 if (sigsetsize != sizeof(sigset_t)) 3033 return -EINVAL; 3034 if (copy_from_user(&kmask, umask, sizeof(sigset_t))) 3035 return -EFAULT; 3036 3037 set_restore_sigmask(); 3038 current->saved_sigmask = current->blocked; 3039 set_current_blocked(&kmask); 3040 3041 return 0; 3042 } 3043 3044 #ifdef CONFIG_COMPAT 3045 int set_compat_user_sigmask(const compat_sigset_t __user *umask, 3046 size_t sigsetsize) 3047 { 3048 sigset_t kmask; 3049 3050 if (!umask) 3051 return 0; 3052 if (sigsetsize != sizeof(compat_sigset_t)) 3053 return -EINVAL; 3054 if (get_compat_sigset(&kmask, umask)) 3055 return -EFAULT; 3056 3057 set_restore_sigmask(); 3058 current->saved_sigmask = current->blocked; 3059 set_current_blocked(&kmask); 3060 3061 return 0; 3062 } 3063 #endif 3064 3065 /** 3066 * sys_rt_sigprocmask - change the list of currently blocked signals 3067 * @how: whether to add, remove, or set signals 3068 * @nset: stores pending signals 3069 * @oset: previous value of signal mask if non-null 3070 * @sigsetsize: size of sigset_t type 3071 */ 3072 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset, 3073 sigset_t __user *, oset, size_t, sigsetsize) 3074 { 3075 sigset_t old_set, new_set; 3076 int error; 3077 3078 /* XXX: Don't preclude handling different sized sigset_t's. */ 3079 if (sigsetsize != sizeof(sigset_t)) 3080 return -EINVAL; 3081 3082 old_set = current->blocked; 3083 3084 if (nset) { 3085 if (copy_from_user(&new_set, nset, sizeof(sigset_t))) 3086 return -EFAULT; 3087 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP)); 3088 3089 error = sigprocmask(how, &new_set, NULL); 3090 if (error) 3091 return error; 3092 } 3093 3094 if (oset) { 3095 if (copy_to_user(oset, &old_set, sizeof(sigset_t))) 3096 return -EFAULT; 3097 } 3098 3099 return 0; 3100 } 3101 3102 #ifdef CONFIG_COMPAT 3103 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset, 3104 compat_sigset_t __user *, oset, compat_size_t, sigsetsize) 3105 { 3106 sigset_t old_set = current->blocked; 3107 3108 /* XXX: Don't preclude handling different sized sigset_t's. */ 3109 if (sigsetsize != sizeof(sigset_t)) 3110 return -EINVAL; 3111 3112 if (nset) { 3113 sigset_t new_set; 3114 int error; 3115 if (get_compat_sigset(&new_set, nset)) 3116 return -EFAULT; 3117 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP)); 3118 3119 error = sigprocmask(how, &new_set, NULL); 3120 if (error) 3121 return error; 3122 } 3123 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0; 3124 } 3125 #endif 3126 3127 static void do_sigpending(sigset_t *set) 3128 { 3129 spin_lock_irq(¤t->sighand->siglock); 3130 sigorsets(set, ¤t->pending.signal, 3131 ¤t->signal->shared_pending.signal); 3132 spin_unlock_irq(¤t->sighand->siglock); 3133 3134 /* Outside the lock because only this thread touches it. */ 3135 sigandsets(set, ¤t->blocked, set); 3136 } 3137 3138 /** 3139 * sys_rt_sigpending - examine a pending signal that has been raised 3140 * while blocked 3141 * @uset: stores pending signals 3142 * @sigsetsize: size of sigset_t type or larger 3143 */ 3144 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize) 3145 { 3146 sigset_t set; 3147 3148 if (sigsetsize > sizeof(*uset)) 3149 return -EINVAL; 3150 3151 do_sigpending(&set); 3152 3153 if (copy_to_user(uset, &set, sigsetsize)) 3154 return -EFAULT; 3155 3156 return 0; 3157 } 3158 3159 #ifdef CONFIG_COMPAT 3160 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset, 3161 compat_size_t, sigsetsize) 3162 { 3163 sigset_t set; 3164 3165 if (sigsetsize > sizeof(*uset)) 3166 return -EINVAL; 3167 3168 do_sigpending(&set); 3169 3170 return put_compat_sigset(uset, &set, sigsetsize); 3171 } 3172 #endif 3173 3174 static const struct { 3175 unsigned char limit, layout; 3176 } sig_sicodes[] = { 3177 [SIGILL] = { NSIGILL, SIL_FAULT }, 3178 [SIGFPE] = { NSIGFPE, SIL_FAULT }, 3179 [SIGSEGV] = { NSIGSEGV, SIL_FAULT }, 3180 [SIGBUS] = { NSIGBUS, SIL_FAULT }, 3181 [SIGTRAP] = { NSIGTRAP, SIL_FAULT }, 3182 #if defined(SIGEMT) 3183 [SIGEMT] = { NSIGEMT, SIL_FAULT }, 3184 #endif 3185 [SIGCHLD] = { NSIGCHLD, SIL_CHLD }, 3186 [SIGPOLL] = { NSIGPOLL, SIL_POLL }, 3187 [SIGSYS] = { NSIGSYS, SIL_SYS }, 3188 }; 3189 3190 static bool known_siginfo_layout(unsigned sig, int si_code) 3191 { 3192 if (si_code == SI_KERNEL) 3193 return true; 3194 else if ((si_code > SI_USER)) { 3195 if (sig_specific_sicodes(sig)) { 3196 if (si_code <= sig_sicodes[sig].limit) 3197 return true; 3198 } 3199 else if (si_code <= NSIGPOLL) 3200 return true; 3201 } 3202 else if (si_code >= SI_DETHREAD) 3203 return true; 3204 else if (si_code == SI_ASYNCNL) 3205 return true; 3206 return false; 3207 } 3208 3209 enum siginfo_layout siginfo_layout(unsigned sig, int si_code) 3210 { 3211 enum siginfo_layout layout = SIL_KILL; 3212 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) { 3213 if ((sig < ARRAY_SIZE(sig_sicodes)) && 3214 (si_code <= sig_sicodes[sig].limit)) { 3215 layout = sig_sicodes[sig].layout; 3216 /* Handle the exceptions */ 3217 if ((sig == SIGBUS) && 3218 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO)) 3219 layout = SIL_FAULT_MCEERR; 3220 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR)) 3221 layout = SIL_FAULT_BNDERR; 3222 #ifdef SEGV_PKUERR 3223 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR)) 3224 layout = SIL_FAULT_PKUERR; 3225 #endif 3226 } 3227 else if (si_code <= NSIGPOLL) 3228 layout = SIL_POLL; 3229 } else { 3230 if (si_code == SI_TIMER) 3231 layout = SIL_TIMER; 3232 else if (si_code == SI_SIGIO) 3233 layout = SIL_POLL; 3234 else if (si_code < 0) 3235 layout = SIL_RT; 3236 } 3237 return layout; 3238 } 3239 3240 static inline char __user *si_expansion(const siginfo_t __user *info) 3241 { 3242 return ((char __user *)info) + sizeof(struct kernel_siginfo); 3243 } 3244 3245 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from) 3246 { 3247 char __user *expansion = si_expansion(to); 3248 if (copy_to_user(to, from , sizeof(struct kernel_siginfo))) 3249 return -EFAULT; 3250 if (clear_user(expansion, SI_EXPANSION_SIZE)) 3251 return -EFAULT; 3252 return 0; 3253 } 3254 3255 static int post_copy_siginfo_from_user(kernel_siginfo_t *info, 3256 const siginfo_t __user *from) 3257 { 3258 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) { 3259 char __user *expansion = si_expansion(from); 3260 char buf[SI_EXPANSION_SIZE]; 3261 int i; 3262 /* 3263 * An unknown si_code might need more than 3264 * sizeof(struct kernel_siginfo) bytes. Verify all of the 3265 * extra bytes are 0. This guarantees copy_siginfo_to_user 3266 * will return this data to userspace exactly. 3267 */ 3268 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE)) 3269 return -EFAULT; 3270 for (i = 0; i < SI_EXPANSION_SIZE; i++) { 3271 if (buf[i] != 0) 3272 return -E2BIG; 3273 } 3274 } 3275 return 0; 3276 } 3277 3278 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to, 3279 const siginfo_t __user *from) 3280 { 3281 if (copy_from_user(to, from, sizeof(struct kernel_siginfo))) 3282 return -EFAULT; 3283 to->si_signo = signo; 3284 return post_copy_siginfo_from_user(to, from); 3285 } 3286 3287 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from) 3288 { 3289 if (copy_from_user(to, from, sizeof(struct kernel_siginfo))) 3290 return -EFAULT; 3291 return post_copy_siginfo_from_user(to, from); 3292 } 3293 3294 #ifdef CONFIG_COMPAT 3295 /** 3296 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo 3297 * @to: compat siginfo destination 3298 * @from: kernel siginfo source 3299 * 3300 * Note: This function does not work properly for the SIGCHLD on x32, but 3301 * fortunately it doesn't have to. The only valid callers for this function are 3302 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code. 3303 * The latter does not care because SIGCHLD will never cause a coredump. 3304 */ 3305 void copy_siginfo_to_external32(struct compat_siginfo *to, 3306 const struct kernel_siginfo *from) 3307 { 3308 memset(to, 0, sizeof(*to)); 3309 3310 to->si_signo = from->si_signo; 3311 to->si_errno = from->si_errno; 3312 to->si_code = from->si_code; 3313 switch(siginfo_layout(from->si_signo, from->si_code)) { 3314 case SIL_KILL: 3315 to->si_pid = from->si_pid; 3316 to->si_uid = from->si_uid; 3317 break; 3318 case SIL_TIMER: 3319 to->si_tid = from->si_tid; 3320 to->si_overrun = from->si_overrun; 3321 to->si_int = from->si_int; 3322 break; 3323 case SIL_POLL: 3324 to->si_band = from->si_band; 3325 to->si_fd = from->si_fd; 3326 break; 3327 case SIL_FAULT: 3328 to->si_addr = ptr_to_compat(from->si_addr); 3329 #ifdef __ARCH_SI_TRAPNO 3330 to->si_trapno = from->si_trapno; 3331 #endif 3332 break; 3333 case SIL_FAULT_MCEERR: 3334 to->si_addr = ptr_to_compat(from->si_addr); 3335 #ifdef __ARCH_SI_TRAPNO 3336 to->si_trapno = from->si_trapno; 3337 #endif 3338 to->si_addr_lsb = from->si_addr_lsb; 3339 break; 3340 case SIL_FAULT_BNDERR: 3341 to->si_addr = ptr_to_compat(from->si_addr); 3342 #ifdef __ARCH_SI_TRAPNO 3343 to->si_trapno = from->si_trapno; 3344 #endif 3345 to->si_lower = ptr_to_compat(from->si_lower); 3346 to->si_upper = ptr_to_compat(from->si_upper); 3347 break; 3348 case SIL_FAULT_PKUERR: 3349 to->si_addr = ptr_to_compat(from->si_addr); 3350 #ifdef __ARCH_SI_TRAPNO 3351 to->si_trapno = from->si_trapno; 3352 #endif 3353 to->si_pkey = from->si_pkey; 3354 break; 3355 case SIL_CHLD: 3356 to->si_pid = from->si_pid; 3357 to->si_uid = from->si_uid; 3358 to->si_status = from->si_status; 3359 to->si_utime = from->si_utime; 3360 to->si_stime = from->si_stime; 3361 break; 3362 case SIL_RT: 3363 to->si_pid = from->si_pid; 3364 to->si_uid = from->si_uid; 3365 to->si_int = from->si_int; 3366 break; 3367 case SIL_SYS: 3368 to->si_call_addr = ptr_to_compat(from->si_call_addr); 3369 to->si_syscall = from->si_syscall; 3370 to->si_arch = from->si_arch; 3371 break; 3372 } 3373 } 3374 3375 int __copy_siginfo_to_user32(struct compat_siginfo __user *to, 3376 const struct kernel_siginfo *from) 3377 { 3378 struct compat_siginfo new; 3379 3380 copy_siginfo_to_external32(&new, from); 3381 if (copy_to_user(to, &new, sizeof(struct compat_siginfo))) 3382 return -EFAULT; 3383 return 0; 3384 } 3385 3386 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to, 3387 const struct compat_siginfo *from) 3388 { 3389 clear_siginfo(to); 3390 to->si_signo = from->si_signo; 3391 to->si_errno = from->si_errno; 3392 to->si_code = from->si_code; 3393 switch(siginfo_layout(from->si_signo, from->si_code)) { 3394 case SIL_KILL: 3395 to->si_pid = from->si_pid; 3396 to->si_uid = from->si_uid; 3397 break; 3398 case SIL_TIMER: 3399 to->si_tid = from->si_tid; 3400 to->si_overrun = from->si_overrun; 3401 to->si_int = from->si_int; 3402 break; 3403 case SIL_POLL: 3404 to->si_band = from->si_band; 3405 to->si_fd = from->si_fd; 3406 break; 3407 case SIL_FAULT: 3408 to->si_addr = compat_ptr(from->si_addr); 3409 #ifdef __ARCH_SI_TRAPNO 3410 to->si_trapno = from->si_trapno; 3411 #endif 3412 break; 3413 case SIL_FAULT_MCEERR: 3414 to->si_addr = compat_ptr(from->si_addr); 3415 #ifdef __ARCH_SI_TRAPNO 3416 to->si_trapno = from->si_trapno; 3417 #endif 3418 to->si_addr_lsb = from->si_addr_lsb; 3419 break; 3420 case SIL_FAULT_BNDERR: 3421 to->si_addr = compat_ptr(from->si_addr); 3422 #ifdef __ARCH_SI_TRAPNO 3423 to->si_trapno = from->si_trapno; 3424 #endif 3425 to->si_lower = compat_ptr(from->si_lower); 3426 to->si_upper = compat_ptr(from->si_upper); 3427 break; 3428 case SIL_FAULT_PKUERR: 3429 to->si_addr = compat_ptr(from->si_addr); 3430 #ifdef __ARCH_SI_TRAPNO 3431 to->si_trapno = from->si_trapno; 3432 #endif 3433 to->si_pkey = from->si_pkey; 3434 break; 3435 case SIL_CHLD: 3436 to->si_pid = from->si_pid; 3437 to->si_uid = from->si_uid; 3438 to->si_status = from->si_status; 3439 #ifdef CONFIG_X86_X32_ABI 3440 if (in_x32_syscall()) { 3441 to->si_utime = from->_sifields._sigchld_x32._utime; 3442 to->si_stime = from->_sifields._sigchld_x32._stime; 3443 } else 3444 #endif 3445 { 3446 to->si_utime = from->si_utime; 3447 to->si_stime = from->si_stime; 3448 } 3449 break; 3450 case SIL_RT: 3451 to->si_pid = from->si_pid; 3452 to->si_uid = from->si_uid; 3453 to->si_int = from->si_int; 3454 break; 3455 case SIL_SYS: 3456 to->si_call_addr = compat_ptr(from->si_call_addr); 3457 to->si_syscall = from->si_syscall; 3458 to->si_arch = from->si_arch; 3459 break; 3460 } 3461 return 0; 3462 } 3463 3464 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to, 3465 const struct compat_siginfo __user *ufrom) 3466 { 3467 struct compat_siginfo from; 3468 3469 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo))) 3470 return -EFAULT; 3471 3472 from.si_signo = signo; 3473 return post_copy_siginfo_from_user32(to, &from); 3474 } 3475 3476 int copy_siginfo_from_user32(struct kernel_siginfo *to, 3477 const struct compat_siginfo __user *ufrom) 3478 { 3479 struct compat_siginfo from; 3480 3481 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo))) 3482 return -EFAULT; 3483 3484 return post_copy_siginfo_from_user32(to, &from); 3485 } 3486 #endif /* CONFIG_COMPAT */ 3487 3488 /** 3489 * do_sigtimedwait - wait for queued signals specified in @which 3490 * @which: queued signals to wait for 3491 * @info: if non-null, the signal's siginfo is returned here 3492 * @ts: upper bound on process time suspension 3493 */ 3494 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info, 3495 const struct timespec64 *ts) 3496 { 3497 ktime_t *to = NULL, timeout = KTIME_MAX; 3498 struct task_struct *tsk = current; 3499 sigset_t mask = *which; 3500 int sig, ret = 0; 3501 3502 if (ts) { 3503 if (!timespec64_valid(ts)) 3504 return -EINVAL; 3505 timeout = timespec64_to_ktime(*ts); 3506 to = &timeout; 3507 } 3508 3509 /* 3510 * Invert the set of allowed signals to get those we want to block. 3511 */ 3512 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP)); 3513 signotset(&mask); 3514 3515 spin_lock_irq(&tsk->sighand->siglock); 3516 sig = dequeue_signal(tsk, &mask, info); 3517 if (!sig && timeout) { 3518 /* 3519 * None ready, temporarily unblock those we're interested 3520 * while we are sleeping in so that we'll be awakened when 3521 * they arrive. Unblocking is always fine, we can avoid 3522 * set_current_blocked(). 3523 */ 3524 tsk->real_blocked = tsk->blocked; 3525 sigandsets(&tsk->blocked, &tsk->blocked, &mask); 3526 recalc_sigpending(); 3527 spin_unlock_irq(&tsk->sighand->siglock); 3528 3529 __set_current_state(TASK_INTERRUPTIBLE); 3530 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns, 3531 HRTIMER_MODE_REL); 3532 spin_lock_irq(&tsk->sighand->siglock); 3533 __set_task_blocked(tsk, &tsk->real_blocked); 3534 sigemptyset(&tsk->real_blocked); 3535 sig = dequeue_signal(tsk, &mask, info); 3536 } 3537 spin_unlock_irq(&tsk->sighand->siglock); 3538 3539 if (sig) 3540 return sig; 3541 return ret ? -EINTR : -EAGAIN; 3542 } 3543 3544 /** 3545 * sys_rt_sigtimedwait - synchronously wait for queued signals specified 3546 * in @uthese 3547 * @uthese: queued signals to wait for 3548 * @uinfo: if non-null, the signal's siginfo is returned here 3549 * @uts: upper bound on process time suspension 3550 * @sigsetsize: size of sigset_t type 3551 */ 3552 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese, 3553 siginfo_t __user *, uinfo, 3554 const struct __kernel_timespec __user *, uts, 3555 size_t, sigsetsize) 3556 { 3557 sigset_t these; 3558 struct timespec64 ts; 3559 kernel_siginfo_t info; 3560 int ret; 3561 3562 /* XXX: Don't preclude handling different sized sigset_t's. */ 3563 if (sigsetsize != sizeof(sigset_t)) 3564 return -EINVAL; 3565 3566 if (copy_from_user(&these, uthese, sizeof(these))) 3567 return -EFAULT; 3568 3569 if (uts) { 3570 if (get_timespec64(&ts, uts)) 3571 return -EFAULT; 3572 } 3573 3574 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL); 3575 3576 if (ret > 0 && uinfo) { 3577 if (copy_siginfo_to_user(uinfo, &info)) 3578 ret = -EFAULT; 3579 } 3580 3581 return ret; 3582 } 3583 3584 #ifdef CONFIG_COMPAT_32BIT_TIME 3585 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese, 3586 siginfo_t __user *, uinfo, 3587 const struct old_timespec32 __user *, uts, 3588 size_t, sigsetsize) 3589 { 3590 sigset_t these; 3591 struct timespec64 ts; 3592 kernel_siginfo_t info; 3593 int ret; 3594 3595 if (sigsetsize != sizeof(sigset_t)) 3596 return -EINVAL; 3597 3598 if (copy_from_user(&these, uthese, sizeof(these))) 3599 return -EFAULT; 3600 3601 if (uts) { 3602 if (get_old_timespec32(&ts, uts)) 3603 return -EFAULT; 3604 } 3605 3606 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL); 3607 3608 if (ret > 0 && uinfo) { 3609 if (copy_siginfo_to_user(uinfo, &info)) 3610 ret = -EFAULT; 3611 } 3612 3613 return ret; 3614 } 3615 #endif 3616 3617 #ifdef CONFIG_COMPAT 3618 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese, 3619 struct compat_siginfo __user *, uinfo, 3620 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize) 3621 { 3622 sigset_t s; 3623 struct timespec64 t; 3624 kernel_siginfo_t info; 3625 long ret; 3626 3627 if (sigsetsize != sizeof(sigset_t)) 3628 return -EINVAL; 3629 3630 if (get_compat_sigset(&s, uthese)) 3631 return -EFAULT; 3632 3633 if (uts) { 3634 if (get_timespec64(&t, uts)) 3635 return -EFAULT; 3636 } 3637 3638 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL); 3639 3640 if (ret > 0 && uinfo) { 3641 if (copy_siginfo_to_user32(uinfo, &info)) 3642 ret = -EFAULT; 3643 } 3644 3645 return ret; 3646 } 3647 3648 #ifdef CONFIG_COMPAT_32BIT_TIME 3649 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese, 3650 struct compat_siginfo __user *, uinfo, 3651 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize) 3652 { 3653 sigset_t s; 3654 struct timespec64 t; 3655 kernel_siginfo_t info; 3656 long ret; 3657 3658 if (sigsetsize != sizeof(sigset_t)) 3659 return -EINVAL; 3660 3661 if (get_compat_sigset(&s, uthese)) 3662 return -EFAULT; 3663 3664 if (uts) { 3665 if (get_old_timespec32(&t, uts)) 3666 return -EFAULT; 3667 } 3668 3669 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL); 3670 3671 if (ret > 0 && uinfo) { 3672 if (copy_siginfo_to_user32(uinfo, &info)) 3673 ret = -EFAULT; 3674 } 3675 3676 return ret; 3677 } 3678 #endif 3679 #endif 3680 3681 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info) 3682 { 3683 clear_siginfo(info); 3684 info->si_signo = sig; 3685 info->si_errno = 0; 3686 info->si_code = SI_USER; 3687 info->si_pid = task_tgid_vnr(current); 3688 info->si_uid = from_kuid_munged(current_user_ns(), current_uid()); 3689 } 3690 3691 /** 3692 * sys_kill - send a signal to a process 3693 * @pid: the PID of the process 3694 * @sig: signal to be sent 3695 */ 3696 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig) 3697 { 3698 struct kernel_siginfo info; 3699 3700 prepare_kill_siginfo(sig, &info); 3701 3702 return kill_something_info(sig, &info, pid); 3703 } 3704 3705 /* 3706 * Verify that the signaler and signalee either are in the same pid namespace 3707 * or that the signaler's pid namespace is an ancestor of the signalee's pid 3708 * namespace. 3709 */ 3710 static bool access_pidfd_pidns(struct pid *pid) 3711 { 3712 struct pid_namespace *active = task_active_pid_ns(current); 3713 struct pid_namespace *p = ns_of_pid(pid); 3714 3715 for (;;) { 3716 if (!p) 3717 return false; 3718 if (p == active) 3719 break; 3720 p = p->parent; 3721 } 3722 3723 return true; 3724 } 3725 3726 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, siginfo_t *info) 3727 { 3728 #ifdef CONFIG_COMPAT 3729 /* 3730 * Avoid hooking up compat syscalls and instead handle necessary 3731 * conversions here. Note, this is a stop-gap measure and should not be 3732 * considered a generic solution. 3733 */ 3734 if (in_compat_syscall()) 3735 return copy_siginfo_from_user32( 3736 kinfo, (struct compat_siginfo __user *)info); 3737 #endif 3738 return copy_siginfo_from_user(kinfo, info); 3739 } 3740 3741 static struct pid *pidfd_to_pid(const struct file *file) 3742 { 3743 struct pid *pid; 3744 3745 pid = pidfd_pid(file); 3746 if (!IS_ERR(pid)) 3747 return pid; 3748 3749 return tgid_pidfd_to_pid(file); 3750 } 3751 3752 /** 3753 * sys_pidfd_send_signal - Signal a process through a pidfd 3754 * @pidfd: file descriptor of the process 3755 * @sig: signal to send 3756 * @info: signal info 3757 * @flags: future flags 3758 * 3759 * The syscall currently only signals via PIDTYPE_PID which covers 3760 * kill(<positive-pid>, <signal>. It does not signal threads or process 3761 * groups. 3762 * In order to extend the syscall to threads and process groups the @flags 3763 * argument should be used. In essence, the @flags argument will determine 3764 * what is signaled and not the file descriptor itself. Put in other words, 3765 * grouping is a property of the flags argument not a property of the file 3766 * descriptor. 3767 * 3768 * Return: 0 on success, negative errno on failure 3769 */ 3770 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig, 3771 siginfo_t __user *, info, unsigned int, flags) 3772 { 3773 int ret; 3774 struct fd f; 3775 struct pid *pid; 3776 kernel_siginfo_t kinfo; 3777 3778 /* Enforce flags be set to 0 until we add an extension. */ 3779 if (flags) 3780 return -EINVAL; 3781 3782 f = fdget(pidfd); 3783 if (!f.file) 3784 return -EBADF; 3785 3786 /* Is this a pidfd? */ 3787 pid = pidfd_to_pid(f.file); 3788 if (IS_ERR(pid)) { 3789 ret = PTR_ERR(pid); 3790 goto err; 3791 } 3792 3793 ret = -EINVAL; 3794 if (!access_pidfd_pidns(pid)) 3795 goto err; 3796 3797 if (info) { 3798 ret = copy_siginfo_from_user_any(&kinfo, info); 3799 if (unlikely(ret)) 3800 goto err; 3801 3802 ret = -EINVAL; 3803 if (unlikely(sig != kinfo.si_signo)) 3804 goto err; 3805 3806 /* Only allow sending arbitrary signals to yourself. */ 3807 ret = -EPERM; 3808 if ((task_pid(current) != pid) && 3809 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL)) 3810 goto err; 3811 } else { 3812 prepare_kill_siginfo(sig, &kinfo); 3813 } 3814 3815 ret = kill_pid_info(sig, &kinfo, pid); 3816 3817 err: 3818 fdput(f); 3819 return ret; 3820 } 3821 3822 static int 3823 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info) 3824 { 3825 struct task_struct *p; 3826 int error = -ESRCH; 3827 3828 rcu_read_lock(); 3829 p = find_task_by_vpid(pid); 3830 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) { 3831 error = check_kill_permission(sig, info, p); 3832 /* 3833 * The null signal is a permissions and process existence 3834 * probe. No signal is actually delivered. 3835 */ 3836 if (!error && sig) { 3837 error = do_send_sig_info(sig, info, p, PIDTYPE_PID); 3838 /* 3839 * If lock_task_sighand() failed we pretend the task 3840 * dies after receiving the signal. The window is tiny, 3841 * and the signal is private anyway. 3842 */ 3843 if (unlikely(error == -ESRCH)) 3844 error = 0; 3845 } 3846 } 3847 rcu_read_unlock(); 3848 3849 return error; 3850 } 3851 3852 static int do_tkill(pid_t tgid, pid_t pid, int sig) 3853 { 3854 struct kernel_siginfo info; 3855 3856 clear_siginfo(&info); 3857 info.si_signo = sig; 3858 info.si_errno = 0; 3859 info.si_code = SI_TKILL; 3860 info.si_pid = task_tgid_vnr(current); 3861 info.si_uid = from_kuid_munged(current_user_ns(), current_uid()); 3862 3863 return do_send_specific(tgid, pid, sig, &info); 3864 } 3865 3866 /** 3867 * sys_tgkill - send signal to one specific thread 3868 * @tgid: the thread group ID of the thread 3869 * @pid: the PID of the thread 3870 * @sig: signal to be sent 3871 * 3872 * This syscall also checks the @tgid and returns -ESRCH even if the PID 3873 * exists but it's not belonging to the target process anymore. This 3874 * method solves the problem of threads exiting and PIDs getting reused. 3875 */ 3876 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig) 3877 { 3878 /* This is only valid for single tasks */ 3879 if (pid <= 0 || tgid <= 0) 3880 return -EINVAL; 3881 3882 return do_tkill(tgid, pid, sig); 3883 } 3884 3885 /** 3886 * sys_tkill - send signal to one specific task 3887 * @pid: the PID of the task 3888 * @sig: signal to be sent 3889 * 3890 * Send a signal to only one task, even if it's a CLONE_THREAD task. 3891 */ 3892 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig) 3893 { 3894 /* This is only valid for single tasks */ 3895 if (pid <= 0) 3896 return -EINVAL; 3897 3898 return do_tkill(0, pid, sig); 3899 } 3900 3901 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info) 3902 { 3903 /* Not even root can pretend to send signals from the kernel. 3904 * Nor can they impersonate a kill()/tgkill(), which adds source info. 3905 */ 3906 if ((info->si_code >= 0 || info->si_code == SI_TKILL) && 3907 (task_pid_vnr(current) != pid)) 3908 return -EPERM; 3909 3910 /* POSIX.1b doesn't mention process groups. */ 3911 return kill_proc_info(sig, info, pid); 3912 } 3913 3914 /** 3915 * sys_rt_sigqueueinfo - send signal information to a signal 3916 * @pid: the PID of the thread 3917 * @sig: signal to be sent 3918 * @uinfo: signal info to be sent 3919 */ 3920 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig, 3921 siginfo_t __user *, uinfo) 3922 { 3923 kernel_siginfo_t info; 3924 int ret = __copy_siginfo_from_user(sig, &info, uinfo); 3925 if (unlikely(ret)) 3926 return ret; 3927 return do_rt_sigqueueinfo(pid, sig, &info); 3928 } 3929 3930 #ifdef CONFIG_COMPAT 3931 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo, 3932 compat_pid_t, pid, 3933 int, sig, 3934 struct compat_siginfo __user *, uinfo) 3935 { 3936 kernel_siginfo_t info; 3937 int ret = __copy_siginfo_from_user32(sig, &info, uinfo); 3938 if (unlikely(ret)) 3939 return ret; 3940 return do_rt_sigqueueinfo(pid, sig, &info); 3941 } 3942 #endif 3943 3944 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info) 3945 { 3946 /* This is only valid for single tasks */ 3947 if (pid <= 0 || tgid <= 0) 3948 return -EINVAL; 3949 3950 /* Not even root can pretend to send signals from the kernel. 3951 * Nor can they impersonate a kill()/tgkill(), which adds source info. 3952 */ 3953 if ((info->si_code >= 0 || info->si_code == SI_TKILL) && 3954 (task_pid_vnr(current) != pid)) 3955 return -EPERM; 3956 3957 return do_send_specific(tgid, pid, sig, info); 3958 } 3959 3960 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig, 3961 siginfo_t __user *, uinfo) 3962 { 3963 kernel_siginfo_t info; 3964 int ret = __copy_siginfo_from_user(sig, &info, uinfo); 3965 if (unlikely(ret)) 3966 return ret; 3967 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); 3968 } 3969 3970 #ifdef CONFIG_COMPAT 3971 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo, 3972 compat_pid_t, tgid, 3973 compat_pid_t, pid, 3974 int, sig, 3975 struct compat_siginfo __user *, uinfo) 3976 { 3977 kernel_siginfo_t info; 3978 int ret = __copy_siginfo_from_user32(sig, &info, uinfo); 3979 if (unlikely(ret)) 3980 return ret; 3981 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); 3982 } 3983 #endif 3984 3985 /* 3986 * For kthreads only, must not be used if cloned with CLONE_SIGHAND 3987 */ 3988 void kernel_sigaction(int sig, __sighandler_t action) 3989 { 3990 spin_lock_irq(¤t->sighand->siglock); 3991 current->sighand->action[sig - 1].sa.sa_handler = action; 3992 if (action == SIG_IGN) { 3993 sigset_t mask; 3994 3995 sigemptyset(&mask); 3996 sigaddset(&mask, sig); 3997 3998 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending); 3999 flush_sigqueue_mask(&mask, ¤t->pending); 4000 recalc_sigpending(); 4001 } 4002 spin_unlock_irq(¤t->sighand->siglock); 4003 } 4004 EXPORT_SYMBOL(kernel_sigaction); 4005 4006 void __weak sigaction_compat_abi(struct k_sigaction *act, 4007 struct k_sigaction *oact) 4008 { 4009 } 4010 4011 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact) 4012 { 4013 struct task_struct *p = current, *t; 4014 struct k_sigaction *k; 4015 sigset_t mask; 4016 4017 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig))) 4018 return -EINVAL; 4019 4020 k = &p->sighand->action[sig-1]; 4021 4022 spin_lock_irq(&p->sighand->siglock); 4023 if (oact) 4024 *oact = *k; 4025 4026 /* 4027 * Make sure that we never accidentally claim to support SA_UNSUPPORTED, 4028 * e.g. by having an architecture use the bit in their uapi. 4029 */ 4030 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED); 4031 4032 /* 4033 * Clear unknown flag bits in order to allow userspace to detect missing 4034 * support for flag bits and to allow the kernel to use non-uapi bits 4035 * internally. 4036 */ 4037 if (act) 4038 act->sa.sa_flags &= UAPI_SA_FLAGS; 4039 if (oact) 4040 oact->sa.sa_flags &= UAPI_SA_FLAGS; 4041 4042 sigaction_compat_abi(act, oact); 4043 4044 if (act) { 4045 sigdelsetmask(&act->sa.sa_mask, 4046 sigmask(SIGKILL) | sigmask(SIGSTOP)); 4047 *k = *act; 4048 /* 4049 * POSIX 3.3.1.3: 4050 * "Setting a signal action to SIG_IGN for a signal that is 4051 * pending shall cause the pending signal to be discarded, 4052 * whether or not it is blocked." 4053 * 4054 * "Setting a signal action to SIG_DFL for a signal that is 4055 * pending and whose default action is to ignore the signal 4056 * (for example, SIGCHLD), shall cause the pending signal to 4057 * be discarded, whether or not it is blocked" 4058 */ 4059 if (sig_handler_ignored(sig_handler(p, sig), sig)) { 4060 sigemptyset(&mask); 4061 sigaddset(&mask, sig); 4062 flush_sigqueue_mask(&mask, &p->signal->shared_pending); 4063 for_each_thread(p, t) 4064 flush_sigqueue_mask(&mask, &t->pending); 4065 } 4066 } 4067 4068 spin_unlock_irq(&p->sighand->siglock); 4069 return 0; 4070 } 4071 4072 static int 4073 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp, 4074 size_t min_ss_size) 4075 { 4076 struct task_struct *t = current; 4077 4078 if (oss) { 4079 memset(oss, 0, sizeof(stack_t)); 4080 oss->ss_sp = (void __user *) t->sas_ss_sp; 4081 oss->ss_size = t->sas_ss_size; 4082 oss->ss_flags = sas_ss_flags(sp) | 4083 (current->sas_ss_flags & SS_FLAG_BITS); 4084 } 4085 4086 if (ss) { 4087 void __user *ss_sp = ss->ss_sp; 4088 size_t ss_size = ss->ss_size; 4089 unsigned ss_flags = ss->ss_flags; 4090 int ss_mode; 4091 4092 if (unlikely(on_sig_stack(sp))) 4093 return -EPERM; 4094 4095 ss_mode = ss_flags & ~SS_FLAG_BITS; 4096 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK && 4097 ss_mode != 0)) 4098 return -EINVAL; 4099 4100 if (ss_mode == SS_DISABLE) { 4101 ss_size = 0; 4102 ss_sp = NULL; 4103 } else { 4104 if (unlikely(ss_size < min_ss_size)) 4105 return -ENOMEM; 4106 } 4107 4108 t->sas_ss_sp = (unsigned long) ss_sp; 4109 t->sas_ss_size = ss_size; 4110 t->sas_ss_flags = ss_flags; 4111 } 4112 return 0; 4113 } 4114 4115 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss) 4116 { 4117 stack_t new, old; 4118 int err; 4119 if (uss && copy_from_user(&new, uss, sizeof(stack_t))) 4120 return -EFAULT; 4121 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL, 4122 current_user_stack_pointer(), 4123 MINSIGSTKSZ); 4124 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t))) 4125 err = -EFAULT; 4126 return err; 4127 } 4128 4129 int restore_altstack(const stack_t __user *uss) 4130 { 4131 stack_t new; 4132 if (copy_from_user(&new, uss, sizeof(stack_t))) 4133 return -EFAULT; 4134 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(), 4135 MINSIGSTKSZ); 4136 /* squash all but EFAULT for now */ 4137 return 0; 4138 } 4139 4140 int __save_altstack(stack_t __user *uss, unsigned long sp) 4141 { 4142 struct task_struct *t = current; 4143 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) | 4144 __put_user(t->sas_ss_flags, &uss->ss_flags) | 4145 __put_user(t->sas_ss_size, &uss->ss_size); 4146 if (err) 4147 return err; 4148 if (t->sas_ss_flags & SS_AUTODISARM) 4149 sas_ss_reset(t); 4150 return 0; 4151 } 4152 4153 #ifdef CONFIG_COMPAT 4154 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr, 4155 compat_stack_t __user *uoss_ptr) 4156 { 4157 stack_t uss, uoss; 4158 int ret; 4159 4160 if (uss_ptr) { 4161 compat_stack_t uss32; 4162 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t))) 4163 return -EFAULT; 4164 uss.ss_sp = compat_ptr(uss32.ss_sp); 4165 uss.ss_flags = uss32.ss_flags; 4166 uss.ss_size = uss32.ss_size; 4167 } 4168 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss, 4169 compat_user_stack_pointer(), 4170 COMPAT_MINSIGSTKSZ); 4171 if (ret >= 0 && uoss_ptr) { 4172 compat_stack_t old; 4173 memset(&old, 0, sizeof(old)); 4174 old.ss_sp = ptr_to_compat(uoss.ss_sp); 4175 old.ss_flags = uoss.ss_flags; 4176 old.ss_size = uoss.ss_size; 4177 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t))) 4178 ret = -EFAULT; 4179 } 4180 return ret; 4181 } 4182 4183 COMPAT_SYSCALL_DEFINE2(sigaltstack, 4184 const compat_stack_t __user *, uss_ptr, 4185 compat_stack_t __user *, uoss_ptr) 4186 { 4187 return do_compat_sigaltstack(uss_ptr, uoss_ptr); 4188 } 4189 4190 int compat_restore_altstack(const compat_stack_t __user *uss) 4191 { 4192 int err = do_compat_sigaltstack(uss, NULL); 4193 /* squash all but -EFAULT for now */ 4194 return err == -EFAULT ? err : 0; 4195 } 4196 4197 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp) 4198 { 4199 int err; 4200 struct task_struct *t = current; 4201 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp), 4202 &uss->ss_sp) | 4203 __put_user(t->sas_ss_flags, &uss->ss_flags) | 4204 __put_user(t->sas_ss_size, &uss->ss_size); 4205 if (err) 4206 return err; 4207 if (t->sas_ss_flags & SS_AUTODISARM) 4208 sas_ss_reset(t); 4209 return 0; 4210 } 4211 #endif 4212 4213 #ifdef __ARCH_WANT_SYS_SIGPENDING 4214 4215 /** 4216 * sys_sigpending - examine pending signals 4217 * @uset: where mask of pending signal is returned 4218 */ 4219 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset) 4220 { 4221 sigset_t set; 4222 4223 if (sizeof(old_sigset_t) > sizeof(*uset)) 4224 return -EINVAL; 4225 4226 do_sigpending(&set); 4227 4228 if (copy_to_user(uset, &set, sizeof(old_sigset_t))) 4229 return -EFAULT; 4230 4231 return 0; 4232 } 4233 4234 #ifdef CONFIG_COMPAT 4235 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32) 4236 { 4237 sigset_t set; 4238 4239 do_sigpending(&set); 4240 4241 return put_user(set.sig[0], set32); 4242 } 4243 #endif 4244 4245 #endif 4246 4247 #ifdef __ARCH_WANT_SYS_SIGPROCMASK 4248 /** 4249 * sys_sigprocmask - examine and change blocked signals 4250 * @how: whether to add, remove, or set signals 4251 * @nset: signals to add or remove (if non-null) 4252 * @oset: previous value of signal mask if non-null 4253 * 4254 * Some platforms have their own version with special arguments; 4255 * others support only sys_rt_sigprocmask. 4256 */ 4257 4258 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset, 4259 old_sigset_t __user *, oset) 4260 { 4261 old_sigset_t old_set, new_set; 4262 sigset_t new_blocked; 4263 4264 old_set = current->blocked.sig[0]; 4265 4266 if (nset) { 4267 if (copy_from_user(&new_set, nset, sizeof(*nset))) 4268 return -EFAULT; 4269 4270 new_blocked = current->blocked; 4271 4272 switch (how) { 4273 case SIG_BLOCK: 4274 sigaddsetmask(&new_blocked, new_set); 4275 break; 4276 case SIG_UNBLOCK: 4277 sigdelsetmask(&new_blocked, new_set); 4278 break; 4279 case SIG_SETMASK: 4280 new_blocked.sig[0] = new_set; 4281 break; 4282 default: 4283 return -EINVAL; 4284 } 4285 4286 set_current_blocked(&new_blocked); 4287 } 4288 4289 if (oset) { 4290 if (copy_to_user(oset, &old_set, sizeof(*oset))) 4291 return -EFAULT; 4292 } 4293 4294 return 0; 4295 } 4296 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */ 4297 4298 #ifndef CONFIG_ODD_RT_SIGACTION 4299 /** 4300 * sys_rt_sigaction - alter an action taken by a process 4301 * @sig: signal to be sent 4302 * @act: new sigaction 4303 * @oact: used to save the previous sigaction 4304 * @sigsetsize: size of sigset_t type 4305 */ 4306 SYSCALL_DEFINE4(rt_sigaction, int, sig, 4307 const struct sigaction __user *, act, 4308 struct sigaction __user *, oact, 4309 size_t, sigsetsize) 4310 { 4311 struct k_sigaction new_sa, old_sa; 4312 int ret; 4313 4314 /* XXX: Don't preclude handling different sized sigset_t's. */ 4315 if (sigsetsize != sizeof(sigset_t)) 4316 return -EINVAL; 4317 4318 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa))) 4319 return -EFAULT; 4320 4321 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL); 4322 if (ret) 4323 return ret; 4324 4325 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa))) 4326 return -EFAULT; 4327 4328 return 0; 4329 } 4330 #ifdef CONFIG_COMPAT 4331 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig, 4332 const struct compat_sigaction __user *, act, 4333 struct compat_sigaction __user *, oact, 4334 compat_size_t, sigsetsize) 4335 { 4336 struct k_sigaction new_ka, old_ka; 4337 #ifdef __ARCH_HAS_SA_RESTORER 4338 compat_uptr_t restorer; 4339 #endif 4340 int ret; 4341 4342 /* XXX: Don't preclude handling different sized sigset_t's. */ 4343 if (sigsetsize != sizeof(compat_sigset_t)) 4344 return -EINVAL; 4345 4346 if (act) { 4347 compat_uptr_t handler; 4348 ret = get_user(handler, &act->sa_handler); 4349 new_ka.sa.sa_handler = compat_ptr(handler); 4350 #ifdef __ARCH_HAS_SA_RESTORER 4351 ret |= get_user(restorer, &act->sa_restorer); 4352 new_ka.sa.sa_restorer = compat_ptr(restorer); 4353 #endif 4354 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask); 4355 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags); 4356 if (ret) 4357 return -EFAULT; 4358 } 4359 4360 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4361 if (!ret && oact) { 4362 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler), 4363 &oact->sa_handler); 4364 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask, 4365 sizeof(oact->sa_mask)); 4366 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags); 4367 #ifdef __ARCH_HAS_SA_RESTORER 4368 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer), 4369 &oact->sa_restorer); 4370 #endif 4371 } 4372 return ret; 4373 } 4374 #endif 4375 #endif /* !CONFIG_ODD_RT_SIGACTION */ 4376 4377 #ifdef CONFIG_OLD_SIGACTION 4378 SYSCALL_DEFINE3(sigaction, int, sig, 4379 const struct old_sigaction __user *, act, 4380 struct old_sigaction __user *, oact) 4381 { 4382 struct k_sigaction new_ka, old_ka; 4383 int ret; 4384 4385 if (act) { 4386 old_sigset_t mask; 4387 if (!access_ok(act, sizeof(*act)) || 4388 __get_user(new_ka.sa.sa_handler, &act->sa_handler) || 4389 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) || 4390 __get_user(new_ka.sa.sa_flags, &act->sa_flags) || 4391 __get_user(mask, &act->sa_mask)) 4392 return -EFAULT; 4393 #ifdef __ARCH_HAS_KA_RESTORER 4394 new_ka.ka_restorer = NULL; 4395 #endif 4396 siginitset(&new_ka.sa.sa_mask, mask); 4397 } 4398 4399 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4400 4401 if (!ret && oact) { 4402 if (!access_ok(oact, sizeof(*oact)) || 4403 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) || 4404 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) || 4405 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) || 4406 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask)) 4407 return -EFAULT; 4408 } 4409 4410 return ret; 4411 } 4412 #endif 4413 #ifdef CONFIG_COMPAT_OLD_SIGACTION 4414 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig, 4415 const struct compat_old_sigaction __user *, act, 4416 struct compat_old_sigaction __user *, oact) 4417 { 4418 struct k_sigaction new_ka, old_ka; 4419 int ret; 4420 compat_old_sigset_t mask; 4421 compat_uptr_t handler, restorer; 4422 4423 if (act) { 4424 if (!access_ok(act, sizeof(*act)) || 4425 __get_user(handler, &act->sa_handler) || 4426 __get_user(restorer, &act->sa_restorer) || 4427 __get_user(new_ka.sa.sa_flags, &act->sa_flags) || 4428 __get_user(mask, &act->sa_mask)) 4429 return -EFAULT; 4430 4431 #ifdef __ARCH_HAS_KA_RESTORER 4432 new_ka.ka_restorer = NULL; 4433 #endif 4434 new_ka.sa.sa_handler = compat_ptr(handler); 4435 new_ka.sa.sa_restorer = compat_ptr(restorer); 4436 siginitset(&new_ka.sa.sa_mask, mask); 4437 } 4438 4439 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4440 4441 if (!ret && oact) { 4442 if (!access_ok(oact, sizeof(*oact)) || 4443 __put_user(ptr_to_compat(old_ka.sa.sa_handler), 4444 &oact->sa_handler) || 4445 __put_user(ptr_to_compat(old_ka.sa.sa_restorer), 4446 &oact->sa_restorer) || 4447 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) || 4448 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask)) 4449 return -EFAULT; 4450 } 4451 return ret; 4452 } 4453 #endif 4454 4455 #ifdef CONFIG_SGETMASK_SYSCALL 4456 4457 /* 4458 * For backwards compatibility. Functionality superseded by sigprocmask. 4459 */ 4460 SYSCALL_DEFINE0(sgetmask) 4461 { 4462 /* SMP safe */ 4463 return current->blocked.sig[0]; 4464 } 4465 4466 SYSCALL_DEFINE1(ssetmask, int, newmask) 4467 { 4468 int old = current->blocked.sig[0]; 4469 sigset_t newset; 4470 4471 siginitset(&newset, newmask); 4472 set_current_blocked(&newset); 4473 4474 return old; 4475 } 4476 #endif /* CONFIG_SGETMASK_SYSCALL */ 4477 4478 #ifdef __ARCH_WANT_SYS_SIGNAL 4479 /* 4480 * For backwards compatibility. Functionality superseded by sigaction. 4481 */ 4482 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler) 4483 { 4484 struct k_sigaction new_sa, old_sa; 4485 int ret; 4486 4487 new_sa.sa.sa_handler = handler; 4488 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK; 4489 sigemptyset(&new_sa.sa.sa_mask); 4490 4491 ret = do_sigaction(sig, &new_sa, &old_sa); 4492 4493 return ret ? ret : (unsigned long)old_sa.sa.sa_handler; 4494 } 4495 #endif /* __ARCH_WANT_SYS_SIGNAL */ 4496 4497 #ifdef __ARCH_WANT_SYS_PAUSE 4498 4499 SYSCALL_DEFINE0(pause) 4500 { 4501 while (!signal_pending(current)) { 4502 __set_current_state(TASK_INTERRUPTIBLE); 4503 schedule(); 4504 } 4505 return -ERESTARTNOHAND; 4506 } 4507 4508 #endif 4509 4510 static int sigsuspend(sigset_t *set) 4511 { 4512 current->saved_sigmask = current->blocked; 4513 set_current_blocked(set); 4514 4515 while (!signal_pending(current)) { 4516 __set_current_state(TASK_INTERRUPTIBLE); 4517 schedule(); 4518 } 4519 set_restore_sigmask(); 4520 return -ERESTARTNOHAND; 4521 } 4522 4523 /** 4524 * sys_rt_sigsuspend - replace the signal mask for a value with the 4525 * @unewset value until a signal is received 4526 * @unewset: new signal mask value 4527 * @sigsetsize: size of sigset_t type 4528 */ 4529 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize) 4530 { 4531 sigset_t newset; 4532 4533 /* XXX: Don't preclude handling different sized sigset_t's. */ 4534 if (sigsetsize != sizeof(sigset_t)) 4535 return -EINVAL; 4536 4537 if (copy_from_user(&newset, unewset, sizeof(newset))) 4538 return -EFAULT; 4539 return sigsuspend(&newset); 4540 } 4541 4542 #ifdef CONFIG_COMPAT 4543 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize) 4544 { 4545 sigset_t newset; 4546 4547 /* XXX: Don't preclude handling different sized sigset_t's. */ 4548 if (sigsetsize != sizeof(sigset_t)) 4549 return -EINVAL; 4550 4551 if (get_compat_sigset(&newset, unewset)) 4552 return -EFAULT; 4553 return sigsuspend(&newset); 4554 } 4555 #endif 4556 4557 #ifdef CONFIG_OLD_SIGSUSPEND 4558 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask) 4559 { 4560 sigset_t blocked; 4561 siginitset(&blocked, mask); 4562 return sigsuspend(&blocked); 4563 } 4564 #endif 4565 #ifdef CONFIG_OLD_SIGSUSPEND3 4566 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask) 4567 { 4568 sigset_t blocked; 4569 siginitset(&blocked, mask); 4570 return sigsuspend(&blocked); 4571 } 4572 #endif 4573 4574 __weak const char *arch_vma_name(struct vm_area_struct *vma) 4575 { 4576 return NULL; 4577 } 4578 4579 static inline void siginfo_buildtime_checks(void) 4580 { 4581 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE); 4582 4583 /* Verify the offsets in the two siginfos match */ 4584 #define CHECK_OFFSET(field) \ 4585 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field)) 4586 4587 /* kill */ 4588 CHECK_OFFSET(si_pid); 4589 CHECK_OFFSET(si_uid); 4590 4591 /* timer */ 4592 CHECK_OFFSET(si_tid); 4593 CHECK_OFFSET(si_overrun); 4594 CHECK_OFFSET(si_value); 4595 4596 /* rt */ 4597 CHECK_OFFSET(si_pid); 4598 CHECK_OFFSET(si_uid); 4599 CHECK_OFFSET(si_value); 4600 4601 /* sigchld */ 4602 CHECK_OFFSET(si_pid); 4603 CHECK_OFFSET(si_uid); 4604 CHECK_OFFSET(si_status); 4605 CHECK_OFFSET(si_utime); 4606 CHECK_OFFSET(si_stime); 4607 4608 /* sigfault */ 4609 CHECK_OFFSET(si_addr); 4610 CHECK_OFFSET(si_addr_lsb); 4611 CHECK_OFFSET(si_lower); 4612 CHECK_OFFSET(si_upper); 4613 CHECK_OFFSET(si_pkey); 4614 4615 /* sigpoll */ 4616 CHECK_OFFSET(si_band); 4617 CHECK_OFFSET(si_fd); 4618 4619 /* sigsys */ 4620 CHECK_OFFSET(si_call_addr); 4621 CHECK_OFFSET(si_syscall); 4622 CHECK_OFFSET(si_arch); 4623 #undef CHECK_OFFSET 4624 4625 /* usb asyncio */ 4626 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) != 4627 offsetof(struct siginfo, si_addr)); 4628 if (sizeof(int) == sizeof(void __user *)) { 4629 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) != 4630 sizeof(void __user *)); 4631 } else { 4632 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) + 4633 sizeof_field(struct siginfo, si_uid)) != 4634 sizeof(void __user *)); 4635 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) != 4636 offsetof(struct siginfo, si_uid)); 4637 } 4638 #ifdef CONFIG_COMPAT 4639 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) != 4640 offsetof(struct compat_siginfo, si_addr)); 4641 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) != 4642 sizeof(compat_uptr_t)); 4643 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) != 4644 sizeof_field(struct siginfo, si_pid)); 4645 #endif 4646 } 4647 4648 void __init signals_init(void) 4649 { 4650 siginfo_buildtime_checks(); 4651 4652 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC); 4653 } 4654 4655 #ifdef CONFIG_KGDB_KDB 4656 #include <linux/kdb.h> 4657 /* 4658 * kdb_send_sig - Allows kdb to send signals without exposing 4659 * signal internals. This function checks if the required locks are 4660 * available before calling the main signal code, to avoid kdb 4661 * deadlocks. 4662 */ 4663 void kdb_send_sig(struct task_struct *t, int sig) 4664 { 4665 static struct task_struct *kdb_prev_t; 4666 int new_t, ret; 4667 if (!spin_trylock(&t->sighand->siglock)) { 4668 kdb_printf("Can't do kill command now.\n" 4669 "The sigmask lock is held somewhere else in " 4670 "kernel, try again later\n"); 4671 return; 4672 } 4673 new_t = kdb_prev_t != t; 4674 kdb_prev_t = t; 4675 if (t->state != TASK_RUNNING && new_t) { 4676 spin_unlock(&t->sighand->siglock); 4677 kdb_printf("Process is not RUNNING, sending a signal from " 4678 "kdb risks deadlock\n" 4679 "on the run queue locks. " 4680 "The signal has _not_ been sent.\n" 4681 "Reissue the kill command if you want to risk " 4682 "the deadlock.\n"); 4683 return; 4684 } 4685 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID); 4686 spin_unlock(&t->sighand->siglock); 4687 if (ret) 4688 kdb_printf("Fail to deliver Signal %d to process %d.\n", 4689 sig, t->pid); 4690 else 4691 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid); 4692 } 4693 #endif /* CONFIG_KGDB_KDB */ 4694