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/mm.h> 26 #include <linux/proc_fs.h> 27 #include <linux/tty.h> 28 #include <linux/binfmts.h> 29 #include <linux/coredump.h> 30 #include <linux/security.h> 31 #include <linux/syscalls.h> 32 #include <linux/ptrace.h> 33 #include <linux/signal.h> 34 #include <linux/signalfd.h> 35 #include <linux/ratelimit.h> 36 #include <linux/task_work.h> 37 #include <linux/capability.h> 38 #include <linux/freezer.h> 39 #include <linux/pid_namespace.h> 40 #include <linux/nsproxy.h> 41 #include <linux/user_namespace.h> 42 #include <linux/uprobes.h> 43 #include <linux/compat.h> 44 #include <linux/cn_proc.h> 45 #include <linux/compiler.h> 46 #include <linux/posix-timers.h> 47 #include <linux/cgroup.h> 48 #include <linux/audit.h> 49 #include <linux/sysctl.h> 50 #include <uapi/linux/pidfd.h> 51 52 #define CREATE_TRACE_POINTS 53 #include <trace/events/signal.h> 54 55 #include <asm/param.h> 56 #include <linux/uaccess.h> 57 #include <asm/unistd.h> 58 #include <asm/siginfo.h> 59 #include <asm/cacheflush.h> 60 #include <asm/syscall.h> /* for syscall_get_* */ 61 62 #include "time/posix-timers.h" 63 64 /* 65 * SLAB caches for signal bits. 66 */ 67 68 static struct kmem_cache *sigqueue_cachep; 69 70 int print_fatal_signals __read_mostly; 71 72 static void __user *sig_handler(struct task_struct *t, int sig) 73 { 74 return t->sighand->action[sig - 1].sa.sa_handler; 75 } 76 77 static inline bool sig_handler_ignored(void __user *handler, int sig) 78 { 79 /* Is it explicitly or implicitly ignored? */ 80 return handler == SIG_IGN || 81 (handler == SIG_DFL && sig_kernel_ignore(sig)); 82 } 83 84 static bool sig_task_ignored(struct task_struct *t, int sig, bool force) 85 { 86 void __user *handler; 87 88 handler = sig_handler(t, sig); 89 90 /* SIGKILL and SIGSTOP may not be sent to the global init */ 91 if (unlikely(is_global_init(t) && sig_kernel_only(sig))) 92 return true; 93 94 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) && 95 handler == SIG_DFL && !(force && sig_kernel_only(sig))) 96 return true; 97 98 /* Only allow kernel generated signals to this kthread */ 99 if (unlikely((t->flags & PF_KTHREAD) && 100 (handler == SIG_KTHREAD_KERNEL) && !force)) 101 return true; 102 103 return sig_handler_ignored(handler, sig); 104 } 105 106 static bool sig_ignored(struct task_struct *t, int sig, bool force) 107 { 108 /* 109 * Blocked signals are never ignored, since the 110 * signal handler may change by the time it is 111 * unblocked. 112 */ 113 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig)) 114 return false; 115 116 /* 117 * Tracers may want to know about even ignored signal unless it 118 * is SIGKILL which can't be reported anyway but can be ignored 119 * by SIGNAL_UNKILLABLE task. 120 */ 121 if (t->ptrace && sig != SIGKILL) 122 return false; 123 124 return sig_task_ignored(t, sig, force); 125 } 126 127 /* 128 * Re-calculate pending state from the set of locally pending 129 * signals, globally pending signals, and blocked signals. 130 */ 131 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked) 132 { 133 unsigned long ready = 0; 134 for (long i = 0; i < _NSIG_WORDS; i++) 135 ready |= signal->sig[i] & ~blocked->sig[i]; 136 return ready != 0; 137 } 138 139 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b)) 140 141 static bool recalc_sigpending_tsk(struct task_struct *t) 142 { 143 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) || 144 PENDING(&t->pending, &t->blocked) || 145 PENDING(&t->signal->shared_pending, &t->blocked) || 146 cgroup_task_frozen(t)) { 147 set_tsk_thread_flag(t, TIF_SIGPENDING); 148 return true; 149 } 150 151 /* 152 * We must never clear the flag in another thread, or in current 153 * when it's possible the current syscall is returning -ERESTART*. 154 * So we don't clear it here, and only callers who know they should do. 155 */ 156 return false; 157 } 158 159 void recalc_sigpending(void) 160 { 161 if (!recalc_sigpending_tsk(current) && !freezing(current)) { 162 if (unlikely(test_thread_flag(TIF_SIGPENDING))) 163 clear_thread_flag(TIF_SIGPENDING); 164 } 165 } 166 EXPORT_SYMBOL(recalc_sigpending); 167 168 void calculate_sigpending(void) 169 { 170 /* Have any signals or users of TIF_SIGPENDING been delayed 171 * until after fork? 172 */ 173 spin_lock_irq(¤t->sighand->siglock); 174 set_tsk_thread_flag(current, TIF_SIGPENDING); 175 recalc_sigpending(); 176 spin_unlock_irq(¤t->sighand->siglock); 177 } 178 179 /* Given the mask, find the first available signal that should be serviced. */ 180 181 #define SYNCHRONOUS_MASK \ 182 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \ 183 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS)) 184 185 int next_signal(struct sigpending *pending, sigset_t *mask) 186 { 187 unsigned long i, *s, *m, x; 188 int sig = 0; 189 190 s = pending->signal.sig; 191 m = mask->sig; 192 193 /* 194 * Handle the first word specially: it contains the 195 * synchronous signals that need to be dequeued first. 196 */ 197 x = *s &~ *m; 198 if (x) { 199 if (x & SYNCHRONOUS_MASK) 200 x &= SYNCHRONOUS_MASK; 201 sig = ffz(~x) + 1; 202 return sig; 203 } 204 205 switch (_NSIG_WORDS) { 206 default: 207 for (i = 1; i < _NSIG_WORDS; ++i) { 208 x = *++s &~ *++m; 209 if (!x) 210 continue; 211 sig = ffz(~x) + i*_NSIG_BPW + 1; 212 break; 213 } 214 break; 215 216 case 2: 217 x = s[1] &~ m[1]; 218 if (!x) 219 break; 220 sig = ffz(~x) + _NSIG_BPW + 1; 221 break; 222 223 case 1: 224 /* Nothing to do */ 225 break; 226 } 227 228 return sig; 229 } 230 231 static inline void print_dropped_signal(int sig) 232 { 233 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10); 234 235 if (!print_fatal_signals) 236 return; 237 238 if (!__ratelimit(&ratelimit_state)) 239 return; 240 241 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n", 242 current->comm, current->pid, sig); 243 } 244 245 /** 246 * task_set_jobctl_pending - set jobctl pending bits 247 * @task: target task 248 * @mask: pending bits to set 249 * 250 * Clear @mask from @task->jobctl. @mask must be subset of 251 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK | 252 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is 253 * cleared. If @task is already being killed or exiting, this function 254 * becomes noop. 255 * 256 * CONTEXT: 257 * Must be called with @task->sighand->siglock held. 258 * 259 * RETURNS: 260 * %true if @mask is set, %false if made noop because @task was dying. 261 */ 262 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask) 263 { 264 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME | 265 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING)); 266 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK)); 267 268 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING))) 269 return false; 270 271 if (mask & JOBCTL_STOP_SIGMASK) 272 task->jobctl &= ~JOBCTL_STOP_SIGMASK; 273 274 task->jobctl |= mask; 275 return true; 276 } 277 278 /** 279 * task_clear_jobctl_trapping - clear jobctl trapping bit 280 * @task: target task 281 * 282 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED. 283 * Clear it and wake up the ptracer. Note that we don't need any further 284 * locking. @task->siglock guarantees that @task->parent points to the 285 * ptracer. 286 * 287 * CONTEXT: 288 * Must be called with @task->sighand->siglock held. 289 */ 290 void task_clear_jobctl_trapping(struct task_struct *task) 291 { 292 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) { 293 task->jobctl &= ~JOBCTL_TRAPPING; 294 smp_mb(); /* advised by wake_up_bit() */ 295 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT); 296 } 297 } 298 299 /** 300 * task_clear_jobctl_pending - clear jobctl pending bits 301 * @task: target task 302 * @mask: pending bits to clear 303 * 304 * Clear @mask from @task->jobctl. @mask must be subset of 305 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other 306 * STOP bits are cleared together. 307 * 308 * If clearing of @mask leaves no stop or trap pending, this function calls 309 * task_clear_jobctl_trapping(). 310 * 311 * CONTEXT: 312 * Must be called with @task->sighand->siglock held. 313 */ 314 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask) 315 { 316 BUG_ON(mask & ~JOBCTL_PENDING_MASK); 317 318 if (mask & JOBCTL_STOP_PENDING) 319 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED; 320 321 task->jobctl &= ~mask; 322 323 if (!(task->jobctl & JOBCTL_PENDING_MASK)) 324 task_clear_jobctl_trapping(task); 325 } 326 327 /** 328 * task_participate_group_stop - participate in a group stop 329 * @task: task participating in a group stop 330 * 331 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop. 332 * Group stop states are cleared and the group stop count is consumed if 333 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group 334 * stop, the appropriate `SIGNAL_*` flags are set. 335 * 336 * CONTEXT: 337 * Must be called with @task->sighand->siglock held. 338 * 339 * RETURNS: 340 * %true if group stop completion should be notified to the parent, %false 341 * otherwise. 342 */ 343 static bool task_participate_group_stop(struct task_struct *task) 344 { 345 struct signal_struct *sig = task->signal; 346 bool consume = task->jobctl & JOBCTL_STOP_CONSUME; 347 348 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING)); 349 350 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING); 351 352 if (!consume) 353 return false; 354 355 if (!WARN_ON_ONCE(sig->group_stop_count == 0)) 356 sig->group_stop_count--; 357 358 /* 359 * Tell the caller to notify completion iff we are entering into a 360 * fresh group stop. Read comment in do_signal_stop() for details. 361 */ 362 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) { 363 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED); 364 return true; 365 } 366 return false; 367 } 368 369 void task_join_group_stop(struct task_struct *task) 370 { 371 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK; 372 struct signal_struct *sig = current->signal; 373 374 if (sig->group_stop_count) { 375 sig->group_stop_count++; 376 mask |= JOBCTL_STOP_CONSUME; 377 } else if (!(sig->flags & SIGNAL_STOP_STOPPED)) 378 return; 379 380 /* Have the new thread join an on-going signal group stop */ 381 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING); 382 } 383 384 static struct ucounts *sig_get_ucounts(struct task_struct *t, int sig, 385 int override_rlimit) 386 { 387 struct ucounts *ucounts; 388 long sigpending; 389 390 /* 391 * Protect access to @t credentials. This can go away when all 392 * callers hold rcu read lock. 393 * 394 * NOTE! A pending signal will hold on to the user refcount, 395 * and we get/put the refcount only when the sigpending count 396 * changes from/to zero. 397 */ 398 rcu_read_lock(); 399 ucounts = task_ucounts(t); 400 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 401 override_rlimit); 402 rcu_read_unlock(); 403 if (!sigpending) 404 return NULL; 405 406 if (unlikely(!override_rlimit && sigpending > task_rlimit(t, RLIMIT_SIGPENDING))) { 407 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING); 408 print_dropped_signal(sig); 409 return NULL; 410 } 411 412 return ucounts; 413 } 414 415 static void __sigqueue_init(struct sigqueue *q, struct ucounts *ucounts, 416 const unsigned int sigqueue_flags) 417 { 418 INIT_LIST_HEAD(&q->list); 419 q->flags = sigqueue_flags; 420 q->ucounts = ucounts; 421 } 422 423 /* 424 * allocate a new signal queue record 425 * - this may be called without locks if and only if t == current, otherwise an 426 * appropriate lock must be held to stop the target task from exiting 427 */ 428 static struct sigqueue *sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags, 429 int override_rlimit) 430 { 431 struct ucounts *ucounts = sig_get_ucounts(t, sig, override_rlimit); 432 struct sigqueue *q; 433 434 if (!ucounts) 435 return NULL; 436 437 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags); 438 if (!q) { 439 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING); 440 return NULL; 441 } 442 443 __sigqueue_init(q, ucounts, 0); 444 return q; 445 } 446 447 static void __sigqueue_free(struct sigqueue *q) 448 { 449 if (q->flags & SIGQUEUE_PREALLOC) { 450 posixtimer_sigqueue_putref(q); 451 return; 452 } 453 if (q->ucounts) { 454 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING); 455 q->ucounts = NULL; 456 } 457 kmem_cache_free(sigqueue_cachep, q); 458 } 459 460 void flush_sigqueue(struct sigpending *queue) 461 { 462 struct sigqueue *q; 463 464 sigemptyset(&queue->signal); 465 while (!list_empty(&queue->list)) { 466 q = list_entry(queue->list.next, struct sigqueue , list); 467 list_del_init(&q->list); 468 __sigqueue_free(q); 469 } 470 } 471 472 /* 473 * Flush all pending signals for this kthread. 474 */ 475 void flush_signals(struct task_struct *t) 476 { 477 unsigned long flags; 478 479 spin_lock_irqsave(&t->sighand->siglock, flags); 480 clear_tsk_thread_flag(t, TIF_SIGPENDING); 481 flush_sigqueue(&t->pending); 482 flush_sigqueue(&t->signal->shared_pending); 483 spin_unlock_irqrestore(&t->sighand->siglock, flags); 484 } 485 EXPORT_SYMBOL(flush_signals); 486 487 void ignore_signals(struct task_struct *t) 488 { 489 int i; 490 491 for (i = 0; i < _NSIG; ++i) 492 t->sighand->action[i].sa.sa_handler = SIG_IGN; 493 494 flush_signals(t); 495 } 496 497 /* 498 * Flush all handlers for a task. 499 */ 500 501 void 502 flush_signal_handlers(struct task_struct *t, int force_default) 503 { 504 int i; 505 struct k_sigaction *ka = &t->sighand->action[0]; 506 for (i = _NSIG ; i != 0 ; i--) { 507 if (force_default || ka->sa.sa_handler != SIG_IGN) 508 ka->sa.sa_handler = SIG_DFL; 509 ka->sa.sa_flags = 0; 510 #ifdef __ARCH_HAS_SA_RESTORER 511 ka->sa.sa_restorer = NULL; 512 #endif 513 sigemptyset(&ka->sa.sa_mask); 514 ka++; 515 } 516 } 517 518 bool unhandled_signal(struct task_struct *tsk, int sig) 519 { 520 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler; 521 if (is_global_init(tsk)) 522 return true; 523 524 if (handler != SIG_IGN && handler != SIG_DFL) 525 return false; 526 527 /* If dying, we handle all new signals by ignoring them */ 528 if (fatal_signal_pending(tsk)) 529 return false; 530 531 /* if ptraced, let the tracer determine */ 532 return !tsk->ptrace; 533 } 534 535 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info, 536 struct sigqueue **timer_sigq) 537 { 538 struct sigqueue *q, *first = NULL; 539 540 /* 541 * Collect the siginfo appropriate to this signal. Check if 542 * there is another siginfo for the same signal. 543 */ 544 list_for_each_entry(q, &list->list, list) { 545 if (q->info.si_signo == sig) { 546 if (first) 547 goto still_pending; 548 first = q; 549 } 550 } 551 552 sigdelset(&list->signal, sig); 553 554 if (first) { 555 still_pending: 556 list_del_init(&first->list); 557 copy_siginfo(info, &first->info); 558 559 /* 560 * posix-timer signals are preallocated and freed when the last 561 * reference count is dropped in posixtimer_deliver_signal() or 562 * immediately on timer deletion when the signal is not pending. 563 * Spare the extra round through __sigqueue_free() which is 564 * ignoring preallocated signals. 565 */ 566 if (unlikely((first->flags & SIGQUEUE_PREALLOC) && (info->si_code == SI_TIMER))) 567 *timer_sigq = first; 568 else 569 __sigqueue_free(first); 570 } else { 571 /* 572 * Ok, it wasn't in the queue. This must be 573 * a fast-pathed signal or we must have been 574 * out of queue space. So zero out the info. 575 */ 576 clear_siginfo(info); 577 info->si_signo = sig; 578 info->si_errno = 0; 579 info->si_code = SI_USER; 580 info->si_pid = 0; 581 info->si_uid = 0; 582 } 583 } 584 585 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask, 586 kernel_siginfo_t *info, struct sigqueue **timer_sigq) 587 { 588 int sig = next_signal(pending, mask); 589 590 if (sig) 591 collect_signal(sig, pending, info, timer_sigq); 592 return sig; 593 } 594 595 /* 596 * Try to dequeue a signal. If a deliverable signal is found fill in the 597 * caller provided siginfo and return the signal number. Otherwise return 598 * 0. 599 */ 600 int dequeue_signal(sigset_t *mask, kernel_siginfo_t *info, enum pid_type *type) 601 { 602 struct task_struct *tsk = current; 603 struct sigqueue *timer_sigq; 604 int signr; 605 606 lockdep_assert_held(&tsk->sighand->siglock); 607 608 again: 609 *type = PIDTYPE_PID; 610 timer_sigq = NULL; 611 signr = __dequeue_signal(&tsk->pending, mask, info, &timer_sigq); 612 if (!signr) { 613 *type = PIDTYPE_TGID; 614 signr = __dequeue_signal(&tsk->signal->shared_pending, 615 mask, info, &timer_sigq); 616 617 if (unlikely(signr == SIGALRM)) 618 posixtimer_rearm_itimer(tsk); 619 } 620 621 recalc_sigpending(); 622 if (!signr) 623 return 0; 624 625 if (unlikely(sig_kernel_stop(signr))) { 626 /* 627 * Set a marker that we have dequeued a stop signal. Our 628 * caller might release the siglock and then the pending 629 * stop signal it is about to process is no longer in the 630 * pending bitmasks, but must still be cleared by a SIGCONT 631 * (and overruled by a SIGKILL). So those cases clear this 632 * shared flag after we've set it. Note that this flag may 633 * remain set after the signal we return is ignored or 634 * handled. That doesn't matter because its only purpose 635 * is to alert stop-signal processing code when another 636 * processor has come along and cleared the flag. 637 */ 638 current->jobctl |= JOBCTL_STOP_DEQUEUED; 639 } 640 641 if (IS_ENABLED(CONFIG_POSIX_TIMERS) && unlikely(timer_sigq)) { 642 if (!posixtimer_deliver_signal(info, timer_sigq)) 643 goto again; 644 } 645 646 return signr; 647 } 648 EXPORT_SYMBOL_GPL(dequeue_signal); 649 650 static int dequeue_synchronous_signal(kernel_siginfo_t *info) 651 { 652 struct task_struct *tsk = current; 653 struct sigpending *pending = &tsk->pending; 654 struct sigqueue *q, *sync = NULL; 655 656 /* 657 * Might a synchronous signal be in the queue? 658 */ 659 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK)) 660 return 0; 661 662 /* 663 * Return the first synchronous signal in the queue. 664 */ 665 list_for_each_entry(q, &pending->list, list) { 666 /* Synchronous signals have a positive si_code */ 667 if ((q->info.si_code > SI_USER) && 668 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) { 669 sync = q; 670 goto next; 671 } 672 } 673 return 0; 674 next: 675 /* 676 * Check if there is another siginfo for the same signal. 677 */ 678 list_for_each_entry_continue(q, &pending->list, list) { 679 if (q->info.si_signo == sync->info.si_signo) 680 goto still_pending; 681 } 682 683 sigdelset(&pending->signal, sync->info.si_signo); 684 recalc_sigpending(); 685 still_pending: 686 list_del_init(&sync->list); 687 copy_siginfo(info, &sync->info); 688 __sigqueue_free(sync); 689 return info->si_signo; 690 } 691 692 /* 693 * Tell a process that it has a new active signal.. 694 * 695 * NOTE! we rely on the previous spin_lock to 696 * lock interrupts for us! We can only be called with 697 * "siglock" held, and the local interrupt must 698 * have been disabled when that got acquired! 699 * 700 * No need to set need_resched since signal event passing 701 * goes through ->blocked 702 */ 703 void signal_wake_up_state(struct task_struct *t, unsigned int state) 704 { 705 lockdep_assert_held(&t->sighand->siglock); 706 707 set_tsk_thread_flag(t, TIF_SIGPENDING); 708 709 /* 710 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable 711 * case. We don't check t->state here because there is a race with it 712 * executing another processor and just now entering stopped state. 713 * By using wake_up_state, we ensure the process will wake up and 714 * handle its death signal. 715 */ 716 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE)) 717 kick_process(t); 718 } 719 720 static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q); 721 722 static void sigqueue_free_ignored(struct task_struct *tsk, struct sigqueue *q) 723 { 724 if (likely(!(q->flags & SIGQUEUE_PREALLOC) || q->info.si_code != SI_TIMER)) 725 __sigqueue_free(q); 726 else 727 posixtimer_sig_ignore(tsk, q); 728 } 729 730 /* Remove signals in mask from the pending set and queue. */ 731 static void flush_sigqueue_mask(struct task_struct *p, sigset_t *mask, struct sigpending *s) 732 { 733 struct sigqueue *q, *n; 734 sigset_t m; 735 736 lockdep_assert_held(&p->sighand->siglock); 737 738 sigandsets(&m, mask, &s->signal); 739 if (sigisemptyset(&m)) 740 return; 741 742 sigandnsets(&s->signal, &s->signal, mask); 743 list_for_each_entry_safe(q, n, &s->list, list) { 744 if (sigismember(mask, q->info.si_signo)) { 745 list_del_init(&q->list); 746 sigqueue_free_ignored(p, q); 747 } 748 } 749 } 750 751 static inline int is_si_special(const struct kernel_siginfo *info) 752 { 753 return info <= SEND_SIG_PRIV; 754 } 755 756 static inline bool si_fromuser(const struct kernel_siginfo *info) 757 { 758 return info == SEND_SIG_NOINFO || 759 (!is_si_special(info) && SI_FROMUSER(info)); 760 } 761 762 /* 763 * called with RCU read lock from check_kill_permission() 764 */ 765 static bool kill_ok_by_cred(struct task_struct *t) 766 { 767 const struct cred *cred = current_cred(); 768 const struct cred *tcred = __task_cred(t); 769 770 return uid_eq(cred->euid, tcred->suid) || 771 uid_eq(cred->euid, tcred->uid) || 772 uid_eq(cred->uid, tcred->suid) || 773 uid_eq(cred->uid, tcred->uid) || 774 ns_capable(tcred->user_ns, CAP_KILL); 775 } 776 777 /* 778 * Bad permissions for sending the signal 779 * - the caller must hold the RCU read lock 780 */ 781 static int check_kill_permission(int sig, struct kernel_siginfo *info, 782 struct task_struct *t) 783 { 784 struct pid *sid; 785 int error; 786 787 if (!valid_signal(sig)) 788 return -EINVAL; 789 790 if (!si_fromuser(info)) 791 return 0; 792 793 error = audit_signal_info(sig, t); /* Let audit system see the signal */ 794 if (error) 795 return error; 796 797 if (!same_thread_group(current, t) && 798 !kill_ok_by_cred(t)) { 799 switch (sig) { 800 case SIGCONT: 801 sid = task_session(t); 802 /* 803 * We don't return the error if sid == NULL. The 804 * task was unhashed, the caller must notice this. 805 */ 806 if (!sid || sid == task_session(current)) 807 break; 808 fallthrough; 809 default: 810 return -EPERM; 811 } 812 } 813 814 return security_task_kill(t, info, sig, NULL); 815 } 816 817 /** 818 * ptrace_trap_notify - schedule trap to notify ptracer 819 * @t: tracee wanting to notify tracer 820 * 821 * This function schedules sticky ptrace trap which is cleared on the next 822 * TRAP_STOP to notify ptracer of an event. @t must have been seized by 823 * ptracer. 824 * 825 * If @t is running, STOP trap will be taken. If trapped for STOP and 826 * ptracer is listening for events, tracee is woken up so that it can 827 * re-trap for the new event. If trapped otherwise, STOP trap will be 828 * eventually taken without returning to userland after the existing traps 829 * are finished by PTRACE_CONT. 830 * 831 * CONTEXT: 832 * Must be called with @task->sighand->siglock held. 833 */ 834 static void ptrace_trap_notify(struct task_struct *t) 835 { 836 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED)); 837 lockdep_assert_held(&t->sighand->siglock); 838 839 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY); 840 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING); 841 } 842 843 /* 844 * Handle magic process-wide effects of stop/continue signals. Unlike 845 * the signal actions, these happen immediately at signal-generation 846 * time regardless of blocking, ignoring, or handling. This does the 847 * actual continuing for SIGCONT, but not the actual stopping for stop 848 * signals. The process stop is done as a signal action for SIG_DFL. 849 * 850 * Returns true if the signal should be actually delivered, otherwise 851 * it should be dropped. 852 */ 853 static bool prepare_signal(int sig, struct task_struct *p, bool force) 854 { 855 struct signal_struct *signal = p->signal; 856 struct task_struct *t; 857 sigset_t flush; 858 859 if (signal->flags & SIGNAL_GROUP_EXIT) { 860 if (signal->core_state) 861 return sig == SIGKILL; 862 /* 863 * The process is in the middle of dying, drop the signal. 864 */ 865 return false; 866 } else if (sig_kernel_stop(sig)) { 867 /* 868 * This is a stop signal. Remove SIGCONT from all queues. 869 */ 870 siginitset(&flush, sigmask(SIGCONT)); 871 flush_sigqueue_mask(p, &flush, &signal->shared_pending); 872 for_each_thread(p, t) 873 flush_sigqueue_mask(p, &flush, &t->pending); 874 } else if (sig == SIGCONT) { 875 unsigned int why; 876 /* 877 * Remove all stop signals from all queues, wake all threads. 878 */ 879 siginitset(&flush, SIG_KERNEL_STOP_MASK); 880 flush_sigqueue_mask(p, &flush, &signal->shared_pending); 881 for_each_thread(p, t) { 882 flush_sigqueue_mask(p, &flush, &t->pending); 883 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING); 884 if (likely(!(t->ptrace & PT_SEIZED))) { 885 t->jobctl &= ~JOBCTL_STOPPED; 886 wake_up_state(t, __TASK_STOPPED); 887 } else 888 ptrace_trap_notify(t); 889 } 890 891 /* 892 * Notify the parent with CLD_CONTINUED if we were stopped. 893 * 894 * If we were in the middle of a group stop, we pretend it 895 * was already finished, and then continued. Since SIGCHLD 896 * doesn't queue we report only CLD_STOPPED, as if the next 897 * CLD_CONTINUED was dropped. 898 */ 899 why = 0; 900 if (signal->flags & SIGNAL_STOP_STOPPED) 901 why |= SIGNAL_CLD_CONTINUED; 902 else if (signal->group_stop_count) 903 why |= SIGNAL_CLD_STOPPED; 904 905 if (why) { 906 /* 907 * The first thread which returns from do_signal_stop() 908 * will take ->siglock, notice SIGNAL_CLD_MASK, and 909 * notify its parent. See get_signal(). 910 */ 911 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED); 912 signal->group_stop_count = 0; 913 signal->group_exit_code = 0; 914 } 915 } 916 917 return !sig_ignored(p, sig, force); 918 } 919 920 /* 921 * Test if P wants to take SIG. After we've checked all threads with this, 922 * it's equivalent to finding no threads not blocking SIG. Any threads not 923 * blocking SIG were ruled out because they are not running and already 924 * have pending signals. Such threads will dequeue from the shared queue 925 * as soon as they're available, so putting the signal on the shared queue 926 * will be equivalent to sending it to one such thread. 927 */ 928 static inline bool wants_signal(int sig, struct task_struct *p) 929 { 930 if (sigismember(&p->blocked, sig)) 931 return false; 932 933 if (p->flags & PF_EXITING) 934 return false; 935 936 if (sig == SIGKILL) 937 return true; 938 939 if (task_is_stopped_or_traced(p)) 940 return false; 941 942 return task_curr(p) || !task_sigpending(p); 943 } 944 945 static void complete_signal(int sig, struct task_struct *p, enum pid_type type) 946 { 947 struct signal_struct *signal = p->signal; 948 struct task_struct *t; 949 950 /* 951 * Now find a thread we can wake up to take the signal off the queue. 952 * 953 * Try the suggested task first (may or may not be the main thread). 954 */ 955 if (wants_signal(sig, p)) 956 t = p; 957 else if ((type == PIDTYPE_PID) || thread_group_empty(p)) 958 /* 959 * There is just one thread and it does not need to be woken. 960 * It will dequeue unblocked signals before it runs again. 961 */ 962 return; 963 else { 964 /* 965 * Otherwise try to find a suitable thread. 966 */ 967 t = signal->curr_target; 968 while (!wants_signal(sig, t)) { 969 t = next_thread(t); 970 if (t == signal->curr_target) 971 /* 972 * No thread needs to be woken. 973 * Any eligible threads will see 974 * the signal in the queue soon. 975 */ 976 return; 977 } 978 signal->curr_target = t; 979 } 980 981 /* 982 * Found a killable thread. If the signal will be fatal, 983 * then start taking the whole group down immediately. 984 */ 985 if (sig_fatal(p, sig) && !sigismember(&t->real_blocked, sig) && 986 (sig == SIGKILL || !p->ptrace)) { 987 /* 988 * This signal will be fatal to the whole group. 989 */ 990 if (!sig_kernel_coredump(sig)) { 991 /* 992 * Start a group exit and wake everybody up. 993 * This way we don't have other threads 994 * running and doing things after a slower 995 * thread has the fatal signal pending. 996 */ 997 signal->flags = SIGNAL_GROUP_EXIT; 998 signal->group_exit_code = sig; 999 signal->group_stop_count = 0; 1000 __for_each_thread(signal, t) { 1001 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); 1002 sigaddset(&t->pending.signal, SIGKILL); 1003 signal_wake_up(t, 1); 1004 } 1005 return; 1006 } 1007 } 1008 1009 /* 1010 * The signal is already in the shared-pending queue. 1011 * Tell the chosen thread to wake up and dequeue it. 1012 */ 1013 signal_wake_up(t, sig == SIGKILL); 1014 return; 1015 } 1016 1017 static inline bool legacy_queue(struct sigpending *signals, int sig) 1018 { 1019 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig); 1020 } 1021 1022 static int __send_signal_locked(int sig, struct kernel_siginfo *info, 1023 struct task_struct *t, enum pid_type type, bool force) 1024 { 1025 struct sigpending *pending; 1026 struct sigqueue *q; 1027 int override_rlimit; 1028 int ret = 0, result; 1029 1030 lockdep_assert_held(&t->sighand->siglock); 1031 1032 result = TRACE_SIGNAL_IGNORED; 1033 if (!prepare_signal(sig, t, force)) 1034 goto ret; 1035 1036 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending; 1037 /* 1038 * Short-circuit ignored signals and support queuing 1039 * exactly one non-rt signal, so that we can get more 1040 * detailed information about the cause of the signal. 1041 */ 1042 result = TRACE_SIGNAL_ALREADY_PENDING; 1043 if (legacy_queue(pending, sig)) 1044 goto ret; 1045 1046 result = TRACE_SIGNAL_DELIVERED; 1047 /* 1048 * Skip useless siginfo allocation for SIGKILL and kernel threads. 1049 */ 1050 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD)) 1051 goto out_set; 1052 1053 /* 1054 * Real-time signals must be queued if sent by sigqueue, or 1055 * some other real-time mechanism. It is implementation 1056 * defined whether kill() does so. We attempt to do so, on 1057 * the principle of least surprise, but since kill is not 1058 * allowed to fail with EAGAIN when low on memory we just 1059 * make sure at least one signal gets delivered and don't 1060 * pass on the info struct. 1061 */ 1062 if (sig < SIGRTMIN) 1063 override_rlimit = (is_si_special(info) || info->si_code >= 0); 1064 else 1065 override_rlimit = 0; 1066 1067 q = sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit); 1068 1069 if (q) { 1070 list_add_tail(&q->list, &pending->list); 1071 switch ((unsigned long) info) { 1072 case (unsigned long) SEND_SIG_NOINFO: 1073 clear_siginfo(&q->info); 1074 q->info.si_signo = sig; 1075 q->info.si_errno = 0; 1076 q->info.si_code = SI_USER; 1077 q->info.si_pid = task_tgid_nr_ns(current, 1078 task_active_pid_ns(t)); 1079 rcu_read_lock(); 1080 q->info.si_uid = 1081 from_kuid_munged(task_cred_xxx(t, user_ns), 1082 current_uid()); 1083 rcu_read_unlock(); 1084 break; 1085 case (unsigned long) SEND_SIG_PRIV: 1086 clear_siginfo(&q->info); 1087 q->info.si_signo = sig; 1088 q->info.si_errno = 0; 1089 q->info.si_code = SI_KERNEL; 1090 q->info.si_pid = 0; 1091 q->info.si_uid = 0; 1092 break; 1093 default: 1094 copy_siginfo(&q->info, info); 1095 break; 1096 } 1097 } else if (!is_si_special(info) && 1098 sig >= SIGRTMIN && info->si_code != SI_USER) { 1099 /* 1100 * Queue overflow, abort. We may abort if the 1101 * signal was rt and sent by user using something 1102 * other than kill(). 1103 */ 1104 result = TRACE_SIGNAL_OVERFLOW_FAIL; 1105 ret = -EAGAIN; 1106 goto ret; 1107 } else { 1108 /* 1109 * This is a silent loss of information. We still 1110 * send the signal, but the *info bits are lost. 1111 */ 1112 result = TRACE_SIGNAL_LOSE_INFO; 1113 } 1114 1115 out_set: 1116 signalfd_notify(t, sig); 1117 sigaddset(&pending->signal, sig); 1118 1119 /* Let multiprocess signals appear after on-going forks */ 1120 if (type > PIDTYPE_TGID) { 1121 struct multiprocess_signals *delayed; 1122 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) { 1123 sigset_t *signal = &delayed->signal; 1124 /* Can't queue both a stop and a continue signal */ 1125 if (sig == SIGCONT) 1126 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK); 1127 else if (sig_kernel_stop(sig)) 1128 sigdelset(signal, SIGCONT); 1129 sigaddset(signal, sig); 1130 } 1131 } 1132 1133 complete_signal(sig, t, type); 1134 ret: 1135 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result); 1136 return ret; 1137 } 1138 1139 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info) 1140 { 1141 bool ret = false; 1142 switch (siginfo_layout(info->si_signo, info->si_code)) { 1143 case SIL_KILL: 1144 case SIL_CHLD: 1145 case SIL_RT: 1146 ret = true; 1147 break; 1148 case SIL_TIMER: 1149 case SIL_POLL: 1150 case SIL_FAULT: 1151 case SIL_FAULT_TRAPNO: 1152 case SIL_FAULT_MCEERR: 1153 case SIL_FAULT_BNDERR: 1154 case SIL_FAULT_PKUERR: 1155 case SIL_FAULT_PERF_EVENT: 1156 case SIL_SYS: 1157 ret = false; 1158 break; 1159 } 1160 return ret; 1161 } 1162 1163 int send_signal_locked(int sig, struct kernel_siginfo *info, 1164 struct task_struct *t, enum pid_type type) 1165 { 1166 struct kernel_siginfo __maybe_unused rewritten; 1167 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */ 1168 bool force = false; 1169 1170 if (info == SEND_SIG_NOINFO) { 1171 /* Force if sent from an ancestor pid namespace */ 1172 force = !task_pid_nr_ns(current, task_active_pid_ns(t)); 1173 } else if (info == SEND_SIG_PRIV) { 1174 /* Don't ignore kernel generated signals */ 1175 force = true; 1176 } else if (has_si_pid_and_uid(info)) { 1177 /* SIGKILL and SIGSTOP is special or has ids */ 1178 #ifdef CONFIG_USER_NS 1179 struct user_namespace *t_user_ns; 1180 kuid_t uid; 1181 1182 rcu_read_lock(); 1183 t_user_ns = task_cred_xxx(t, user_ns); 1184 if (current_user_ns() != t_user_ns) { 1185 rewritten = *info; 1186 info = &rewritten; 1187 uid = make_kuid(current_user_ns(), info->si_uid); 1188 rewritten.si_uid = from_kuid_munged(t_user_ns, uid); 1189 } 1190 rcu_read_unlock(); 1191 #endif 1192 /* A kernel generated signal? */ 1193 force = (info->si_code == SI_KERNEL); 1194 1195 #ifdef CONFIG_PID_NS 1196 /* From an ancestor pid namespace? */ 1197 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) { 1198 if (info != &rewritten) { 1199 rewritten = *info; 1200 info = &rewritten; 1201 } 1202 rewritten.si_pid = 0; 1203 force = true; 1204 } 1205 #endif 1206 } 1207 return __send_signal_locked(sig, info, t, type, force); 1208 } 1209 1210 static void print_fatal_signal(int signr) 1211 { 1212 struct pt_regs *regs = task_pt_regs(current); 1213 struct file *exe_file; 1214 1215 exe_file = get_task_exe_file(current); 1216 if (exe_file) { 1217 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n", 1218 exe_file, current->comm, signr); 1219 fput(exe_file); 1220 } else { 1221 pr_info("%s: potentially unexpected fatal signal %d.\n", 1222 current->comm, signr); 1223 } 1224 1225 #if defined(__i386__) && !defined(__arch_um__) 1226 pr_info("code at %08lx: ", regs->ip); 1227 { 1228 int i; 1229 for (i = 0; i < 16; i++) { 1230 unsigned char insn; 1231 1232 if (get_user(insn, (unsigned char *)(regs->ip + i))) 1233 break; 1234 pr_cont("%02x ", insn); 1235 } 1236 } 1237 pr_cont("\n"); 1238 #endif 1239 preempt_disable(); 1240 show_regs(regs); 1241 preempt_enable(); 1242 } 1243 1244 static int __init setup_print_fatal_signals(char *str) 1245 { 1246 get_option (&str, &print_fatal_signals); 1247 1248 return 1; 1249 } 1250 1251 __setup("print-fatal-signals=", setup_print_fatal_signals); 1252 1253 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p, 1254 enum pid_type type) 1255 { 1256 unsigned long flags; 1257 int ret = -ESRCH; 1258 1259 if (lock_task_sighand(p, &flags)) { 1260 ret = send_signal_locked(sig, info, p, type); 1261 unlock_task_sighand(p, &flags); 1262 } 1263 1264 return ret; 1265 } 1266 1267 enum sig_handler { 1268 HANDLER_CURRENT, /* If reachable use the current handler */ 1269 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */ 1270 HANDLER_EXIT, /* Only visible as the process exit code */ 1271 }; 1272 1273 /* 1274 * Force a signal that the process can't ignore: if necessary 1275 * we unblock the signal and change any SIG_IGN to SIG_DFL. 1276 * 1277 * Note: If we unblock the signal, we always reset it to SIG_DFL, 1278 * since we do not want to have a signal handler that was blocked 1279 * be invoked when user space had explicitly blocked it. 1280 * 1281 * We don't want to have recursive SIGSEGV's etc, for example, 1282 * that is why we also clear SIGNAL_UNKILLABLE. 1283 */ 1284 static int 1285 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t, 1286 enum sig_handler handler) 1287 { 1288 unsigned long int flags; 1289 int ret, blocked, ignored; 1290 struct k_sigaction *action; 1291 int sig = info->si_signo; 1292 1293 spin_lock_irqsave(&t->sighand->siglock, flags); 1294 action = &t->sighand->action[sig-1]; 1295 ignored = action->sa.sa_handler == SIG_IGN; 1296 blocked = sigismember(&t->blocked, sig); 1297 if (blocked || ignored || (handler != HANDLER_CURRENT)) { 1298 action->sa.sa_handler = SIG_DFL; 1299 if (handler == HANDLER_EXIT) 1300 action->sa.sa_flags |= SA_IMMUTABLE; 1301 if (blocked) 1302 sigdelset(&t->blocked, sig); 1303 } 1304 /* 1305 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect 1306 * debugging to leave init killable. But HANDLER_EXIT is always fatal. 1307 */ 1308 if (action->sa.sa_handler == SIG_DFL && 1309 (!t->ptrace || (handler == HANDLER_EXIT))) 1310 t->signal->flags &= ~SIGNAL_UNKILLABLE; 1311 ret = send_signal_locked(sig, info, t, PIDTYPE_PID); 1312 /* This can happen if the signal was already pending and blocked */ 1313 if (!task_sigpending(t)) 1314 signal_wake_up(t, 0); 1315 spin_unlock_irqrestore(&t->sighand->siglock, flags); 1316 1317 return ret; 1318 } 1319 1320 int force_sig_info(struct kernel_siginfo *info) 1321 { 1322 return force_sig_info_to_task(info, current, HANDLER_CURRENT); 1323 } 1324 1325 /* 1326 * Nuke all other threads in the group. 1327 */ 1328 int zap_other_threads(struct task_struct *p) 1329 { 1330 struct task_struct *t; 1331 int count = 0; 1332 1333 p->signal->group_stop_count = 0; 1334 task_clear_jobctl_pending(p, JOBCTL_PENDING_MASK); 1335 1336 for_other_threads(p, t) { 1337 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); 1338 count++; 1339 1340 /* Don't bother with already dead threads */ 1341 if (t->exit_state) 1342 continue; 1343 sigaddset(&t->pending.signal, SIGKILL); 1344 signal_wake_up(t, 1); 1345 } 1346 1347 return count; 1348 } 1349 1350 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, 1351 unsigned long *flags) 1352 { 1353 struct sighand_struct *sighand; 1354 1355 rcu_read_lock(); 1356 for (;;) { 1357 sighand = rcu_dereference(tsk->sighand); 1358 if (unlikely(sighand == NULL)) { 1359 /* 1360 * Pairs with the smp_store_release() in 1361 * __exit_signal(). It ensures that all state 1362 * modifications to the task preceeding the store are 1363 * visible to the callers of lock_task_sighand(). 1364 */ 1365 smp_acquire__after_ctrl_dep(); 1366 break; 1367 } 1368 1369 /* 1370 * This sighand can be already freed and even reused, but 1371 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which 1372 * initializes ->siglock: this slab can't go away, it has 1373 * the same object type, ->siglock can't be reinitialized. 1374 * 1375 * We need to ensure that tsk->sighand is still the same 1376 * after we take the lock, we can race with de_thread() or 1377 * __exit_signal(). In the latter case the next iteration 1378 * must see ->sighand == NULL. 1379 */ 1380 spin_lock_irqsave(&sighand->siglock, *flags); 1381 if (likely(sighand == rcu_access_pointer(tsk->sighand))) 1382 break; 1383 spin_unlock_irqrestore(&sighand->siglock, *flags); 1384 } 1385 rcu_read_unlock(); 1386 1387 return sighand; 1388 } 1389 1390 #ifdef CONFIG_LOCKDEP 1391 void lockdep_assert_task_sighand_held(struct task_struct *task) 1392 { 1393 struct sighand_struct *sighand; 1394 1395 rcu_read_lock(); 1396 sighand = rcu_dereference(task->sighand); 1397 if (sighand) 1398 lockdep_assert_held(&sighand->siglock); 1399 else 1400 WARN_ON_ONCE(1); 1401 rcu_read_unlock(); 1402 } 1403 #endif 1404 1405 /* 1406 * send signal info to all the members of a thread group or to the 1407 * individual thread if type == PIDTYPE_PID. 1408 */ 1409 int group_send_sig_info(int sig, struct kernel_siginfo *info, 1410 struct task_struct *p, enum pid_type type) 1411 { 1412 int ret; 1413 1414 rcu_read_lock(); 1415 ret = check_kill_permission(sig, info, p); 1416 rcu_read_unlock(); 1417 1418 if (!ret && sig) 1419 ret = do_send_sig_info(sig, info, p, type); 1420 1421 return ret; 1422 } 1423 1424 /* 1425 * __kill_pgrp_info() sends a signal to a process group: this is what the tty 1426 * control characters do (^C, ^Z etc) 1427 * - the caller must hold at least a readlock on tasklist_lock 1428 */ 1429 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp) 1430 { 1431 struct task_struct *p = NULL; 1432 int ret = -ESRCH; 1433 1434 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 1435 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID); 1436 /* 1437 * If group_send_sig_info() succeeds at least once ret 1438 * becomes 0 and after that the code below has no effect. 1439 * Otherwise we return the last err or -ESRCH if this 1440 * process group is empty. 1441 */ 1442 if (ret) 1443 ret = err; 1444 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 1445 1446 return ret; 1447 } 1448 1449 static int kill_pid_info_type(int sig, struct kernel_siginfo *info, 1450 struct pid *pid, enum pid_type type) 1451 { 1452 int error = -ESRCH; 1453 struct task_struct *p; 1454 1455 for (;;) { 1456 rcu_read_lock(); 1457 p = pid_task(pid, PIDTYPE_PID); 1458 if (p) 1459 error = group_send_sig_info(sig, info, p, type); 1460 rcu_read_unlock(); 1461 if (likely(!p || error != -ESRCH)) 1462 return error; 1463 /* 1464 * The task was unhashed in between, try again. If it 1465 * is dead, pid_task() will return NULL, if we race with 1466 * de_thread() it will find the new leader. 1467 */ 1468 } 1469 } 1470 1471 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid) 1472 { 1473 return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID); 1474 } 1475 1476 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid) 1477 { 1478 int error; 1479 rcu_read_lock(); 1480 error = kill_pid_info(sig, info, find_vpid(pid)); 1481 rcu_read_unlock(); 1482 return error; 1483 } 1484 1485 static inline bool kill_as_cred_perm(const struct cred *cred, 1486 struct task_struct *target) 1487 { 1488 const struct cred *pcred = __task_cred(target); 1489 1490 return uid_eq(cred->euid, pcred->suid) || 1491 uid_eq(cred->euid, pcred->uid) || 1492 uid_eq(cred->uid, pcred->suid) || 1493 uid_eq(cred->uid, pcred->uid); 1494 } 1495 1496 /* 1497 * The usb asyncio usage of siginfo is wrong. The glibc support 1498 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT. 1499 * AKA after the generic fields: 1500 * kernel_pid_t si_pid; 1501 * kernel_uid32_t si_uid; 1502 * sigval_t si_value; 1503 * 1504 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout 1505 * after the generic fields is: 1506 * void __user *si_addr; 1507 * 1508 * This is a practical problem when there is a 64bit big endian kernel 1509 * and a 32bit userspace. As the 32bit address will encoded in the low 1510 * 32bits of the pointer. Those low 32bits will be stored at higher 1511 * address than appear in a 32 bit pointer. So userspace will not 1512 * see the address it was expecting for it's completions. 1513 * 1514 * There is nothing in the encoding that can allow 1515 * copy_siginfo_to_user32 to detect this confusion of formats, so 1516 * handle this by requiring the caller of kill_pid_usb_asyncio to 1517 * notice when this situration takes place and to store the 32bit 1518 * pointer in sival_int, instead of sival_addr of the sigval_t addr 1519 * parameter. 1520 */ 1521 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr, 1522 struct pid *pid, const struct cred *cred) 1523 { 1524 struct kernel_siginfo info; 1525 struct task_struct *p; 1526 unsigned long flags; 1527 int ret = -EINVAL; 1528 1529 if (!valid_signal(sig)) 1530 return ret; 1531 1532 clear_siginfo(&info); 1533 info.si_signo = sig; 1534 info.si_errno = errno; 1535 info.si_code = SI_ASYNCIO; 1536 *((sigval_t *)&info.si_pid) = addr; 1537 1538 rcu_read_lock(); 1539 p = pid_task(pid, PIDTYPE_PID); 1540 if (!p) { 1541 ret = -ESRCH; 1542 goto out_unlock; 1543 } 1544 if (!kill_as_cred_perm(cred, p)) { 1545 ret = -EPERM; 1546 goto out_unlock; 1547 } 1548 ret = security_task_kill(p, &info, sig, cred); 1549 if (ret) 1550 goto out_unlock; 1551 1552 if (sig) { 1553 if (lock_task_sighand(p, &flags)) { 1554 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false); 1555 unlock_task_sighand(p, &flags); 1556 } else 1557 ret = -ESRCH; 1558 } 1559 out_unlock: 1560 rcu_read_unlock(); 1561 return ret; 1562 } 1563 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio); 1564 1565 /* 1566 * kill_something_info() interprets pid in interesting ways just like kill(2). 1567 * 1568 * POSIX specifies that kill(-1,sig) is unspecified, but what we have 1569 * is probably wrong. Should make it like BSD or SYSV. 1570 */ 1571 1572 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid) 1573 { 1574 int ret; 1575 1576 if (pid > 0) 1577 return kill_proc_info(sig, info, pid); 1578 1579 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */ 1580 if (pid == INT_MIN) 1581 return -ESRCH; 1582 1583 read_lock(&tasklist_lock); 1584 if (pid != -1) { 1585 ret = __kill_pgrp_info(sig, info, 1586 pid ? find_vpid(-pid) : task_pgrp(current)); 1587 } else { 1588 int retval = 0, count = 0; 1589 struct task_struct * p; 1590 1591 for_each_process(p) { 1592 if (task_pid_vnr(p) > 1 && 1593 !same_thread_group(p, current)) { 1594 int err = group_send_sig_info(sig, info, p, 1595 PIDTYPE_MAX); 1596 ++count; 1597 if (err != -EPERM) 1598 retval = err; 1599 } 1600 } 1601 ret = count ? retval : -ESRCH; 1602 } 1603 read_unlock(&tasklist_lock); 1604 1605 return ret; 1606 } 1607 1608 /* 1609 * These are for backward compatibility with the rest of the kernel source. 1610 */ 1611 1612 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p) 1613 { 1614 /* 1615 * Make sure legacy kernel users don't send in bad values 1616 * (normal paths check this in check_kill_permission). 1617 */ 1618 if (!valid_signal(sig)) 1619 return -EINVAL; 1620 1621 return do_send_sig_info(sig, info, p, PIDTYPE_PID); 1622 } 1623 EXPORT_SYMBOL(send_sig_info); 1624 1625 #define __si_special(priv) \ 1626 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO) 1627 1628 int 1629 send_sig(int sig, struct task_struct *p, int priv) 1630 { 1631 return send_sig_info(sig, __si_special(priv), p); 1632 } 1633 EXPORT_SYMBOL(send_sig); 1634 1635 void force_sig(int sig) 1636 { 1637 struct kernel_siginfo info; 1638 1639 clear_siginfo(&info); 1640 info.si_signo = sig; 1641 info.si_errno = 0; 1642 info.si_code = SI_KERNEL; 1643 info.si_pid = 0; 1644 info.si_uid = 0; 1645 force_sig_info(&info); 1646 } 1647 EXPORT_SYMBOL(force_sig); 1648 1649 void force_fatal_sig(int sig) 1650 { 1651 struct kernel_siginfo info; 1652 1653 clear_siginfo(&info); 1654 info.si_signo = sig; 1655 info.si_errno = 0; 1656 info.si_code = SI_KERNEL; 1657 info.si_pid = 0; 1658 info.si_uid = 0; 1659 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL); 1660 } 1661 1662 void force_exit_sig(int sig) 1663 { 1664 struct kernel_siginfo info; 1665 1666 clear_siginfo(&info); 1667 info.si_signo = sig; 1668 info.si_errno = 0; 1669 info.si_code = SI_KERNEL; 1670 info.si_pid = 0; 1671 info.si_uid = 0; 1672 force_sig_info_to_task(&info, current, HANDLER_EXIT); 1673 } 1674 1675 /* 1676 * When things go south during signal handling, we 1677 * will force a SIGSEGV. And if the signal that caused 1678 * the problem was already a SIGSEGV, we'll want to 1679 * make sure we don't even try to deliver the signal.. 1680 */ 1681 void force_sigsegv(int sig) 1682 { 1683 if (sig == SIGSEGV) 1684 force_fatal_sig(SIGSEGV); 1685 else 1686 force_sig(SIGSEGV); 1687 } 1688 1689 int force_sig_fault_to_task(int sig, int code, void __user *addr, 1690 struct task_struct *t) 1691 { 1692 struct kernel_siginfo info; 1693 1694 clear_siginfo(&info); 1695 info.si_signo = sig; 1696 info.si_errno = 0; 1697 info.si_code = code; 1698 info.si_addr = addr; 1699 return force_sig_info_to_task(&info, t, HANDLER_CURRENT); 1700 } 1701 1702 int force_sig_fault(int sig, int code, void __user *addr) 1703 { 1704 return force_sig_fault_to_task(sig, code, addr, current); 1705 } 1706 1707 int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t) 1708 { 1709 struct kernel_siginfo info; 1710 1711 clear_siginfo(&info); 1712 info.si_signo = sig; 1713 info.si_errno = 0; 1714 info.si_code = code; 1715 info.si_addr = addr; 1716 return send_sig_info(info.si_signo, &info, t); 1717 } 1718 1719 int force_sig_mceerr(int code, void __user *addr, short lsb) 1720 { 1721 struct kernel_siginfo info; 1722 1723 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR)); 1724 clear_siginfo(&info); 1725 info.si_signo = SIGBUS; 1726 info.si_errno = 0; 1727 info.si_code = code; 1728 info.si_addr = addr; 1729 info.si_addr_lsb = lsb; 1730 return force_sig_info(&info); 1731 } 1732 1733 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t) 1734 { 1735 struct kernel_siginfo info; 1736 1737 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR)); 1738 clear_siginfo(&info); 1739 info.si_signo = SIGBUS; 1740 info.si_errno = 0; 1741 info.si_code = code; 1742 info.si_addr = addr; 1743 info.si_addr_lsb = lsb; 1744 return send_sig_info(info.si_signo, &info, t); 1745 } 1746 EXPORT_SYMBOL(send_sig_mceerr); 1747 1748 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper) 1749 { 1750 struct kernel_siginfo info; 1751 1752 clear_siginfo(&info); 1753 info.si_signo = SIGSEGV; 1754 info.si_errno = 0; 1755 info.si_code = SEGV_BNDERR; 1756 info.si_addr = addr; 1757 info.si_lower = lower; 1758 info.si_upper = upper; 1759 return force_sig_info(&info); 1760 } 1761 1762 #ifdef SEGV_PKUERR 1763 int force_sig_pkuerr(void __user *addr, u32 pkey) 1764 { 1765 struct kernel_siginfo info; 1766 1767 clear_siginfo(&info); 1768 info.si_signo = SIGSEGV; 1769 info.si_errno = 0; 1770 info.si_code = SEGV_PKUERR; 1771 info.si_addr = addr; 1772 info.si_pkey = pkey; 1773 return force_sig_info(&info); 1774 } 1775 #endif 1776 1777 int send_sig_perf(void __user *addr, u32 type, u64 sig_data) 1778 { 1779 struct kernel_siginfo info; 1780 1781 clear_siginfo(&info); 1782 info.si_signo = SIGTRAP; 1783 info.si_errno = 0; 1784 info.si_code = TRAP_PERF; 1785 info.si_addr = addr; 1786 info.si_perf_data = sig_data; 1787 info.si_perf_type = type; 1788 1789 /* 1790 * Signals generated by perf events should not terminate the whole 1791 * process if SIGTRAP is blocked, however, delivering the signal 1792 * asynchronously is better than not delivering at all. But tell user 1793 * space if the signal was asynchronous, so it can clearly be 1794 * distinguished from normal synchronous ones. 1795 */ 1796 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ? 1797 TRAP_PERF_FLAG_ASYNC : 1798 0; 1799 1800 return send_sig_info(info.si_signo, &info, current); 1801 } 1802 1803 /** 1804 * force_sig_seccomp - signals the task to allow in-process syscall emulation 1805 * @syscall: syscall number to send to userland 1806 * @reason: filter-supplied reason code to send to userland (via si_errno) 1807 * @force_coredump: true to trigger a coredump 1808 * 1809 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. 1810 */ 1811 int force_sig_seccomp(int syscall, int reason, bool force_coredump) 1812 { 1813 struct kernel_siginfo info; 1814 1815 clear_siginfo(&info); 1816 info.si_signo = SIGSYS; 1817 info.si_code = SYS_SECCOMP; 1818 info.si_call_addr = (void __user *)KSTK_EIP(current); 1819 info.si_errno = reason; 1820 info.si_arch = syscall_get_arch(current); 1821 info.si_syscall = syscall; 1822 return force_sig_info_to_task(&info, current, 1823 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT); 1824 } 1825 1826 /* For the crazy architectures that include trap information in 1827 * the errno field, instead of an actual errno value. 1828 */ 1829 int force_sig_ptrace_errno_trap(int errno, void __user *addr) 1830 { 1831 struct kernel_siginfo info; 1832 1833 clear_siginfo(&info); 1834 info.si_signo = SIGTRAP; 1835 info.si_errno = errno; 1836 info.si_code = TRAP_HWBKPT; 1837 info.si_addr = addr; 1838 return force_sig_info(&info); 1839 } 1840 1841 /* For the rare architectures that include trap information using 1842 * si_trapno. 1843 */ 1844 int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno) 1845 { 1846 struct kernel_siginfo info; 1847 1848 clear_siginfo(&info); 1849 info.si_signo = sig; 1850 info.si_errno = 0; 1851 info.si_code = code; 1852 info.si_addr = addr; 1853 info.si_trapno = trapno; 1854 return force_sig_info(&info); 1855 } 1856 1857 /* For the rare architectures that include trap information using 1858 * si_trapno. 1859 */ 1860 int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno, 1861 struct task_struct *t) 1862 { 1863 struct kernel_siginfo info; 1864 1865 clear_siginfo(&info); 1866 info.si_signo = sig; 1867 info.si_errno = 0; 1868 info.si_code = code; 1869 info.si_addr = addr; 1870 info.si_trapno = trapno; 1871 return send_sig_info(info.si_signo, &info, t); 1872 } 1873 1874 static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp) 1875 { 1876 int ret; 1877 read_lock(&tasklist_lock); 1878 ret = __kill_pgrp_info(sig, info, pgrp); 1879 read_unlock(&tasklist_lock); 1880 return ret; 1881 } 1882 1883 int kill_pgrp(struct pid *pid, int sig, int priv) 1884 { 1885 return kill_pgrp_info(sig, __si_special(priv), pid); 1886 } 1887 EXPORT_SYMBOL(kill_pgrp); 1888 1889 int kill_pid(struct pid *pid, int sig, int priv) 1890 { 1891 return kill_pid_info(sig, __si_special(priv), pid); 1892 } 1893 EXPORT_SYMBOL(kill_pid); 1894 1895 #ifdef CONFIG_POSIX_TIMERS 1896 /* 1897 * These functions handle POSIX timer signals. POSIX timers use 1898 * preallocated sigqueue structs for sending signals. 1899 */ 1900 static void __flush_itimer_signals(struct sigpending *pending) 1901 { 1902 sigset_t signal, retain; 1903 struct sigqueue *q, *n; 1904 1905 signal = pending->signal; 1906 sigemptyset(&retain); 1907 1908 list_for_each_entry_safe(q, n, &pending->list, list) { 1909 int sig = q->info.si_signo; 1910 1911 if (likely(q->info.si_code != SI_TIMER)) { 1912 sigaddset(&retain, sig); 1913 } else { 1914 sigdelset(&signal, sig); 1915 list_del_init(&q->list); 1916 __sigqueue_free(q); 1917 } 1918 } 1919 1920 sigorsets(&pending->signal, &signal, &retain); 1921 } 1922 1923 void flush_itimer_signals(void) 1924 { 1925 struct task_struct *tsk = current; 1926 1927 guard(spinlock_irqsave)(&tsk->sighand->siglock); 1928 __flush_itimer_signals(&tsk->pending); 1929 __flush_itimer_signals(&tsk->signal->shared_pending); 1930 } 1931 1932 bool posixtimer_init_sigqueue(struct sigqueue *q) 1933 { 1934 struct ucounts *ucounts = sig_get_ucounts(current, -1, 0); 1935 1936 if (!ucounts) 1937 return false; 1938 clear_siginfo(&q->info); 1939 __sigqueue_init(q, ucounts, SIGQUEUE_PREALLOC); 1940 return true; 1941 } 1942 1943 static void posixtimer_queue_sigqueue(struct sigqueue *q, struct task_struct *t, enum pid_type type) 1944 { 1945 struct sigpending *pending; 1946 int sig = q->info.si_signo; 1947 1948 signalfd_notify(t, sig); 1949 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending; 1950 list_add_tail(&q->list, &pending->list); 1951 sigaddset(&pending->signal, sig); 1952 complete_signal(sig, t, type); 1953 } 1954 1955 /* 1956 * This function is used by POSIX timers to deliver a timer signal. 1957 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID 1958 * set), the signal must be delivered to the specific thread (queues 1959 * into t->pending). 1960 * 1961 * Where type is not PIDTYPE_PID, signals must be delivered to the 1962 * process. In this case, prefer to deliver to current if it is in 1963 * the same thread group as the target process and its sighand is 1964 * stable, which avoids unnecessarily waking up a potentially idle task. 1965 */ 1966 static inline struct task_struct *posixtimer_get_target(struct k_itimer *tmr) 1967 { 1968 struct task_struct *t = pid_task(tmr->it_pid, tmr->it_pid_type); 1969 1970 if (t && tmr->it_pid_type != PIDTYPE_PID && 1971 same_thread_group(t, current) && !current->exit_state) 1972 t = current; 1973 return t; 1974 } 1975 1976 void posixtimer_send_sigqueue(struct k_itimer *tmr) 1977 { 1978 struct sigqueue *q = &tmr->sigq; 1979 int sig = q->info.si_signo; 1980 struct task_struct *t; 1981 unsigned long flags; 1982 int result; 1983 1984 guard(rcu)(); 1985 1986 t = posixtimer_get_target(tmr); 1987 if (!t) 1988 return; 1989 1990 if (!likely(lock_task_sighand(t, &flags))) 1991 return; 1992 1993 /* 1994 * Update @tmr::sigqueue_seq for posix timer signals with sighand 1995 * locked to prevent a race against dequeue_signal(). 1996 */ 1997 tmr->it_sigqueue_seq = tmr->it_signal_seq; 1998 1999 /* 2000 * Set the signal delivery status under sighand lock, so that the 2001 * ignored signal handling can distinguish between a periodic and a 2002 * non-periodic timer. 2003 */ 2004 tmr->it_sig_periodic = tmr->it_status == POSIX_TIMER_REQUEUE_PENDING; 2005 2006 if (!prepare_signal(sig, t, false)) { 2007 result = TRACE_SIGNAL_IGNORED; 2008 2009 if (!list_empty(&q->list)) { 2010 /* 2011 * The signal was ignored and blocked. The timer 2012 * expiry queued it because blocked signals are 2013 * queued independent of the ignored state. 2014 * 2015 * The unblocking set SIGPENDING, but the signal 2016 * was not yet dequeued from the pending list. 2017 * So prepare_signal() sees unblocked and ignored, 2018 * which ends up here. Leave it queued like a 2019 * regular signal. 2020 * 2021 * The same happens when the task group is exiting 2022 * and the signal is already queued. 2023 * prepare_signal() treats SIGNAL_GROUP_EXIT as 2024 * ignored independent of its queued state. This 2025 * gets cleaned up in __exit_signal(). 2026 */ 2027 goto out; 2028 } 2029 2030 /* Periodic timers with SIG_IGN are queued on the ignored list */ 2031 if (tmr->it_sig_periodic) { 2032 /* 2033 * Already queued means the timer was rearmed after 2034 * the previous expiry got it on the ignore list. 2035 * Nothing to do for that case. 2036 */ 2037 if (hlist_unhashed(&tmr->ignored_list)) { 2038 /* 2039 * Take a signal reference and queue it on 2040 * the ignored list. 2041 */ 2042 posixtimer_sigqueue_getref(q); 2043 posixtimer_sig_ignore(t, q); 2044 } 2045 } else if (!hlist_unhashed(&tmr->ignored_list)) { 2046 /* 2047 * Covers the case where a timer was periodic and 2048 * then the signal was ignored. Later it was rearmed 2049 * as oneshot timer. The previous signal is invalid 2050 * now, and this oneshot signal has to be dropped. 2051 * Remove it from the ignored list and drop the 2052 * reference count as the signal is not longer 2053 * queued. 2054 */ 2055 hlist_del_init(&tmr->ignored_list); 2056 posixtimer_putref(tmr); 2057 } 2058 goto out; 2059 } 2060 2061 if (unlikely(!list_empty(&q->list))) { 2062 /* This holds a reference count already */ 2063 result = TRACE_SIGNAL_ALREADY_PENDING; 2064 goto out; 2065 } 2066 2067 /* 2068 * If the signal is on the ignore list, it got blocked after it was 2069 * ignored earlier. But nothing lifted the ignore. Move it back to 2070 * the pending list to be consistent with the regular signal 2071 * handling. This already holds a reference count. 2072 * 2073 * If it's not on the ignore list acquire a reference count. 2074 */ 2075 if (likely(hlist_unhashed(&tmr->ignored_list))) 2076 posixtimer_sigqueue_getref(q); 2077 else 2078 hlist_del_init(&tmr->ignored_list); 2079 2080 posixtimer_queue_sigqueue(q, t, tmr->it_pid_type); 2081 result = TRACE_SIGNAL_DELIVERED; 2082 out: 2083 trace_signal_generate(sig, &q->info, t, tmr->it_pid_type != PIDTYPE_PID, result); 2084 unlock_task_sighand(t, &flags); 2085 } 2086 2087 static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q) 2088 { 2089 struct k_itimer *tmr = container_of(q, struct k_itimer, sigq); 2090 2091 /* 2092 * If the timer is marked deleted already or the signal originates 2093 * from a non-periodic timer, then just drop the reference 2094 * count. Otherwise queue it on the ignored list. 2095 */ 2096 if (posixtimer_valid(tmr) && tmr->it_sig_periodic) 2097 hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers); 2098 else 2099 posixtimer_putref(tmr); 2100 } 2101 2102 static void posixtimer_sig_unignore(struct task_struct *tsk, int sig) 2103 { 2104 struct hlist_head *head = &tsk->signal->ignored_posix_timers; 2105 struct hlist_node *tmp; 2106 struct k_itimer *tmr; 2107 2108 if (likely(hlist_empty(head))) 2109 return; 2110 2111 /* 2112 * Rearming a timer with sighand lock held is not possible due to 2113 * lock ordering vs. tmr::it_lock. Just stick the sigqueue back and 2114 * let the signal delivery path deal with it whether it needs to be 2115 * rearmed or not. This cannot be decided here w/o dropping sighand 2116 * lock and creating a loop retry horror show. 2117 */ 2118 hlist_for_each_entry_safe(tmr, tmp , head, ignored_list) { 2119 struct task_struct *target; 2120 2121 /* 2122 * tmr::sigq.info.si_signo is immutable, so accessing it 2123 * without holding tmr::it_lock is safe. 2124 */ 2125 if (tmr->sigq.info.si_signo != sig) 2126 continue; 2127 2128 hlist_del_init(&tmr->ignored_list); 2129 2130 /* This should never happen and leaks a reference count */ 2131 if (WARN_ON_ONCE(!list_empty(&tmr->sigq.list))) 2132 continue; 2133 2134 /* 2135 * Get the target for the signal. If target is a thread and 2136 * has exited by now, drop the reference count. 2137 */ 2138 guard(rcu)(); 2139 target = posixtimer_get_target(tmr); 2140 if (target) 2141 posixtimer_queue_sigqueue(&tmr->sigq, target, tmr->it_pid_type); 2142 else 2143 posixtimer_putref(tmr); 2144 } 2145 } 2146 #else /* CONFIG_POSIX_TIMERS */ 2147 static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q) { } 2148 static inline void posixtimer_sig_unignore(struct task_struct *tsk, int sig) { } 2149 #endif /* !CONFIG_POSIX_TIMERS */ 2150 2151 void do_notify_pidfd(struct task_struct *task) 2152 { 2153 struct pid *pid = task_pid(task); 2154 2155 WARN_ON(task->exit_state == 0); 2156 2157 __wake_up(&pid->wait_pidfd, TASK_NORMAL, 0, 2158 poll_to_key(EPOLLIN | EPOLLRDNORM)); 2159 } 2160 2161 /* 2162 * Let a parent know about the death of a child. 2163 * For a stopped/continued status change, use do_notify_parent_cldstop instead. 2164 * 2165 * Returns true if our parent ignored us and so we've switched to 2166 * self-reaping. 2167 */ 2168 bool do_notify_parent(struct task_struct *tsk, int sig) 2169 { 2170 struct kernel_siginfo info; 2171 unsigned long flags; 2172 struct sighand_struct *psig; 2173 bool autoreap = false; 2174 u64 utime, stime; 2175 2176 if (WARN_ON_ONCE(!valid_signal(sig))) 2177 return false; 2178 2179 /* do_notify_parent_cldstop should have been called instead. */ 2180 WARN_ON_ONCE(task_is_stopped_or_traced(tsk)); 2181 2182 WARN_ON_ONCE(!tsk->ptrace && !thread_group_empty(tsk)); 2183 2184 /* ptraced, or group-leader without sub-threads */ 2185 do_notify_pidfd(tsk); 2186 2187 if (sig != SIGCHLD) { 2188 /* 2189 * This is only possible if parent == real_parent. 2190 * Check if it has changed security domain. 2191 */ 2192 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id)) 2193 sig = SIGCHLD; 2194 } 2195 2196 clear_siginfo(&info); 2197 info.si_signo = sig; 2198 info.si_errno = 0; 2199 /* 2200 * We are under tasklist_lock here so our parent is tied to 2201 * us and cannot change. 2202 * 2203 * task_active_pid_ns will always return the same pid namespace 2204 * until a task passes through release_task. 2205 * 2206 * write_lock() currently calls preempt_disable() which is the 2207 * same as rcu_read_lock(), but according to Oleg, this is not 2208 * correct to rely on this 2209 */ 2210 rcu_read_lock(); 2211 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent)); 2212 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns), 2213 task_uid(tsk)); 2214 rcu_read_unlock(); 2215 2216 task_cputime(tsk, &utime, &stime); 2217 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime); 2218 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime); 2219 2220 info.si_status = tsk->exit_code & 0x7f; 2221 if (tsk->exit_code & 0x80) 2222 info.si_code = CLD_DUMPED; 2223 else if (tsk->exit_code & 0x7f) 2224 info.si_code = CLD_KILLED; 2225 else { 2226 info.si_code = CLD_EXITED; 2227 info.si_status = tsk->exit_code >> 8; 2228 } 2229 2230 psig = tsk->parent->sighand; 2231 spin_lock_irqsave(&psig->siglock, flags); 2232 if (!tsk->ptrace && sig == SIGCHLD && 2233 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN || 2234 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) { 2235 /* 2236 * We are exiting and our parent doesn't care. POSIX.1 2237 * defines special semantics for setting SIGCHLD to SIG_IGN 2238 * or setting the SA_NOCLDWAIT flag: we should be reaped 2239 * automatically and not left for our parent's wait4 call. 2240 * Rather than having the parent do it as a magic kind of 2241 * signal handler, we just set this to tell do_exit that we 2242 * can be cleaned up without becoming a zombie. Note that 2243 * we still call __wake_up_parent in this case, because a 2244 * blocked sys_wait4 might now return -ECHILD. 2245 * 2246 * Whether we send SIGCHLD or not for SA_NOCLDWAIT 2247 * is implementation-defined: we do (if you don't want 2248 * it, just use SIG_IGN instead). 2249 */ 2250 autoreap = true; 2251 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) 2252 sig = 0; 2253 } 2254 if (!tsk->ptrace && tsk->signal->autoreap) { 2255 autoreap = true; 2256 sig = 0; 2257 } 2258 /* 2259 * Send with __send_signal as si_pid and si_uid are in the 2260 * parent's namespaces. 2261 */ 2262 if (sig) 2263 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false); 2264 __wake_up_parent(tsk, tsk->parent); 2265 spin_unlock_irqrestore(&psig->siglock, flags); 2266 2267 return autoreap; 2268 } 2269 2270 /** 2271 * do_notify_parent_cldstop - notify parent of stopped/continued state change 2272 * @tsk: task reporting the state change 2273 * @for_ptracer: the notification is for ptracer 2274 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report 2275 * 2276 * Notify @tsk's parent that the stopped/continued state has changed. If 2277 * @for_ptracer is %false, @tsk's group leader notifies to its real parent. 2278 * If %true, @tsk reports to @tsk->parent which should be the ptracer. 2279 * 2280 * CONTEXT: 2281 * Must be called with tasklist_lock at least read locked. 2282 */ 2283 static void do_notify_parent_cldstop(struct task_struct *tsk, 2284 bool for_ptracer, int why) 2285 { 2286 struct kernel_siginfo info; 2287 unsigned long flags; 2288 struct task_struct *parent; 2289 struct sighand_struct *sighand; 2290 u64 utime, stime; 2291 2292 if (for_ptracer) { 2293 parent = tsk->parent; 2294 } else { 2295 tsk = tsk->group_leader; 2296 parent = tsk->real_parent; 2297 } 2298 2299 clear_siginfo(&info); 2300 info.si_signo = SIGCHLD; 2301 info.si_errno = 0; 2302 /* 2303 * see comment in do_notify_parent() about the following 4 lines 2304 */ 2305 rcu_read_lock(); 2306 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent)); 2307 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk)); 2308 rcu_read_unlock(); 2309 2310 task_cputime(tsk, &utime, &stime); 2311 info.si_utime = nsec_to_clock_t(utime); 2312 info.si_stime = nsec_to_clock_t(stime); 2313 2314 info.si_code = why; 2315 switch (why) { 2316 case CLD_CONTINUED: 2317 info.si_status = SIGCONT; 2318 break; 2319 case CLD_STOPPED: 2320 info.si_status = tsk->signal->group_exit_code & 0x7f; 2321 break; 2322 case CLD_TRAPPED: 2323 info.si_status = tsk->exit_code & 0x7f; 2324 break; 2325 default: 2326 BUG(); 2327 } 2328 2329 sighand = parent->sighand; 2330 spin_lock_irqsave(&sighand->siglock, flags); 2331 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN && 2332 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP)) 2333 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID); 2334 /* 2335 * Even if SIGCHLD is not generated, we must wake up wait4 calls. 2336 */ 2337 __wake_up_parent(tsk, parent); 2338 spin_unlock_irqrestore(&sighand->siglock, flags); 2339 } 2340 2341 /* 2342 * This must be called with current->sighand->siglock held. 2343 * 2344 * This should be the path for all ptrace stops. 2345 * We always set current->last_siginfo while stopped here. 2346 * That makes it a way to test a stopped process for 2347 * being ptrace-stopped vs being job-control-stopped. 2348 * 2349 * Returns the signal the ptracer requested the code resume 2350 * with. If the code did not stop because the tracer is gone, 2351 * the stop signal remains unchanged unless clear_code. 2352 */ 2353 static int ptrace_stop(int exit_code, int why, unsigned long message, 2354 kernel_siginfo_t *info) 2355 __releases(¤t->sighand->siglock) 2356 __acquires(¤t->sighand->siglock) 2357 { 2358 bool gstop_done = false; 2359 2360 if (arch_ptrace_stop_needed()) { 2361 /* 2362 * The arch code has something special to do before a 2363 * ptrace stop. This is allowed to block, e.g. for faults 2364 * on user stack pages. We can't keep the siglock while 2365 * calling arch_ptrace_stop, so we must release it now. 2366 * To preserve proper semantics, we must do this before 2367 * any signal bookkeeping like checking group_stop_count. 2368 */ 2369 spin_unlock_irq(¤t->sighand->siglock); 2370 arch_ptrace_stop(); 2371 spin_lock_irq(¤t->sighand->siglock); 2372 } 2373 2374 /* 2375 * After this point ptrace_signal_wake_up or signal_wake_up 2376 * will clear TASK_TRACED if ptrace_unlink happens or a fatal 2377 * signal comes in. Handle previous ptrace_unlinks and fatal 2378 * signals here to prevent ptrace_stop sleeping in schedule. 2379 */ 2380 if (!current->ptrace || __fatal_signal_pending(current)) 2381 return exit_code; 2382 2383 set_special_state(TASK_TRACED); 2384 current->jobctl |= JOBCTL_TRACED; 2385 2386 /* 2387 * We're committing to trapping. TRACED should be visible before 2388 * TRAPPING is cleared; otherwise, the tracer might fail do_wait(). 2389 * Also, transition to TRACED and updates to ->jobctl should be 2390 * atomic with respect to siglock and should be done after the arch 2391 * hook as siglock is released and regrabbed across it. 2392 * 2393 * TRACER TRACEE 2394 * 2395 * ptrace_attach() 2396 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED) 2397 * do_wait() 2398 * set_current_state() smp_wmb(); 2399 * ptrace_do_wait() 2400 * wait_task_stopped() 2401 * task_stopped_code() 2402 * [L] task_is_traced() [S] task_clear_jobctl_trapping(); 2403 */ 2404 smp_wmb(); 2405 2406 current->ptrace_message = message; 2407 current->last_siginfo = info; 2408 current->exit_code = exit_code; 2409 2410 /* 2411 * If @why is CLD_STOPPED, we're trapping to participate in a group 2412 * stop. Do the bookkeeping. Note that if SIGCONT was delievered 2413 * across siglock relocks since INTERRUPT was scheduled, PENDING 2414 * could be clear now. We act as if SIGCONT is received after 2415 * TASK_TRACED is entered - ignore it. 2416 */ 2417 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING)) 2418 gstop_done = task_participate_group_stop(current); 2419 2420 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */ 2421 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP); 2422 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP) 2423 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY); 2424 2425 /* entering a trap, clear TRAPPING */ 2426 task_clear_jobctl_trapping(current); 2427 2428 spin_unlock_irq(¤t->sighand->siglock); 2429 read_lock(&tasklist_lock); 2430 /* 2431 * Notify parents of the stop. 2432 * 2433 * While ptraced, there are two parents - the ptracer and 2434 * the real_parent of the group_leader. The ptracer should 2435 * know about every stop while the real parent is only 2436 * interested in the completion of group stop. The states 2437 * for the two don't interact with each other. Notify 2438 * separately unless they're gonna be duplicates. 2439 */ 2440 if (current->ptrace) 2441 do_notify_parent_cldstop(current, true, why); 2442 if (gstop_done && (!current->ptrace || ptrace_reparented(current))) 2443 do_notify_parent_cldstop(current, false, why); 2444 2445 /* 2446 * The previous do_notify_parent_cldstop() invocation woke ptracer. 2447 * One a PREEMPTION kernel this can result in preemption requirement 2448 * which will be fulfilled after read_unlock() and the ptracer will be 2449 * put on the CPU. 2450 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for 2451 * this task wait in schedule(). If this task gets preempted then it 2452 * remains enqueued on the runqueue. The ptracer will observe this and 2453 * then sleep for a delay of one HZ tick. In the meantime this task 2454 * gets scheduled, enters schedule() and will wait for the ptracer. 2455 * 2456 * This preemption point is not bad from a correctness point of 2457 * view but extends the runtime by one HZ tick time due to the 2458 * ptracer's sleep. The preempt-disable section ensures that there 2459 * will be no preemption between unlock and schedule() and so 2460 * improving the performance since the ptracer will observe that 2461 * the tracee is scheduled out once it gets on the CPU. 2462 * 2463 * On PREEMPT_RT locking tasklist_lock does not disable preemption. 2464 * Therefore the task can be preempted after do_notify_parent_cldstop() 2465 * before unlocking tasklist_lock so there is no benefit in doing this. 2466 * 2467 * In fact disabling preemption is harmful on PREEMPT_RT because 2468 * the spinlock_t in cgroup_enter_frozen() must not be acquired 2469 * with preemption disabled due to the 'sleeping' spinlock 2470 * substitution of RT. 2471 */ 2472 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 2473 preempt_disable(); 2474 read_unlock(&tasklist_lock); 2475 cgroup_enter_frozen(); 2476 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 2477 preempt_enable_no_resched(); 2478 schedule(); 2479 cgroup_leave_frozen(true); 2480 2481 /* 2482 * We are back. Now reacquire the siglock before touching 2483 * last_siginfo, so that we are sure to have synchronized with 2484 * any signal-sending on another CPU that wants to examine it. 2485 */ 2486 spin_lock_irq(¤t->sighand->siglock); 2487 exit_code = current->exit_code; 2488 current->last_siginfo = NULL; 2489 current->ptrace_message = 0; 2490 current->exit_code = 0; 2491 2492 /* LISTENING can be set only during STOP traps, clear it */ 2493 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN); 2494 2495 /* 2496 * Queued signals ignored us while we were stopped for tracing. 2497 * So check for any that we should take before resuming user mode. 2498 * This sets TIF_SIGPENDING, but never clears it. 2499 */ 2500 recalc_sigpending_tsk(current); 2501 return exit_code; 2502 } 2503 2504 static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message) 2505 { 2506 kernel_siginfo_t info; 2507 2508 clear_siginfo(&info); 2509 info.si_signo = signr; 2510 info.si_code = exit_code; 2511 info.si_pid = task_pid_vnr(current); 2512 info.si_uid = from_kuid_munged(current_user_ns(), current_uid()); 2513 2514 /* Let the debugger run. */ 2515 return ptrace_stop(exit_code, why, message, &info); 2516 } 2517 2518 int ptrace_notify(int exit_code, unsigned long message) 2519 { 2520 int signr; 2521 2522 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP); 2523 if (unlikely(task_work_pending(current))) 2524 task_work_run(); 2525 2526 spin_lock_irq(¤t->sighand->siglock); 2527 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message); 2528 spin_unlock_irq(¤t->sighand->siglock); 2529 return signr; 2530 } 2531 2532 /** 2533 * do_signal_stop - handle group stop for SIGSTOP and other stop signals 2534 * @signr: signr causing group stop if initiating 2535 * 2536 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr 2537 * and participate in it. If already set, participate in the existing 2538 * group stop. If participated in a group stop (and thus slept), %true is 2539 * returned with siglock released. 2540 * 2541 * If ptraced, this function doesn't handle stop itself. Instead, 2542 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock 2543 * untouched. The caller must ensure that INTERRUPT trap handling takes 2544 * places afterwards. 2545 * 2546 * CONTEXT: 2547 * Must be called with @current->sighand->siglock held, which is released 2548 * on %true return. 2549 * 2550 * RETURNS: 2551 * %false if group stop is already cancelled or ptrace trap is scheduled. 2552 * %true if participated in group stop. 2553 */ 2554 static bool do_signal_stop(int signr) 2555 __releases(¤t->sighand->siglock) 2556 { 2557 struct signal_struct *sig = current->signal; 2558 2559 if (!(current->jobctl & JOBCTL_STOP_PENDING)) { 2560 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME; 2561 struct task_struct *t; 2562 2563 /* signr will be recorded in task->jobctl for retries */ 2564 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK); 2565 2566 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) || 2567 unlikely(sig->flags & SIGNAL_GROUP_EXIT) || 2568 unlikely(sig->group_exec_task)) 2569 return false; 2570 /* 2571 * There is no group stop already in progress. We must 2572 * initiate one now. 2573 * 2574 * While ptraced, a task may be resumed while group stop is 2575 * still in effect and then receive a stop signal and 2576 * initiate another group stop. This deviates from the 2577 * usual behavior as two consecutive stop signals can't 2578 * cause two group stops when !ptraced. That is why we 2579 * also check !task_is_stopped(t) below. 2580 * 2581 * The condition can be distinguished by testing whether 2582 * SIGNAL_STOP_STOPPED is already set. Don't generate 2583 * group_exit_code in such case. 2584 * 2585 * This is not necessary for SIGNAL_STOP_CONTINUED because 2586 * an intervening stop signal is required to cause two 2587 * continued events regardless of ptrace. 2588 */ 2589 if (!(sig->flags & SIGNAL_STOP_STOPPED)) 2590 sig->group_exit_code = signr; 2591 2592 sig->group_stop_count = 0; 2593 if (task_set_jobctl_pending(current, signr | gstop)) 2594 sig->group_stop_count++; 2595 2596 for_other_threads(current, t) { 2597 /* 2598 * Setting state to TASK_STOPPED for a group 2599 * stop is always done with the siglock held, 2600 * so this check has no races. 2601 */ 2602 if (!task_is_stopped(t) && 2603 task_set_jobctl_pending(t, signr | gstop)) { 2604 sig->group_stop_count++; 2605 if (likely(!(t->ptrace & PT_SEIZED))) 2606 signal_wake_up(t, 0); 2607 else 2608 ptrace_trap_notify(t); 2609 } 2610 } 2611 } 2612 2613 if (likely(!current->ptrace)) { 2614 int notify = 0; 2615 2616 /* 2617 * If there are no other threads in the group, or if there 2618 * is a group stop in progress and we are the last to stop, 2619 * report to the parent. 2620 */ 2621 if (task_participate_group_stop(current)) 2622 notify = CLD_STOPPED; 2623 2624 current->jobctl |= JOBCTL_STOPPED; 2625 set_special_state(TASK_STOPPED); 2626 spin_unlock_irq(¤t->sighand->siglock); 2627 2628 /* 2629 * Notify the parent of the group stop completion. Because 2630 * we're not holding either the siglock or tasklist_lock 2631 * here, ptracer may attach inbetween; however, this is for 2632 * group stop and should always be delivered to the real 2633 * parent of the group leader. The new ptracer will get 2634 * its notification when this task transitions into 2635 * TASK_TRACED. 2636 */ 2637 if (notify) { 2638 read_lock(&tasklist_lock); 2639 do_notify_parent_cldstop(current, false, notify); 2640 read_unlock(&tasklist_lock); 2641 } 2642 2643 /* Now we don't run again until woken by SIGCONT or SIGKILL */ 2644 cgroup_enter_frozen(); 2645 schedule(); 2646 return true; 2647 } else { 2648 /* 2649 * While ptraced, group stop is handled by STOP trap. 2650 * Schedule it and let the caller deal with it. 2651 */ 2652 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP); 2653 return false; 2654 } 2655 } 2656 2657 /** 2658 * do_jobctl_trap - take care of ptrace jobctl traps 2659 * 2660 * When PT_SEIZED, it's used for both group stop and explicit 2661 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with 2662 * accompanying siginfo. If stopped, lower eight bits of exit_code contain 2663 * the stop signal; otherwise, %SIGTRAP. 2664 * 2665 * When !PT_SEIZED, it's used only for group stop trap with stop signal 2666 * number as exit_code and no siginfo. 2667 * 2668 * CONTEXT: 2669 * Must be called with @current->sighand->siglock held, which may be 2670 * released and re-acquired before returning with intervening sleep. 2671 */ 2672 static void do_jobctl_trap(void) 2673 { 2674 struct signal_struct *signal = current->signal; 2675 int signr = current->jobctl & JOBCTL_STOP_SIGMASK; 2676 2677 if (current->ptrace & PT_SEIZED) { 2678 if (!signal->group_stop_count && 2679 !(signal->flags & SIGNAL_STOP_STOPPED)) 2680 signr = SIGTRAP; 2681 WARN_ON_ONCE(!signr); 2682 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8), 2683 CLD_STOPPED, 0); 2684 } else { 2685 WARN_ON_ONCE(!signr); 2686 ptrace_stop(signr, CLD_STOPPED, 0, NULL); 2687 } 2688 } 2689 2690 /** 2691 * do_freezer_trap - handle the freezer jobctl trap 2692 * 2693 * Puts the task into frozen state, if only the task is not about to quit. 2694 * In this case it drops JOBCTL_TRAP_FREEZE. 2695 * 2696 * CONTEXT: 2697 * Must be called with @current->sighand->siglock held, 2698 * which is always released before returning. 2699 */ 2700 static void do_freezer_trap(void) 2701 __releases(¤t->sighand->siglock) 2702 { 2703 /* 2704 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE, 2705 * let's make another loop to give it a chance to be handled. 2706 * In any case, we'll return back. 2707 */ 2708 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) != 2709 JOBCTL_TRAP_FREEZE) { 2710 spin_unlock_irq(¤t->sighand->siglock); 2711 return; 2712 } 2713 2714 /* 2715 * Now we're sure that there is no pending fatal signal and no 2716 * pending traps. Clear TIF_SIGPENDING to not get out of schedule() 2717 * immediately (if there is a non-fatal signal pending), and 2718 * put the task into sleep. 2719 */ 2720 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE); 2721 clear_thread_flag(TIF_SIGPENDING); 2722 spin_unlock_irq(¤t->sighand->siglock); 2723 cgroup_enter_frozen(); 2724 schedule(); 2725 2726 /* 2727 * We could've been woken by task_work, run it to clear 2728 * TIF_NOTIFY_SIGNAL. The caller will retry if necessary. 2729 */ 2730 clear_notify_signal(); 2731 if (unlikely(task_work_pending(current))) 2732 task_work_run(); 2733 } 2734 2735 static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type) 2736 { 2737 /* 2738 * We do not check sig_kernel_stop(signr) but set this marker 2739 * unconditionally because we do not know whether debugger will 2740 * change signr. This flag has no meaning unless we are going 2741 * to stop after return from ptrace_stop(). In this case it will 2742 * be checked in do_signal_stop(), we should only stop if it was 2743 * not cleared by SIGCONT while we were sleeping. See also the 2744 * comment in dequeue_signal(). 2745 */ 2746 current->jobctl |= JOBCTL_STOP_DEQUEUED; 2747 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info); 2748 2749 /* We're back. Did the debugger cancel the sig? */ 2750 if (signr == 0) 2751 return signr; 2752 2753 /* 2754 * Update the siginfo structure if the signal has 2755 * changed. If the debugger wanted something 2756 * specific in the siginfo structure then it should 2757 * have updated *info via PTRACE_SETSIGINFO. 2758 */ 2759 if (signr != info->si_signo) { 2760 clear_siginfo(info); 2761 info->si_signo = signr; 2762 info->si_errno = 0; 2763 info->si_code = SI_USER; 2764 rcu_read_lock(); 2765 info->si_pid = task_pid_vnr(current->parent); 2766 info->si_uid = from_kuid_munged(current_user_ns(), 2767 task_uid(current->parent)); 2768 rcu_read_unlock(); 2769 } 2770 2771 /* If the (new) signal is now blocked, requeue it. */ 2772 if (sigismember(¤t->blocked, signr) || 2773 fatal_signal_pending(current)) { 2774 send_signal_locked(signr, info, current, type); 2775 signr = 0; 2776 } 2777 2778 return signr; 2779 } 2780 2781 static void hide_si_addr_tag_bits(struct ksignal *ksig) 2782 { 2783 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) { 2784 case SIL_FAULT: 2785 case SIL_FAULT_TRAPNO: 2786 case SIL_FAULT_MCEERR: 2787 case SIL_FAULT_BNDERR: 2788 case SIL_FAULT_PKUERR: 2789 case SIL_FAULT_PERF_EVENT: 2790 ksig->info.si_addr = arch_untagged_si_addr( 2791 ksig->info.si_addr, ksig->sig, ksig->info.si_code); 2792 break; 2793 case SIL_KILL: 2794 case SIL_TIMER: 2795 case SIL_POLL: 2796 case SIL_CHLD: 2797 case SIL_RT: 2798 case SIL_SYS: 2799 break; 2800 } 2801 } 2802 2803 bool get_signal(struct ksignal *ksig) 2804 { 2805 struct sighand_struct *sighand = current->sighand; 2806 struct signal_struct *signal = current->signal; 2807 int signr; 2808 2809 clear_notify_signal(); 2810 if (unlikely(task_work_pending(current))) 2811 task_work_run(); 2812 2813 if (!task_sigpending(current)) 2814 return false; 2815 2816 if (unlikely(uprobe_deny_signal())) 2817 return false; 2818 2819 /* 2820 * Do this once, we can't return to user-mode if freezing() == T. 2821 * do_signal_stop() and ptrace_stop() set TASK_STOPPED/TASK_TRACED 2822 * and the freezer handles those states via TASK_FROZEN, thus they 2823 * do not need another check after return. 2824 */ 2825 try_to_freeze(); 2826 2827 relock: 2828 spin_lock_irq(&sighand->siglock); 2829 2830 /* 2831 * Every stopped thread goes here after wakeup. Check to see if 2832 * we should notify the parent, prepare_signal(SIGCONT) encodes 2833 * the CLD_ si_code into SIGNAL_CLD_MASK bits. 2834 */ 2835 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) { 2836 int why; 2837 2838 if (signal->flags & SIGNAL_CLD_CONTINUED) 2839 why = CLD_CONTINUED; 2840 else 2841 why = CLD_STOPPED; 2842 2843 signal->flags &= ~SIGNAL_CLD_MASK; 2844 2845 spin_unlock_irq(&sighand->siglock); 2846 2847 /* 2848 * Notify the parent that we're continuing. This event is 2849 * always per-process and doesn't make whole lot of sense 2850 * for ptracers, who shouldn't consume the state via 2851 * wait(2) either, but, for backward compatibility, notify 2852 * the ptracer of the group leader too unless it's gonna be 2853 * a duplicate. 2854 */ 2855 read_lock(&tasklist_lock); 2856 do_notify_parent_cldstop(current, false, why); 2857 2858 if (ptrace_reparented(current->group_leader)) 2859 do_notify_parent_cldstop(current->group_leader, 2860 true, why); 2861 read_unlock(&tasklist_lock); 2862 2863 goto relock; 2864 } 2865 2866 for (;;) { 2867 struct k_sigaction *ka; 2868 enum pid_type type; 2869 2870 /* Has this task already been marked for death? */ 2871 if ((signal->flags & SIGNAL_GROUP_EXIT) || 2872 signal->group_exec_task) { 2873 signr = SIGKILL; 2874 sigdelset(¤t->pending.signal, SIGKILL); 2875 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO, 2876 &sighand->action[SIGKILL-1]); 2877 recalc_sigpending(); 2878 /* 2879 * implies do_group_exit() or return to PF_USER_WORKER, 2880 * no need to initialize ksig->info/etc. 2881 */ 2882 goto fatal; 2883 } 2884 2885 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) && 2886 do_signal_stop(0)) 2887 goto relock; 2888 2889 if (unlikely(current->jobctl & 2890 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) { 2891 if (current->jobctl & JOBCTL_TRAP_MASK) { 2892 do_jobctl_trap(); 2893 spin_unlock_irq(&sighand->siglock); 2894 } else if (current->jobctl & JOBCTL_TRAP_FREEZE) 2895 do_freezer_trap(); 2896 2897 goto relock; 2898 } 2899 2900 /* 2901 * If the task is leaving the frozen state, let's update 2902 * cgroup counters and reset the frozen bit. 2903 */ 2904 if (unlikely(cgroup_task_frozen(current))) { 2905 spin_unlock_irq(&sighand->siglock); 2906 cgroup_leave_frozen(false); 2907 goto relock; 2908 } 2909 2910 /* 2911 * Signals generated by the execution of an instruction 2912 * need to be delivered before any other pending signals 2913 * so that the instruction pointer in the signal stack 2914 * frame points to the faulting instruction. 2915 */ 2916 type = PIDTYPE_PID; 2917 signr = dequeue_synchronous_signal(&ksig->info); 2918 if (!signr) 2919 signr = dequeue_signal(¤t->blocked, &ksig->info, &type); 2920 2921 if (!signr) 2922 break; /* will return 0 */ 2923 2924 if (unlikely(current->ptrace) && (signr != SIGKILL) && 2925 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) { 2926 signr = ptrace_signal(signr, &ksig->info, type); 2927 if (!signr) 2928 continue; 2929 } 2930 2931 ka = &sighand->action[signr-1]; 2932 2933 /* Trace actually delivered signals. */ 2934 trace_signal_deliver(signr, &ksig->info, ka); 2935 2936 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */ 2937 continue; 2938 if (ka->sa.sa_handler != SIG_DFL) { 2939 /* Run the handler. */ 2940 ksig->ka = *ka; 2941 2942 if (ka->sa.sa_flags & SA_ONESHOT) 2943 ka->sa.sa_handler = SIG_DFL; 2944 2945 break; /* will return non-zero "signr" value */ 2946 } 2947 2948 /* 2949 * Now we are doing the default action for this signal. 2950 */ 2951 if (sig_kernel_ignore(signr)) /* Default is nothing. */ 2952 continue; 2953 2954 /* 2955 * Global init gets no signals it doesn't want. 2956 * Container-init gets no signals it doesn't want from same 2957 * container. 2958 * 2959 * Note that if global/container-init sees a sig_kernel_only() 2960 * signal here, the signal must have been generated internally 2961 * or must have come from an ancestor namespace. In either 2962 * case, the signal cannot be dropped. 2963 */ 2964 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) && 2965 !sig_kernel_only(signr)) 2966 continue; 2967 2968 if (sig_kernel_stop(signr)) { 2969 /* 2970 * The default action is to stop all threads in 2971 * the thread group. The job control signals 2972 * do nothing in an orphaned pgrp, but SIGSTOP 2973 * always works. Note that siglock needs to be 2974 * dropped during the call to is_orphaned_pgrp() 2975 * because of lock ordering with tasklist_lock. 2976 * This allows an intervening SIGCONT to be posted. 2977 * We need to check for that and bail out if necessary. 2978 */ 2979 if (signr != SIGSTOP) { 2980 spin_unlock_irq(&sighand->siglock); 2981 2982 /* signals can be posted during this window */ 2983 2984 if (is_current_pgrp_orphaned()) 2985 goto relock; 2986 2987 spin_lock_irq(&sighand->siglock); 2988 } 2989 2990 if (likely(do_signal_stop(signr))) { 2991 /* It released the siglock. */ 2992 goto relock; 2993 } 2994 2995 /* 2996 * We didn't actually stop, due to a race 2997 * with SIGCONT or something like that. 2998 */ 2999 continue; 3000 } 3001 3002 fatal: 3003 spin_unlock_irq(&sighand->siglock); 3004 if (unlikely(cgroup_task_frozen(current))) 3005 cgroup_leave_frozen(true); 3006 3007 /* 3008 * Anything else is fatal, maybe with a core dump. 3009 */ 3010 current->flags |= PF_SIGNALED; 3011 3012 if (sig_kernel_coredump(signr)) { 3013 if (print_fatal_signals) 3014 print_fatal_signal(signr); 3015 proc_coredump_connector(current); 3016 /* 3017 * If it was able to dump core, this kills all 3018 * other threads in the group and synchronizes with 3019 * their demise. If we lost the race with another 3020 * thread getting here, it set group_exit_code 3021 * first and our do_group_exit call below will use 3022 * that value and ignore the one we pass it. 3023 */ 3024 vfs_coredump(&ksig->info); 3025 } 3026 3027 /* 3028 * PF_USER_WORKER threads will catch and exit on fatal signals 3029 * themselves. They have cleanup that must be performed, so we 3030 * cannot call do_exit() on their behalf. Note that ksig won't 3031 * be properly initialized, PF_USER_WORKER's shouldn't use it. 3032 */ 3033 if (current->flags & PF_USER_WORKER) 3034 goto out; 3035 3036 /* 3037 * Death signals, no core dump. 3038 */ 3039 do_group_exit(signr); 3040 /* NOTREACHED */ 3041 } 3042 spin_unlock_irq(&sighand->siglock); 3043 3044 ksig->sig = signr; 3045 3046 if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS)) 3047 hide_si_addr_tag_bits(ksig); 3048 out: 3049 return signr > 0; 3050 } 3051 3052 /** 3053 * signal_delivered - called after signal delivery to update blocked signals 3054 * @ksig: kernel signal struct 3055 * @stepping: nonzero if debugger single-step or block-step in use 3056 * 3057 * This function should be called when a signal has successfully been 3058 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask 3059 * is always blocked), and the signal itself is blocked unless %SA_NODEFER 3060 * is set in @ksig->ka.sa.sa_flags. Tracing is notified. 3061 */ 3062 static void signal_delivered(struct ksignal *ksig, int stepping) 3063 { 3064 sigset_t blocked; 3065 3066 /* A signal was successfully delivered, and the 3067 saved sigmask was stored on the signal frame, 3068 and will be restored by sigreturn. So we can 3069 simply clear the restore sigmask flag. */ 3070 clear_restore_sigmask(); 3071 3072 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask); 3073 if (!(ksig->ka.sa.sa_flags & SA_NODEFER)) 3074 sigaddset(&blocked, ksig->sig); 3075 set_current_blocked(&blocked); 3076 if (current->sas_ss_flags & SS_AUTODISARM) 3077 sas_ss_reset(current); 3078 if (stepping) 3079 ptrace_notify(SIGTRAP, 0); 3080 } 3081 3082 void signal_setup_done(int failed, struct ksignal *ksig, int stepping) 3083 { 3084 if (failed) 3085 force_sigsegv(ksig->sig); 3086 else 3087 signal_delivered(ksig, stepping); 3088 } 3089 3090 /* 3091 * It could be that complete_signal() picked us to notify about the 3092 * group-wide signal. Other threads should be notified now to take 3093 * the shared signals in @which since we will not. 3094 */ 3095 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which) 3096 { 3097 sigset_t retarget; 3098 struct task_struct *t; 3099 3100 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which); 3101 if (sigisemptyset(&retarget)) 3102 return; 3103 3104 for_other_threads(tsk, t) { 3105 if (t->flags & PF_EXITING) 3106 continue; 3107 3108 if (!has_pending_signals(&retarget, &t->blocked)) 3109 continue; 3110 /* Remove the signals this thread can handle. */ 3111 sigandsets(&retarget, &retarget, &t->blocked); 3112 3113 if (!task_sigpending(t)) 3114 signal_wake_up(t, 0); 3115 3116 if (sigisemptyset(&retarget)) 3117 break; 3118 } 3119 } 3120 3121 void exit_signals(struct task_struct *tsk) 3122 { 3123 int group_stop = 0; 3124 sigset_t unblocked; 3125 3126 /* 3127 * @tsk is about to have PF_EXITING set - lock out users which 3128 * expect stable threadgroup. 3129 */ 3130 cgroup_threadgroup_change_begin(tsk); 3131 3132 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) { 3133 tsk->flags |= PF_EXITING; 3134 cgroup_threadgroup_change_end(tsk); 3135 return; 3136 } 3137 3138 spin_lock_irq(&tsk->sighand->siglock); 3139 /* 3140 * From now this task is not visible for group-wide signals, 3141 * see wants_signal(), do_signal_stop(). 3142 */ 3143 tsk->flags |= PF_EXITING; 3144 3145 cgroup_threadgroup_change_end(tsk); 3146 3147 if (!task_sigpending(tsk)) 3148 goto out; 3149 3150 unblocked = tsk->blocked; 3151 signotset(&unblocked); 3152 retarget_shared_pending(tsk, &unblocked); 3153 3154 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) && 3155 task_participate_group_stop(tsk)) 3156 group_stop = CLD_STOPPED; 3157 out: 3158 spin_unlock_irq(&tsk->sighand->siglock); 3159 3160 /* 3161 * If group stop has completed, deliver the notification. This 3162 * should always go to the real parent of the group leader. 3163 */ 3164 if (unlikely(group_stop)) { 3165 read_lock(&tasklist_lock); 3166 do_notify_parent_cldstop(tsk, false, group_stop); 3167 read_unlock(&tasklist_lock); 3168 } 3169 } 3170 3171 /* 3172 * System call entry points. 3173 */ 3174 3175 /** 3176 * sys_restart_syscall - restart a system call 3177 */ 3178 SYSCALL_DEFINE0(restart_syscall) 3179 { 3180 struct restart_block *restart = ¤t->restart_block; 3181 return restart->fn(restart); 3182 } 3183 3184 long do_no_restart_syscall(struct restart_block *param) 3185 { 3186 return -EINTR; 3187 } 3188 3189 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset) 3190 { 3191 if (task_sigpending(tsk) && !thread_group_empty(tsk)) { 3192 sigset_t newblocked; 3193 /* A set of now blocked but previously unblocked signals. */ 3194 sigandnsets(&newblocked, newset, ¤t->blocked); 3195 retarget_shared_pending(tsk, &newblocked); 3196 } 3197 tsk->blocked = *newset; 3198 recalc_sigpending(); 3199 } 3200 3201 /** 3202 * set_current_blocked - change current->blocked mask 3203 * @newset: new mask 3204 * 3205 * It is wrong to change ->blocked directly, this helper should be used 3206 * to ensure the process can't miss a shared signal we are going to block. 3207 */ 3208 void set_current_blocked(sigset_t *newset) 3209 { 3210 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP)); 3211 __set_current_blocked(newset); 3212 } 3213 3214 void __set_current_blocked(const sigset_t *newset) 3215 { 3216 struct task_struct *tsk = current; 3217 3218 /* 3219 * In case the signal mask hasn't changed, there is nothing we need 3220 * to do. The current->blocked shouldn't be modified by other task. 3221 */ 3222 if (sigequalsets(&tsk->blocked, newset)) 3223 return; 3224 3225 spin_lock_irq(&tsk->sighand->siglock); 3226 __set_task_blocked(tsk, newset); 3227 spin_unlock_irq(&tsk->sighand->siglock); 3228 } 3229 3230 /* 3231 * This is also useful for kernel threads that want to temporarily 3232 * (or permanently) block certain signals. 3233 * 3234 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel 3235 * interface happily blocks "unblockable" signals like SIGKILL 3236 * and friends. 3237 */ 3238 int sigprocmask(int how, sigset_t *set, sigset_t *oldset) 3239 { 3240 struct task_struct *tsk = current; 3241 sigset_t newset; 3242 3243 /* Lockless, only current can change ->blocked, never from irq */ 3244 if (oldset) 3245 *oldset = tsk->blocked; 3246 3247 switch (how) { 3248 case SIG_BLOCK: 3249 sigorsets(&newset, &tsk->blocked, set); 3250 break; 3251 case SIG_UNBLOCK: 3252 sigandnsets(&newset, &tsk->blocked, set); 3253 break; 3254 case SIG_SETMASK: 3255 newset = *set; 3256 break; 3257 default: 3258 return -EINVAL; 3259 } 3260 3261 __set_current_blocked(&newset); 3262 return 0; 3263 } 3264 EXPORT_SYMBOL(sigprocmask); 3265 3266 /* 3267 * The api helps set app-provided sigmasks. 3268 * 3269 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and 3270 * epoll_pwait where a new sigmask is passed from userland for the syscalls. 3271 * 3272 * Note that it does set_restore_sigmask() in advance, so it must be always 3273 * paired with restore_saved_sigmask_unless() before return from syscall. 3274 */ 3275 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize) 3276 { 3277 sigset_t kmask; 3278 3279 if (!umask) 3280 return 0; 3281 if (sigsetsize != sizeof(sigset_t)) 3282 return -EINVAL; 3283 if (copy_from_user(&kmask, umask, sizeof(sigset_t))) 3284 return -EFAULT; 3285 3286 set_restore_sigmask(); 3287 current->saved_sigmask = current->blocked; 3288 set_current_blocked(&kmask); 3289 3290 return 0; 3291 } 3292 3293 #ifdef CONFIG_COMPAT 3294 int set_compat_user_sigmask(const compat_sigset_t __user *umask, 3295 size_t sigsetsize) 3296 { 3297 sigset_t kmask; 3298 3299 if (!umask) 3300 return 0; 3301 if (sigsetsize != sizeof(compat_sigset_t)) 3302 return -EINVAL; 3303 if (get_compat_sigset(&kmask, umask)) 3304 return -EFAULT; 3305 3306 set_restore_sigmask(); 3307 current->saved_sigmask = current->blocked; 3308 set_current_blocked(&kmask); 3309 3310 return 0; 3311 } 3312 #endif 3313 3314 /** 3315 * sys_rt_sigprocmask - change the list of currently blocked signals 3316 * @how: whether to add, remove, or set signals 3317 * @nset: stores pending signals 3318 * @oset: previous value of signal mask if non-null 3319 * @sigsetsize: size of sigset_t type 3320 */ 3321 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset, 3322 sigset_t __user *, oset, size_t, sigsetsize) 3323 { 3324 sigset_t old_set, new_set; 3325 int error; 3326 3327 /* XXX: Don't preclude handling different sized sigset_t's. */ 3328 if (sigsetsize != sizeof(sigset_t)) 3329 return -EINVAL; 3330 3331 old_set = current->blocked; 3332 3333 if (nset) { 3334 if (copy_from_user(&new_set, nset, sizeof(sigset_t))) 3335 return -EFAULT; 3336 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP)); 3337 3338 error = sigprocmask(how, &new_set, NULL); 3339 if (error) 3340 return error; 3341 } 3342 3343 if (oset) { 3344 if (copy_to_user(oset, &old_set, sizeof(sigset_t))) 3345 return -EFAULT; 3346 } 3347 3348 return 0; 3349 } 3350 3351 #ifdef CONFIG_COMPAT 3352 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset, 3353 compat_sigset_t __user *, oset, compat_size_t, sigsetsize) 3354 { 3355 sigset_t old_set = current->blocked; 3356 3357 /* XXX: Don't preclude handling different sized sigset_t's. */ 3358 if (sigsetsize != sizeof(sigset_t)) 3359 return -EINVAL; 3360 3361 if (nset) { 3362 sigset_t new_set; 3363 int error; 3364 if (get_compat_sigset(&new_set, nset)) 3365 return -EFAULT; 3366 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP)); 3367 3368 error = sigprocmask(how, &new_set, NULL); 3369 if (error) 3370 return error; 3371 } 3372 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0; 3373 } 3374 #endif 3375 3376 static void do_sigpending(sigset_t *set) 3377 { 3378 spin_lock_irq(¤t->sighand->siglock); 3379 sigorsets(set, ¤t->pending.signal, 3380 ¤t->signal->shared_pending.signal); 3381 spin_unlock_irq(¤t->sighand->siglock); 3382 3383 /* Outside the lock because only this thread touches it. */ 3384 sigandsets(set, ¤t->blocked, set); 3385 } 3386 3387 /** 3388 * sys_rt_sigpending - examine a pending signal that has been raised 3389 * while blocked 3390 * @uset: stores pending signals 3391 * @sigsetsize: size of sigset_t type or larger 3392 */ 3393 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize) 3394 { 3395 sigset_t set; 3396 3397 if (sigsetsize > sizeof(*uset)) 3398 return -EINVAL; 3399 3400 do_sigpending(&set); 3401 3402 if (copy_to_user(uset, &set, sigsetsize)) 3403 return -EFAULT; 3404 3405 return 0; 3406 } 3407 3408 #ifdef CONFIG_COMPAT 3409 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset, 3410 compat_size_t, sigsetsize) 3411 { 3412 sigset_t set; 3413 3414 if (sigsetsize > sizeof(*uset)) 3415 return -EINVAL; 3416 3417 do_sigpending(&set); 3418 3419 return put_compat_sigset(uset, &set, sigsetsize); 3420 } 3421 #endif 3422 3423 static const struct { 3424 unsigned char limit, layout; 3425 } sig_sicodes[] = { 3426 [SIGILL] = { NSIGILL, SIL_FAULT }, 3427 [SIGFPE] = { NSIGFPE, SIL_FAULT }, 3428 [SIGSEGV] = { NSIGSEGV, SIL_FAULT }, 3429 [SIGBUS] = { NSIGBUS, SIL_FAULT }, 3430 [SIGTRAP] = { NSIGTRAP, SIL_FAULT }, 3431 #if defined(SIGEMT) 3432 [SIGEMT] = { NSIGEMT, SIL_FAULT }, 3433 #endif 3434 [SIGCHLD] = { NSIGCHLD, SIL_CHLD }, 3435 [SIGPOLL] = { NSIGPOLL, SIL_POLL }, 3436 [SIGSYS] = { NSIGSYS, SIL_SYS }, 3437 }; 3438 3439 static bool known_siginfo_layout(unsigned sig, int si_code) 3440 { 3441 if (si_code == SI_KERNEL) 3442 return true; 3443 else if ((si_code > SI_USER)) { 3444 if (sig_specific_sicodes(sig)) { 3445 if (si_code <= sig_sicodes[sig].limit) 3446 return true; 3447 } 3448 else if (si_code <= NSIGPOLL) 3449 return true; 3450 } 3451 else if (si_code >= SI_DETHREAD) 3452 return true; 3453 else if (si_code == SI_ASYNCNL) 3454 return true; 3455 return false; 3456 } 3457 3458 enum siginfo_layout siginfo_layout(unsigned sig, int si_code) 3459 { 3460 enum siginfo_layout layout = SIL_KILL; 3461 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) { 3462 if ((sig < ARRAY_SIZE(sig_sicodes)) && 3463 (si_code <= sig_sicodes[sig].limit)) { 3464 layout = sig_sicodes[sig].layout; 3465 /* Handle the exceptions */ 3466 if ((sig == SIGBUS) && 3467 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO)) 3468 layout = SIL_FAULT_MCEERR; 3469 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR)) 3470 layout = SIL_FAULT_BNDERR; 3471 #ifdef SEGV_PKUERR 3472 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR)) 3473 layout = SIL_FAULT_PKUERR; 3474 #endif 3475 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF)) 3476 layout = SIL_FAULT_PERF_EVENT; 3477 else if (IS_ENABLED(CONFIG_SPARC) && 3478 (sig == SIGILL) && (si_code == ILL_ILLTRP)) 3479 layout = SIL_FAULT_TRAPNO; 3480 else if (IS_ENABLED(CONFIG_ALPHA) && 3481 ((sig == SIGFPE) || 3482 ((sig == SIGTRAP) && (si_code == TRAP_UNK)))) 3483 layout = SIL_FAULT_TRAPNO; 3484 } 3485 else if (si_code <= NSIGPOLL) 3486 layout = SIL_POLL; 3487 } else { 3488 if (si_code == SI_TIMER) 3489 layout = SIL_TIMER; 3490 else if (si_code == SI_SIGIO) 3491 layout = SIL_POLL; 3492 else if (si_code < 0) 3493 layout = SIL_RT; 3494 } 3495 return layout; 3496 } 3497 3498 static inline char __user *si_expansion(const siginfo_t __user *info) 3499 { 3500 return ((char __user *)info) + sizeof(struct kernel_siginfo); 3501 } 3502 3503 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from) 3504 { 3505 char __user *expansion = si_expansion(to); 3506 if (copy_to_user(to, from , sizeof(struct kernel_siginfo))) 3507 return -EFAULT; 3508 if (clear_user(expansion, SI_EXPANSION_SIZE)) 3509 return -EFAULT; 3510 return 0; 3511 } 3512 3513 static int post_copy_siginfo_from_user(kernel_siginfo_t *info, 3514 const siginfo_t __user *from) 3515 { 3516 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) { 3517 char __user *expansion = si_expansion(from); 3518 char buf[SI_EXPANSION_SIZE]; 3519 int i; 3520 /* 3521 * An unknown si_code might need more than 3522 * sizeof(struct kernel_siginfo) bytes. Verify all of the 3523 * extra bytes are 0. This guarantees copy_siginfo_to_user 3524 * will return this data to userspace exactly. 3525 */ 3526 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE)) 3527 return -EFAULT; 3528 for (i = 0; i < SI_EXPANSION_SIZE; i++) { 3529 if (buf[i] != 0) 3530 return -E2BIG; 3531 } 3532 } 3533 return 0; 3534 } 3535 3536 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to, 3537 const siginfo_t __user *from) 3538 { 3539 if (copy_from_user(to, from, sizeof(struct kernel_siginfo))) 3540 return -EFAULT; 3541 to->si_signo = signo; 3542 return post_copy_siginfo_from_user(to, from); 3543 } 3544 3545 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from) 3546 { 3547 if (copy_from_user(to, from, sizeof(struct kernel_siginfo))) 3548 return -EFAULT; 3549 return post_copy_siginfo_from_user(to, from); 3550 } 3551 3552 #ifdef CONFIG_COMPAT 3553 /** 3554 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo 3555 * @to: compat siginfo destination 3556 * @from: kernel siginfo source 3557 * 3558 * Note: This function does not work properly for the SIGCHLD on x32, but 3559 * fortunately it doesn't have to. The only valid callers for this function are 3560 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code. 3561 * The latter does not care because SIGCHLD will never cause a coredump. 3562 */ 3563 void copy_siginfo_to_external32(struct compat_siginfo *to, 3564 const struct kernel_siginfo *from) 3565 { 3566 memset(to, 0, sizeof(*to)); 3567 3568 to->si_signo = from->si_signo; 3569 to->si_errno = from->si_errno; 3570 to->si_code = from->si_code; 3571 switch(siginfo_layout(from->si_signo, from->si_code)) { 3572 case SIL_KILL: 3573 to->si_pid = from->si_pid; 3574 to->si_uid = from->si_uid; 3575 break; 3576 case SIL_TIMER: 3577 to->si_tid = from->si_tid; 3578 to->si_overrun = from->si_overrun; 3579 to->si_int = from->si_int; 3580 break; 3581 case SIL_POLL: 3582 to->si_band = from->si_band; 3583 to->si_fd = from->si_fd; 3584 break; 3585 case SIL_FAULT: 3586 to->si_addr = ptr_to_compat(from->si_addr); 3587 break; 3588 case SIL_FAULT_TRAPNO: 3589 to->si_addr = ptr_to_compat(from->si_addr); 3590 to->si_trapno = from->si_trapno; 3591 break; 3592 case SIL_FAULT_MCEERR: 3593 to->si_addr = ptr_to_compat(from->si_addr); 3594 to->si_addr_lsb = from->si_addr_lsb; 3595 break; 3596 case SIL_FAULT_BNDERR: 3597 to->si_addr = ptr_to_compat(from->si_addr); 3598 to->si_lower = ptr_to_compat(from->si_lower); 3599 to->si_upper = ptr_to_compat(from->si_upper); 3600 break; 3601 case SIL_FAULT_PKUERR: 3602 to->si_addr = ptr_to_compat(from->si_addr); 3603 to->si_pkey = from->si_pkey; 3604 break; 3605 case SIL_FAULT_PERF_EVENT: 3606 to->si_addr = ptr_to_compat(from->si_addr); 3607 to->si_perf_data = from->si_perf_data; 3608 to->si_perf_type = from->si_perf_type; 3609 to->si_perf_flags = from->si_perf_flags; 3610 break; 3611 case SIL_CHLD: 3612 to->si_pid = from->si_pid; 3613 to->si_uid = from->si_uid; 3614 to->si_status = from->si_status; 3615 to->si_utime = from->si_utime; 3616 to->si_stime = from->si_stime; 3617 break; 3618 case SIL_RT: 3619 to->si_pid = from->si_pid; 3620 to->si_uid = from->si_uid; 3621 to->si_int = from->si_int; 3622 break; 3623 case SIL_SYS: 3624 to->si_call_addr = ptr_to_compat(from->si_call_addr); 3625 to->si_syscall = from->si_syscall; 3626 to->si_arch = from->si_arch; 3627 break; 3628 } 3629 } 3630 3631 int __copy_siginfo_to_user32(struct compat_siginfo __user *to, 3632 const struct kernel_siginfo *from) 3633 { 3634 struct compat_siginfo new; 3635 3636 copy_siginfo_to_external32(&new, from); 3637 if (copy_to_user(to, &new, sizeof(struct compat_siginfo))) 3638 return -EFAULT; 3639 return 0; 3640 } 3641 3642 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to, 3643 const struct compat_siginfo *from) 3644 { 3645 clear_siginfo(to); 3646 to->si_signo = from->si_signo; 3647 to->si_errno = from->si_errno; 3648 to->si_code = from->si_code; 3649 switch(siginfo_layout(from->si_signo, from->si_code)) { 3650 case SIL_KILL: 3651 to->si_pid = from->si_pid; 3652 to->si_uid = from->si_uid; 3653 break; 3654 case SIL_TIMER: 3655 to->si_tid = from->si_tid; 3656 to->si_overrun = from->si_overrun; 3657 to->si_int = from->si_int; 3658 break; 3659 case SIL_POLL: 3660 to->si_band = from->si_band; 3661 to->si_fd = from->si_fd; 3662 break; 3663 case SIL_FAULT: 3664 to->si_addr = compat_ptr(from->si_addr); 3665 break; 3666 case SIL_FAULT_TRAPNO: 3667 to->si_addr = compat_ptr(from->si_addr); 3668 to->si_trapno = from->si_trapno; 3669 break; 3670 case SIL_FAULT_MCEERR: 3671 to->si_addr = compat_ptr(from->si_addr); 3672 to->si_addr_lsb = from->si_addr_lsb; 3673 break; 3674 case SIL_FAULT_BNDERR: 3675 to->si_addr = compat_ptr(from->si_addr); 3676 to->si_lower = compat_ptr(from->si_lower); 3677 to->si_upper = compat_ptr(from->si_upper); 3678 break; 3679 case SIL_FAULT_PKUERR: 3680 to->si_addr = compat_ptr(from->si_addr); 3681 to->si_pkey = from->si_pkey; 3682 break; 3683 case SIL_FAULT_PERF_EVENT: 3684 to->si_addr = compat_ptr(from->si_addr); 3685 to->si_perf_data = from->si_perf_data; 3686 to->si_perf_type = from->si_perf_type; 3687 to->si_perf_flags = from->si_perf_flags; 3688 break; 3689 case SIL_CHLD: 3690 to->si_pid = from->si_pid; 3691 to->si_uid = from->si_uid; 3692 to->si_status = from->si_status; 3693 #ifdef CONFIG_X86_X32_ABI 3694 if (in_x32_syscall()) { 3695 to->si_utime = from->_sifields._sigchld_x32._utime; 3696 to->si_stime = from->_sifields._sigchld_x32._stime; 3697 } else 3698 #endif 3699 { 3700 to->si_utime = from->si_utime; 3701 to->si_stime = from->si_stime; 3702 } 3703 break; 3704 case SIL_RT: 3705 to->si_pid = from->si_pid; 3706 to->si_uid = from->si_uid; 3707 to->si_int = from->si_int; 3708 break; 3709 case SIL_SYS: 3710 to->si_call_addr = compat_ptr(from->si_call_addr); 3711 to->si_syscall = from->si_syscall; 3712 to->si_arch = from->si_arch; 3713 break; 3714 } 3715 return 0; 3716 } 3717 3718 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to, 3719 const struct compat_siginfo __user *ufrom) 3720 { 3721 struct compat_siginfo from; 3722 3723 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo))) 3724 return -EFAULT; 3725 3726 from.si_signo = signo; 3727 return post_copy_siginfo_from_user32(to, &from); 3728 } 3729 3730 int copy_siginfo_from_user32(struct kernel_siginfo *to, 3731 const struct compat_siginfo __user *ufrom) 3732 { 3733 struct compat_siginfo from; 3734 3735 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo))) 3736 return -EFAULT; 3737 3738 return post_copy_siginfo_from_user32(to, &from); 3739 } 3740 #endif /* CONFIG_COMPAT */ 3741 3742 /** 3743 * do_sigtimedwait - wait for queued signals specified in @which 3744 * @which: queued signals to wait for 3745 * @info: if non-null, the signal's siginfo is returned here 3746 * @ts: upper bound on process time suspension 3747 */ 3748 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info, 3749 const struct timespec64 *ts) 3750 { 3751 ktime_t *to = NULL, timeout = KTIME_MAX; 3752 struct task_struct *tsk = current; 3753 sigset_t mask = *which; 3754 enum pid_type type; 3755 int sig, ret = 0; 3756 3757 if (ts) { 3758 if (!timespec64_valid(ts)) 3759 return -EINVAL; 3760 timeout = timespec64_to_ktime(*ts); 3761 to = &timeout; 3762 } 3763 3764 /* 3765 * Invert the set of allowed signals to get those we want to block. 3766 */ 3767 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP)); 3768 signotset(&mask); 3769 3770 spin_lock_irq(&tsk->sighand->siglock); 3771 sig = dequeue_signal(&mask, info, &type); 3772 if (!sig && timeout) { 3773 /* 3774 * None ready, temporarily unblock those we're interested 3775 * while we are sleeping in so that we'll be awakened when 3776 * they arrive. Unblocking is always fine, we can avoid 3777 * set_current_blocked(). 3778 */ 3779 tsk->real_blocked = tsk->blocked; 3780 sigandsets(&tsk->blocked, &tsk->blocked, &mask); 3781 recalc_sigpending(); 3782 spin_unlock_irq(&tsk->sighand->siglock); 3783 3784 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE); 3785 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns, 3786 HRTIMER_MODE_REL); 3787 spin_lock_irq(&tsk->sighand->siglock); 3788 __set_task_blocked(tsk, &tsk->real_blocked); 3789 sigemptyset(&tsk->real_blocked); 3790 sig = dequeue_signal(&mask, info, &type); 3791 } 3792 spin_unlock_irq(&tsk->sighand->siglock); 3793 3794 if (sig) 3795 return sig; 3796 return ret ? -EINTR : -EAGAIN; 3797 } 3798 3799 /** 3800 * sys_rt_sigtimedwait - synchronously wait for queued signals specified 3801 * in @uthese 3802 * @uthese: queued signals to wait for 3803 * @uinfo: if non-null, the signal's siginfo is returned here 3804 * @uts: upper bound on process time suspension 3805 * @sigsetsize: size of sigset_t type 3806 */ 3807 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese, 3808 siginfo_t __user *, uinfo, 3809 const struct __kernel_timespec __user *, uts, 3810 size_t, sigsetsize) 3811 { 3812 sigset_t these; 3813 struct timespec64 ts; 3814 kernel_siginfo_t info; 3815 int ret; 3816 3817 /* XXX: Don't preclude handling different sized sigset_t's. */ 3818 if (sigsetsize != sizeof(sigset_t)) 3819 return -EINVAL; 3820 3821 if (copy_from_user(&these, uthese, sizeof(these))) 3822 return -EFAULT; 3823 3824 if (uts) { 3825 if (get_timespec64(&ts, uts)) 3826 return -EFAULT; 3827 } 3828 3829 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL); 3830 3831 if (ret > 0 && uinfo) { 3832 if (copy_siginfo_to_user(uinfo, &info)) 3833 ret = -EFAULT; 3834 } 3835 3836 return ret; 3837 } 3838 3839 #ifdef CONFIG_COMPAT_32BIT_TIME 3840 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese, 3841 siginfo_t __user *, uinfo, 3842 const struct old_timespec32 __user *, uts, 3843 size_t, sigsetsize) 3844 { 3845 sigset_t these; 3846 struct timespec64 ts; 3847 kernel_siginfo_t info; 3848 int ret; 3849 3850 if (sigsetsize != sizeof(sigset_t)) 3851 return -EINVAL; 3852 3853 if (copy_from_user(&these, uthese, sizeof(these))) 3854 return -EFAULT; 3855 3856 if (uts) { 3857 if (get_old_timespec32(&ts, uts)) 3858 return -EFAULT; 3859 } 3860 3861 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL); 3862 3863 if (ret > 0 && uinfo) { 3864 if (copy_siginfo_to_user(uinfo, &info)) 3865 ret = -EFAULT; 3866 } 3867 3868 return ret; 3869 } 3870 #endif 3871 3872 #ifdef CONFIG_COMPAT 3873 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese, 3874 struct compat_siginfo __user *, uinfo, 3875 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize) 3876 { 3877 sigset_t s; 3878 struct timespec64 t; 3879 kernel_siginfo_t info; 3880 long ret; 3881 3882 if (sigsetsize != sizeof(sigset_t)) 3883 return -EINVAL; 3884 3885 if (get_compat_sigset(&s, uthese)) 3886 return -EFAULT; 3887 3888 if (uts) { 3889 if (get_timespec64(&t, uts)) 3890 return -EFAULT; 3891 } 3892 3893 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL); 3894 3895 if (ret > 0 && uinfo) { 3896 if (copy_siginfo_to_user32(uinfo, &info)) 3897 ret = -EFAULT; 3898 } 3899 3900 return ret; 3901 } 3902 3903 #ifdef CONFIG_COMPAT_32BIT_TIME 3904 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese, 3905 struct compat_siginfo __user *, uinfo, 3906 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize) 3907 { 3908 sigset_t s; 3909 struct timespec64 t; 3910 kernel_siginfo_t info; 3911 long ret; 3912 3913 if (sigsetsize != sizeof(sigset_t)) 3914 return -EINVAL; 3915 3916 if (get_compat_sigset(&s, uthese)) 3917 return -EFAULT; 3918 3919 if (uts) { 3920 if (get_old_timespec32(&t, uts)) 3921 return -EFAULT; 3922 } 3923 3924 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL); 3925 3926 if (ret > 0 && uinfo) { 3927 if (copy_siginfo_to_user32(uinfo, &info)) 3928 ret = -EFAULT; 3929 } 3930 3931 return ret; 3932 } 3933 #endif 3934 #endif 3935 3936 static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info, 3937 enum pid_type type) 3938 { 3939 clear_siginfo(info); 3940 info->si_signo = sig; 3941 info->si_errno = 0; 3942 info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER; 3943 info->si_pid = task_tgid_vnr(current); 3944 info->si_uid = from_kuid_munged(current_user_ns(), current_uid()); 3945 } 3946 3947 /** 3948 * sys_kill - send a signal to a process 3949 * @pid: the PID of the process 3950 * @sig: signal to be sent 3951 */ 3952 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig) 3953 { 3954 return kill_something_info(sig, SEND_SIG_NOINFO, pid); 3955 } 3956 3957 /* 3958 * Verify that the signaler and signalee either are in the same pid namespace 3959 * or that the signaler's pid namespace is an ancestor of the signalee's pid 3960 * namespace. 3961 */ 3962 static bool access_pidfd_pidns(struct pid *pid) 3963 { 3964 struct pid_namespace *active = task_active_pid_ns(current); 3965 struct pid_namespace *p = ns_of_pid(pid); 3966 3967 for (;;) { 3968 if (!p) 3969 return false; 3970 if (p == active) 3971 break; 3972 p = p->parent; 3973 } 3974 3975 return true; 3976 } 3977 3978 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, 3979 siginfo_t __user *info) 3980 { 3981 #ifdef CONFIG_COMPAT 3982 /* 3983 * Avoid hooking up compat syscalls and instead handle necessary 3984 * conversions here. Note, this is a stop-gap measure and should not be 3985 * considered a generic solution. 3986 */ 3987 if (in_compat_syscall()) 3988 return copy_siginfo_from_user32( 3989 kinfo, (struct compat_siginfo __user *)info); 3990 #endif 3991 return copy_siginfo_from_user(kinfo, info); 3992 } 3993 3994 static struct pid *pidfd_to_pid(const struct file *file) 3995 { 3996 struct pid *pid; 3997 3998 pid = pidfd_pid(file); 3999 if (!IS_ERR(pid)) 4000 return pid; 4001 4002 return tgid_pidfd_to_pid(file); 4003 } 4004 4005 #define PIDFD_SEND_SIGNAL_FLAGS \ 4006 (PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \ 4007 PIDFD_SIGNAL_PROCESS_GROUP) 4008 4009 static int do_pidfd_send_signal(struct pid *pid, int sig, enum pid_type type, 4010 siginfo_t __user *info, unsigned int flags) 4011 { 4012 kernel_siginfo_t kinfo; 4013 4014 switch (flags) { 4015 case PIDFD_SIGNAL_THREAD: 4016 type = PIDTYPE_PID; 4017 break; 4018 case PIDFD_SIGNAL_THREAD_GROUP: 4019 type = PIDTYPE_TGID; 4020 break; 4021 case PIDFD_SIGNAL_PROCESS_GROUP: 4022 type = PIDTYPE_PGID; 4023 break; 4024 } 4025 4026 if (info) { 4027 int ret; 4028 4029 ret = copy_siginfo_from_user_any(&kinfo, info); 4030 if (unlikely(ret)) 4031 return ret; 4032 4033 if (unlikely(sig != kinfo.si_signo)) 4034 return -EINVAL; 4035 4036 /* Only allow sending arbitrary signals to yourself. */ 4037 if ((task_pid(current) != pid || type > PIDTYPE_TGID) && 4038 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL)) 4039 return -EPERM; 4040 } else { 4041 prepare_kill_siginfo(sig, &kinfo, type); 4042 } 4043 4044 if (type == PIDTYPE_PGID) 4045 return kill_pgrp_info(sig, &kinfo, pid); 4046 4047 return kill_pid_info_type(sig, &kinfo, pid, type); 4048 } 4049 4050 /** 4051 * sys_pidfd_send_signal - Signal a process through a pidfd 4052 * @pidfd: file descriptor of the process 4053 * @sig: signal to send 4054 * @info: signal info 4055 * @flags: future flags 4056 * 4057 * Send the signal to the thread group or to the individual thread depending 4058 * on PIDFD_THREAD. 4059 * In the future extension to @flags may be used to override the default scope 4060 * of @pidfd. 4061 * 4062 * Return: 0 on success, negative errno on failure 4063 */ 4064 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig, 4065 siginfo_t __user *, info, unsigned int, flags) 4066 { 4067 struct pid *pid; 4068 enum pid_type type; 4069 int ret; 4070 4071 /* Enforce flags be set to 0 until we add an extension. */ 4072 if (flags & ~PIDFD_SEND_SIGNAL_FLAGS) 4073 return -EINVAL; 4074 4075 /* Ensure that only a single signal scope determining flag is set. */ 4076 if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1) 4077 return -EINVAL; 4078 4079 switch (pidfd) { 4080 case PIDFD_SELF_THREAD: 4081 pid = get_task_pid(current, PIDTYPE_PID); 4082 type = PIDTYPE_PID; 4083 break; 4084 case PIDFD_SELF_THREAD_GROUP: 4085 pid = get_task_pid(current, PIDTYPE_TGID); 4086 type = PIDTYPE_TGID; 4087 break; 4088 default: { 4089 CLASS(fd, f)(pidfd); 4090 if (fd_empty(f)) 4091 return -EBADF; 4092 4093 /* Is this a pidfd? */ 4094 pid = pidfd_to_pid(fd_file(f)); 4095 if (IS_ERR(pid)) 4096 return PTR_ERR(pid); 4097 4098 if (!access_pidfd_pidns(pid)) 4099 return -EINVAL; 4100 4101 /* Infer scope from the type of pidfd. */ 4102 if (fd_file(f)->f_flags & PIDFD_THREAD) 4103 type = PIDTYPE_PID; 4104 else 4105 type = PIDTYPE_TGID; 4106 4107 return do_pidfd_send_signal(pid, sig, type, info, flags); 4108 } 4109 } 4110 4111 ret = do_pidfd_send_signal(pid, sig, type, info, flags); 4112 put_pid(pid); 4113 4114 return ret; 4115 } 4116 4117 static int 4118 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info) 4119 { 4120 struct task_struct *p; 4121 int error = -ESRCH; 4122 4123 rcu_read_lock(); 4124 p = find_task_by_vpid(pid); 4125 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) { 4126 error = check_kill_permission(sig, info, p); 4127 /* 4128 * The null signal is a permissions and process existence 4129 * probe. No signal is actually delivered. 4130 */ 4131 if (!error && sig) { 4132 error = do_send_sig_info(sig, info, p, PIDTYPE_PID); 4133 /* 4134 * If lock_task_sighand() failed we pretend the task 4135 * dies after receiving the signal. The window is tiny, 4136 * and the signal is private anyway. 4137 */ 4138 if (unlikely(error == -ESRCH)) 4139 error = 0; 4140 } 4141 } 4142 rcu_read_unlock(); 4143 4144 return error; 4145 } 4146 4147 static int do_tkill(pid_t tgid, pid_t pid, int sig) 4148 { 4149 struct kernel_siginfo info; 4150 4151 prepare_kill_siginfo(sig, &info, PIDTYPE_PID); 4152 4153 return do_send_specific(tgid, pid, sig, &info); 4154 } 4155 4156 /** 4157 * sys_tgkill - send signal to one specific thread 4158 * @tgid: the thread group ID of the thread 4159 * @pid: the PID of the thread 4160 * @sig: signal to be sent 4161 * 4162 * This syscall also checks the @tgid and returns -ESRCH even if the PID 4163 * exists but it's not belonging to the target process anymore. This 4164 * method solves the problem of threads exiting and PIDs getting reused. 4165 */ 4166 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig) 4167 { 4168 /* This is only valid for single tasks */ 4169 if (pid <= 0 || tgid <= 0) 4170 return -EINVAL; 4171 4172 return do_tkill(tgid, pid, sig); 4173 } 4174 4175 /** 4176 * sys_tkill - send signal to one specific task 4177 * @pid: the PID of the task 4178 * @sig: signal to be sent 4179 * 4180 * Send a signal to only one task, even if it's a CLONE_THREAD task. 4181 */ 4182 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig) 4183 { 4184 /* This is only valid for single tasks */ 4185 if (pid <= 0) 4186 return -EINVAL; 4187 4188 return do_tkill(0, pid, sig); 4189 } 4190 4191 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info) 4192 { 4193 /* Not even root can pretend to send signals from the kernel. 4194 * Nor can they impersonate a kill()/tgkill(), which adds source info. 4195 */ 4196 if ((info->si_code >= 0 || info->si_code == SI_TKILL) && 4197 (task_pid_vnr(current) != pid)) 4198 return -EPERM; 4199 4200 /* POSIX.1b doesn't mention process groups. */ 4201 return kill_proc_info(sig, info, pid); 4202 } 4203 4204 /** 4205 * sys_rt_sigqueueinfo - send signal information to a signal 4206 * @pid: the PID of the thread 4207 * @sig: signal to be sent 4208 * @uinfo: signal info to be sent 4209 */ 4210 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig, 4211 siginfo_t __user *, uinfo) 4212 { 4213 kernel_siginfo_t info; 4214 int ret = __copy_siginfo_from_user(sig, &info, uinfo); 4215 if (unlikely(ret)) 4216 return ret; 4217 return do_rt_sigqueueinfo(pid, sig, &info); 4218 } 4219 4220 #ifdef CONFIG_COMPAT 4221 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo, 4222 compat_pid_t, pid, 4223 int, sig, 4224 struct compat_siginfo __user *, uinfo) 4225 { 4226 kernel_siginfo_t info; 4227 int ret = __copy_siginfo_from_user32(sig, &info, uinfo); 4228 if (unlikely(ret)) 4229 return ret; 4230 return do_rt_sigqueueinfo(pid, sig, &info); 4231 } 4232 #endif 4233 4234 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info) 4235 { 4236 /* This is only valid for single tasks */ 4237 if (pid <= 0 || tgid <= 0) 4238 return -EINVAL; 4239 4240 /* Not even root can pretend to send signals from the kernel. 4241 * Nor can they impersonate a kill()/tgkill(), which adds source info. 4242 */ 4243 if ((info->si_code >= 0 || info->si_code == SI_TKILL) && 4244 (task_pid_vnr(current) != pid)) 4245 return -EPERM; 4246 4247 return do_send_specific(tgid, pid, sig, info); 4248 } 4249 4250 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig, 4251 siginfo_t __user *, uinfo) 4252 { 4253 kernel_siginfo_t info; 4254 int ret = __copy_siginfo_from_user(sig, &info, uinfo); 4255 if (unlikely(ret)) 4256 return ret; 4257 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); 4258 } 4259 4260 #ifdef CONFIG_COMPAT 4261 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo, 4262 compat_pid_t, tgid, 4263 compat_pid_t, pid, 4264 int, sig, 4265 struct compat_siginfo __user *, uinfo) 4266 { 4267 kernel_siginfo_t info; 4268 int ret = __copy_siginfo_from_user32(sig, &info, uinfo); 4269 if (unlikely(ret)) 4270 return ret; 4271 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); 4272 } 4273 #endif 4274 4275 /* 4276 * For kthreads only, must not be used if cloned with CLONE_SIGHAND 4277 */ 4278 void kernel_sigaction(int sig, __sighandler_t action) 4279 { 4280 spin_lock_irq(¤t->sighand->siglock); 4281 current->sighand->action[sig - 1].sa.sa_handler = action; 4282 if (action == SIG_IGN) { 4283 sigset_t mask; 4284 4285 sigemptyset(&mask); 4286 sigaddset(&mask, sig); 4287 4288 flush_sigqueue_mask(current, &mask, ¤t->signal->shared_pending); 4289 flush_sigqueue_mask(current, &mask, ¤t->pending); 4290 recalc_sigpending(); 4291 } 4292 spin_unlock_irq(¤t->sighand->siglock); 4293 } 4294 EXPORT_SYMBOL(kernel_sigaction); 4295 4296 void __weak sigaction_compat_abi(struct k_sigaction *act, 4297 struct k_sigaction *oact) 4298 { 4299 } 4300 4301 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact) 4302 { 4303 struct task_struct *p = current, *t; 4304 struct k_sigaction *k; 4305 sigset_t mask; 4306 4307 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig))) 4308 return -EINVAL; 4309 4310 k = &p->sighand->action[sig-1]; 4311 4312 spin_lock_irq(&p->sighand->siglock); 4313 if (k->sa.sa_flags & SA_IMMUTABLE) { 4314 spin_unlock_irq(&p->sighand->siglock); 4315 return -EINVAL; 4316 } 4317 if (oact) 4318 *oact = *k; 4319 4320 /* 4321 * Make sure that we never accidentally claim to support SA_UNSUPPORTED, 4322 * e.g. by having an architecture use the bit in their uapi. 4323 */ 4324 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED); 4325 4326 /* 4327 * Clear unknown flag bits in order to allow userspace to detect missing 4328 * support for flag bits and to allow the kernel to use non-uapi bits 4329 * internally. 4330 */ 4331 if (act) 4332 act->sa.sa_flags &= UAPI_SA_FLAGS; 4333 if (oact) 4334 oact->sa.sa_flags &= UAPI_SA_FLAGS; 4335 4336 sigaction_compat_abi(act, oact); 4337 4338 if (act) { 4339 bool was_ignored = k->sa.sa_handler == SIG_IGN; 4340 4341 sigdelsetmask(&act->sa.sa_mask, 4342 sigmask(SIGKILL) | sigmask(SIGSTOP)); 4343 *k = *act; 4344 /* 4345 * POSIX 3.3.1.3: 4346 * "Setting a signal action to SIG_IGN for a signal that is 4347 * pending shall cause the pending signal to be discarded, 4348 * whether or not it is blocked." 4349 * 4350 * "Setting a signal action to SIG_DFL for a signal that is 4351 * pending and whose default action is to ignore the signal 4352 * (for example, SIGCHLD), shall cause the pending signal to 4353 * be discarded, whether or not it is blocked" 4354 */ 4355 if (sig_handler_ignored(sig_handler(p, sig), sig)) { 4356 sigemptyset(&mask); 4357 sigaddset(&mask, sig); 4358 flush_sigqueue_mask(p, &mask, &p->signal->shared_pending); 4359 for_each_thread(p, t) 4360 flush_sigqueue_mask(p, &mask, &t->pending); 4361 } else if (was_ignored) { 4362 posixtimer_sig_unignore(p, sig); 4363 } 4364 } 4365 4366 spin_unlock_irq(&p->sighand->siglock); 4367 return 0; 4368 } 4369 4370 #ifdef CONFIG_DYNAMIC_SIGFRAME 4371 static inline void sigaltstack_lock(void) 4372 __acquires(¤t->sighand->siglock) 4373 { 4374 spin_lock_irq(¤t->sighand->siglock); 4375 } 4376 4377 static inline void sigaltstack_unlock(void) 4378 __releases(¤t->sighand->siglock) 4379 { 4380 spin_unlock_irq(¤t->sighand->siglock); 4381 } 4382 #else 4383 static inline void sigaltstack_lock(void) { } 4384 static inline void sigaltstack_unlock(void) { } 4385 #endif 4386 4387 static int 4388 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp, 4389 size_t min_ss_size) 4390 { 4391 struct task_struct *t = current; 4392 int ret = 0; 4393 4394 if (oss) { 4395 memset(oss, 0, sizeof(stack_t)); 4396 oss->ss_sp = (void __user *) t->sas_ss_sp; 4397 oss->ss_size = t->sas_ss_size; 4398 oss->ss_flags = sas_ss_flags(sp) | 4399 (current->sas_ss_flags & SS_FLAG_BITS); 4400 } 4401 4402 if (ss) { 4403 void __user *ss_sp = ss->ss_sp; 4404 size_t ss_size = ss->ss_size; 4405 unsigned ss_flags = ss->ss_flags; 4406 int ss_mode; 4407 4408 if (unlikely(on_sig_stack(sp))) 4409 return -EPERM; 4410 4411 ss_mode = ss_flags & ~SS_FLAG_BITS; 4412 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK && 4413 ss_mode != 0)) 4414 return -EINVAL; 4415 4416 /* 4417 * Return before taking any locks if no actual 4418 * sigaltstack changes were requested. 4419 */ 4420 if (t->sas_ss_sp == (unsigned long)ss_sp && 4421 t->sas_ss_size == ss_size && 4422 t->sas_ss_flags == ss_flags) 4423 return 0; 4424 4425 sigaltstack_lock(); 4426 if (ss_mode == SS_DISABLE) { 4427 ss_size = 0; 4428 ss_sp = NULL; 4429 } else { 4430 if (unlikely(ss_size < min_ss_size)) 4431 ret = -ENOMEM; 4432 if (!sigaltstack_size_valid(ss_size)) 4433 ret = -ENOMEM; 4434 } 4435 if (!ret) { 4436 t->sas_ss_sp = (unsigned long) ss_sp; 4437 t->sas_ss_size = ss_size; 4438 t->sas_ss_flags = ss_flags; 4439 } 4440 sigaltstack_unlock(); 4441 } 4442 return ret; 4443 } 4444 4445 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss) 4446 { 4447 stack_t new, old; 4448 int err; 4449 if (uss && copy_from_user(&new, uss, sizeof(stack_t))) 4450 return -EFAULT; 4451 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL, 4452 current_user_stack_pointer(), 4453 MINSIGSTKSZ); 4454 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t))) 4455 err = -EFAULT; 4456 return err; 4457 } 4458 4459 int restore_altstack(const stack_t __user *uss) 4460 { 4461 stack_t new; 4462 if (copy_from_user(&new, uss, sizeof(stack_t))) 4463 return -EFAULT; 4464 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(), 4465 MINSIGSTKSZ); 4466 /* squash all but EFAULT for now */ 4467 return 0; 4468 } 4469 4470 int __save_altstack(stack_t __user *uss, unsigned long sp) 4471 { 4472 struct task_struct *t = current; 4473 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) | 4474 __put_user(t->sas_ss_flags, &uss->ss_flags) | 4475 __put_user(t->sas_ss_size, &uss->ss_size); 4476 return err; 4477 } 4478 4479 #ifdef CONFIG_COMPAT 4480 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr, 4481 compat_stack_t __user *uoss_ptr) 4482 { 4483 stack_t uss, uoss; 4484 int ret; 4485 4486 if (uss_ptr) { 4487 compat_stack_t uss32; 4488 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t))) 4489 return -EFAULT; 4490 uss.ss_sp = compat_ptr(uss32.ss_sp); 4491 uss.ss_flags = uss32.ss_flags; 4492 uss.ss_size = uss32.ss_size; 4493 } 4494 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss, 4495 compat_user_stack_pointer(), 4496 COMPAT_MINSIGSTKSZ); 4497 if (ret >= 0 && uoss_ptr) { 4498 compat_stack_t old; 4499 memset(&old, 0, sizeof(old)); 4500 old.ss_sp = ptr_to_compat(uoss.ss_sp); 4501 old.ss_flags = uoss.ss_flags; 4502 old.ss_size = uoss.ss_size; 4503 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t))) 4504 ret = -EFAULT; 4505 } 4506 return ret; 4507 } 4508 4509 COMPAT_SYSCALL_DEFINE2(sigaltstack, 4510 const compat_stack_t __user *, uss_ptr, 4511 compat_stack_t __user *, uoss_ptr) 4512 { 4513 return do_compat_sigaltstack(uss_ptr, uoss_ptr); 4514 } 4515 4516 int compat_restore_altstack(const compat_stack_t __user *uss) 4517 { 4518 int err = do_compat_sigaltstack(uss, NULL); 4519 /* squash all but -EFAULT for now */ 4520 return err == -EFAULT ? err : 0; 4521 } 4522 4523 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp) 4524 { 4525 int err; 4526 struct task_struct *t = current; 4527 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp), 4528 &uss->ss_sp) | 4529 __put_user(t->sas_ss_flags, &uss->ss_flags) | 4530 __put_user(t->sas_ss_size, &uss->ss_size); 4531 return err; 4532 } 4533 #endif 4534 4535 #ifdef __ARCH_WANT_SYS_SIGPENDING 4536 4537 /** 4538 * sys_sigpending - examine pending signals 4539 * @uset: where mask of pending signal is returned 4540 */ 4541 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset) 4542 { 4543 sigset_t set; 4544 4545 if (sizeof(old_sigset_t) > sizeof(*uset)) 4546 return -EINVAL; 4547 4548 do_sigpending(&set); 4549 4550 if (copy_to_user(uset, &set, sizeof(old_sigset_t))) 4551 return -EFAULT; 4552 4553 return 0; 4554 } 4555 4556 #ifdef CONFIG_COMPAT 4557 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32) 4558 { 4559 sigset_t set; 4560 4561 do_sigpending(&set); 4562 4563 return put_user(set.sig[0], set32); 4564 } 4565 #endif 4566 4567 #endif 4568 4569 #ifdef __ARCH_WANT_SYS_SIGPROCMASK 4570 /** 4571 * sys_sigprocmask - examine and change blocked signals 4572 * @how: whether to add, remove, or set signals 4573 * @nset: signals to add or remove (if non-null) 4574 * @oset: previous value of signal mask if non-null 4575 * 4576 * Some platforms have their own version with special arguments; 4577 * others support only sys_rt_sigprocmask. 4578 */ 4579 4580 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset, 4581 old_sigset_t __user *, oset) 4582 { 4583 old_sigset_t old_set, new_set; 4584 sigset_t new_blocked; 4585 4586 old_set = current->blocked.sig[0]; 4587 4588 if (nset) { 4589 if (copy_from_user(&new_set, nset, sizeof(*nset))) 4590 return -EFAULT; 4591 4592 new_blocked = current->blocked; 4593 4594 switch (how) { 4595 case SIG_BLOCK: 4596 sigaddsetmask(&new_blocked, new_set); 4597 break; 4598 case SIG_UNBLOCK: 4599 sigdelsetmask(&new_blocked, new_set); 4600 break; 4601 case SIG_SETMASK: 4602 new_blocked.sig[0] = new_set; 4603 break; 4604 default: 4605 return -EINVAL; 4606 } 4607 4608 set_current_blocked(&new_blocked); 4609 } 4610 4611 if (oset) { 4612 if (copy_to_user(oset, &old_set, sizeof(*oset))) 4613 return -EFAULT; 4614 } 4615 4616 return 0; 4617 } 4618 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */ 4619 4620 #ifndef CONFIG_ODD_RT_SIGACTION 4621 /** 4622 * sys_rt_sigaction - alter an action taken by a process 4623 * @sig: signal to be sent 4624 * @act: new sigaction 4625 * @oact: used to save the previous sigaction 4626 * @sigsetsize: size of sigset_t type 4627 */ 4628 SYSCALL_DEFINE4(rt_sigaction, int, sig, 4629 const struct sigaction __user *, act, 4630 struct sigaction __user *, oact, 4631 size_t, sigsetsize) 4632 { 4633 struct k_sigaction new_sa, old_sa; 4634 int ret; 4635 4636 /* XXX: Don't preclude handling different sized sigset_t's. */ 4637 if (sigsetsize != sizeof(sigset_t)) 4638 return -EINVAL; 4639 4640 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa))) 4641 return -EFAULT; 4642 4643 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL); 4644 if (ret) 4645 return ret; 4646 4647 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa))) 4648 return -EFAULT; 4649 4650 return 0; 4651 } 4652 #ifdef CONFIG_COMPAT 4653 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig, 4654 const struct compat_sigaction __user *, act, 4655 struct compat_sigaction __user *, oact, 4656 compat_size_t, sigsetsize) 4657 { 4658 struct k_sigaction new_ka, old_ka; 4659 #ifdef __ARCH_HAS_SA_RESTORER 4660 compat_uptr_t restorer; 4661 #endif 4662 int ret; 4663 4664 /* XXX: Don't preclude handling different sized sigset_t's. */ 4665 if (sigsetsize != sizeof(compat_sigset_t)) 4666 return -EINVAL; 4667 4668 if (act) { 4669 compat_uptr_t handler; 4670 ret = get_user(handler, &act->sa_handler); 4671 new_ka.sa.sa_handler = compat_ptr(handler); 4672 #ifdef __ARCH_HAS_SA_RESTORER 4673 ret |= get_user(restorer, &act->sa_restorer); 4674 new_ka.sa.sa_restorer = compat_ptr(restorer); 4675 #endif 4676 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask); 4677 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags); 4678 if (ret) 4679 return -EFAULT; 4680 } 4681 4682 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4683 if (!ret && oact) { 4684 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler), 4685 &oact->sa_handler); 4686 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask, 4687 sizeof(oact->sa_mask)); 4688 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags); 4689 #ifdef __ARCH_HAS_SA_RESTORER 4690 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer), 4691 &oact->sa_restorer); 4692 #endif 4693 } 4694 return ret; 4695 } 4696 #endif 4697 #endif /* !CONFIG_ODD_RT_SIGACTION */ 4698 4699 #ifdef CONFIG_OLD_SIGACTION 4700 SYSCALL_DEFINE3(sigaction, int, sig, 4701 const struct old_sigaction __user *, act, 4702 struct old_sigaction __user *, oact) 4703 { 4704 struct k_sigaction new_ka, old_ka; 4705 int ret; 4706 4707 if (act) { 4708 old_sigset_t mask; 4709 if (!access_ok(act, sizeof(*act)) || 4710 __get_user(new_ka.sa.sa_handler, &act->sa_handler) || 4711 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) || 4712 __get_user(new_ka.sa.sa_flags, &act->sa_flags) || 4713 __get_user(mask, &act->sa_mask)) 4714 return -EFAULT; 4715 #ifdef __ARCH_HAS_KA_RESTORER 4716 new_ka.ka_restorer = NULL; 4717 #endif 4718 siginitset(&new_ka.sa.sa_mask, mask); 4719 } 4720 4721 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4722 4723 if (!ret && oact) { 4724 if (!access_ok(oact, sizeof(*oact)) || 4725 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) || 4726 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) || 4727 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) || 4728 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask)) 4729 return -EFAULT; 4730 } 4731 4732 return ret; 4733 } 4734 #endif 4735 #ifdef CONFIG_COMPAT_OLD_SIGACTION 4736 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig, 4737 const struct compat_old_sigaction __user *, act, 4738 struct compat_old_sigaction __user *, oact) 4739 { 4740 struct k_sigaction new_ka, old_ka; 4741 int ret; 4742 compat_old_sigset_t mask; 4743 compat_uptr_t handler, restorer; 4744 4745 if (act) { 4746 if (!access_ok(act, sizeof(*act)) || 4747 __get_user(handler, &act->sa_handler) || 4748 __get_user(restorer, &act->sa_restorer) || 4749 __get_user(new_ka.sa.sa_flags, &act->sa_flags) || 4750 __get_user(mask, &act->sa_mask)) 4751 return -EFAULT; 4752 4753 #ifdef __ARCH_HAS_KA_RESTORER 4754 new_ka.ka_restorer = NULL; 4755 #endif 4756 new_ka.sa.sa_handler = compat_ptr(handler); 4757 new_ka.sa.sa_restorer = compat_ptr(restorer); 4758 siginitset(&new_ka.sa.sa_mask, mask); 4759 } 4760 4761 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL); 4762 4763 if (!ret && oact) { 4764 if (!access_ok(oact, sizeof(*oact)) || 4765 __put_user(ptr_to_compat(old_ka.sa.sa_handler), 4766 &oact->sa_handler) || 4767 __put_user(ptr_to_compat(old_ka.sa.sa_restorer), 4768 &oact->sa_restorer) || 4769 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) || 4770 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask)) 4771 return -EFAULT; 4772 } 4773 return ret; 4774 } 4775 #endif 4776 4777 #ifdef CONFIG_SGETMASK_SYSCALL 4778 4779 /* 4780 * For backwards compatibility. Functionality superseded by sigprocmask. 4781 */ 4782 SYSCALL_DEFINE0(sgetmask) 4783 { 4784 /* SMP safe */ 4785 return current->blocked.sig[0]; 4786 } 4787 4788 SYSCALL_DEFINE1(ssetmask, int, newmask) 4789 { 4790 int old = current->blocked.sig[0]; 4791 sigset_t newset; 4792 4793 siginitset(&newset, newmask); 4794 set_current_blocked(&newset); 4795 4796 return old; 4797 } 4798 #endif /* CONFIG_SGETMASK_SYSCALL */ 4799 4800 #ifdef __ARCH_WANT_SYS_SIGNAL 4801 /* 4802 * For backwards compatibility. Functionality superseded by sigaction. 4803 */ 4804 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler) 4805 { 4806 struct k_sigaction new_sa, old_sa; 4807 int ret; 4808 4809 new_sa.sa.sa_handler = handler; 4810 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK; 4811 sigemptyset(&new_sa.sa.sa_mask); 4812 4813 ret = do_sigaction(sig, &new_sa, &old_sa); 4814 4815 return ret ? ret : (unsigned long)old_sa.sa.sa_handler; 4816 } 4817 #endif /* __ARCH_WANT_SYS_SIGNAL */ 4818 4819 #ifdef __ARCH_WANT_SYS_PAUSE 4820 4821 SYSCALL_DEFINE0(pause) 4822 { 4823 while (!signal_pending(current)) { 4824 __set_current_state(TASK_INTERRUPTIBLE); 4825 schedule(); 4826 } 4827 return -ERESTARTNOHAND; 4828 } 4829 4830 #endif 4831 4832 static int sigsuspend(sigset_t *set) 4833 { 4834 current->saved_sigmask = current->blocked; 4835 set_current_blocked(set); 4836 4837 while (!signal_pending(current)) { 4838 __set_current_state(TASK_INTERRUPTIBLE); 4839 schedule(); 4840 } 4841 set_restore_sigmask(); 4842 return -ERESTARTNOHAND; 4843 } 4844 4845 /** 4846 * sys_rt_sigsuspend - replace the signal mask for a value with the 4847 * @unewset value until a signal is received 4848 * @unewset: new signal mask value 4849 * @sigsetsize: size of sigset_t type 4850 */ 4851 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize) 4852 { 4853 sigset_t newset; 4854 4855 /* XXX: Don't preclude handling different sized sigset_t's. */ 4856 if (sigsetsize != sizeof(sigset_t)) 4857 return -EINVAL; 4858 4859 if (copy_from_user(&newset, unewset, sizeof(newset))) 4860 return -EFAULT; 4861 return sigsuspend(&newset); 4862 } 4863 4864 #ifdef CONFIG_COMPAT 4865 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize) 4866 { 4867 sigset_t newset; 4868 4869 /* XXX: Don't preclude handling different sized sigset_t's. */ 4870 if (sigsetsize != sizeof(sigset_t)) 4871 return -EINVAL; 4872 4873 if (get_compat_sigset(&newset, unewset)) 4874 return -EFAULT; 4875 return sigsuspend(&newset); 4876 } 4877 #endif 4878 4879 #ifdef CONFIG_OLD_SIGSUSPEND 4880 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask) 4881 { 4882 sigset_t blocked; 4883 siginitset(&blocked, mask); 4884 return sigsuspend(&blocked); 4885 } 4886 #endif 4887 #ifdef CONFIG_OLD_SIGSUSPEND3 4888 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask) 4889 { 4890 sigset_t blocked; 4891 siginitset(&blocked, mask); 4892 return sigsuspend(&blocked); 4893 } 4894 #endif 4895 4896 __weak const char *arch_vma_name(struct vm_area_struct *vma) 4897 { 4898 return NULL; 4899 } 4900 4901 static inline void siginfo_buildtime_checks(void) 4902 { 4903 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE); 4904 4905 /* Verify the offsets in the two siginfos match */ 4906 #define CHECK_OFFSET(field) \ 4907 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field)) 4908 4909 /* kill */ 4910 CHECK_OFFSET(si_pid); 4911 CHECK_OFFSET(si_uid); 4912 4913 /* timer */ 4914 CHECK_OFFSET(si_tid); 4915 CHECK_OFFSET(si_overrun); 4916 CHECK_OFFSET(si_value); 4917 4918 /* rt */ 4919 CHECK_OFFSET(si_pid); 4920 CHECK_OFFSET(si_uid); 4921 CHECK_OFFSET(si_value); 4922 4923 /* sigchld */ 4924 CHECK_OFFSET(si_pid); 4925 CHECK_OFFSET(si_uid); 4926 CHECK_OFFSET(si_status); 4927 CHECK_OFFSET(si_utime); 4928 CHECK_OFFSET(si_stime); 4929 4930 /* sigfault */ 4931 CHECK_OFFSET(si_addr); 4932 CHECK_OFFSET(si_trapno); 4933 CHECK_OFFSET(si_addr_lsb); 4934 CHECK_OFFSET(si_lower); 4935 CHECK_OFFSET(si_upper); 4936 CHECK_OFFSET(si_pkey); 4937 CHECK_OFFSET(si_perf_data); 4938 CHECK_OFFSET(si_perf_type); 4939 CHECK_OFFSET(si_perf_flags); 4940 4941 /* sigpoll */ 4942 CHECK_OFFSET(si_band); 4943 CHECK_OFFSET(si_fd); 4944 4945 /* sigsys */ 4946 CHECK_OFFSET(si_call_addr); 4947 CHECK_OFFSET(si_syscall); 4948 CHECK_OFFSET(si_arch); 4949 #undef CHECK_OFFSET 4950 4951 /* usb asyncio */ 4952 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) != 4953 offsetof(struct siginfo, si_addr)); 4954 if (sizeof(int) == sizeof(void __user *)) { 4955 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) != 4956 sizeof(void __user *)); 4957 } else { 4958 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) + 4959 sizeof_field(struct siginfo, si_uid)) != 4960 sizeof(void __user *)); 4961 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) != 4962 offsetof(struct siginfo, si_uid)); 4963 } 4964 #ifdef CONFIG_COMPAT 4965 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) != 4966 offsetof(struct compat_siginfo, si_addr)); 4967 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) != 4968 sizeof(compat_uptr_t)); 4969 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) != 4970 sizeof_field(struct siginfo, si_pid)); 4971 #endif 4972 } 4973 4974 #if defined(CONFIG_SYSCTL) 4975 static const struct ctl_table signal_debug_table[] = { 4976 #ifdef CONFIG_SYSCTL_EXCEPTION_TRACE 4977 { 4978 .procname = "exception-trace", 4979 .data = &show_unhandled_signals, 4980 .maxlen = sizeof(int), 4981 .mode = 0644, 4982 .proc_handler = proc_dointvec 4983 }, 4984 #endif 4985 }; 4986 4987 static const struct ctl_table signal_table[] = { 4988 { 4989 .procname = "print-fatal-signals", 4990 .data = &print_fatal_signals, 4991 .maxlen = sizeof(int), 4992 .mode = 0644, 4993 .proc_handler = proc_dointvec, 4994 }, 4995 }; 4996 4997 static int __init init_signal_sysctls(void) 4998 { 4999 register_sysctl_init("debug", signal_debug_table); 5000 register_sysctl_init("kernel", signal_table); 5001 return 0; 5002 } 5003 early_initcall(init_signal_sysctls); 5004 #endif /* CONFIG_SYSCTL */ 5005 5006 void __init signals_init(void) 5007 { 5008 siginfo_buildtime_checks(); 5009 5010 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT); 5011 } 5012 5013 #ifdef CONFIG_KGDB_KDB 5014 #include <linux/kdb.h> 5015 /* 5016 * kdb_send_sig - Allows kdb to send signals without exposing 5017 * signal internals. This function checks if the required locks are 5018 * available before calling the main signal code, to avoid kdb 5019 * deadlocks. 5020 */ 5021 void kdb_send_sig(struct task_struct *t, int sig) 5022 { 5023 static struct task_struct *kdb_prev_t; 5024 int new_t, ret; 5025 if (!spin_trylock(&t->sighand->siglock)) { 5026 kdb_printf("Can't do kill command now.\n" 5027 "The sigmask lock is held somewhere else in " 5028 "kernel, try again later\n"); 5029 return; 5030 } 5031 new_t = kdb_prev_t != t; 5032 kdb_prev_t = t; 5033 if (!task_is_running(t) && new_t) { 5034 spin_unlock(&t->sighand->siglock); 5035 kdb_printf("Process is not RUNNING, sending a signal from " 5036 "kdb risks deadlock\n" 5037 "on the run queue locks. " 5038 "The signal has _not_ been sent.\n" 5039 "Reissue the kill command if you want to risk " 5040 "the deadlock.\n"); 5041 return; 5042 } 5043 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID); 5044 spin_unlock(&t->sighand->siglock); 5045 if (ret) 5046 kdb_printf("Fail to deliver Signal %d to process %d.\n", 5047 sig, t->pid); 5048 else 5049 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid); 5050 } 5051 #endif /* CONFIG_KGDB_KDB */ 5052