linux/kernel/signal.c
<<
>>
Prefs
   1/*
   2 *  linux/kernel/signal.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
   7 *
   8 *  2003-06-02  Jim Houston - Concurrent Computer Corp.
   9 *              Changes to use preallocated sigqueue structures
  10 *              to allow signals to be sent reliably.
  11 */
  12
  13#include <linux/slab.h>
  14#include <linux/export.h>
  15#include <linux/init.h>
  16#include <linux/sched.h>
  17#include <linux/fs.h>
  18#include <linux/tty.h>
  19#include <linux/binfmts.h>
  20#include <linux/security.h>
  21#include <linux/syscalls.h>
  22#include <linux/ptrace.h>
  23#include <linux/signal.h>
  24#include <linux/signalfd.h>
  25#include <linux/ratelimit.h>
  26#include <linux/tracehook.h>
  27#include <linux/capability.h>
  28#include <linux/freezer.h>
  29#include <linux/pid_namespace.h>
  30#include <linux/nsproxy.h>
  31#include <linux/user_namespace.h>
  32#define CREATE_TRACE_POINTS
  33#include <trace/events/signal.h>
  34
  35#include <asm/param.h>
  36#include <asm/uaccess.h>
  37#include <asm/unistd.h>
  38#include <asm/siginfo.h>
  39#include <asm/cacheflush.h>
  40#include "audit.h"      /* audit_signal_info() */
  41
  42/*
  43 * SLAB caches for signal bits.
  44 */
  45
  46static struct kmem_cache *sigqueue_cachep;
  47
  48int print_fatal_signals __read_mostly;
  49
  50static void __user *sig_handler(struct task_struct *t, int sig)
  51{
  52        return t->sighand->action[sig - 1].sa.sa_handler;
  53}
  54
  55static int sig_handler_ignored(void __user *handler, int sig)
  56{
  57        /* Is it explicitly or implicitly ignored? */
  58        return handler == SIG_IGN ||
  59                (handler == SIG_DFL && sig_kernel_ignore(sig));
  60}
  61
  62static int sig_task_ignored(struct task_struct *t, int sig, bool force)
  63{
  64        void __user *handler;
  65
  66        handler = sig_handler(t, sig);
  67
  68        if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
  69                        handler == SIG_DFL && !force)
  70                return 1;
  71
  72        return sig_handler_ignored(handler, sig);
  73}
  74
  75static int sig_ignored(struct task_struct *t, int sig, bool force)
  76{
  77        /*
  78         * Blocked signals are never ignored, since the
  79         * signal handler may change by the time it is
  80         * unblocked.
  81         */
  82        if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
  83                return 0;
  84
  85        if (!sig_task_ignored(t, sig, force))
  86                return 0;
  87
  88        /*
  89         * Tracers may want to know about even ignored signals.
  90         */
  91        return !t->ptrace;
  92}
  93
  94/*
  95 * Re-calculate pending state from the set of locally pending
  96 * signals, globally pending signals, and blocked signals.
  97 */
  98static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
  99{
 100        unsigned long ready;
 101        long i;
 102
 103        switch (_NSIG_WORDS) {
 104        default:
 105                for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
 106                        ready |= signal->sig[i] &~ blocked->sig[i];
 107                break;
 108
 109        case 4: ready  = signal->sig[3] &~ blocked->sig[3];
 110                ready |= signal->sig[2] &~ blocked->sig[2];
 111                ready |= signal->sig[1] &~ blocked->sig[1];
 112                ready |= signal->sig[0] &~ blocked->sig[0];
 113                break;
 114
 115        case 2: ready  = signal->sig[1] &~ blocked->sig[1];
 116                ready |= signal->sig[0] &~ blocked->sig[0];
 117                break;
 118
 119        case 1: ready  = signal->sig[0] &~ blocked->sig[0];
 120        }
 121        return ready != 0;
 122}
 123
 124#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
 125
 126static int recalc_sigpending_tsk(struct task_struct *t)
 127{
 128        if ((t->jobctl & JOBCTL_PENDING_MASK) ||
 129            PENDING(&t->pending, &t->blocked) ||
 130            PENDING(&t->signal->shared_pending, &t->blocked)) {
 131                set_tsk_thread_flag(t, TIF_SIGPENDING);
 132                return 1;
 133        }
 134        /*
 135         * We must never clear the flag in another thread, or in current
 136         * when it's possible the current syscall is returning -ERESTART*.
 137         * So we don't clear it here, and only callers who know they should do.
 138         */
 139        return 0;
 140}
 141
 142/*
 143 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
 144 * This is superfluous when called on current, the wakeup is a harmless no-op.
 145 */
 146void recalc_sigpending_and_wake(struct task_struct *t)
 147{
 148        if (recalc_sigpending_tsk(t))
 149                signal_wake_up(t, 0);
 150}
 151
 152void recalc_sigpending(void)
 153{
 154        if (!recalc_sigpending_tsk(current) && !freezing(current))
 155                clear_thread_flag(TIF_SIGPENDING);
 156
 157}
 158
 159/* Given the mask, find the first available signal that should be serviced. */
 160
 161#define SYNCHRONOUS_MASK \
 162        (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
 163         sigmask(SIGTRAP) | sigmask(SIGFPE))
 164
 165int next_signal(struct sigpending *pending, sigset_t *mask)
 166{
 167        unsigned long i, *s, *m, x;
 168        int sig = 0;
 169
 170        s = pending->signal.sig;
 171        m = mask->sig;
 172
 173        /*
 174         * Handle the first word specially: it contains the
 175         * synchronous signals that need to be dequeued first.
 176         */
 177        x = *s &~ *m;
 178        if (x) {
 179                if (x & SYNCHRONOUS_MASK)
 180                        x &= SYNCHRONOUS_MASK;
 181                sig = ffz(~x) + 1;
 182                return sig;
 183        }
 184
 185        switch (_NSIG_WORDS) {
 186        default:
 187                for (i = 1; i < _NSIG_WORDS; ++i) {
 188                        x = *++s &~ *++m;
 189                        if (!x)
 190                                continue;
 191                        sig = ffz(~x) + i*_NSIG_BPW + 1;
 192                        break;
 193                }
 194                break;
 195
 196        case 2:
 197                x = s[1] &~ m[1];
 198                if (!x)
 199                        break;
 200                sig = ffz(~x) + _NSIG_BPW + 1;
 201                break;
 202
 203        case 1:
 204                /* Nothing to do */
 205                break;
 206        }
 207
 208        return sig;
 209}
 210
 211static inline void print_dropped_signal(int sig)
 212{
 213        static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
 214
 215        if (!print_fatal_signals)
 216                return;
 217
 218        if (!__ratelimit(&ratelimit_state))
 219                return;
 220
 221        printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
 222                                current->comm, current->pid, sig);
 223}
 224
 225/**
 226 * task_set_jobctl_pending - set jobctl pending bits
 227 * @task: target task
 228 * @mask: pending bits to set
 229 *
 230 * Clear @mask from @task->jobctl.  @mask must be subset of
 231 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
 232 * %JOBCTL_TRAPPING.  If stop signo is being set, the existing signo is
 233 * cleared.  If @task is already being killed or exiting, this function
 234 * becomes noop.
 235 *
 236 * CONTEXT:
 237 * Must be called with @task->sighand->siglock held.
 238 *
 239 * RETURNS:
 240 * %true if @mask is set, %false if made noop because @task was dying.
 241 */
 242bool task_set_jobctl_pending(struct task_struct *task, unsigned int mask)
 243{
 244        BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
 245                        JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
 246        BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
 247
 248        if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
 249                return false;
 250
 251        if (mask & JOBCTL_STOP_SIGMASK)
 252                task->jobctl &= ~JOBCTL_STOP_SIGMASK;
 253
 254        task->jobctl |= mask;
 255        return true;
 256}
 257
 258/**
 259 * task_clear_jobctl_trapping - clear jobctl trapping bit
 260 * @task: target task
 261 *
 262 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
 263 * Clear it and wake up the ptracer.  Note that we don't need any further
 264 * locking.  @task->siglock guarantees that @task->parent points to the
 265 * ptracer.
 266 *
 267 * CONTEXT:
 268 * Must be called with @task->sighand->siglock held.
 269 */
 270void task_clear_jobctl_trapping(struct task_struct *task)
 271{
 272        if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
 273                task->jobctl &= ~JOBCTL_TRAPPING;
 274                wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
 275        }
 276}
 277
 278/**
 279 * task_clear_jobctl_pending - clear jobctl pending bits
 280 * @task: target task
 281 * @mask: pending bits to clear
 282 *
 283 * Clear @mask from @task->jobctl.  @mask must be subset of
 284 * %JOBCTL_PENDING_MASK.  If %JOBCTL_STOP_PENDING is being cleared, other
 285 * STOP bits are cleared together.
 286 *
 287 * If clearing of @mask leaves no stop or trap pending, this function calls
 288 * task_clear_jobctl_trapping().
 289 *
 290 * CONTEXT:
 291 * Must be called with @task->sighand->siglock held.
 292 */
 293void task_clear_jobctl_pending(struct task_struct *task, unsigned int mask)
 294{
 295        BUG_ON(mask & ~JOBCTL_PENDING_MASK);
 296
 297        if (mask & JOBCTL_STOP_PENDING)
 298                mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
 299
 300        task->jobctl &= ~mask;
 301
 302        if (!(task->jobctl & JOBCTL_PENDING_MASK))
 303                task_clear_jobctl_trapping(task);
 304}
 305
 306/**
 307 * task_participate_group_stop - participate in a group stop
 308 * @task: task participating in a group stop
 309 *
 310 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
 311 * Group stop states are cleared and the group stop count is consumed if
 312 * %JOBCTL_STOP_CONSUME was set.  If the consumption completes the group
 313 * stop, the appropriate %SIGNAL_* flags are set.
 314 *
 315 * CONTEXT:
 316 * Must be called with @task->sighand->siglock held.
 317 *
 318 * RETURNS:
 319 * %true if group stop completion should be notified to the parent, %false
 320 * otherwise.
 321 */
 322static bool task_participate_group_stop(struct task_struct *task)
 323{
 324        struct signal_struct *sig = task->signal;
 325        bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
 326
 327        WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
 328
 329        task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
 330
 331        if (!consume)
 332                return false;
 333
 334        if (!WARN_ON_ONCE(sig->group_stop_count == 0))
 335                sig->group_stop_count--;
 336
 337        /*
 338         * Tell the caller to notify completion iff we are entering into a
 339         * fresh group stop.  Read comment in do_signal_stop() for details.
 340         */
 341        if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
 342                sig->flags = SIGNAL_STOP_STOPPED;
 343                return true;
 344        }
 345        return false;
 346}
 347
 348/*
 349 * allocate a new signal queue record
 350 * - this may be called without locks if and only if t == current, otherwise an
 351 *   appropriate lock must be held to stop the target task from exiting
 352 */
 353static struct sigqueue *
 354__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
 355{
 356        struct sigqueue *q = NULL;
 357        struct user_struct *user;
 358
 359        /*
 360         * Protect access to @t credentials. This can go away when all
 361         * callers hold rcu read lock.
 362         */
 363        rcu_read_lock();
 364        user = get_uid(__task_cred(t)->user);
 365        atomic_inc(&user->sigpending);
 366        rcu_read_unlock();
 367
 368        if (override_rlimit ||
 369            atomic_read(&user->sigpending) <=
 370                        task_rlimit(t, RLIMIT_SIGPENDING)) {
 371                q = kmem_cache_alloc(sigqueue_cachep, flags);
 372        } else {
 373                print_dropped_signal(sig);
 374        }
 375
 376        if (unlikely(q == NULL)) {
 377                atomic_dec(&user->sigpending);
 378                free_uid(user);
 379        } else {
 380                INIT_LIST_HEAD(&q->list);
 381                q->flags = 0;
 382                q->user = user;
 383        }
 384
 385        return q;
 386}
 387
 388static void __sigqueue_free(struct sigqueue *q)
 389{
 390        if (q->flags & SIGQUEUE_PREALLOC)
 391                return;
 392        atomic_dec(&q->user->sigpending);
 393        free_uid(q->user);
 394        kmem_cache_free(sigqueue_cachep, q);
 395}
 396
 397void flush_sigqueue(struct sigpending *queue)
 398{
 399        struct sigqueue *q;
 400
 401        sigemptyset(&queue->signal);
 402        while (!list_empty(&queue->list)) {
 403                q = list_entry(queue->list.next, struct sigqueue , list);
 404                list_del_init(&q->list);
 405                __sigqueue_free(q);
 406        }
 407}
 408
 409/*
 410 * Flush all pending signals for a task.
 411 */
 412void __flush_signals(struct task_struct *t)
 413{
 414        clear_tsk_thread_flag(t, TIF_SIGPENDING);
 415        flush_sigqueue(&t->pending);
 416        flush_sigqueue(&t->signal->shared_pending);
 417}
 418
 419void flush_signals(struct task_struct *t)
 420{
 421        unsigned long flags;
 422
 423        spin_lock_irqsave(&t->sighand->siglock, flags);
 424        __flush_signals(t);
 425        spin_unlock_irqrestore(&t->sighand->siglock, flags);
 426}
 427
 428static void __flush_itimer_signals(struct sigpending *pending)
 429{
 430        sigset_t signal, retain;
 431        struct sigqueue *q, *n;
 432
 433        signal = pending->signal;
 434        sigemptyset(&retain);
 435
 436        list_for_each_entry_safe(q, n, &pending->list, list) {
 437                int sig = q->info.si_signo;
 438
 439                if (likely(q->info.si_code != SI_TIMER)) {
 440                        sigaddset(&retain, sig);
 441                } else {
 442                        sigdelset(&signal, sig);
 443                        list_del_init(&q->list);
 444                        __sigqueue_free(q);
 445                }
 446        }
 447
 448        sigorsets(&pending->signal, &signal, &retain);
 449}
 450
 451void flush_itimer_signals(void)
 452{
 453        struct task_struct *tsk = current;
 454        unsigned long flags;
 455
 456        spin_lock_irqsave(&tsk->sighand->siglock, flags);
 457        __flush_itimer_signals(&tsk->pending);
 458        __flush_itimer_signals(&tsk->signal->shared_pending);
 459        spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 460}
 461
 462void ignore_signals(struct task_struct *t)
 463{
 464        int i;
 465
 466        for (i = 0; i < _NSIG; ++i)
 467                t->sighand->action[i].sa.sa_handler = SIG_IGN;
 468
 469        flush_signals(t);
 470}
 471
 472/*
 473 * Flush all handlers for a task.
 474 */
 475
 476void
 477flush_signal_handlers(struct task_struct *t, int force_default)
 478{
 479        int i;
 480        struct k_sigaction *ka = &t->sighand->action[0];
 481        for (i = _NSIG ; i != 0 ; i--) {
 482                if (force_default || ka->sa.sa_handler != SIG_IGN)
 483                        ka->sa.sa_handler = SIG_DFL;
 484                ka->sa.sa_flags = 0;
 485                sigemptyset(&ka->sa.sa_mask);
 486                ka++;
 487        }
 488}
 489
 490int unhandled_signal(struct task_struct *tsk, int sig)
 491{
 492        void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
 493        if (is_global_init(tsk))
 494                return 1;
 495        if (handler != SIG_IGN && handler != SIG_DFL)
 496                return 0;
 497        /* if ptraced, let the tracer determine */
 498        return !tsk->ptrace;
 499}
 500
 501/*
 502 * Notify the system that a driver wants to block all signals for this
 503 * process, and wants to be notified if any signals at all were to be
 504 * sent/acted upon.  If the notifier routine returns non-zero, then the
 505 * signal will be acted upon after all.  If the notifier routine returns 0,
 506 * then then signal will be blocked.  Only one block per process is
 507 * allowed.  priv is a pointer to private data that the notifier routine
 508 * can use to determine if the signal should be blocked or not.
 509 */
 510void
 511block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
 512{
 513        unsigned long flags;
 514
 515        spin_lock_irqsave(&current->sighand->siglock, flags);
 516        current->notifier_mask = mask;
 517        current->notifier_data = priv;
 518        current->notifier = notifier;
 519        spin_unlock_irqrestore(&current->sighand->siglock, flags);
 520}
 521
 522/* Notify the system that blocking has ended. */
 523
 524void
 525unblock_all_signals(void)
 526{
 527        unsigned long flags;
 528
 529        spin_lock_irqsave(&current->sighand->siglock, flags);
 530        current->notifier = NULL;
 531        current->notifier_data = NULL;
 532        recalc_sigpending();
 533        spin_unlock_irqrestore(&current->sighand->siglock, flags);
 534}
 535
 536static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
 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) {
 555still_pending:
 556                list_del_init(&first->list);
 557                copy_siginfo(info, &first->info);
 558                __sigqueue_free(first);
 559        } else {
 560                /*
 561                 * Ok, it wasn't in the queue.  This must be
 562                 * a fast-pathed signal or we must have been
 563                 * out of queue space.  So zero out the info.
 564                 */
 565                info->si_signo = sig;
 566                info->si_errno = 0;
 567                info->si_code = SI_USER;
 568                info->si_pid = 0;
 569                info->si_uid = 0;
 570        }
 571}
 572
 573static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
 574                        siginfo_t *info)
 575{
 576        int sig = next_signal(pending, mask);
 577
 578        if (sig) {
 579                if (current->notifier) {
 580                        if (sigismember(current->notifier_mask, sig)) {
 581                                if (!(current->notifier)(current->notifier_data)) {
 582                                        clear_thread_flag(TIF_SIGPENDING);
 583                                        return 0;
 584                                }
 585                        }
 586                }
 587
 588                collect_signal(sig, pending, info);
 589        }
 590
 591        return sig;
 592}
 593
 594/*
 595 * Dequeue a signal and return the element to the caller, which is
 596 * expected to free it.
 597 *
 598 * All callers have to hold the siglock.
 599 */
 600int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
 601{
 602        int signr;
 603
 604        /* We only dequeue private signals from ourselves, we don't let
 605         * signalfd steal them
 606         */
 607        signr = __dequeue_signal(&tsk->pending, mask, info);
 608        if (!signr) {
 609                signr = __dequeue_signal(&tsk->signal->shared_pending,
 610                                         mask, info);
 611                /*
 612                 * itimer signal ?
 613                 *
 614                 * itimers are process shared and we restart periodic
 615                 * itimers in the signal delivery path to prevent DoS
 616                 * attacks in the high resolution timer case. This is
 617                 * compliant with the old way of self-restarting
 618                 * itimers, as the SIGALRM is a legacy signal and only
 619                 * queued once. Changing the restart behaviour to
 620                 * restart the timer in the signal dequeue path is
 621                 * reducing the timer noise on heavy loaded !highres
 622                 * systems too.
 623                 */
 624                if (unlikely(signr == SIGALRM)) {
 625                        struct hrtimer *tmr = &tsk->signal->real_timer;
 626
 627                        if (!hrtimer_is_queued(tmr) &&
 628                            tsk->signal->it_real_incr.tv64 != 0) {
 629                                hrtimer_forward(tmr, tmr->base->get_time(),
 630                                                tsk->signal->it_real_incr);
 631                                hrtimer_restart(tmr);
 632                        }
 633                }
 634        }
 635
 636        recalc_sigpending();
 637        if (!signr)
 638                return 0;
 639
 640        if (unlikely(sig_kernel_stop(signr))) {
 641                /*
 642                 * Set a marker that we have dequeued a stop signal.  Our
 643                 * caller might release the siglock and then the pending
 644                 * stop signal it is about to process is no longer in the
 645                 * pending bitmasks, but must still be cleared by a SIGCONT
 646                 * (and overruled by a SIGKILL).  So those cases clear this
 647                 * shared flag after we've set it.  Note that this flag may
 648                 * remain set after the signal we return is ignored or
 649                 * handled.  That doesn't matter because its only purpose
 650                 * is to alert stop-signal processing code when another
 651                 * processor has come along and cleared the flag.
 652                 */
 653                current->jobctl |= JOBCTL_STOP_DEQUEUED;
 654        }
 655        if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
 656                /*
 657                 * Release the siglock to ensure proper locking order
 658                 * of timer locks outside of siglocks.  Note, we leave
 659                 * irqs disabled here, since the posix-timers code is
 660                 * about to disable them again anyway.
 661                 */
 662                spin_unlock(&tsk->sighand->siglock);
 663                do_schedule_next_timer(info);
 664                spin_lock(&tsk->sighand->siglock);
 665        }
 666        return signr;
 667}
 668
 669/*
 670 * Tell a process that it has a new active signal..
 671 *
 672 * NOTE! we rely on the previous spin_lock to
 673 * lock interrupts for us! We can only be called with
 674 * "siglock" held, and the local interrupt must
 675 * have been disabled when that got acquired!
 676 *
 677 * No need to set need_resched since signal event passing
 678 * goes through ->blocked
 679 */
 680void signal_wake_up(struct task_struct *t, int resume)
 681{
 682        unsigned int mask;
 683
 684        set_tsk_thread_flag(t, TIF_SIGPENDING);
 685
 686        /*
 687         * For SIGKILL, we want to wake it up in the stopped/traced/killable
 688         * case. We don't check t->state here because there is a race with it
 689         * executing another processor and just now entering stopped state.
 690         * By using wake_up_state, we ensure the process will wake up and
 691         * handle its death signal.
 692         */
 693        mask = TASK_INTERRUPTIBLE;
 694        if (resume)
 695                mask |= TASK_WAKEKILL;
 696        if (!wake_up_state(t, mask))
 697                kick_process(t);
 698}
 699
 700/*
 701 * Remove signals in mask from the pending set and queue.
 702 * Returns 1 if any signals were found.
 703 *
 704 * All callers must be holding the siglock.
 705 *
 706 * This version takes a sigset mask and looks at all signals,
 707 * not just those in the first mask word.
 708 */
 709static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
 710{
 711        struct sigqueue *q, *n;
 712        sigset_t m;
 713
 714        sigandsets(&m, mask, &s->signal);
 715        if (sigisemptyset(&m))
 716                return 0;
 717
 718        sigandnsets(&s->signal, &s->signal, mask);
 719        list_for_each_entry_safe(q, n, &s->list, list) {
 720                if (sigismember(mask, q->info.si_signo)) {
 721                        list_del_init(&q->list);
 722                        __sigqueue_free(q);
 723                }
 724        }
 725        return 1;
 726}
 727/*
 728 * Remove signals in mask from the pending set and queue.
 729 * Returns 1 if any signals were found.
 730 *
 731 * All callers must be holding the siglock.
 732 */
 733static int rm_from_queue(unsigned long mask, struct sigpending *s)
 734{
 735        struct sigqueue *q, *n;
 736
 737        if (!sigtestsetmask(&s->signal, mask))
 738                return 0;
 739
 740        sigdelsetmask(&s->signal, mask);
 741        list_for_each_entry_safe(q, n, &s->list, list) {
 742                if (q->info.si_signo < SIGRTMIN &&
 743                    (mask & sigmask(q->info.si_signo))) {
 744                        list_del_init(&q->list);
 745                        __sigqueue_free(q);
 746                }
 747        }
 748        return 1;
 749}
 750
 751static inline int is_si_special(const struct siginfo *info)
 752{
 753        return info <= SEND_SIG_FORCED;
 754}
 755
 756static inline bool si_fromuser(const struct 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 */
 765static int 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        if (cred->user->user_ns == tcred->user->user_ns &&
 771            (cred->euid == tcred->suid ||
 772             cred->euid == tcred->uid ||
 773             cred->uid  == tcred->suid ||
 774             cred->uid  == tcred->uid))
 775                return 1;
 776
 777        if (ns_capable(tcred->user->user_ns, CAP_KILL))
 778                return 1;
 779
 780        return 0;
 781}
 782
 783/*
 784 * Bad permissions for sending the signal
 785 * - the caller must hold the RCU read lock
 786 */
 787static int check_kill_permission(int sig, struct siginfo *info,
 788                                 struct task_struct *t)
 789{
 790        struct pid *sid;
 791        int error;
 792
 793        if (!valid_signal(sig))
 794                return -EINVAL;
 795
 796        if (!si_fromuser(info))
 797                return 0;
 798
 799        error = audit_signal_info(sig, t); /* Let audit system see the signal */
 800        if (error)
 801                return error;
 802
 803        if (!same_thread_group(current, t) &&
 804            !kill_ok_by_cred(t)) {
 805                switch (sig) {
 806                case SIGCONT:
 807                        sid = task_session(t);
 808                        /*
 809                         * We don't return the error if sid == NULL. The
 810                         * task was unhashed, the caller must notice this.
 811                         */
 812                        if (!sid || sid == task_session(current))
 813                                break;
 814                default:
 815                        return -EPERM;
 816                }
 817        }
 818
 819        return security_task_kill(t, info, sig, 0);
 820}
 821
 822/**
 823 * ptrace_trap_notify - schedule trap to notify ptracer
 824 * @t: tracee wanting to notify tracer
 825 *
 826 * This function schedules sticky ptrace trap which is cleared on the next
 827 * TRAP_STOP to notify ptracer of an event.  @t must have been seized by
 828 * ptracer.
 829 *
 830 * If @t is running, STOP trap will be taken.  If trapped for STOP and
 831 * ptracer is listening for events, tracee is woken up so that it can
 832 * re-trap for the new event.  If trapped otherwise, STOP trap will be
 833 * eventually taken without returning to userland after the existing traps
 834 * are finished by PTRACE_CONT.
 835 *
 836 * CONTEXT:
 837 * Must be called with @task->sighand->siglock held.
 838 */
 839static void ptrace_trap_notify(struct task_struct *t)
 840{
 841        WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
 842        assert_spin_locked(&t->sighand->siglock);
 843
 844        task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
 845        signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
 846}
 847
 848/*
 849 * Handle magic process-wide effects of stop/continue signals. Unlike
 850 * the signal actions, these happen immediately at signal-generation
 851 * time regardless of blocking, ignoring, or handling.  This does the
 852 * actual continuing for SIGCONT, but not the actual stopping for stop
 853 * signals. The process stop is done as a signal action for SIG_DFL.
 854 *
 855 * Returns true if the signal should be actually delivered, otherwise
 856 * it should be dropped.
 857 */
 858static int prepare_signal(int sig, struct task_struct *p, bool force)
 859{
 860        struct signal_struct *signal = p->signal;
 861        struct task_struct *t;
 862
 863        if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
 864                /*
 865                 * The process is in the middle of dying, nothing to do.
 866                 */
 867        } else if (sig_kernel_stop(sig)) {
 868                /*
 869                 * This is a stop signal.  Remove SIGCONT from all queues.
 870                 */
 871                rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
 872                t = p;
 873                do {
 874                        rm_from_queue(sigmask(SIGCONT), &t->pending);
 875                } while_each_thread(p, t);
 876        } else if (sig == SIGCONT) {
 877                unsigned int why;
 878                /*
 879                 * Remove all stop signals from all queues, wake all threads.
 880                 */
 881                rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
 882                t = p;
 883                do {
 884                        task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
 885                        rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
 886                        if (likely(!(t->ptrace & PT_SEIZED)))
 887                                wake_up_state(t, __TASK_STOPPED);
 888                        else
 889                                ptrace_trap_notify(t);
 890                } while_each_thread(p, t);
 891
 892                /*
 893                 * Notify the parent with CLD_CONTINUED if we were stopped.
 894                 *
 895                 * If we were in the middle of a group stop, we pretend it
 896                 * was already finished, and then continued. Since SIGCHLD
 897                 * doesn't queue we report only CLD_STOPPED, as if the next
 898                 * CLD_CONTINUED was dropped.
 899                 */
 900                why = 0;
 901                if (signal->flags & SIGNAL_STOP_STOPPED)
 902                        why |= SIGNAL_CLD_CONTINUED;
 903                else if (signal->group_stop_count)
 904                        why |= SIGNAL_CLD_STOPPED;
 905
 906                if (why) {
 907                        /*
 908                         * The first thread which returns from do_signal_stop()
 909                         * will take ->siglock, notice SIGNAL_CLD_MASK, and
 910                         * notify its parent. See get_signal_to_deliver().
 911                         */
 912                        signal->flags = why | SIGNAL_STOP_CONTINUED;
 913                        signal->group_stop_count = 0;
 914                        signal->group_exit_code = 0;
 915                }
 916        }
 917
 918        return !sig_ignored(p, sig, force);
 919}
 920
 921/*
 922 * Test if P wants to take SIG.  After we've checked all threads with this,
 923 * it's equivalent to finding no threads not blocking SIG.  Any threads not
 924 * blocking SIG were ruled out because they are not running and already
 925 * have pending signals.  Such threads will dequeue from the shared queue
 926 * as soon as they're available, so putting the signal on the shared queue
 927 * will be equivalent to sending it to one such thread.
 928 */
 929static inline int wants_signal(int sig, struct task_struct *p)
 930{
 931        if (sigismember(&p->blocked, sig))
 932                return 0;
 933        if (p->flags & PF_EXITING)
 934                return 0;
 935        if (sig == SIGKILL)
 936                return 1;
 937        if (task_is_stopped_or_traced(p))
 938                return 0;
 939        return task_curr(p) || !signal_pending(p);
 940}
 941
 942static void complete_signal(int sig, struct task_struct *p, int group)
 943{
 944        struct signal_struct *signal = p->signal;
 945        struct task_struct *t;
 946
 947        /*
 948         * Now find a thread we can wake up to take the signal off the queue.
 949         *
 950         * If the main thread wants the signal, it gets first crack.
 951         * Probably the least surprising to the average bear.
 952         */
 953        if (wants_signal(sig, p))
 954                t = p;
 955        else if (!group || thread_group_empty(p))
 956                /*
 957                 * There is just one thread and it does not need to be woken.
 958                 * It will dequeue unblocked signals before it runs again.
 959                 */
 960                return;
 961        else {
 962                /*
 963                 * Otherwise try to find a suitable thread.
 964                 */
 965                t = signal->curr_target;
 966                while (!wants_signal(sig, t)) {
 967                        t = next_thread(t);
 968                        if (t == signal->curr_target)
 969                                /*
 970                                 * No thread needs to be woken.
 971                                 * Any eligible threads will see
 972                                 * the signal in the queue soon.
 973                                 */
 974                                return;
 975                }
 976                signal->curr_target = t;
 977        }
 978
 979        /*
 980         * Found a killable thread.  If the signal will be fatal,
 981         * then start taking the whole group down immediately.
 982         */
 983        if (sig_fatal(p, sig) &&
 984            !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
 985            !sigismember(&t->real_blocked, sig) &&
 986            (sig == SIGKILL || !t->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                        t = p;
1001                        do {
1002                                task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1003                                sigaddset(&t->pending.signal, SIGKILL);
1004                                signal_wake_up(t, 1);
1005                        } while_each_thread(p, t);
1006                        return;
1007                }
1008        }
1009
1010        /*
1011         * The signal is already in the shared-pending queue.
1012         * Tell the chosen thread to wake up and dequeue it.
1013         */
1014        signal_wake_up(t, sig == SIGKILL);
1015        return;
1016}
1017
1018static inline int legacy_queue(struct sigpending *signals, int sig)
1019{
1020        return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1021}
1022
1023/*
1024 * map the uid in struct cred into user namespace *ns
1025 */
1026static inline uid_t map_cred_ns(const struct cred *cred,
1027                                struct user_namespace *ns)
1028{
1029        return user_ns_map_uid(ns, cred, cred->uid);
1030}
1031
1032#ifdef CONFIG_USER_NS
1033static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1034{
1035        if (current_user_ns() == task_cred_xxx(t, user_ns))
1036                return;
1037
1038        if (SI_FROMKERNEL(info))
1039                return;
1040
1041        info->si_uid = user_ns_map_uid(task_cred_xxx(t, user_ns),
1042                                        current_cred(), info->si_uid);
1043}
1044#else
1045static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1046{
1047        return;
1048}
1049#endif
1050
1051static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
1052                        int group, int from_ancestor_ns)
1053{
1054        struct sigpending *pending;
1055        struct sigqueue *q;
1056        int override_rlimit;
1057        int ret = 0, result;
1058
1059        assert_spin_locked(&t->sighand->siglock);
1060
1061        result = TRACE_SIGNAL_IGNORED;
1062        if (!prepare_signal(sig, t,
1063                        from_ancestor_ns || (info == SEND_SIG_FORCED)))
1064                goto ret;
1065
1066        pending = group ? &t->signal->shared_pending : &t->pending;
1067        /*
1068         * Short-circuit ignored signals and support queuing
1069         * exactly one non-rt signal, so that we can get more
1070         * detailed information about the cause of the signal.
1071         */
1072        result = TRACE_SIGNAL_ALREADY_PENDING;
1073        if (legacy_queue(pending, sig))
1074                goto ret;
1075
1076        result = TRACE_SIGNAL_DELIVERED;
1077        /*
1078         * fast-pathed signals for kernel-internal things like SIGSTOP
1079         * or SIGKILL.
1080         */
1081        if (info == SEND_SIG_FORCED)
1082                goto out_set;
1083
1084        /*
1085         * Real-time signals must be queued if sent by sigqueue, or
1086         * some other real-time mechanism.  It is implementation
1087         * defined whether kill() does so.  We attempt to do so, on
1088         * the principle of least surprise, but since kill is not
1089         * allowed to fail with EAGAIN when low on memory we just
1090         * make sure at least one signal gets delivered and don't
1091         * pass on the info struct.
1092         */
1093        if (sig < SIGRTMIN)
1094                override_rlimit = (is_si_special(info) || info->si_code >= 0);
1095        else
1096                override_rlimit = 0;
1097
1098        q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
1099                override_rlimit);
1100        if (q) {
1101                list_add_tail(&q->list, &pending->list);
1102                switch ((unsigned long) info) {
1103                case (unsigned long) SEND_SIG_NOINFO:
1104                        q->info.si_signo = sig;
1105                        q->info.si_errno = 0;
1106                        q->info.si_code = SI_USER;
1107                        q->info.si_pid = task_tgid_nr_ns(current,
1108                                                        task_active_pid_ns(t));
1109                        q->info.si_uid = current_uid();
1110                        break;
1111                case (unsigned long) SEND_SIG_PRIV:
1112                        q->info.si_signo = sig;
1113                        q->info.si_errno = 0;
1114                        q->info.si_code = SI_KERNEL;
1115                        q->info.si_pid = 0;
1116                        q->info.si_uid = 0;
1117                        break;
1118                default:
1119                        copy_siginfo(&q->info, info);
1120                        if (from_ancestor_ns)
1121                                q->info.si_pid = 0;
1122                        break;
1123                }
1124
1125                userns_fixup_signal_uid(&q->info, t);
1126
1127        } else if (!is_si_special(info)) {
1128                if (sig >= SIGRTMIN && info->si_code != SI_USER) {
1129                        /*
1130                         * Queue overflow, abort.  We may abort if the
1131                         * signal was rt and sent by user using something
1132                         * other than kill().
1133                         */
1134                        result = TRACE_SIGNAL_OVERFLOW_FAIL;
1135                        ret = -EAGAIN;
1136                        goto ret;
1137                } else {
1138                        /*
1139                         * This is a silent loss of information.  We still
1140                         * send the signal, but the *info bits are lost.
1141                         */
1142                        result = TRACE_SIGNAL_LOSE_INFO;
1143                }
1144        }
1145
1146out_set:
1147        signalfd_notify(t, sig);
1148        sigaddset(&pending->signal, sig);
1149        complete_signal(sig, t, group);
1150ret:
1151        trace_signal_generate(sig, info, t, group, result);
1152        return ret;
1153}
1154
1155static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
1156                        int group)
1157{
1158        int from_ancestor_ns = 0;
1159
1160#ifdef CONFIG_PID_NS
1161        from_ancestor_ns = si_fromuser(info) &&
1162                           !task_pid_nr_ns(current, task_active_pid_ns(t));
1163#endif
1164
1165        return __send_signal(sig, info, t, group, from_ancestor_ns);
1166}
1167
1168static void print_fatal_signal(struct pt_regs *regs, int signr)
1169{
1170        printk("%s/%d: potentially unexpected fatal signal %d.\n",
1171                current->comm, task_pid_nr(current), signr);
1172
1173#if defined(__i386__) && !defined(__arch_um__)
1174        printk("code at %08lx: ", regs->ip);
1175        {
1176                int i;
1177                for (i = 0; i < 16; i++) {
1178                        unsigned char insn;
1179
1180                        if (get_user(insn, (unsigned char *)(regs->ip + i)))
1181                                break;
1182                        printk("%02x ", insn);
1183                }
1184        }
1185#endif
1186        printk("\n");
1187        preempt_disable();
1188        show_regs(regs);
1189        preempt_enable();
1190}
1191
1192static int __init setup_print_fatal_signals(char *str)
1193{
1194        get_option (&str, &print_fatal_signals);
1195
1196        return 1;
1197}
1198
1199__setup("print-fatal-signals=", setup_print_fatal_signals);
1200
1201int
1202__group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1203{
1204        return send_signal(sig, info, p, 1);
1205}
1206
1207static int
1208specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1209{
1210        return send_signal(sig, info, t, 0);
1211}
1212
1213int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1214                        bool group)
1215{
1216        unsigned long flags;
1217        int ret = -ESRCH;
1218
1219        if (lock_task_sighand(p, &flags)) {
1220                ret = send_signal(sig, info, p, group);
1221                unlock_task_sighand(p, &flags);
1222        }
1223
1224        return ret;
1225}
1226
1227/*
1228 * Force a signal that the process can't ignore: if necessary
1229 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1230 *
1231 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1232 * since we do not want to have a signal handler that was blocked
1233 * be invoked when user space had explicitly blocked it.
1234 *
1235 * We don't want to have recursive SIGSEGV's etc, for example,
1236 * that is why we also clear SIGNAL_UNKILLABLE.
1237 */
1238int
1239force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1240{
1241        unsigned long int flags;
1242        int ret, blocked, ignored;
1243        struct k_sigaction *action;
1244
1245        spin_lock_irqsave(&t->sighand->siglock, flags);
1246        action = &t->sighand->action[sig-1];
1247        ignored = action->sa.sa_handler == SIG_IGN;
1248        blocked = sigismember(&t->blocked, sig);
1249        if (blocked || ignored) {
1250                action->sa.sa_handler = SIG_DFL;
1251                if (blocked) {
1252                        sigdelset(&t->blocked, sig);
1253                        recalc_sigpending_and_wake(t);
1254                }
1255        }
1256        if (action->sa.sa_handler == SIG_DFL)
1257                t->signal->flags &= ~SIGNAL_UNKILLABLE;
1258        ret = specific_send_sig_info(sig, info, t);
1259        spin_unlock_irqrestore(&t->sighand->siglock, flags);
1260
1261        return ret;
1262}
1263
1264/*
1265 * Nuke all other threads in the group.
1266 */
1267int zap_other_threads(struct task_struct *p)
1268{
1269        struct task_struct *t = p;
1270        int count = 0;
1271
1272        p->signal->group_stop_count = 0;
1273
1274        while_each_thread(p, t) {
1275                task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1276                count++;
1277
1278                /* Don't bother with already dead threads */
1279                if (t->exit_state)
1280                        continue;
1281                sigaddset(&t->pending.signal, SIGKILL);
1282                signal_wake_up(t, 1);
1283        }
1284
1285        return count;
1286}
1287
1288struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1289                                           unsigned long *flags)
1290{
1291        struct sighand_struct *sighand;
1292
1293        for (;;) {
1294                local_irq_save(*flags);
1295                rcu_read_lock();
1296                sighand = rcu_dereference(tsk->sighand);
1297                if (unlikely(sighand == NULL)) {
1298                        rcu_read_unlock();
1299                        local_irq_restore(*flags);
1300                        break;
1301                }
1302
1303                spin_lock(&sighand->siglock);
1304                if (likely(sighand == tsk->sighand)) {
1305                        rcu_read_unlock();
1306                        break;
1307                }
1308                spin_unlock(&sighand->siglock);
1309                rcu_read_unlock();
1310                local_irq_restore(*flags);
1311        }
1312
1313        return sighand;
1314}
1315
1316/*
1317 * send signal info to all the members of a group
1318 */
1319int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1320{
1321        int ret;
1322
1323        rcu_read_lock();
1324        ret = check_kill_permission(sig, info, p);
1325        rcu_read_unlock();
1326
1327        if (!ret && sig)
1328                ret = do_send_sig_info(sig, info, p, true);
1329
1330        return ret;
1331}
1332
1333/*
1334 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1335 * control characters do (^C, ^Z etc)
1336 * - the caller must hold at least a readlock on tasklist_lock
1337 */
1338int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1339{
1340        struct task_struct *p = NULL;
1341        int retval, success;
1342
1343        success = 0;
1344        retval = -ESRCH;
1345        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1346                int err = group_send_sig_info(sig, info, p);
1347                success |= !err;
1348                retval = err;
1349        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1350        return success ? 0 : retval;
1351}
1352
1353int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1354{
1355        int error = -ESRCH;
1356        struct task_struct *p;
1357
1358        rcu_read_lock();
1359retry:
1360        p = pid_task(pid, PIDTYPE_PID);
1361        if (p) {
1362                error = group_send_sig_info(sig, info, p);
1363                if (unlikely(error == -ESRCH))
1364                        /*
1365                         * The task was unhashed in between, try again.
1366                         * If it is dead, pid_task() will return NULL,
1367                         * if we race with de_thread() it will find the
1368                         * new leader.
1369                         */
1370                        goto retry;
1371        }
1372        rcu_read_unlock();
1373
1374        return error;
1375}
1376
1377int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1378{
1379        int error;
1380        rcu_read_lock();
1381        error = kill_pid_info(sig, info, find_vpid(pid));
1382        rcu_read_unlock();
1383        return error;
1384}
1385
1386static int kill_as_cred_perm(const struct cred *cred,
1387                             struct task_struct *target)
1388{
1389        const struct cred *pcred = __task_cred(target);
1390        if (cred->user_ns != pcred->user_ns)
1391                return 0;
1392        if (cred->euid != pcred->suid && cred->euid != pcred->uid &&
1393            cred->uid  != pcred->suid && cred->uid  != pcred->uid)
1394                return 0;
1395        return 1;
1396}
1397
1398/* like kill_pid_info(), but doesn't use uid/euid of "current" */
1399int kill_pid_info_as_cred(int sig, struct siginfo *info, struct pid *pid,
1400                         const struct cred *cred, u32 secid)
1401{
1402        int ret = -EINVAL;
1403        struct task_struct *p;
1404        unsigned long flags;
1405
1406        if (!valid_signal(sig))
1407                return ret;
1408
1409        rcu_read_lock();
1410        p = pid_task(pid, PIDTYPE_PID);
1411        if (!p) {
1412                ret = -ESRCH;
1413                goto out_unlock;
1414        }
1415        if (si_fromuser(info) && !kill_as_cred_perm(cred, p)) {
1416                ret = -EPERM;
1417                goto out_unlock;
1418        }
1419        ret = security_task_kill(p, info, sig, secid);
1420        if (ret)
1421                goto out_unlock;
1422
1423        if (sig) {
1424                if (lock_task_sighand(p, &flags)) {
1425                        ret = __send_signal(sig, info, p, 1, 0);
1426                        unlock_task_sighand(p, &flags);
1427                } else
1428                        ret = -ESRCH;
1429        }
1430out_unlock:
1431        rcu_read_unlock();
1432        return ret;
1433}
1434EXPORT_SYMBOL_GPL(kill_pid_info_as_cred);
1435
1436/*
1437 * kill_something_info() interprets pid in interesting ways just like kill(2).
1438 *
1439 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1440 * is probably wrong.  Should make it like BSD or SYSV.
1441 */
1442
1443static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1444{
1445        int ret;
1446
1447        if (pid > 0) {
1448                rcu_read_lock();
1449                ret = kill_pid_info(sig, info, find_vpid(pid));
1450                rcu_read_unlock();
1451                return ret;
1452        }
1453
1454        read_lock(&tasklist_lock);
1455        if (pid != -1) {
1456                ret = __kill_pgrp_info(sig, info,
1457                                pid ? find_vpid(-pid) : task_pgrp(current));
1458        } else {
1459                int retval = 0, count = 0;
1460                struct task_struct * p;
1461
1462                for_each_process(p) {
1463                        if (task_pid_vnr(p) > 1 &&
1464                                        !same_thread_group(p, current)) {
1465                                int err = group_send_sig_info(sig, info, p);
1466                                ++count;
1467                                if (err != -EPERM)
1468                                        retval = err;
1469                        }
1470                }
1471                ret = count ? retval : -ESRCH;
1472        }
1473        read_unlock(&tasklist_lock);
1474
1475        return ret;
1476}
1477
1478/*
1479 * These are for backward compatibility with the rest of the kernel source.
1480 */
1481
1482int send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1483{
1484        /*
1485         * Make sure legacy kernel users don't send in bad values
1486         * (normal paths check this in check_kill_permission).
1487         */
1488        if (!valid_signal(sig))
1489                return -EINVAL;
1490
1491        return do_send_sig_info(sig, info, p, false);
1492}
1493
1494#define __si_special(priv) \
1495        ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1496
1497int
1498send_sig(int sig, struct task_struct *p, int priv)
1499{
1500        return send_sig_info(sig, __si_special(priv), p);
1501}
1502
1503void
1504force_sig(int sig, struct task_struct *p)
1505{
1506        force_sig_info(sig, SEND_SIG_PRIV, p);
1507}
1508
1509/*
1510 * When things go south during signal handling, we
1511 * will force a SIGSEGV. And if the signal that caused
1512 * the problem was already a SIGSEGV, we'll want to
1513 * make sure we don't even try to deliver the signal..
1514 */
1515int
1516force_sigsegv(int sig, struct task_struct *p)
1517{
1518        if (sig == SIGSEGV) {
1519                unsigned long flags;
1520                spin_lock_irqsave(&p->sighand->siglock, flags);
1521                p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1522                spin_unlock_irqrestore(&p->sighand->siglock, flags);
1523        }
1524        force_sig(SIGSEGV, p);
1525        return 0;
1526}
1527
1528int kill_pgrp(struct pid *pid, int sig, int priv)
1529{
1530        int ret;
1531
1532        read_lock(&tasklist_lock);
1533        ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1534        read_unlock(&tasklist_lock);
1535
1536        return ret;
1537}
1538EXPORT_SYMBOL(kill_pgrp);
1539
1540int kill_pid(struct pid *pid, int sig, int priv)
1541{
1542        return kill_pid_info(sig, __si_special(priv), pid);
1543}
1544EXPORT_SYMBOL(kill_pid);
1545
1546/*
1547 * These functions support sending signals using preallocated sigqueue
1548 * structures.  This is needed "because realtime applications cannot
1549 * afford to lose notifications of asynchronous events, like timer
1550 * expirations or I/O completions".  In the case of POSIX Timers
1551 * we allocate the sigqueue structure from the timer_create.  If this
1552 * allocation fails we are able to report the failure to the application
1553 * with an EAGAIN error.
1554 */
1555struct sigqueue *sigqueue_alloc(void)
1556{
1557        struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1558
1559        if (q)
1560                q->flags |= SIGQUEUE_PREALLOC;
1561
1562        return q;
1563}
1564
1565void sigqueue_free(struct sigqueue *q)
1566{
1567        unsigned long flags;
1568        spinlock_t *lock = &current->sighand->siglock;
1569
1570        BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1571        /*
1572         * We must hold ->siglock while testing q->list
1573         * to serialize with collect_signal() or with
1574         * __exit_signal()->flush_sigqueue().
1575         */
1576        spin_lock_irqsave(lock, flags);
1577        q->flags &= ~SIGQUEUE_PREALLOC;
1578        /*
1579         * If it is queued it will be freed when dequeued,
1580         * like the "regular" sigqueue.
1581         */
1582        if (!list_empty(&q->list))
1583                q = NULL;
1584        spin_unlock_irqrestore(lock, flags);
1585
1586        if (q)
1587                __sigqueue_free(q);
1588}
1589
1590int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1591{
1592        int sig = q->info.si_signo;
1593        struct sigpending *pending;
1594        unsigned long flags;
1595        int ret, result;
1596
1597        BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1598
1599        ret = -1;
1600        if (!likely(lock_task_sighand(t, &flags)))
1601                goto ret;
1602
1603        ret = 1; /* the signal is ignored */
1604        result = TRACE_SIGNAL_IGNORED;
1605        if (!prepare_signal(sig, t, false))
1606                goto out;
1607
1608        ret = 0;
1609        if (unlikely(!list_empty(&q->list))) {
1610                /*
1611                 * If an SI_TIMER entry is already queue just increment
1612                 * the overrun count.
1613                 */
1614                BUG_ON(q->info.si_code != SI_TIMER);
1615                q->info.si_overrun++;
1616                result = TRACE_SIGNAL_ALREADY_PENDING;
1617                goto out;
1618        }
1619        q->info.si_overrun = 0;
1620
1621        signalfd_notify(t, sig);
1622        pending = group ? &t->signal->shared_pending : &t->pending;
1623        list_add_tail(&q->list, &pending->list);
1624        sigaddset(&pending->signal, sig);
1625        complete_signal(sig, t, group);
1626        result = TRACE_SIGNAL_DELIVERED;
1627out:
1628        trace_signal_generate(sig, &q->info, t, group, result);
1629        unlock_task_sighand(t, &flags);
1630ret:
1631        return ret;
1632}
1633
1634/*
1635 * Let a parent know about the death of a child.
1636 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1637 *
1638 * Returns true if our parent ignored us and so we've switched to
1639 * self-reaping.
1640 */
1641bool do_notify_parent(struct task_struct *tsk, int sig)
1642{
1643        struct siginfo info;
1644        unsigned long flags;
1645        struct sighand_struct *psig;
1646        bool autoreap = false;
1647
1648        BUG_ON(sig == -1);
1649
1650        /* do_notify_parent_cldstop should have been called instead.  */
1651        BUG_ON(task_is_stopped_or_traced(tsk));
1652
1653        BUG_ON(!tsk->ptrace &&
1654               (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1655
1656        if (sig != SIGCHLD) {
1657                /*
1658                 * This is only possible if parent == real_parent.
1659                 * Check if it has changed security domain.
1660                 */
1661                if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1662                        sig = SIGCHLD;
1663        }
1664
1665        info.si_signo = sig;
1666        info.si_errno = 0;
1667        /*
1668         * we are under tasklist_lock here so our parent is tied to
1669         * us and cannot exit and release its namespace.
1670         *
1671         * the only it can is to switch its nsproxy with sys_unshare,
1672         * bu uncharing pid namespaces is not allowed, so we'll always
1673         * see relevant namespace
1674         *
1675         * write_lock() currently calls preempt_disable() which is the
1676         * same as rcu_read_lock(), but according to Oleg, this is not
1677         * correct to rely on this
1678         */
1679        rcu_read_lock();
1680        info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1681        info.si_uid = map_cred_ns(__task_cred(tsk),
1682                        task_cred_xxx(tsk->parent, user_ns));
1683        rcu_read_unlock();
1684
1685        info.si_utime = cputime_to_clock_t(tsk->utime + tsk->signal->utime);
1686        info.si_stime = cputime_to_clock_t(tsk->stime + tsk->signal->stime);
1687
1688        info.si_status = tsk->exit_code & 0x7f;
1689        if (tsk->exit_code & 0x80)
1690                info.si_code = CLD_DUMPED;
1691        else if (tsk->exit_code & 0x7f)
1692                info.si_code = CLD_KILLED;
1693        else {
1694                info.si_code = CLD_EXITED;
1695                info.si_status = tsk->exit_code >> 8;
1696        }
1697
1698        psig = tsk->parent->sighand;
1699        spin_lock_irqsave(&psig->siglock, flags);
1700        if (!tsk->ptrace && sig == SIGCHLD &&
1701            (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1702             (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1703                /*
1704                 * We are exiting and our parent doesn't care.  POSIX.1
1705                 * defines special semantics for setting SIGCHLD to SIG_IGN
1706                 * or setting the SA_NOCLDWAIT flag: we should be reaped
1707                 * automatically and not left for our parent's wait4 call.
1708                 * Rather than having the parent do it as a magic kind of
1709                 * signal handler, we just set this to tell do_exit that we
1710                 * can be cleaned up without becoming a zombie.  Note that
1711                 * we still call __wake_up_parent in this case, because a
1712                 * blocked sys_wait4 might now return -ECHILD.
1713                 *
1714                 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1715                 * is implementation-defined: we do (if you don't want
1716                 * it, just use SIG_IGN instead).
1717                 */
1718                autoreap = true;
1719                if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1720                        sig = 0;
1721        }
1722        if (valid_signal(sig) && sig)
1723                __group_send_sig_info(sig, &info, tsk->parent);
1724        __wake_up_parent(tsk, tsk->parent);
1725        spin_unlock_irqrestore(&psig->siglock, flags);
1726
1727        return autoreap;
1728}
1729
1730/**
1731 * do_notify_parent_cldstop - notify parent of stopped/continued state change
1732 * @tsk: task reporting the state change
1733 * @for_ptracer: the notification is for ptracer
1734 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
1735 *
1736 * Notify @tsk's parent that the stopped/continued state has changed.  If
1737 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
1738 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
1739 *
1740 * CONTEXT:
1741 * Must be called with tasklist_lock at least read locked.
1742 */
1743static void do_notify_parent_cldstop(struct task_struct *tsk,
1744                                     bool for_ptracer, int why)
1745{
1746        struct siginfo info;
1747        unsigned long flags;
1748        struct task_struct *parent;
1749        struct sighand_struct *sighand;
1750
1751        if (for_ptracer) {
1752                parent = tsk->parent;
1753        } else {
1754                tsk = tsk->group_leader;
1755                parent = tsk->real_parent;
1756        }
1757
1758        info.si_signo = SIGCHLD;
1759        info.si_errno = 0;
1760        /*
1761         * see comment in do_notify_parent() about the following 4 lines
1762         */
1763        rcu_read_lock();
1764        info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1765        info.si_uid = map_cred_ns(__task_cred(tsk),
1766                        task_cred_xxx(parent, user_ns));
1767        rcu_read_unlock();
1768
1769        info.si_utime = cputime_to_clock_t(tsk->utime);
1770        info.si_stime = cputime_to_clock_t(tsk->stime);
1771
1772        info.si_code = why;
1773        switch (why) {
1774        case CLD_CONTINUED:
1775                info.si_status = SIGCONT;
1776                break;
1777        case CLD_STOPPED:
1778                info.si_status = tsk->signal->group_exit_code & 0x7f;
1779                break;
1780        case CLD_TRAPPED:
1781                info.si_status = tsk->exit_code & 0x7f;
1782                break;
1783        default:
1784                BUG();
1785        }
1786
1787        sighand = parent->sighand;
1788        spin_lock_irqsave(&sighand->siglock, flags);
1789        if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1790            !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1791                __group_send_sig_info(SIGCHLD, &info, parent);
1792        /*
1793         * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1794         */
1795        __wake_up_parent(tsk, parent);
1796        spin_unlock_irqrestore(&sighand->siglock, flags);
1797}
1798
1799static inline int may_ptrace_stop(void)
1800{
1801        if (!likely(current->ptrace))
1802                return 0;
1803        /*
1804         * Are we in the middle of do_coredump?
1805         * If so and our tracer is also part of the coredump stopping
1806         * is a deadlock situation, and pointless because our tracer
1807         * is dead so don't allow us to stop.
1808         * If SIGKILL was already sent before the caller unlocked
1809         * ->siglock we must see ->core_state != NULL. Otherwise it
1810         * is safe to enter schedule().
1811         */
1812        if (unlikely(current->mm->core_state) &&
1813            unlikely(current->mm == current->parent->mm))
1814                return 0;
1815
1816        return 1;
1817}
1818
1819/*
1820 * Return non-zero if there is a SIGKILL that should be waking us up.
1821 * Called with the siglock held.
1822 */
1823static int sigkill_pending(struct task_struct *tsk)
1824{
1825        return  sigismember(&tsk->pending.signal, SIGKILL) ||
1826                sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1827}
1828
1829/*
1830 * This must be called with current->sighand->siglock held.
1831 *
1832 * This should be the path for all ptrace stops.
1833 * We always set current->last_siginfo while stopped here.
1834 * That makes it a way to test a stopped process for
1835 * being ptrace-stopped vs being job-control-stopped.
1836 *
1837 * If we actually decide not to stop at all because the tracer
1838 * is gone, we keep current->exit_code unless clear_code.
1839 */
1840static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
1841        __releases(&current->sighand->siglock)
1842        __acquires(&current->sighand->siglock)
1843{
1844        bool gstop_done = false;
1845
1846        if (arch_ptrace_stop_needed(exit_code, info)) {
1847                /*
1848                 * The arch code has something special to do before a
1849                 * ptrace stop.  This is allowed to block, e.g. for faults
1850                 * on user stack pages.  We can't keep the siglock while
1851                 * calling arch_ptrace_stop, so we must release it now.
1852                 * To preserve proper semantics, we must do this before
1853                 * any signal bookkeeping like checking group_stop_count.
1854                 * Meanwhile, a SIGKILL could come in before we retake the
1855                 * siglock.  That must prevent us from sleeping in TASK_TRACED.
1856                 * So after regaining the lock, we must check for SIGKILL.
1857                 */
1858                spin_unlock_irq(&current->sighand->siglock);
1859                arch_ptrace_stop(exit_code, info);
1860                spin_lock_irq(&current->sighand->siglock);
1861                if (sigkill_pending(current))
1862                        return;
1863        }
1864
1865        /*
1866         * We're committing to trapping.  TRACED should be visible before
1867         * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
1868         * Also, transition to TRACED and updates to ->jobctl should be
1869         * atomic with respect to siglock and should be done after the arch
1870         * hook as siglock is released and regrabbed across it.
1871         */
1872        set_current_state(TASK_TRACED);
1873
1874        current->last_siginfo = info;
1875        current->exit_code = exit_code;
1876
1877        /*
1878         * If @why is CLD_STOPPED, we're trapping to participate in a group
1879         * stop.  Do the bookkeeping.  Note that if SIGCONT was delievered
1880         * across siglock relocks since INTERRUPT was scheduled, PENDING
1881         * could be clear now.  We act as if SIGCONT is received after
1882         * TASK_TRACED is entered - ignore it.
1883         */
1884        if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
1885                gstop_done = task_participate_group_stop(current);
1886
1887        /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
1888        task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
1889        if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
1890                task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
1891
1892        /* entering a trap, clear TRAPPING */
1893        task_clear_jobctl_trapping(current);
1894
1895        spin_unlock_irq(&current->sighand->siglock);
1896        read_lock(&tasklist_lock);
1897        if (may_ptrace_stop()) {
1898                /*
1899                 * Notify parents of the stop.
1900                 *
1901                 * While ptraced, there are two parents - the ptracer and
1902                 * the real_parent of the group_leader.  The ptracer should
1903                 * know about every stop while the real parent is only
1904                 * interested in the completion of group stop.  The states
1905                 * for the two don't interact with each other.  Notify
1906                 * separately unless they're gonna be duplicates.
1907                 */
1908                do_notify_parent_cldstop(current, true, why);
1909                if (gstop_done && ptrace_reparented(current))
1910                        do_notify_parent_cldstop(current, false, why);
1911
1912                /*
1913                 * Don't want to allow preemption here, because
1914                 * sys_ptrace() needs this task to be inactive.
1915                 *
1916                 * XXX: implement read_unlock_no_resched().
1917                 */
1918                preempt_disable();
1919                read_unlock(&tasklist_lock);
1920                preempt_enable_no_resched();
1921                schedule();
1922        } else {
1923                /*
1924                 * By the time we got the lock, our tracer went away.
1925                 * Don't drop the lock yet, another tracer may come.
1926                 *
1927                 * If @gstop_done, the ptracer went away between group stop
1928                 * completion and here.  During detach, it would have set
1929                 * JOBCTL_STOP_PENDING on us and we'll re-enter
1930                 * TASK_STOPPED in do_signal_stop() on return, so notifying
1931                 * the real parent of the group stop completion is enough.
1932                 */
1933                if (gstop_done)
1934                        do_notify_parent_cldstop(current, false, why);
1935
1936                __set_current_state(TASK_RUNNING);
1937                if (clear_code)
1938                        current->exit_code = 0;
1939                read_unlock(&tasklist_lock);
1940        }
1941
1942        /*
1943         * While in TASK_TRACED, we were considered "frozen enough".
1944         * Now that we woke up, it's crucial if we're supposed to be
1945         * frozen that we freeze now before running anything substantial.
1946         */
1947        try_to_freeze();
1948
1949        /*
1950         * We are back.  Now reacquire the siglock before touching
1951         * last_siginfo, so that we are sure to have synchronized with
1952         * any signal-sending on another CPU that wants to examine it.
1953         */
1954        spin_lock_irq(&current->sighand->siglock);
1955        current->last_siginfo = NULL;
1956
1957        /* LISTENING can be set only during STOP traps, clear it */
1958        current->jobctl &= ~JOBCTL_LISTENING;
1959
1960        /*
1961         * Queued signals ignored us while we were stopped for tracing.
1962         * So check for any that we should take before resuming user mode.
1963         * This sets TIF_SIGPENDING, but never clears it.
1964         */
1965        recalc_sigpending_tsk(current);
1966}
1967
1968static void ptrace_do_notify(int signr, int exit_code, int why)
1969{
1970        siginfo_t info;
1971
1972        memset(&info, 0, sizeof info);
1973        info.si_signo = signr;
1974        info.si_code = exit_code;
1975        info.si_pid = task_pid_vnr(current);
1976        info.si_uid = current_uid();
1977
1978        /* Let the debugger run.  */
1979        ptrace_stop(exit_code, why, 1, &info);
1980}
1981
1982void ptrace_notify(int exit_code)
1983{
1984        BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1985
1986        spin_lock_irq(&current->sighand->siglock);
1987        ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
1988        spin_unlock_irq(&current->sighand->siglock);
1989}
1990
1991/**
1992 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
1993 * @signr: signr causing group stop if initiating
1994 *
1995 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
1996 * and participate in it.  If already set, participate in the existing
1997 * group stop.  If participated in a group stop (and thus slept), %true is
1998 * returned with siglock released.
1999 *
2000 * If ptraced, this function doesn't handle stop itself.  Instead,
2001 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2002 * untouched.  The caller must ensure that INTERRUPT trap handling takes
2003 * places afterwards.
2004 *
2005 * CONTEXT:
2006 * Must be called with @current->sighand->siglock held, which is released
2007 * on %true return.
2008 *
2009 * RETURNS:
2010 * %false if group stop is already cancelled or ptrace trap is scheduled.
2011 * %true if participated in group stop.
2012 */
2013static bool do_signal_stop(int signr)
2014        __releases(&current->sighand->siglock)
2015{
2016        struct signal_struct *sig = current->signal;
2017
2018        if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2019                unsigned int gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2020                struct task_struct *t;
2021
2022                /* signr will be recorded in task->jobctl for retries */
2023                WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2024
2025                if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2026                    unlikely(signal_group_exit(sig)))
2027                        return false;
2028                /*
2029                 * There is no group stop already in progress.  We must
2030                 * initiate one now.
2031                 *
2032                 * While ptraced, a task may be resumed while group stop is
2033                 * still in effect and then receive a stop signal and
2034                 * initiate another group stop.  This deviates from the
2035                 * usual behavior as two consecutive stop signals can't
2036                 * cause two group stops when !ptraced.  That is why we
2037                 * also check !task_is_stopped(t) below.
2038                 *
2039                 * The condition can be distinguished by testing whether
2040                 * SIGNAL_STOP_STOPPED is already set.  Don't generate
2041                 * group_exit_code in such case.
2042                 *
2043                 * This is not necessary for SIGNAL_STOP_CONTINUED because
2044                 * an intervening stop signal is required to cause two
2045                 * continued events regardless of ptrace.
2046                 */
2047                if (!(sig->flags & SIGNAL_STOP_STOPPED))
2048                        sig->group_exit_code = signr;
2049
2050                sig->group_stop_count = 0;
2051
2052                if (task_set_jobctl_pending(current, signr | gstop))
2053                        sig->group_stop_count++;
2054
2055                for (t = next_thread(current); t != current;
2056                     t = next_thread(t)) {
2057                        /*
2058                         * Setting state to TASK_STOPPED for a group
2059                         * stop is always done with the siglock held,
2060                         * so this check has no races.
2061                         */
2062                        if (!task_is_stopped(t) &&
2063                            task_set_jobctl_pending(t, signr | gstop)) {
2064                                sig->group_stop_count++;
2065                                if (likely(!(t->ptrace & PT_SEIZED)))
2066                                        signal_wake_up(t, 0);
2067                                else
2068                                        ptrace_trap_notify(t);
2069                        }
2070                }
2071        }
2072
2073        if (likely(!current->ptrace)) {
2074                int notify = 0;
2075
2076                /*
2077                 * If there are no other threads in the group, or if there
2078                 * is a group stop in progress and we are the last to stop,
2079                 * report to the parent.
2080                 */
2081                if (task_participate_group_stop(current))
2082                        notify = CLD_STOPPED;
2083
2084                __set_current_state(TASK_STOPPED);
2085                spin_unlock_irq(&current->sighand->siglock);
2086
2087                /*
2088                 * Notify the parent of the group stop completion.  Because
2089                 * we're not holding either the siglock or tasklist_lock
2090                 * here, ptracer may attach inbetween; however, this is for
2091                 * group stop and should always be delivered to the real
2092                 * parent of the group leader.  The new ptracer will get
2093                 * its notification when this task transitions into
2094                 * TASK_TRACED.
2095                 */
2096                if (notify) {
2097                        read_lock(&tasklist_lock);
2098                        do_notify_parent_cldstop(current, false, notify);
2099                        read_unlock(&tasklist_lock);
2100                }
2101
2102                /* Now we don't run again until woken by SIGCONT or SIGKILL */
2103                schedule();
2104                return true;
2105        } else {
2106                /*
2107                 * While ptraced, group stop is handled by STOP trap.
2108                 * Schedule it and let the caller deal with it.
2109                 */
2110                task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2111                return false;
2112        }
2113}
2114
2115/**
2116 * do_jobctl_trap - take care of ptrace jobctl traps
2117 *
2118 * When PT_SEIZED, it's used for both group stop and explicit
2119 * SEIZE/INTERRUPT traps.  Both generate PTRACE_EVENT_STOP trap with
2120 * accompanying siginfo.  If stopped, lower eight bits of exit_code contain
2121 * the stop signal; otherwise, %SIGTRAP.
2122 *
2123 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2124 * number as exit_code and no siginfo.
2125 *
2126 * CONTEXT:
2127 * Must be called with @current->sighand->siglock held, which may be
2128 * released and re-acquired before returning with intervening sleep.
2129 */
2130static void do_jobctl_trap(void)
2131{
2132        struct signal_struct *signal = current->signal;
2133        int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2134
2135        if (current->ptrace & PT_SEIZED) {
2136                if (!signal->group_stop_count &&
2137                    !(signal->flags & SIGNAL_STOP_STOPPED))
2138                        signr = SIGTRAP;
2139                WARN_ON_ONCE(!signr);
2140                ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2141                                 CLD_STOPPED);
2142        } else {
2143                WARN_ON_ONCE(!signr);
2144                ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2145                current->exit_code = 0;
2146        }
2147}
2148
2149static int ptrace_signal(int signr, siginfo_t *info,
2150                         struct pt_regs *regs, void *cookie)
2151{
2152        ptrace_signal_deliver(regs, cookie);
2153        /*
2154         * We do not check sig_kernel_stop(signr) but set this marker
2155         * unconditionally because we do not know whether debugger will
2156         * change signr. This flag has no meaning unless we are going
2157         * to stop after return from ptrace_stop(). In this case it will
2158         * be checked in do_signal_stop(), we should only stop if it was
2159         * not cleared by SIGCONT while we were sleeping. See also the
2160         * comment in dequeue_signal().
2161         */
2162        current->jobctl |= JOBCTL_STOP_DEQUEUED;
2163        ptrace_stop(signr, CLD_TRAPPED, 0, info);
2164
2165        /* We're back.  Did the debugger cancel the sig?  */
2166        signr = current->exit_code;
2167        if (signr == 0)
2168                return signr;
2169
2170        current->exit_code = 0;
2171
2172        /*
2173         * Update the siginfo structure if the signal has
2174         * changed.  If the debugger wanted something
2175         * specific in the siginfo structure then it should
2176         * have updated *info via PTRACE_SETSIGINFO.
2177         */
2178        if (signr != info->si_signo) {
2179                info->si_signo = signr;
2180                info->si_errno = 0;
2181                info->si_code = SI_USER;
2182                rcu_read_lock();
2183                info->si_pid = task_pid_vnr(current->parent);
2184                info->si_uid = map_cred_ns(__task_cred(current->parent),
2185                                current_user_ns());
2186                rcu_read_unlock();
2187        }
2188
2189        /* If the (new) signal is now blocked, requeue it.  */
2190        if (sigismember(&current->blocked, signr)) {
2191                specific_send_sig_info(signr, info, current);
2192                signr = 0;
2193        }
2194
2195        return signr;
2196}
2197
2198int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2199                          struct pt_regs *regs, void *cookie)
2200{
2201        struct sighand_struct *sighand = current->sighand;
2202        struct signal_struct *signal = current->signal;
2203        int signr;
2204
2205relock:
2206        /*
2207         * We'll jump back here after any time we were stopped in TASK_STOPPED.
2208         * While in TASK_STOPPED, we were considered "frozen enough".
2209         * Now that we woke up, it's crucial if we're supposed to be
2210         * frozen that we freeze now before running anything substantial.
2211         */
2212        try_to_freeze();
2213
2214        spin_lock_irq(&sighand->siglock);
2215        /*
2216         * Every stopped thread goes here after wakeup. Check to see if
2217         * we should notify the parent, prepare_signal(SIGCONT) encodes
2218         * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2219         */
2220        if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2221                int why;
2222
2223                if (signal->flags & SIGNAL_CLD_CONTINUED)
2224                        why = CLD_CONTINUED;
2225                else
2226                        why = CLD_STOPPED;
2227
2228                signal->flags &= ~SIGNAL_CLD_MASK;
2229
2230                spin_unlock_irq(&sighand->siglock);
2231
2232                /*
2233                 * Notify the parent that we're continuing.  This event is
2234                 * always per-process and doesn't make whole lot of sense
2235                 * for ptracers, who shouldn't consume the state via
2236                 * wait(2) either, but, for backward compatibility, notify
2237                 * the ptracer of the group leader too unless it's gonna be
2238                 * a duplicate.
2239                 */
2240                read_lock(&tasklist_lock);
2241                do_notify_parent_cldstop(current, false, why);
2242
2243                if (ptrace_reparented(current->group_leader))
2244                        do_notify_parent_cldstop(current->group_leader,
2245                                                true, why);
2246                read_unlock(&tasklist_lock);
2247
2248                goto relock;
2249        }
2250
2251        for (;;) {
2252                struct k_sigaction *ka;
2253
2254                if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2255                    do_signal_stop(0))
2256                        goto relock;
2257
2258                if (unlikely(current->jobctl & JOBCTL_TRAP_MASK)) {
2259                        do_jobctl_trap();
2260                        spin_unlock_irq(&sighand->siglock);
2261                        goto relock;
2262                }
2263
2264                signr = dequeue_signal(current, &current->blocked, info);
2265
2266                if (!signr)
2267                        break; /* will return 0 */
2268
2269                if (unlikely(current->ptrace) && signr != SIGKILL) {
2270                        signr = ptrace_signal(signr, info,
2271                                              regs, cookie);
2272                        if (!signr)
2273                                continue;
2274                }
2275
2276                ka = &sighand->action[signr-1];
2277
2278                /* Trace actually delivered signals. */
2279                trace_signal_deliver(signr, info, ka);
2280
2281                if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
2282                        continue;
2283                if (ka->sa.sa_handler != SIG_DFL) {
2284                        /* Run the handler.  */
2285                        *return_ka = *ka;
2286
2287                        if (ka->sa.sa_flags & SA_ONESHOT)
2288                                ka->sa.sa_handler = SIG_DFL;
2289
2290                        break; /* will return non-zero "signr" value */
2291                }
2292
2293                /*
2294                 * Now we are doing the default action for this signal.
2295                 */
2296                if (sig_kernel_ignore(signr)) /* Default is nothing. */
2297                        continue;
2298
2299                /*
2300                 * Global init gets no signals it doesn't want.
2301                 * Container-init gets no signals it doesn't want from same
2302                 * container.
2303                 *
2304                 * Note that if global/container-init sees a sig_kernel_only()
2305                 * signal here, the signal must have been generated internally
2306                 * or must have come from an ancestor namespace. In either
2307                 * case, the signal cannot be dropped.
2308                 */
2309                if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2310                                !sig_kernel_only(signr))
2311                        continue;
2312
2313                if (sig_kernel_stop(signr)) {
2314                        /*
2315                         * The default action is to stop all threads in
2316                         * the thread group.  The job control signals
2317                         * do nothing in an orphaned pgrp, but SIGSTOP
2318                         * always works.  Note that siglock needs to be
2319                         * dropped during the call to is_orphaned_pgrp()
2320                         * because of lock ordering with tasklist_lock.
2321                         * This allows an intervening SIGCONT to be posted.
2322                         * We need to check for that and bail out if necessary.
2323                         */
2324                        if (signr != SIGSTOP) {
2325                                spin_unlock_irq(&sighand->siglock);
2326
2327                                /* signals can be posted during this window */
2328
2329                                if (is_current_pgrp_orphaned())
2330                                        goto relock;
2331
2332                                spin_lock_irq(&sighand->siglock);
2333                        }
2334
2335                        if (likely(do_signal_stop(info->si_signo))) {
2336                                /* It released the siglock.  */
2337                                goto relock;
2338                        }
2339
2340                        /*
2341                         * We didn't actually stop, due to a race
2342                         * with SIGCONT or something like that.
2343                         */
2344                        continue;
2345                }
2346
2347                spin_unlock_irq(&sighand->siglock);
2348
2349                /*
2350                 * Anything else is fatal, maybe with a core dump.
2351                 */
2352                current->flags |= PF_SIGNALED;
2353
2354                if (sig_kernel_coredump(signr)) {
2355                        if (print_fatal_signals)
2356                                print_fatal_signal(regs, info->si_signo);
2357                        /*
2358                         * If it was able to dump core, this kills all
2359                         * other threads in the group and synchronizes with
2360                         * their demise.  If we lost the race with another
2361                         * thread getting here, it set group_exit_code
2362                         * first and our do_group_exit call below will use
2363                         * that value and ignore the one we pass it.
2364                         */
2365                        do_coredump(info->si_signo, info->si_signo, regs);
2366                }
2367
2368                /*
2369                 * Death signals, no core dump.
2370                 */
2371                do_group_exit(info->si_signo);
2372                /* NOTREACHED */
2373        }
2374        spin_unlock_irq(&sighand->siglock);
2375        return signr;
2376}
2377
2378/**
2379 * block_sigmask - add @ka's signal mask to current->blocked
2380 * @ka: action for @signr
2381 * @signr: signal that has been successfully delivered
2382 *
2383 * This function should be called when a signal has succesfully been
2384 * delivered. It adds the mask of signals for @ka to current->blocked
2385 * so that they are blocked during the execution of the signal
2386 * handler. In addition, @signr will be blocked unless %SA_NODEFER is
2387 * set in @ka->sa.sa_flags.
2388 */
2389void block_sigmask(struct k_sigaction *ka, int signr)
2390{
2391        sigset_t blocked;
2392
2393        sigorsets(&blocked, &current->blocked, &ka->sa.sa_mask);
2394        if (!(ka->sa.sa_flags & SA_NODEFER))
2395                sigaddset(&blocked, signr);
2396        set_current_blocked(&blocked);
2397}
2398
2399/*
2400 * It could be that complete_signal() picked us to notify about the
2401 * group-wide signal. Other threads should be notified now to take
2402 * the shared signals in @which since we will not.
2403 */
2404static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2405{
2406        sigset_t retarget;
2407        struct task_struct *t;
2408
2409        sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2410        if (sigisemptyset(&retarget))
2411                return;
2412
2413        t = tsk;
2414        while_each_thread(tsk, t) {
2415                if (t->flags & PF_EXITING)
2416                        continue;
2417
2418                if (!has_pending_signals(&retarget, &t->blocked))
2419                        continue;
2420                /* Remove the signals this thread can handle. */
2421                sigandsets(&retarget, &retarget, &t->blocked);
2422
2423                if (!signal_pending(t))
2424                        signal_wake_up(t, 0);
2425
2426                if (sigisemptyset(&retarget))
2427                        break;
2428        }
2429}
2430
2431void exit_signals(struct task_struct *tsk)
2432{
2433        int group_stop = 0;
2434        sigset_t unblocked;
2435
2436        /*
2437         * @tsk is about to have PF_EXITING set - lock out users which
2438         * expect stable threadgroup.
2439         */
2440        threadgroup_change_begin(tsk);
2441
2442        if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2443                tsk->flags |= PF_EXITING;
2444                threadgroup_change_end(tsk);
2445                return;
2446        }
2447
2448        spin_lock_irq(&tsk->sighand->siglock);
2449        /*
2450         * From now this task is not visible for group-wide signals,
2451         * see wants_signal(), do_signal_stop().
2452         */
2453        tsk->flags |= PF_EXITING;
2454
2455        threadgroup_change_end(tsk);
2456
2457        if (!signal_pending(tsk))
2458                goto out;
2459
2460        unblocked = tsk->blocked;
2461        signotset(&unblocked);
2462        retarget_shared_pending(tsk, &unblocked);
2463
2464        if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2465            task_participate_group_stop(tsk))
2466                group_stop = CLD_STOPPED;
2467out:
2468        spin_unlock_irq(&tsk->sighand->siglock);
2469
2470        /*
2471         * If group stop has completed, deliver the notification.  This
2472         * should always go to the real parent of the group leader.
2473         */
2474        if (unlikely(group_stop)) {
2475                read_lock(&tasklist_lock);
2476                do_notify_parent_cldstop(tsk, false, group_stop);
2477                read_unlock(&tasklist_lock);
2478        }
2479}
2480
2481EXPORT_SYMBOL(recalc_sigpending);
2482EXPORT_SYMBOL_GPL(dequeue_signal);
2483EXPORT_SYMBOL(flush_signals);
2484EXPORT_SYMBOL(force_sig);
2485EXPORT_SYMBOL(send_sig);
2486EXPORT_SYMBOL(send_sig_info);
2487EXPORT_SYMBOL(sigprocmask);
2488EXPORT_SYMBOL(block_all_signals);
2489EXPORT_SYMBOL(unblock_all_signals);
2490
2491
2492/*
2493 * System call entry points.
2494 */
2495
2496/**
2497 *  sys_restart_syscall - restart a system call
2498 */
2499SYSCALL_DEFINE0(restart_syscall)
2500{
2501        struct restart_block *restart = &current_thread_info()->restart_block;
2502        return restart->fn(restart);
2503}
2504
2505long do_no_restart_syscall(struct restart_block *param)
2506{
2507        return -EINTR;
2508}
2509
2510static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2511{
2512        if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2513                sigset_t newblocked;
2514                /* A set of now blocked but previously unblocked signals. */
2515                sigandnsets(&newblocked, newset, &current->blocked);
2516                retarget_shared_pending(tsk, &newblocked);
2517        }
2518        tsk->blocked = *newset;
2519        recalc_sigpending();
2520}
2521
2522/**
2523 * set_current_blocked - change current->blocked mask
2524 * @newset: new mask
2525 *
2526 * It is wrong to change ->blocked directly, this helper should be used
2527 * to ensure the process can't miss a shared signal we are going to block.
2528 */
2529void set_current_blocked(const sigset_t *newset)
2530{
2531        struct task_struct *tsk = current;
2532
2533        spin_lock_irq(&tsk->sighand->siglock);
2534        __set_task_blocked(tsk, newset);
2535        spin_unlock_irq(&tsk->sighand->siglock);
2536}
2537
2538/*
2539 * This is also useful for kernel threads that want to temporarily
2540 * (or permanently) block certain signals.
2541 *
2542 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2543 * interface happily blocks "unblockable" signals like SIGKILL
2544 * and friends.
2545 */
2546int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2547{
2548        struct task_struct *tsk = current;
2549        sigset_t newset;
2550
2551        /* Lockless, only current can change ->blocked, never from irq */
2552        if (oldset)
2553                *oldset = tsk->blocked;
2554
2555        switch (how) {
2556        case SIG_BLOCK:
2557                sigorsets(&newset, &tsk->blocked, set);
2558                break;
2559        case SIG_UNBLOCK:
2560                sigandnsets(&newset, &tsk->blocked, set);
2561                break;
2562        case SIG_SETMASK:
2563                newset = *set;
2564                break;
2565        default:
2566                return -EINVAL;
2567        }
2568
2569        set_current_blocked(&newset);
2570        return 0;
2571}
2572
2573/**
2574 *  sys_rt_sigprocmask - change the list of currently blocked signals
2575 *  @how: whether to add, remove, or set signals
2576 *  @nset: stores pending signals
2577 *  @oset: previous value of signal mask if non-null
2578 *  @sigsetsize: size of sigset_t type
2579 */
2580SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
2581                sigset_t __user *, oset, size_t, sigsetsize)
2582{
2583        sigset_t old_set, new_set;
2584        int error;
2585
2586        /* XXX: Don't preclude handling different sized sigset_t's.  */
2587        if (sigsetsize != sizeof(sigset_t))
2588                return -EINVAL;
2589
2590        old_set = current->blocked;
2591
2592        if (nset) {
2593                if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
2594                        return -EFAULT;
2595                sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2596
2597                error = sigprocmask(how, &new_set, NULL);
2598                if (error)
2599                        return error;
2600        }
2601
2602        if (oset) {
2603                if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
2604                        return -EFAULT;
2605        }
2606
2607        return 0;
2608}
2609
2610long do_sigpending(void __user *set, unsigned long sigsetsize)
2611{
2612        long error = -EINVAL;
2613        sigset_t pending;
2614
2615        if (sigsetsize > sizeof(sigset_t))
2616                goto out;
2617
2618        spin_lock_irq(&current->sighand->siglock);
2619        sigorsets(&pending, &current->pending.signal,
2620                  &current->signal->shared_pending.signal);
2621        spin_unlock_irq(&current->sighand->siglock);
2622
2623        /* Outside the lock because only this thread touches it.  */
2624        sigandsets(&pending, &current->blocked, &pending);
2625
2626        error = -EFAULT;
2627        if (!copy_to_user(set, &pending, sigsetsize))
2628                error = 0;
2629
2630out:
2631        return error;
2632}
2633
2634/**
2635 *  sys_rt_sigpending - examine a pending signal that has been raised
2636 *                      while blocked
2637 *  @set: stores pending signals
2638 *  @sigsetsize: size of sigset_t type or larger
2639 */
2640SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2641{
2642        return do_sigpending(set, sigsetsize);
2643}
2644
2645#ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2646
2647int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2648{
2649        int err;
2650
2651        if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2652                return -EFAULT;
2653        if (from->si_code < 0)
2654                return __copy_to_user(to, from, sizeof(siginfo_t))
2655                        ? -EFAULT : 0;
2656        /*
2657         * If you change siginfo_t structure, please be sure
2658         * this code is fixed accordingly.
2659         * Please remember to update the signalfd_copyinfo() function
2660         * inside fs/signalfd.c too, in case siginfo_t changes.
2661         * It should never copy any pad contained in the structure
2662         * to avoid security leaks, but must copy the generic
2663         * 3 ints plus the relevant union member.
2664         */
2665        err = __put_user(from->si_signo, &to->si_signo);
2666        err |= __put_user(from->si_errno, &to->si_errno);
2667        err |= __put_user((short)from->si_code, &to->si_code);
2668        switch (from->si_code & __SI_MASK) {
2669        case __SI_KILL:
2670                err |= __put_user(from->si_pid, &to->si_pid);
2671                err |= __put_user(from->si_uid, &to->si_uid);
2672                break;
2673        case __SI_TIMER:
2674                 err |= __put_user(from->si_tid, &to->si_tid);
2675                 err |= __put_user(from->si_overrun, &to->si_overrun);
2676                 err |= __put_user(from->si_ptr, &to->si_ptr);
2677                break;
2678        case __SI_POLL:
2679                err |= __put_user(from->si_band, &to->si_band);
2680                err |= __put_user(from->si_fd, &to->si_fd);
2681                break;
2682        case __SI_FAULT:
2683                err |= __put_user(from->si_addr, &to->si_addr);
2684#ifdef __ARCH_SI_TRAPNO
2685                err |= __put_user(from->si_trapno, &to->si_trapno);
2686#endif
2687#ifdef BUS_MCEERR_AO
2688                /*
2689                 * Other callers might not initialize the si_lsb field,
2690                 * so check explicitly for the right codes here.
2691                 */
2692                if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2693                        err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2694#endif
2695                break;
2696        case __SI_CHLD:
2697                err |= __put_user(from->si_pid, &to->si_pid);
2698                err |= __put_user(from->si_uid, &to->si_uid);
2699                err |= __put_user(from->si_status, &to->si_status);
2700                err |= __put_user(from->si_utime, &to->si_utime);
2701                err |= __put_user(from->si_stime, &to->si_stime);
2702                break;
2703        case __SI_RT: /* This is not generated by the kernel as of now. */
2704        case __SI_MESGQ: /* But this is */
2705                err |= __put_user(from->si_pid, &to->si_pid);
2706                err |= __put_user(from->si_uid, &to->si_uid);
2707                err |= __put_user(from->si_ptr, &to->si_ptr);
2708                break;
2709        default: /* this is just in case for now ... */
2710                err |= __put_user(from->si_pid, &to->si_pid);
2711                err |= __put_user(from->si_uid, &to->si_uid);
2712                break;
2713        }
2714        return err;
2715}
2716
2717#endif
2718
2719/**
2720 *  do_sigtimedwait - wait for queued signals specified in @which
2721 *  @which: queued signals to wait for
2722 *  @info: if non-null, the signal's siginfo is returned here
2723 *  @ts: upper bound on process time suspension
2724 */
2725int do_sigtimedwait(const sigset_t *which, siginfo_t *info,
2726                        const struct timespec *ts)
2727{
2728        struct task_struct *tsk = current;
2729        long timeout = MAX_SCHEDULE_TIMEOUT;
2730        sigset_t mask = *which;
2731        int sig;
2732
2733        if (ts) {
2734                if (!timespec_valid(ts))
2735                        return -EINVAL;
2736                timeout = timespec_to_jiffies(ts);
2737                /*
2738                 * We can be close to the next tick, add another one
2739                 * to ensure we will wait at least the time asked for.
2740                 */
2741                if (ts->tv_sec || ts->tv_nsec)
2742                        timeout++;
2743        }
2744
2745        /*
2746         * Invert the set of allowed signals to get those we want to block.
2747         */
2748        sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2749        signotset(&mask);
2750
2751        spin_lock_irq(&tsk->sighand->siglock);
2752        sig = dequeue_signal(tsk, &mask, info);
2753        if (!sig && timeout) {
2754                /*
2755                 * None ready, temporarily unblock those we're interested
2756                 * while we are sleeping in so that we'll be awakened when
2757                 * they arrive. Unblocking is always fine, we can avoid
2758                 * set_current_blocked().
2759                 */
2760                tsk->real_blocked = tsk->blocked;
2761                sigandsets(&tsk->blocked, &tsk->blocked, &mask);
2762                recalc_sigpending();
2763                spin_unlock_irq(&tsk->sighand->siglock);
2764
2765                timeout = schedule_timeout_interruptible(timeout);
2766
2767                spin_lock_irq(&tsk->sighand->siglock);
2768                __set_task_blocked(tsk, &tsk->real_blocked);
2769                siginitset(&tsk->real_blocked, 0);
2770                sig = dequeue_signal(tsk, &mask, info);
2771        }
2772        spin_unlock_irq(&tsk->sighand->siglock);
2773
2774        if (sig)
2775                return sig;
2776        return timeout ? -EINTR : -EAGAIN;
2777}
2778
2779/**
2780 *  sys_rt_sigtimedwait - synchronously wait for queued signals specified
2781 *                      in @uthese
2782 *  @uthese: queued signals to wait for
2783 *  @uinfo: if non-null, the signal's siginfo is returned here
2784 *  @uts: upper bound on process time suspension
2785 *  @sigsetsize: size of sigset_t type
2786 */
2787SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2788                siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2789                size_t, sigsetsize)
2790{
2791        sigset_t these;
2792        struct timespec ts;
2793        siginfo_t info;
2794        int ret;
2795
2796        /* XXX: Don't preclude handling different sized sigset_t's.  */
2797        if (sigsetsize != sizeof(sigset_t))
2798                return -EINVAL;
2799
2800        if (copy_from_user(&these, uthese, sizeof(these)))
2801                return -EFAULT;
2802
2803        if (uts) {
2804                if (copy_from_user(&ts, uts, sizeof(ts)))
2805                        return -EFAULT;
2806        }
2807
2808        ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
2809
2810        if (ret > 0 && uinfo) {
2811                if (copy_siginfo_to_user(uinfo, &info))
2812                        ret = -EFAULT;
2813        }
2814
2815        return ret;
2816}
2817
2818/**
2819 *  sys_kill - send a signal to a process
2820 *  @pid: the PID of the process
2821 *  @sig: signal to be sent
2822 */
2823SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2824{
2825        struct siginfo info;
2826
2827        info.si_signo = sig;
2828        info.si_errno = 0;
2829        info.si_code = SI_USER;
2830        info.si_pid = task_tgid_vnr(current);
2831        info.si_uid = current_uid();
2832
2833        return kill_something_info(sig, &info, pid);
2834}
2835
2836static int
2837do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2838{
2839        struct task_struct *p;
2840        int error = -ESRCH;
2841
2842        rcu_read_lock();
2843        p = find_task_by_vpid(pid);
2844        if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2845                error = check_kill_permission(sig, info, p);
2846                /*
2847                 * The null signal is a permissions and process existence
2848                 * probe.  No signal is actually delivered.
2849                 */
2850                if (!error && sig) {
2851                        error = do_send_sig_info(sig, info, p, false);
2852                        /*
2853                         * If lock_task_sighand() failed we pretend the task
2854                         * dies after receiving the signal. The window is tiny,
2855                         * and the signal is private anyway.
2856                         */
2857                        if (unlikely(error == -ESRCH))
2858                                error = 0;
2859                }
2860        }
2861        rcu_read_unlock();
2862
2863        return error;
2864}
2865
2866static int do_tkill(pid_t tgid, pid_t pid, int sig)
2867{
2868        struct siginfo info;
2869
2870        info.si_signo = sig;
2871        info.si_errno = 0;
2872        info.si_code = SI_TKILL;
2873        info.si_pid = task_tgid_vnr(current);
2874        info.si_uid = current_uid();
2875
2876        return do_send_specific(tgid, pid, sig, &info);
2877}
2878
2879/**
2880 *  sys_tgkill - send signal to one specific thread
2881 *  @tgid: the thread group ID of the thread
2882 *  @pid: the PID of the thread
2883 *  @sig: signal to be sent
2884 *
2885 *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2886 *  exists but it's not belonging to the target process anymore. This
2887 *  method solves the problem of threads exiting and PIDs getting reused.
2888 */
2889SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2890{
2891        /* This is only valid for single tasks */
2892        if (pid <= 0 || tgid <= 0)
2893                return -EINVAL;
2894
2895        return do_tkill(tgid, pid, sig);
2896}
2897
2898/**
2899 *  sys_tkill - send signal to one specific task
2900 *  @pid: the PID of the task
2901 *  @sig: signal to be sent
2902 *
2903 *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2904 */
2905SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2906{
2907        /* This is only valid for single tasks */
2908        if (pid <= 0)
2909                return -EINVAL;
2910
2911        return do_tkill(0, pid, sig);
2912}
2913
2914/**
2915 *  sys_rt_sigqueueinfo - send signal information to a signal
2916 *  @pid: the PID of the thread
2917 *  @sig: signal to be sent
2918 *  @uinfo: signal info to be sent
2919 */
2920SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2921                siginfo_t __user *, uinfo)
2922{
2923        siginfo_t info;
2924
2925        if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2926                return -EFAULT;
2927
2928        /* Not even root can pretend to send signals from the kernel.
2929         * Nor can they impersonate a kill()/tgkill(), which adds source info.
2930         */
2931        if (info.si_code >= 0 || info.si_code == SI_TKILL) {
2932                /* We used to allow any < 0 si_code */
2933                WARN_ON_ONCE(info.si_code < 0);
2934                return -EPERM;
2935        }
2936        info.si_signo = sig;
2937
2938        /* POSIX.1b doesn't mention process groups.  */
2939        return kill_proc_info(sig, &info, pid);
2940}
2941
2942long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2943{
2944        /* This is only valid for single tasks */
2945        if (pid <= 0 || tgid <= 0)
2946                return -EINVAL;
2947
2948        /* Not even root can pretend to send signals from the kernel.
2949         * Nor can they impersonate a kill()/tgkill(), which adds source info.
2950         */
2951        if (info->si_code >= 0 || info->si_code == SI_TKILL) {
2952                /* We used to allow any < 0 si_code */
2953                WARN_ON_ONCE(info->si_code < 0);
2954                return -EPERM;
2955        }
2956        info->si_signo = sig;
2957
2958        return do_send_specific(tgid, pid, sig, info);
2959}
2960
2961SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2962                siginfo_t __user *, uinfo)
2963{
2964        siginfo_t info;
2965
2966        if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2967                return -EFAULT;
2968
2969        return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2970}
2971
2972int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2973{
2974        struct task_struct *t = current;
2975        struct k_sigaction *k;
2976        sigset_t mask;
2977
2978        if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2979                return -EINVAL;
2980
2981        k = &t->sighand->action[sig-1];
2982
2983        spin_lock_irq(&current->sighand->siglock);
2984        if (oact)
2985                *oact = *k;
2986
2987        if (act) {
2988                sigdelsetmask(&act->sa.sa_mask,
2989                              sigmask(SIGKILL) | sigmask(SIGSTOP));
2990                *k = *act;
2991                /*
2992                 * POSIX 3.3.1.3:
2993                 *  "Setting a signal action to SIG_IGN for a signal that is
2994                 *   pending shall cause the pending signal to be discarded,
2995                 *   whether or not it is blocked."
2996                 *
2997                 *  "Setting a signal action to SIG_DFL for a signal that is
2998                 *   pending and whose default action is to ignore the signal
2999                 *   (for example, SIGCHLD), shall cause the pending signal to
3000                 *   be discarded, whether or not it is blocked"
3001                 */
3002                if (sig_handler_ignored(sig_handler(t, sig), sig)) {
3003                        sigemptyset(&mask);
3004                        sigaddset(&mask, sig);
3005                        rm_from_queue_full(&mask, &t->signal->shared_pending);
3006                        do {
3007                                rm_from_queue_full(&mask, &t->pending);
3008                                t = next_thread(t);
3009                        } while (t != current);
3010                }
3011        }
3012
3013        spin_unlock_irq(&current->sighand->siglock);
3014        return 0;
3015}
3016
3017int 
3018do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
3019{
3020        stack_t oss;
3021        int error;
3022
3023        oss.ss_sp = (void __user *) current->sas_ss_sp;
3024        oss.ss_size = current->sas_ss_size;
3025        oss.ss_flags = sas_ss_flags(sp);
3026
3027        if (uss) {
3028                void __user *ss_sp;
3029                size_t ss_size;
3030                int ss_flags;
3031
3032                error = -EFAULT;
3033                if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
3034                        goto out;
3035                error = __get_user(ss_sp, &uss->ss_sp) |
3036                        __get_user(ss_flags, &uss->ss_flags) |
3037                        __get_user(ss_size, &uss->ss_size);
3038                if (error)
3039                        goto out;
3040
3041                error = -EPERM;
3042                if (on_sig_stack(sp))
3043                        goto out;
3044
3045                error = -EINVAL;
3046                /*
3047                 * Note - this code used to test ss_flags incorrectly:
3048                 *        old code may have been written using ss_flags==0
3049                 *        to mean ss_flags==SS_ONSTACK (as this was the only
3050                 *        way that worked) - this fix preserves that older
3051                 *        mechanism.
3052                 */
3053                if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
3054                        goto out;
3055
3056                if (ss_flags == SS_DISABLE) {
3057                        ss_size = 0;
3058                        ss_sp = NULL;
3059                } else {
3060                        error = -ENOMEM;
3061                        if (ss_size < MINSIGSTKSZ)
3062                                goto out;
3063                }
3064
3065                current->sas_ss_sp = (unsigned long) ss_sp;
3066                current->sas_ss_size = ss_size;
3067        }
3068
3069        error = 0;
3070        if (uoss) {
3071                error = -EFAULT;
3072                if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
3073                        goto out;
3074                error = __put_user(oss.ss_sp, &uoss->ss_sp) |
3075                        __put_user(oss.ss_size, &uoss->ss_size) |
3076                        __put_user(oss.ss_flags, &uoss->ss_flags);
3077        }
3078
3079out:
3080        return error;
3081}
3082
3083#ifdef __ARCH_WANT_SYS_SIGPENDING
3084
3085/**
3086 *  sys_sigpending - examine pending signals
3087 *  @set: where mask of pending signal is returned
3088 */
3089SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
3090{
3091        return do_sigpending(set, sizeof(*set));
3092}
3093
3094#endif
3095
3096#ifdef __ARCH_WANT_SYS_SIGPROCMASK
3097/**
3098 *  sys_sigprocmask - examine and change blocked signals
3099 *  @how: whether to add, remove, or set signals
3100 *  @nset: signals to add or remove (if non-null)
3101 *  @oset: previous value of signal mask if non-null
3102 *
3103 * Some platforms have their own version with special arguments;
3104 * others support only sys_rt_sigprocmask.
3105 */
3106
3107SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
3108                old_sigset_t __user *, oset)
3109{
3110        old_sigset_t old_set, new_set;
3111        sigset_t new_blocked;
3112
3113        old_set = current->blocked.sig[0];
3114
3115        if (nset) {
3116                if (copy_from_user(&new_set, nset, sizeof(*nset)))
3117                        return -EFAULT;
3118                new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
3119
3120                new_blocked = current->blocked;
3121
3122                switch (how) {
3123                case SIG_BLOCK:
3124                        sigaddsetmask(&new_blocked, new_set);
3125                        break;
3126                case SIG_UNBLOCK:
3127                        sigdelsetmask(&new_blocked, new_set);
3128                        break;
3129                case SIG_SETMASK:
3130                        new_blocked.sig[0] = new_set;
3131                        break;
3132                default:
3133                        return -EINVAL;
3134                }
3135
3136                set_current_blocked(&new_blocked);
3137        }
3138
3139        if (oset) {
3140                if (copy_to_user(oset, &old_set, sizeof(*oset)))
3141                        return -EFAULT;
3142        }
3143
3144        return 0;
3145}
3146#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
3147
3148#ifdef __ARCH_WANT_SYS_RT_SIGACTION
3149/**
3150 *  sys_rt_sigaction - alter an action taken by a process
3151 *  @sig: signal to be sent
3152 *  @act: new sigaction
3153 *  @oact: used to save the previous sigaction
3154 *  @sigsetsize: size of sigset_t type
3155 */
3156SYSCALL_DEFINE4(rt_sigaction, int, sig,
3157                const struct sigaction __user *, act,
3158                struct sigaction __user *, oact,
3159                size_t, sigsetsize)
3160{
3161        struct k_sigaction new_sa, old_sa;
3162        int ret = -EINVAL;
3163
3164        /* XXX: Don't preclude handling different sized sigset_t's.  */
3165        if (sigsetsize != sizeof(sigset_t))
3166                goto out;
3167
3168        if (act) {
3169                if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
3170                        return -EFAULT;
3171        }
3172
3173        ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
3174
3175        if (!ret && oact) {
3176                if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
3177                        return -EFAULT;
3178        }
3179out:
3180        return ret;
3181}
3182#endif /* __ARCH_WANT_SYS_RT_SIGACTION */
3183
3184#ifdef __ARCH_WANT_SYS_SGETMASK
3185
3186/*
3187 * For backwards compatibility.  Functionality superseded by sigprocmask.
3188 */
3189SYSCALL_DEFINE0(sgetmask)
3190{
3191        /* SMP safe */
3192        return current->blocked.sig[0];
3193}
3194
3195SYSCALL_DEFINE1(ssetmask, int, newmask)
3196{
3197        int old = current->blocked.sig[0];
3198        sigset_t newset;
3199
3200        siginitset(&newset, newmask & ~(sigmask(SIGKILL) | sigmask(SIGSTOP)));
3201        set_current_blocked(&newset);
3202
3203        return old;
3204}
3205#endif /* __ARCH_WANT_SGETMASK */
3206
3207#ifdef __ARCH_WANT_SYS_SIGNAL
3208/*
3209 * For backwards compatibility.  Functionality superseded by sigaction.
3210 */
3211SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
3212{
3213        struct k_sigaction new_sa, old_sa;
3214        int ret;
3215
3216        new_sa.sa.sa_handler = handler;
3217        new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
3218        sigemptyset(&new_sa.sa.sa_mask);
3219
3220        ret = do_sigaction(sig, &new_sa, &old_sa);
3221
3222        return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
3223}
3224#endif /* __ARCH_WANT_SYS_SIGNAL */
3225
3226#ifdef __ARCH_WANT_SYS_PAUSE
3227
3228SYSCALL_DEFINE0(pause)
3229{
3230        while (!signal_pending(current)) {
3231                current->state = TASK_INTERRUPTIBLE;
3232                schedule();
3233        }
3234        return -ERESTARTNOHAND;
3235}
3236
3237#endif
3238
3239#ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
3240/**
3241 *  sys_rt_sigsuspend - replace the signal mask for a value with the
3242 *      @unewset value until a signal is received
3243 *  @unewset: new signal mask value
3244 *  @sigsetsize: size of sigset_t type
3245 */
3246SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
3247{
3248        sigset_t newset;
3249
3250        /* XXX: Don't preclude handling different sized sigset_t's.  */
3251        if (sigsetsize != sizeof(sigset_t))
3252                return -EINVAL;
3253
3254        if (copy_from_user(&newset, unewset, sizeof(newset)))
3255                return -EFAULT;
3256        sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3257
3258        current->saved_sigmask = current->blocked;
3259        set_current_blocked(&newset);
3260
3261        current->state = TASK_INTERRUPTIBLE;
3262        schedule();
3263        set_restore_sigmask();
3264        return -ERESTARTNOHAND;
3265}
3266#endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
3267
3268__attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
3269{
3270        return NULL;
3271}
3272
3273void __init signals_init(void)
3274{
3275        sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
3276}
3277
3278#ifdef CONFIG_KGDB_KDB
3279#include <linux/kdb.h>
3280/*
3281 * kdb_send_sig_info - Allows kdb to send signals without exposing
3282 * signal internals.  This function checks if the required locks are
3283 * available before calling the main signal code, to avoid kdb
3284 * deadlocks.
3285 */
3286void
3287kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
3288{
3289        static struct task_struct *kdb_prev_t;
3290        int sig, new_t;
3291        if (!spin_trylock(&t->sighand->siglock)) {
3292                kdb_printf("Can't do kill command now.\n"
3293                           "The sigmask lock is held somewhere else in "
3294                           "kernel, try again later\n");
3295                return;
3296        }
3297        spin_unlock(&t->sighand->siglock);
3298        new_t = kdb_prev_t != t;
3299        kdb_prev_t = t;
3300        if (t->state != TASK_RUNNING && new_t) {
3301                kdb_printf("Process is not RUNNING, sending a signal from "
3302                           "kdb risks deadlock\n"
3303                           "on the run queue locks. "
3304                           "The signal has _not_ been sent.\n"
3305                           "Reissue the kill command if you want to risk "
3306                           "the deadlock.\n");
3307                return;
3308        }
3309        sig = info->si_signo;
3310        if (send_sig_info(sig, info, t))
3311                kdb_printf("Fail to deliver Signal %d to process %d.\n",
3312                           sig, t->pid);
3313        else
3314                kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
3315}
3316#endif  /* CONFIG_KGDB_KDB */
3317