1/* 2 * linux/kernel/exit.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7#include <linux/mm.h> 8#include <linux/slab.h> 9#include <linux/interrupt.h> 10#include <linux/module.h> 11#include <linux/capability.h> 12#include <linux/completion.h> 13#include <linux/personality.h> 14#include <linux/tty.h> 15#include <linux/iocontext.h> 16#include <linux/key.h> 17#include <linux/security.h> 18#include <linux/cpu.h> 19#include <linux/acct.h> 20#include <linux/tsacct_kern.h> 21#include <linux/file.h> 22#include <linux/fdtable.h> 23#include <linux/binfmts.h> 24#include <linux/nsproxy.h> 25#include <linux/pid_namespace.h> 26#include <linux/ptrace.h> 27#include <linux/profile.h> 28#include <linux/mount.h> 29#include <linux/proc_fs.h> 30#include <linux/kthread.h> 31#include <linux/mempolicy.h> 32#include <linux/taskstats_kern.h> 33#include <linux/delayacct.h> 34#include <linux/freezer.h> 35#include <linux/cgroup.h> 36#include <linux/syscalls.h> 37#include <linux/signal.h> 38#include <linux/posix-timers.h> 39#include <linux/cn_proc.h> 40#include <linux/mutex.h> 41#include <linux/futex.h> 42#include <linux/pipe_fs_i.h> 43#include <linux/audit.h> /* for audit_free() */ 44#include <linux/resource.h> 45#include <linux/blkdev.h> 46#include <linux/task_io_accounting_ops.h> 47#include <linux/tracehook.h> 48#include <linux/fs_struct.h> 49#include <linux/init_task.h> 50#include <linux/perf_event.h> 51#include <trace/events/sched.h> 52#include <linux/hw_breakpoint.h> 53 54#include <asm/uaccess.h> 55#include <asm/unistd.h> 56#include <asm/pgtable.h> 57#include <asm/mmu_context.h> 58 59static void exit_mm(struct task_struct * tsk); 60 61static void __unhash_process(struct task_struct *p, bool group_dead) 62{ 63 nr_threads--; 64 detach_pid(p, PIDTYPE_PID); 65 if (group_dead) { 66 detach_pid(p, PIDTYPE_PGID); 67 detach_pid(p, PIDTYPE_SID); 68 69 list_del_rcu(&p->tasks); 70 list_del_init(&p->sibling); 71 __get_cpu_var(process_counts)--; 72 } 73 list_del_rcu(&p->thread_group); 74} 75 76/* 77 * This function expects the tasklist_lock write-locked. 78 */ 79static void __exit_signal(struct task_struct *tsk) 80{ 81 struct signal_struct *sig = tsk->signal; 82 bool group_dead = thread_group_leader(tsk); 83 struct sighand_struct *sighand; 84 struct tty_struct *uninitialized_var(tty); 85 86 sighand = rcu_dereference_check(tsk->sighand, 87 rcu_read_lock_held() || 88 lockdep_tasklist_lock_is_held()); 89 spin_lock(&sighand->siglock); 90 91 posix_cpu_timers_exit(tsk); 92 if (group_dead) { 93 posix_cpu_timers_exit_group(tsk); 94 tty = sig->tty; 95 sig->tty = NULL; 96 } else { 97 if (unlikely(has_group_leader_pid(tsk))) 98 posix_cpu_timers_exit_group(tsk); 99 100 /* 101 * If there is any task waiting for the group exit 102 * then notify it: 103 */ 104 if (sig->notify_count > 0 && !--sig->notify_count) 105 wake_up_process(sig->group_exit_task); 106 107 if (tsk == sig->curr_target) 108 sig->curr_target = next_thread(tsk); 109 /* 110 * Accumulate here the counters for all threads but the 111 * group leader as they die, so they can be added into 112 * the process-wide totals when those are taken. 113 * The group leader stays around as a zombie as long 114 * as there are other threads. When it gets reaped, 115 * the exit.c code will add its counts into these totals. 116 * We won't ever get here for the group leader, since it 117 * will have been the last reference on the signal_struct. 118 */ 119 sig->utime = cputime_add(sig->utime, tsk->utime); 120 sig->stime = cputime_add(sig->stime, tsk->stime); 121 sig->gtime = cputime_add(sig->gtime, tsk->gtime); 122 sig->min_flt += tsk->min_flt; 123 sig->maj_flt += tsk->maj_flt; 124 sig->nvcsw += tsk->nvcsw; 125 sig->nivcsw += tsk->nivcsw; 126 sig->inblock += task_io_get_inblock(tsk); 127 sig->oublock += task_io_get_oublock(tsk); 128 task_io_accounting_add(&sig->ioac, &tsk->ioac); 129 sig->sum_sched_runtime += tsk->se.sum_exec_runtime; 130 } 131 132 sig->nr_threads--; 133 __unhash_process(tsk, group_dead); 134 135 /* 136 * Do this under ->siglock, we can race with another thread 137 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. 138 */ 139 flush_sigqueue(&tsk->pending); 140 tsk->sighand = NULL; 141 spin_unlock(&sighand->siglock); 142 143 __cleanup_sighand(sighand); 144 clear_tsk_thread_flag(tsk,TIF_SIGPENDING); 145 if (group_dead) { 146 flush_sigqueue(&sig->shared_pending); 147 tty_kref_put(tty); 148 } 149} 150 151static void delayed_put_task_struct(struct rcu_head *rhp) 152{ 153 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 154 155#ifdef CONFIG_PERF_EVENTS 156 WARN_ON_ONCE(tsk->perf_event_ctxp); 157#endif 158 trace_sched_process_free(tsk); 159 put_task_struct(tsk); 160} 161 162 163void release_task(struct task_struct * p) 164{ 165 struct task_struct *leader; 166 int zap_leader; 167repeat: 168 tracehook_prepare_release_task(p); 169 /* don't need to get the RCU readlock here - the process is dead and 170 * can't be modifying its own credentials. But shut RCU-lockdep up */ 171 rcu_read_lock(); 172 atomic_dec(&__task_cred(p)->user->processes); 173 rcu_read_unlock(); 174 175 proc_flush_task(p); 176 177 write_lock_irq(&tasklist_lock); 178 tracehook_finish_release_task(p); 179 __exit_signal(p); 180 181 /* 182 * If we are the last non-leader member of the thread 183 * group, and the leader is zombie, then notify the 184 * group leader's parent process. (if it wants notification.) 185 */ 186 zap_leader = 0; 187 leader = p->group_leader; 188 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) { 189 BUG_ON(task_detached(leader)); 190 do_notify_parent(leader, leader->exit_signal); 191 /* 192 * If we were the last child thread and the leader has 193 * exited already, and the leader's parent ignores SIGCHLD, 194 * then we are the one who should release the leader. 195 * 196 * do_notify_parent() will have marked it self-reaping in 197 * that case. 198 */ 199 zap_leader = task_detached(leader); 200 201 /* 202 * This maintains the invariant that release_task() 203 * only runs on a task in EXIT_DEAD, just for sanity. 204 */ 205 if (zap_leader) 206 leader->exit_state = EXIT_DEAD; 207 } 208 209 write_unlock_irq(&tasklist_lock); 210 release_thread(p); 211 call_rcu(&p->rcu, delayed_put_task_struct); 212 213 p = leader; 214 if (unlikely(zap_leader)) 215 goto repeat; 216} 217 218/* 219 * This checks not only the pgrp, but falls back on the pid if no 220 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly 221 * without this... 222 * 223 * The caller must hold rcu lock or the tasklist lock. 224 */ 225struct pid *session_of_pgrp(struct pid *pgrp) 226{ 227 struct task_struct *p; 228 struct pid *sid = NULL; 229 230 p = pid_task(pgrp, PIDTYPE_PGID); 231 if (p == NULL) 232 p = pid_task(pgrp, PIDTYPE_PID); 233 if (p != NULL) 234 sid = task_session(p); 235 236 return sid; 237} 238 239/* 240 * Determine if a process group is "orphaned", according to the POSIX 241 * definition in 2.2.2.52. Orphaned process groups are not to be affected 242 * by terminal-generated stop signals. Newly orphaned process groups are 243 * to receive a SIGHUP and a SIGCONT. 244 * 245 * "I ask you, have you ever known what it is to be an orphan?" 246 */ 247static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task) 248{ 249 struct task_struct *p; 250 251 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 252 if ((p == ignored_task) || 253 (p->exit_state && thread_group_empty(p)) || 254 is_global_init(p->real_parent)) 255 continue; 256 257 if (task_pgrp(p->real_parent) != pgrp && 258 task_session(p->real_parent) == task_session(p)) 259 return 0; 260 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 261 262 return 1; 263} 264 265int is_current_pgrp_orphaned(void) 266{ 267 int retval; 268 269 read_lock(&tasklist_lock); 270 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); 271 read_unlock(&tasklist_lock); 272 273 return retval; 274} 275 276static int has_stopped_jobs(struct pid *pgrp) 277{ 278 int retval = 0; 279 struct task_struct *p; 280 281 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 282 if (!task_is_stopped(p)) 283 continue; 284 retval = 1; 285 break; 286 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 287 return retval; 288} 289 290/* 291 * Check to see if any process groups have become orphaned as 292 * a result of our exiting, and if they have any stopped jobs, 293 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 294 */ 295static void 296kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) 297{ 298 struct pid *pgrp = task_pgrp(tsk); 299 struct task_struct *ignored_task = tsk; 300 301 if (!parent) 302 /* exit: our father is in a different pgrp than 303 * we are and we were the only connection outside. 304 */ 305 parent = tsk->real_parent; 306 else 307 /* reparent: our child is in a different pgrp than 308 * we are, and it was the only connection outside. 309 */ 310 ignored_task = NULL; 311 312 if (task_pgrp(parent) != pgrp && 313 task_session(parent) == task_session(tsk) && 314 will_become_orphaned_pgrp(pgrp, ignored_task) && 315 has_stopped_jobs(pgrp)) { 316 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); 317 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); 318 } 319} 320 321/** 322 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd 323 * 324 * If a kernel thread is launched as a result of a system call, or if 325 * it ever exits, it should generally reparent itself to kthreadd so it 326 * isn't in the way of other processes and is correctly cleaned up on exit. 327 * 328 * The various task state such as scheduling policy and priority may have 329 * been inherited from a user process, so we reset them to sane values here. 330 * 331 * NOTE that reparent_to_kthreadd() gives the caller full capabilities. 332 */ 333static void reparent_to_kthreadd(void) 334{ 335 write_lock_irq(&tasklist_lock); 336 337 ptrace_unlink(current); 338 /* Reparent to init */ 339 current->real_parent = current->parent = kthreadd_task; 340 list_move_tail(¤t->sibling, ¤t->real_parent->children); 341 342 /* Set the exit signal to SIGCHLD so we signal init on exit */ 343 current->exit_signal = SIGCHLD; 344 345 if (task_nice(current) < 0) 346 set_user_nice(current, 0); 347 /* cpus_allowed? */ 348 /* rt_priority? */ 349 /* signals? */ 350 memcpy(current->signal->rlim, init_task.signal->rlim, 351 sizeof(current->signal->rlim)); 352 353 atomic_inc(&init_cred.usage); 354 commit_creds(&init_cred); 355 write_unlock_irq(&tasklist_lock); 356} 357 358void __set_special_pids(struct pid *pid) 359{ 360 struct task_struct *curr = current->group_leader; 361 362 if (task_session(curr) != pid) 363 change_pid(curr, PIDTYPE_SID, pid); 364 365 if (task_pgrp(curr) != pid) 366 change_pid(curr, PIDTYPE_PGID, pid); 367} 368 369static void set_special_pids(struct pid *pid) 370{ 371 write_lock_irq(&tasklist_lock); 372 __set_special_pids(pid); 373 write_unlock_irq(&tasklist_lock); 374} 375 376/* 377 * Let kernel threads use this to say that they allow a certain signal. 378 * Must not be used if kthread was cloned with CLONE_SIGHAND. 379 */ 380int allow_signal(int sig) 381{ 382 if (!valid_signal(sig) || sig < 1) 383 return -EINVAL; 384 385 spin_lock_irq(¤t->sighand->siglock); 386 /* This is only needed for daemonize()'ed kthreads */ 387 sigdelset(¤t->blocked, sig); 388 /* 389 * Kernel threads handle their own signals. Let the signal code 390 * know it'll be handled, so that they don't get converted to 391 * SIGKILL or just silently dropped. 392 */ 393 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2; 394 recalc_sigpending(); 395 spin_unlock_irq(¤t->sighand->siglock); 396 return 0; 397} 398 399EXPORT_SYMBOL(allow_signal); 400 401int disallow_signal(int sig) 402{ 403 if (!valid_signal(sig) || sig < 1) 404 return -EINVAL; 405 406 spin_lock_irq(¤t->sighand->siglock); 407 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN; 408 recalc_sigpending(); 409 spin_unlock_irq(¤t->sighand->siglock); 410 return 0; 411} 412 413EXPORT_SYMBOL(disallow_signal); 414 415/* 416 * Put all the gunge required to become a kernel thread without 417 * attached user resources in one place where it belongs. 418 */ 419 420void daemonize(const char *name, ...) 421{ 422 va_list args; 423 sigset_t blocked; 424 425 va_start(args, name); 426 vsnprintf(current->comm, sizeof(current->comm), name, args); 427 va_end(args); 428 429 /* 430 * If we were started as result of loading a module, close all of the 431 * user space pages. We don't need them, and if we didn't close them 432 * they would be locked into memory. 433 */ 434 exit_mm(current); 435 /* 436 * We don't want to have TIF_FREEZE set if the system-wide hibernation 437 * or suspend transition begins right now. 438 */ 439 current->flags |= (PF_NOFREEZE | PF_KTHREAD); 440 441 if (current->nsproxy != &init_nsproxy) { 442 get_nsproxy(&init_nsproxy); 443 switch_task_namespaces(current, &init_nsproxy); 444 } 445 set_special_pids(&init_struct_pid); 446 proc_clear_tty(current); 447 448 /* Block and flush all signals */ 449 sigfillset(&blocked); 450 sigprocmask(SIG_BLOCK, &blocked, NULL); 451 flush_signals(current); 452 453 /* Become as one with the init task */ 454 455 daemonize_fs_struct(); 456 exit_files(current); 457 current->files = init_task.files; 458 atomic_inc(¤t->files->count); 459 460 reparent_to_kthreadd(); 461} 462 463EXPORT_SYMBOL(daemonize); 464 465static void close_files(struct files_struct * files) 466{ 467 int i, j; 468 struct fdtable *fdt; 469 470 j = 0; 471 472 /* 473 * It is safe to dereference the fd table without RCU or 474 * ->file_lock because this is the last reference to the 475 * files structure. But use RCU to shut RCU-lockdep up. 476 */ 477 rcu_read_lock(); 478 fdt = files_fdtable(files); 479 rcu_read_unlock(); 480 for (;;) { 481 unsigned long set; 482 i = j * __NFDBITS; 483 if (i >= fdt->max_fds) 484 break; 485 set = fdt->open_fds->fds_bits[j++]; 486 while (set) { 487 if (set & 1) { 488 struct file * file = xchg(&fdt->fd[i], NULL); 489 if (file) { 490 filp_close(file, files); 491 cond_resched(); 492 } 493 } 494 i++; 495 set >>= 1; 496 } 497 } 498} 499 500struct files_struct *get_files_struct(struct task_struct *task) 501{ 502 struct files_struct *files; 503 504 task_lock(task); 505 files = task->files; 506 if (files) 507 atomic_inc(&files->count); 508 task_unlock(task); 509 510 return files; 511} 512 513void put_files_struct(struct files_struct *files) 514{ 515 struct fdtable *fdt; 516 517 if (atomic_dec_and_test(&files->count)) { 518 close_files(files); 519 /* 520 * Free the fd and fdset arrays if we expanded them. 521 * If the fdtable was embedded, pass files for freeing 522 * at the end of the RCU grace period. Otherwise, 523 * you can free files immediately. 524 */ 525 rcu_read_lock(); 526 fdt = files_fdtable(files); 527 if (fdt != &files->fdtab) 528 kmem_cache_free(files_cachep, files); 529 free_fdtable(fdt); 530 rcu_read_unlock(); 531 } 532} 533 534void reset_files_struct(struct files_struct *files) 535{ 536 struct task_struct *tsk = current; 537 struct files_struct *old; 538 539 old = tsk->files; 540 task_lock(tsk); 541 tsk->files = files; 542 task_unlock(tsk); 543 put_files_struct(old); 544} 545 546void exit_files(struct task_struct *tsk) 547{ 548 struct files_struct * files = tsk->files; 549 550 if (files) { 551 task_lock(tsk); 552 tsk->files = NULL; 553 task_unlock(tsk); 554 put_files_struct(files); 555 } 556} 557 558#ifdef CONFIG_MM_OWNER 559/* 560 * Task p is exiting and it owned mm, lets find a new owner for it 561 */ 562static inline int 563mm_need_new_owner(struct mm_struct *mm, struct task_struct *p) 564{ 565 /* 566 * If there are other users of the mm and the owner (us) is exiting 567 * we need to find a new owner to take on the responsibility. 568 */ 569 if (atomic_read(&mm->mm_users) <= 1) 570 return 0; 571 if (mm->owner != p) 572 return 0; 573 return 1; 574} 575 576void mm_update_next_owner(struct mm_struct *mm) 577{ 578 struct task_struct *c, *g, *p = current; 579 580retry: 581 if (!mm_need_new_owner(mm, p)) 582 return; 583 584 read_lock(&tasklist_lock); 585 /* 586 * Search in the children 587 */ 588 list_for_each_entry(c, &p->children, sibling) { 589 if (c->mm == mm) 590 goto assign_new_owner; 591 } 592 593 /* 594 * Search in the siblings 595 */ 596 list_for_each_entry(c, &p->real_parent->children, sibling) { 597 if (c->mm == mm) 598 goto assign_new_owner; 599 } 600 601 /* 602 * Search through everything else. We should not get 603 * here often 604 */ 605 do_each_thread(g, c) { 606 if (c->mm == mm) 607 goto assign_new_owner; 608 } while_each_thread(g, c); 609 610 read_unlock(&tasklist_lock); 611 /* 612 * We found no owner yet mm_users > 1: this implies that we are 613 * most likely racing with swapoff (try_to_unuse()) or /proc or 614 * ptrace or page migration (get_task_mm()). Mark owner as NULL. 615 */ 616 mm->owner = NULL; 617 return; 618 619assign_new_owner: 620 BUG_ON(c == p); 621 get_task_struct(c); 622 /* 623 * The task_lock protects c->mm from changing. 624 * We always want mm->owner->mm == mm 625 */ 626 task_lock(c); 627 /* 628 * Delay read_unlock() till we have the task_lock() 629 * to ensure that c does not slip away underneath us 630 */ 631 read_unlock(&tasklist_lock); 632 if (c->mm != mm) { 633 task_unlock(c); 634 put_task_struct(c); 635 goto retry; 636 } 637 mm->owner = c; 638 task_unlock(c); 639 put_task_struct(c); 640} 641#endif /* CONFIG_MM_OWNER */ 642 643/* 644 * Turn us into a lazy TLB process if we 645 * aren't already.. 646 */ 647static void exit_mm(struct task_struct * tsk) 648{ 649 struct mm_struct *mm = tsk->mm; 650 struct core_state *core_state; 651 652 mm_release(tsk, mm); 653 if (!mm) 654 return; 655 /* 656 * Serialize with any possible pending coredump. 657 * We must hold mmap_sem around checking core_state 658 * and clearing tsk->mm. The core-inducing thread 659 * will increment ->nr_threads for each thread in the 660 * group with ->mm != NULL. 661 */ 662 down_read(&mm->mmap_sem); 663 core_state = mm->core_state; 664 if (core_state) { 665 struct core_thread self; 666 up_read(&mm->mmap_sem); 667 668 self.task = tsk; 669 self.next = xchg(&core_state->dumper.next, &self); 670 /* 671 * Implies mb(), the result of xchg() must be visible 672 * to core_state->dumper. 673 */ 674 if (atomic_dec_and_test(&core_state->nr_threads)) 675 complete(&core_state->startup); 676 677 for (;;) { 678 set_task_state(tsk, TASK_UNINTERRUPTIBLE); 679 if (!self.task) /* see coredump_finish() */ 680 break; 681 schedule(); 682 } 683 __set_task_state(tsk, TASK_RUNNING); 684 down_read(&mm->mmap_sem); 685 } 686 atomic_inc(&mm->mm_count); 687 BUG_ON(mm != tsk->active_mm); 688 /* more a memory barrier than a real lock */ 689 task_lock(tsk); 690 tsk->mm = NULL; 691 up_read(&mm->mmap_sem); 692 enter_lazy_tlb(mm, current); 693 /* We don't want this task to be frozen prematurely */ 694 clear_freeze_flag(tsk); 695 task_unlock(tsk); 696 mm_update_next_owner(mm); 697 mmput(mm); 698} 699 700/* 701 * When we die, we re-parent all our children. 702 * Try to give them to another thread in our thread 703 * group, and if no such member exists, give it to 704 * the child reaper process (ie "init") in our pid 705 * space. 706 */ 707static struct task_struct *find_new_reaper(struct task_struct *father) 708{ 709 struct pid_namespace *pid_ns = task_active_pid_ns(father); 710 struct task_struct *thread; 711 712 thread = father; 713 while_each_thread(father, thread) { 714 if (thread->flags & PF_EXITING) 715 continue; 716 if (unlikely(pid_ns->child_reaper == father)) 717 pid_ns->child_reaper = thread; 718 return thread; 719 } 720 721 if (unlikely(pid_ns->child_reaper == father)) { 722 write_unlock_irq(&tasklist_lock); 723 if (unlikely(pid_ns == &init_pid_ns)) 724 panic("Attempted to kill init!"); 725 726 zap_pid_ns_processes(pid_ns); 727 write_lock_irq(&tasklist_lock); 728 /* 729 * We can not clear ->child_reaper or leave it alone. 730 * There may by stealth EXIT_DEAD tasks on ->children, 731 * forget_original_parent() must move them somewhere. 732 */ 733 pid_ns->child_reaper = init_pid_ns.child_reaper; 734 } 735 736 return pid_ns->child_reaper; 737} 738 739/* 740* Any that need to be release_task'd are put on the @dead list. 741 */ 742static void reparent_leader(struct task_struct *father, struct task_struct *p, 743 struct list_head *dead) 744{ 745 list_move_tail(&p->sibling, &p->real_parent->children); 746 747 if (task_detached(p)) 748 return; 749 /* 750 * If this is a threaded reparent there is no need to 751 * notify anyone anything has happened. 752 */ 753 if (same_thread_group(p->real_parent, father)) 754 return; 755 756 /* We don't want people slaying init. */ 757 p->exit_signal = SIGCHLD; 758 759 /* If it has exited notify the new parent about this child's death. */ 760 if (!task_ptrace(p) && 761 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { 762 do_notify_parent(p, p->exit_signal); 763 if (task_detached(p)) { 764 p->exit_state = EXIT_DEAD; 765 list_move_tail(&p->sibling, dead); 766 } 767 } 768 769 kill_orphaned_pgrp(p, father); 770} 771 772static void forget_original_parent(struct task_struct *father) 773{ 774 struct task_struct *p, *n, *reaper; 775 LIST_HEAD(dead_children); 776 777 write_lock_irq(&tasklist_lock); 778 /* 779 * Note that exit_ptrace() and find_new_reaper() might 780 * drop tasklist_lock and reacquire it. 781 */ 782 exit_ptrace(father); 783 reaper = find_new_reaper(father); 784 785 list_for_each_entry_safe(p, n, &father->children, sibling) { 786 struct task_struct *t = p; 787 do { 788 t->real_parent = reaper; 789 if (t->parent == father) { 790 BUG_ON(task_ptrace(t)); 791 t->parent = t->real_parent; 792 } 793 if (t->pdeath_signal) 794 group_send_sig_info(t->pdeath_signal, 795 SEND_SIG_NOINFO, t); 796 } while_each_thread(p, t); 797 reparent_leader(father, p, &dead_children); 798 } 799 write_unlock_irq(&tasklist_lock); 800 801 BUG_ON(!list_empty(&father->children)); 802 803 list_for_each_entry_safe(p, n, &dead_children, sibling) { 804 list_del_init(&p->sibling); 805 release_task(p); 806 } 807} 808 809/* 810 * Send signals to all our closest relatives so that they know 811 * to properly mourn us.. 812 */ 813static void exit_notify(struct task_struct *tsk, int group_dead) 814{ 815 int signal; 816 void *cookie; 817 818 /* 819 * This does two things: 820 * 821 * A. Make init inherit all the child processes 822 * B. Check to see if any process groups have become orphaned 823 * as a result of our exiting, and if they have any stopped 824 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 825 */ 826 forget_original_parent(tsk); 827 exit_task_namespaces(tsk); 828 829 write_lock_irq(&tasklist_lock); 830 if (group_dead) 831 kill_orphaned_pgrp(tsk->group_leader, NULL); 832 833 /* Let father know we died 834 * 835 * Thread signals are configurable, but you aren't going to use 836 * that to send signals to arbitary processes. 837 * That stops right now. 838 * 839 * If the parent exec id doesn't match the exec id we saved 840 * when we started then we know the parent has changed security 841 * domain. 842 * 843 * If our self_exec id doesn't match our parent_exec_id then 844 * we have changed execution domain as these two values started 845 * the same after a fork. 846 */ 847 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) && 848 (tsk->parent_exec_id != tsk->real_parent->self_exec_id || 849 tsk->self_exec_id != tsk->parent_exec_id)) 850 tsk->exit_signal = SIGCHLD; 851 852 signal = tracehook_notify_death(tsk, &cookie, group_dead); 853 if (signal >= 0) 854 signal = do_notify_parent(tsk, signal); 855 856 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE; 857 858 /* mt-exec, de_thread() is waiting for group leader */ 859 if (unlikely(tsk->signal->notify_count < 0)) 860 wake_up_process(tsk->signal->group_exit_task); 861 write_unlock_irq(&tasklist_lock); 862 863 tracehook_report_death(tsk, signal, cookie, group_dead); 864 865 /* If the process is dead, release it - nobody will wait for it */ 866 if (signal == DEATH_REAP) 867 release_task(tsk); 868} 869 870#ifdef CONFIG_DEBUG_STACK_USAGE 871static void check_stack_usage(void) 872{ 873 static DEFINE_SPINLOCK(low_water_lock); 874 static int lowest_to_date = THREAD_SIZE; 875 unsigned long free; 876 877 free = stack_not_used(current); 878 879 if (free >= lowest_to_date) 880 return; 881 882 spin_lock(&low_water_lock); 883 if (free < lowest_to_date) { 884 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes " 885 "left\n", 886 current->comm, free); 887 lowest_to_date = free; 888 } 889 spin_unlock(&low_water_lock); 890} 891#else 892static inline void check_stack_usage(void) {} 893#endif 894 895NORET_TYPE void do_exit(long code) 896{ 897 struct task_struct *tsk = current; 898 int group_dead; 899 900 profile_task_exit(tsk); 901 902 WARN_ON(atomic_read(&tsk->fs_excl)); 903 904 if (unlikely(in_interrupt())) 905 panic("Aiee, killing interrupt handler!"); 906 if (unlikely(!tsk->pid)) 907 panic("Attempted to kill the idle task!"); 908 909 /* 910 * If do_exit is called because this processes oopsed, it's possible 911 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before 912 * continuing. Amongst other possible reasons, this is to prevent 913 * mm_release()->clear_child_tid() from writing to a user-controlled 914 * kernel address. 915 */ 916 set_fs(USER_DS); 917 918 tracehook_report_exit(&code); 919 920 validate_creds_for_do_exit(tsk); 921 922 /* 923 * We're taking recursive faults here in do_exit. Safest is to just 924 * leave this task alone and wait for reboot. 925 */ 926 if (unlikely(tsk->flags & PF_EXITING)) { 927 printk(KERN_ALERT 928 "Fixing recursive fault but reboot is needed!\n"); 929 /* 930 * We can do this unlocked here. The futex code uses 931 * this flag just to verify whether the pi state 932 * cleanup has been done or not. In the worst case it 933 * loops once more. We pretend that the cleanup was 934 * done as there is no way to return. Either the 935 * OWNER_DIED bit is set by now or we push the blocked 936 * task into the wait for ever nirwana as well. 937 */ 938 tsk->flags |= PF_EXITPIDONE; 939 set_current_state(TASK_UNINTERRUPTIBLE); 940 schedule(); 941 } 942 943 exit_irq_thread(); 944 945 exit_signals(tsk); /* sets PF_EXITING */ 946 /* 947 * tsk->flags are checked in the futex code to protect against 948 * an exiting task cleaning up the robust pi futexes. 949 */ 950 smp_mb(); 951 raw_spin_unlock_wait(&tsk->pi_lock); 952 953 if (unlikely(in_atomic())) 954 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n", 955 current->comm, task_pid_nr(current), 956 preempt_count()); 957 958 acct_update_integrals(tsk); 959 /* sync mm's RSS info before statistics gathering */ 960 if (tsk->mm) 961 sync_mm_rss(tsk, tsk->mm); 962 group_dead = atomic_dec_and_test(&tsk->signal->live); 963 if (group_dead) { 964 hrtimer_cancel(&tsk->signal->real_timer); 965 exit_itimers(tsk->signal); 966 if (tsk->mm) 967 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 968 } 969 acct_collect(code, group_dead); 970 if (group_dead) 971 tty_audit_exit(); 972 if (unlikely(tsk->audit_context)) 973 audit_free(tsk); 974 975 tsk->exit_code = code; 976 taskstats_exit(tsk, group_dead); 977 978 exit_mm(tsk); 979 980 if (group_dead) 981 acct_process(); 982 trace_sched_process_exit(tsk); 983 984 exit_sem(tsk); 985 exit_files(tsk); 986 exit_fs(tsk); 987 check_stack_usage(); 988 exit_thread(); 989 cgroup_exit(tsk, 1); 990 991 if (group_dead) 992 disassociate_ctty(1); 993 994 module_put(task_thread_info(tsk)->exec_domain->module); 995 996 proc_exit_connector(tsk); 997 998 flush_ptrace_hw_breakpoint(tsk); 999 /* 1000 * Flush inherited counters to the parent - before the parent 1001 * gets woken up by child-exit notifications. 1002 */ 1003 perf_event_exit_task(tsk); 1004 1005 exit_notify(tsk, group_dead); 1006#ifdef CONFIG_NUMA 1007 task_lock(tsk); 1008 mpol_put(tsk->mempolicy); 1009 tsk->mempolicy = NULL; 1010 task_unlock(tsk); 1011#endif 1012#ifdef CONFIG_FUTEX 1013 if (unlikely(current->pi_state_cache)) 1014 kfree(current->pi_state_cache); 1015#endif 1016 /* 1017 * Make sure we are holding no locks: 1018 */ 1019 debug_check_no_locks_held(tsk); 1020 /* 1021 * We can do this unlocked here. The futex code uses this flag 1022 * just to verify whether the pi state cleanup has been done 1023 * or not. In the worst case it loops once more. 1024 */ 1025 tsk->flags |= PF_EXITPIDONE; 1026 1027 if (tsk->io_context) 1028 exit_io_context(tsk); 1029 1030 if (tsk->splice_pipe) 1031 __free_pipe_info(tsk->splice_pipe); 1032 1033 validate_creds_for_do_exit(tsk); 1034 1035 preempt_disable(); 1036 exit_rcu(); 1037 /* causes final put_task_struct in finish_task_switch(). */ 1038 tsk->state = TASK_DEAD; 1039 schedule(); 1040 BUG(); 1041 /* Avoid "noreturn function does return". */ 1042 for (;;) 1043 cpu_relax(); /* For when BUG is null */ 1044} 1045 1046EXPORT_SYMBOL_GPL(do_exit); 1047 1048NORET_TYPE void complete_and_exit(struct completion *comp, long code) 1049{ 1050 if (comp) 1051 complete(comp); 1052 1053 do_exit(code); 1054} 1055 1056EXPORT_SYMBOL(complete_and_exit); 1057 1058SYSCALL_DEFINE1(exit, int, error_code) 1059{ 1060 do_exit((error_code&0xff)<<8); 1061} 1062 1063/* 1064 * Take down every thread in the group. This is called by fatal signals 1065 * as well as by sys_exit_group (below). 1066 */ 1067NORET_TYPE void 1068do_group_exit(int exit_code) 1069{ 1070 struct signal_struct *sig = current->signal; 1071 1072 BUG_ON(exit_code & 0x80); /* core dumps don't get here */ 1073 1074 if (signal_group_exit(sig)) 1075 exit_code = sig->group_exit_code; 1076 else if (!thread_group_empty(current)) { 1077 struct sighand_struct *const sighand = current->sighand; 1078 spin_lock_irq(&sighand->siglock); 1079 if (signal_group_exit(sig)) 1080 /* Another thread got here before we took the lock. */ 1081 exit_code = sig->group_exit_code; 1082 else { 1083 sig->group_exit_code = exit_code; 1084 sig->flags = SIGNAL_GROUP_EXIT; 1085 zap_other_threads(current); 1086 } 1087 spin_unlock_irq(&sighand->siglock); 1088 } 1089 1090 do_exit(exit_code); 1091 /* NOTREACHED */ 1092} 1093 1094/* 1095 * this kills every thread in the thread group. Note that any externally 1096 * wait4()-ing process will get the correct exit code - even if this 1097 * thread is not the thread group leader. 1098 */ 1099SYSCALL_DEFINE1(exit_group, int, error_code) 1100{ 1101 do_group_exit((error_code & 0xff) << 8); 1102 /* NOTREACHED */ 1103 return 0; 1104} 1105 1106struct wait_opts { 1107 enum pid_type wo_type; 1108 int wo_flags; 1109 struct pid *wo_pid; 1110 1111 struct siginfo __user *wo_info; 1112 int __user *wo_stat; 1113 struct rusage __user *wo_rusage; 1114 1115 wait_queue_t child_wait; 1116 int notask_error; 1117}; 1118 1119static inline 1120struct pid *task_pid_type(struct task_struct *task, enum pid_type type) 1121{ 1122 if (type != PIDTYPE_PID) 1123 task = task->group_leader; 1124 return task->pids[type].pid; 1125} 1126 1127static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1128{ 1129 return wo->wo_type == PIDTYPE_MAX || 1130 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1131} 1132 1133static int eligible_child(struct wait_opts *wo, struct task_struct *p) 1134{ 1135 if (!eligible_pid(wo, p)) 1136 return 0; 1137 /* Wait for all children (clone and not) if __WALL is set; 1138 * otherwise, wait for clone children *only* if __WCLONE is 1139 * set; otherwise, wait for non-clone children *only*. (Note: 1140 * A "clone" child here is one that reports to its parent 1141 * using a signal other than SIGCHLD.) */ 1142 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1143 && !(wo->wo_flags & __WALL)) 1144 return 0; 1145 1146 return 1; 1147} 1148 1149static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p, 1150 pid_t pid, uid_t uid, int why, int status) 1151{ 1152 struct siginfo __user *infop; 1153 int retval = wo->wo_rusage 1154 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1155 1156 put_task_struct(p); 1157 infop = wo->wo_info; 1158 if (infop) { 1159 if (!retval) 1160 retval = put_user(SIGCHLD, &infop->si_signo); 1161 if (!retval) 1162 retval = put_user(0, &infop->si_errno); 1163 if (!retval) 1164 retval = put_user((short)why, &infop->si_code); 1165 if (!retval) 1166 retval = put_user(pid, &infop->si_pid); 1167 if (!retval) 1168 retval = put_user(uid, &infop->si_uid); 1169 if (!retval) 1170 retval = put_user(status, &infop->si_status); 1171 } 1172 if (!retval) 1173 retval = pid; 1174 return retval; 1175} 1176 1177/* 1178 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1179 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1180 * the lock and this task is uninteresting. If we return nonzero, we have 1181 * released the lock and the system call should return. 1182 */ 1183static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1184{ 1185 unsigned long state; 1186 int retval, status, traced; 1187 pid_t pid = task_pid_vnr(p); 1188 uid_t uid = __task_cred(p)->uid; 1189 struct siginfo __user *infop; 1190 1191 if (!likely(wo->wo_flags & WEXITED)) 1192 return 0; 1193 1194 if (unlikely(wo->wo_flags & WNOWAIT)) { 1195 int exit_code = p->exit_code; 1196 int why; 1197 1198 get_task_struct(p); 1199 read_unlock(&tasklist_lock); 1200 if ((exit_code & 0x7f) == 0) { 1201 why = CLD_EXITED; 1202 status = exit_code >> 8; 1203 } else { 1204 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; 1205 status = exit_code & 0x7f; 1206 } 1207 return wait_noreap_copyout(wo, p, pid, uid, why, status); 1208 } 1209 1210 /* 1211 * Try to move the task's state to DEAD 1212 * only one thread is allowed to do this: 1213 */ 1214 state = xchg(&p->exit_state, EXIT_DEAD); 1215 if (state != EXIT_ZOMBIE) { 1216 BUG_ON(state != EXIT_DEAD); 1217 return 0; 1218 } 1219 1220 traced = ptrace_reparented(p); 1221 /* 1222 * It can be ptraced but not reparented, check 1223 * !task_detached() to filter out sub-threads. 1224 */ 1225 if (likely(!traced) && likely(!task_detached(p))) { 1226 struct signal_struct *psig; 1227 struct signal_struct *sig; 1228 unsigned long maxrss; 1229 cputime_t tgutime, tgstime; 1230 1231 /* 1232 * The resource counters for the group leader are in its 1233 * own task_struct. Those for dead threads in the group 1234 * are in its signal_struct, as are those for the child 1235 * processes it has previously reaped. All these 1236 * accumulate in the parent's signal_struct c* fields. 1237 * 1238 * We don't bother to take a lock here to protect these 1239 * p->signal fields, because they are only touched by 1240 * __exit_signal, which runs with tasklist_lock 1241 * write-locked anyway, and so is excluded here. We do 1242 * need to protect the access to parent->signal fields, 1243 * as other threads in the parent group can be right 1244 * here reaping other children at the same time. 1245 * 1246 * We use thread_group_times() to get times for the thread 1247 * group, which consolidates times for all threads in the 1248 * group including the group leader. 1249 */ 1250 thread_group_times(p, &tgutime, &tgstime); 1251 spin_lock_irq(&p->real_parent->sighand->siglock); 1252 psig = p->real_parent->signal; 1253 sig = p->signal; 1254 psig->cutime = 1255 cputime_add(psig->cutime, 1256 cputime_add(tgutime, 1257 sig->cutime)); 1258 psig->cstime = 1259 cputime_add(psig->cstime, 1260 cputime_add(tgstime, 1261 sig->cstime)); 1262 psig->cgtime = 1263 cputime_add(psig->cgtime, 1264 cputime_add(p->gtime, 1265 cputime_add(sig->gtime, 1266 sig->cgtime))); 1267 psig->cmin_flt += 1268 p->min_flt + sig->min_flt + sig->cmin_flt; 1269 psig->cmaj_flt += 1270 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1271 psig->cnvcsw += 1272 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1273 psig->cnivcsw += 1274 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1275 psig->cinblock += 1276 task_io_get_inblock(p) + 1277 sig->inblock + sig->cinblock; 1278 psig->coublock += 1279 task_io_get_oublock(p) + 1280 sig->oublock + sig->coublock; 1281 maxrss = max(sig->maxrss, sig->cmaxrss); 1282 if (psig->cmaxrss < maxrss) 1283 psig->cmaxrss = maxrss; 1284 task_io_accounting_add(&psig->ioac, &p->ioac); 1285 task_io_accounting_add(&psig->ioac, &sig->ioac); 1286 spin_unlock_irq(&p->real_parent->sighand->siglock); 1287 } 1288 1289 /* 1290 * Now we are sure this task is interesting, and no other 1291 * thread can reap it because we set its state to EXIT_DEAD. 1292 */ 1293 read_unlock(&tasklist_lock); 1294 1295 retval = wo->wo_rusage 1296 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1297 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1298 ? p->signal->group_exit_code : p->exit_code; 1299 if (!retval && wo->wo_stat) 1300 retval = put_user(status, wo->wo_stat); 1301 1302 infop = wo->wo_info; 1303 if (!retval && infop) 1304 retval = put_user(SIGCHLD, &infop->si_signo); 1305 if (!retval && infop) 1306 retval = put_user(0, &infop->si_errno); 1307 if (!retval && infop) { 1308 int why; 1309 1310 if ((status & 0x7f) == 0) { 1311 why = CLD_EXITED; 1312 status >>= 8; 1313 } else { 1314 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1315 status &= 0x7f; 1316 } 1317 retval = put_user((short)why, &infop->si_code); 1318 if (!retval) 1319 retval = put_user(status, &infop->si_status); 1320 } 1321 if (!retval && infop) 1322 retval = put_user(pid, &infop->si_pid); 1323 if (!retval && infop) 1324 retval = put_user(uid, &infop->si_uid); 1325 if (!retval) 1326 retval = pid; 1327 1328 if (traced) { 1329 write_lock_irq(&tasklist_lock); 1330 /* We dropped tasklist, ptracer could die and untrace */ 1331 ptrace_unlink(p); 1332 /* 1333 * If this is not a detached task, notify the parent. 1334 * If it's still not detached after that, don't release 1335 * it now. 1336 */ 1337 if (!task_detached(p)) { 1338 do_notify_parent(p, p->exit_signal); 1339 if (!task_detached(p)) { 1340 p->exit_state = EXIT_ZOMBIE; 1341 p = NULL; 1342 } 1343 } 1344 write_unlock_irq(&tasklist_lock); 1345 } 1346 if (p != NULL) 1347 release_task(p); 1348 1349 return retval; 1350} 1351 1352static int *task_stopped_code(struct task_struct *p, bool ptrace) 1353{ 1354 if (ptrace) { 1355 if (task_is_stopped_or_traced(p)) 1356 return &p->exit_code; 1357 } else { 1358 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1359 return &p->signal->group_exit_code; 1360 } 1361 return NULL; 1362} 1363 1364/* 1365 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold 1366 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1367 * the lock and this task is uninteresting. If we return nonzero, we have 1368 * released the lock and the system call should return. 1369 */ 1370static int wait_task_stopped(struct wait_opts *wo, 1371 int ptrace, struct task_struct *p) 1372{ 1373 struct siginfo __user *infop; 1374 int retval, exit_code, *p_code, why; 1375 uid_t uid = 0; /* unneeded, required by compiler */ 1376 pid_t pid; 1377 1378 /* 1379 * Traditionally we see ptrace'd stopped tasks regardless of options. 1380 */ 1381 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1382 return 0; 1383 1384 exit_code = 0; 1385 spin_lock_irq(&p->sighand->siglock); 1386 1387 p_code = task_stopped_code(p, ptrace); 1388 if (unlikely(!p_code)) 1389 goto unlock_sig; 1390 1391 exit_code = *p_code; 1392 if (!exit_code) 1393 goto unlock_sig; 1394 1395 if (!unlikely(wo->wo_flags & WNOWAIT)) 1396 *p_code = 0; 1397 1398 uid = task_uid(p); 1399unlock_sig: 1400 spin_unlock_irq(&p->sighand->siglock); 1401 if (!exit_code) 1402 return 0; 1403 1404 /* 1405 * Now we are pretty sure this task is interesting. 1406 * Make sure it doesn't get reaped out from under us while we 1407 * give up the lock and then examine it below. We don't want to 1408 * keep holding onto the tasklist_lock while we call getrusage and 1409 * possibly take page faults for user memory. 1410 */ 1411 get_task_struct(p); 1412 pid = task_pid_vnr(p); 1413 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1414 read_unlock(&tasklist_lock); 1415 1416 if (unlikely(wo->wo_flags & WNOWAIT)) 1417 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code); 1418 1419 retval = wo->wo_rusage 1420 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1421 if (!retval && wo->wo_stat) 1422 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat); 1423 1424 infop = wo->wo_info; 1425 if (!retval && infop) 1426 retval = put_user(SIGCHLD, &infop->si_signo); 1427 if (!retval && infop) 1428 retval = put_user(0, &infop->si_errno); 1429 if (!retval && infop) 1430 retval = put_user((short)why, &infop->si_code); 1431 if (!retval && infop) 1432 retval = put_user(exit_code, &infop->si_status); 1433 if (!retval && infop) 1434 retval = put_user(pid, &infop->si_pid); 1435 if (!retval && infop) 1436 retval = put_user(uid, &infop->si_uid); 1437 if (!retval) 1438 retval = pid; 1439 put_task_struct(p); 1440 1441 BUG_ON(!retval); 1442 return retval; 1443} 1444 1445/* 1446 * Handle do_wait work for one task in a live, non-stopped state. 1447 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1448 * the lock and this task is uninteresting. If we return nonzero, we have 1449 * released the lock and the system call should return. 1450 */ 1451static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1452{ 1453 int retval; 1454 pid_t pid; 1455 uid_t uid; 1456 1457 if (!unlikely(wo->wo_flags & WCONTINUED)) 1458 return 0; 1459 1460 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1461 return 0; 1462 1463 spin_lock_irq(&p->sighand->siglock); 1464 /* Re-check with the lock held. */ 1465 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1466 spin_unlock_irq(&p->sighand->siglock); 1467 return 0; 1468 } 1469 if (!unlikely(wo->wo_flags & WNOWAIT)) 1470 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1471 uid = task_uid(p); 1472 spin_unlock_irq(&p->sighand->siglock); 1473 1474 pid = task_pid_vnr(p); 1475 get_task_struct(p); 1476 read_unlock(&tasklist_lock); 1477 1478 if (!wo->wo_info) { 1479 retval = wo->wo_rusage 1480 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1481 put_task_struct(p); 1482 if (!retval && wo->wo_stat) 1483 retval = put_user(0xffff, wo->wo_stat); 1484 if (!retval) 1485 retval = pid; 1486 } else { 1487 retval = wait_noreap_copyout(wo, p, pid, uid, 1488 CLD_CONTINUED, SIGCONT); 1489 BUG_ON(retval == 0); 1490 } 1491 1492 return retval; 1493} 1494 1495/* 1496 * Consider @p for a wait by @parent. 1497 * 1498 * -ECHILD should be in ->notask_error before the first call. 1499 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1500 * Returns zero if the search for a child should continue; 1501 * then ->notask_error is 0 if @p is an eligible child, 1502 * or another error from security_task_wait(), or still -ECHILD. 1503 */ 1504static int wait_consider_task(struct wait_opts *wo, int ptrace, 1505 struct task_struct *p) 1506{ 1507 int ret = eligible_child(wo, p); 1508 if (!ret) 1509 return ret; 1510 1511 ret = security_task_wait(p); 1512 if (unlikely(ret < 0)) { 1513 /* 1514 * If we have not yet seen any eligible child, 1515 * then let this error code replace -ECHILD. 1516 * A permission error will give the user a clue 1517 * to look for security policy problems, rather 1518 * than for mysterious wait bugs. 1519 */ 1520 if (wo->notask_error) 1521 wo->notask_error = ret; 1522 return 0; 1523 } 1524 1525 if (likely(!ptrace) && unlikely(task_ptrace(p))) { 1526 /* 1527 * This child is hidden by ptrace. 1528 * We aren't allowed to see it now, but eventually we will. 1529 */ 1530 wo->notask_error = 0; 1531 return 0; 1532 } 1533 1534 if (p->exit_state == EXIT_DEAD) 1535 return 0; 1536 1537 /* 1538 * We don't reap group leaders with subthreads. 1539 */ 1540 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p)) 1541 return wait_task_zombie(wo, p); 1542 1543 /* 1544 * It's stopped or running now, so it might 1545 * later continue, exit, or stop again. 1546 */ 1547 wo->notask_error = 0; 1548 1549 if (task_stopped_code(p, ptrace)) 1550 return wait_task_stopped(wo, ptrace, p); 1551 1552 return wait_task_continued(wo, p); 1553} 1554 1555/* 1556 * Do the work of do_wait() for one thread in the group, @tsk. 1557 * 1558 * -ECHILD should be in ->notask_error before the first call. 1559 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1560 * Returns zero if the search for a child should continue; then 1561 * ->notask_error is 0 if there were any eligible children, 1562 * or another error from security_task_wait(), or still -ECHILD. 1563 */ 1564static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1565{ 1566 struct task_struct *p; 1567 1568 list_for_each_entry(p, &tsk->children, sibling) { 1569 int ret = wait_consider_task(wo, 0, p); 1570 if (ret) 1571 return ret; 1572 } 1573 1574 return 0; 1575} 1576 1577static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1578{ 1579 struct task_struct *p; 1580 1581 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1582 int ret = wait_consider_task(wo, 1, p); 1583 if (ret) 1584 return ret; 1585 } 1586 1587 return 0; 1588} 1589 1590static int child_wait_callback(wait_queue_t *wait, unsigned mode, 1591 int sync, void *key) 1592{ 1593 struct wait_opts *wo = container_of(wait, struct wait_opts, 1594 child_wait); 1595 struct task_struct *p = key; 1596 1597 if (!eligible_pid(wo, p)) 1598 return 0; 1599 1600 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) 1601 return 0; 1602 1603 return default_wake_function(wait, mode, sync, key); 1604} 1605 1606void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1607{ 1608 __wake_up_sync_key(&parent->signal->wait_chldexit, 1609 TASK_INTERRUPTIBLE, 1, p); 1610} 1611 1612static long do_wait(struct wait_opts *wo) 1613{ 1614 struct task_struct *tsk; 1615 int retval; 1616 1617 trace_sched_process_wait(wo->wo_pid); 1618 1619 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1620 wo->child_wait.private = current; 1621 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1622repeat: 1623 /* 1624 * If there is nothing that can match our critiera just get out. 1625 * We will clear ->notask_error to zero if we see any child that 1626 * might later match our criteria, even if we are not able to reap 1627 * it yet. 1628 */ 1629 wo->notask_error = -ECHILD; 1630 if ((wo->wo_type < PIDTYPE_MAX) && 1631 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) 1632 goto notask; 1633 1634 set_current_state(TASK_INTERRUPTIBLE); 1635 read_lock(&tasklist_lock); 1636 tsk = current; 1637 do { 1638 retval = do_wait_thread(wo, tsk); 1639 if (retval) 1640 goto end; 1641 1642 retval = ptrace_do_wait(wo, tsk); 1643 if (retval) 1644 goto end; 1645 1646 if (wo->wo_flags & __WNOTHREAD) 1647 break; 1648 } while_each_thread(current, tsk); 1649 read_unlock(&tasklist_lock); 1650 1651notask: 1652 retval = wo->notask_error; 1653 if (!retval && !(wo->wo_flags & WNOHANG)) { 1654 retval = -ERESTARTSYS; 1655 if (!signal_pending(current)) { 1656 schedule(); 1657 goto repeat; 1658 } 1659 } 1660end: 1661 __set_current_state(TASK_RUNNING); 1662 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1663 return retval; 1664} 1665 1666SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1667 infop, int, options, struct rusage __user *, ru) 1668{ 1669 struct wait_opts wo; 1670 struct pid *pid = NULL; 1671 enum pid_type type; 1672 long ret; 1673 1674 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED)) 1675 return -EINVAL; 1676 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1677 return -EINVAL; 1678 1679 switch (which) { 1680 case P_ALL: 1681 type = PIDTYPE_MAX; 1682 break; 1683 case P_PID: 1684 type = PIDTYPE_PID; 1685 if (upid <= 0) 1686 return -EINVAL; 1687 break; 1688 case P_PGID: 1689 type = PIDTYPE_PGID; 1690 if (upid <= 0) 1691 return -EINVAL; 1692 break; 1693 default: 1694 return -EINVAL; 1695 } 1696 1697 if (type < PIDTYPE_MAX) 1698 pid = find_get_pid(upid); 1699 1700 wo.wo_type = type; 1701 wo.wo_pid = pid; 1702 wo.wo_flags = options; 1703 wo.wo_info = infop; 1704 wo.wo_stat = NULL; 1705 wo.wo_rusage = ru; 1706 ret = do_wait(&wo); 1707 1708 if (ret > 0) { 1709 ret = 0; 1710 } else if (infop) { 1711 /* 1712 * For a WNOHANG return, clear out all the fields 1713 * we would set so the user can easily tell the 1714 * difference. 1715 */ 1716 if (!ret) 1717 ret = put_user(0, &infop->si_signo); 1718 if (!ret) 1719 ret = put_user(0, &infop->si_errno); 1720 if (!ret) 1721 ret = put_user(0, &infop->si_code); 1722 if (!ret) 1723 ret = put_user(0, &infop->si_pid); 1724 if (!ret) 1725 ret = put_user(0, &infop->si_uid); 1726 if (!ret) 1727 ret = put_user(0, &infop->si_status); 1728 } 1729 1730 put_pid(pid); 1731 1732 /* avoid REGPARM breakage on x86: */ 1733 asmlinkage_protect(5, ret, which, upid, infop, options, ru); 1734 return ret; 1735} 1736 1737SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1738 int, options, struct rusage __user *, ru) 1739{ 1740 struct wait_opts wo; 1741 struct pid *pid = NULL; 1742 enum pid_type type; 1743 long ret; 1744 1745 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1746 __WNOTHREAD|__WCLONE|__WALL)) 1747 return -EINVAL; 1748 1749 if (upid == -1) 1750 type = PIDTYPE_MAX; 1751 else if (upid < 0) { 1752 type = PIDTYPE_PGID; 1753 pid = find_get_pid(-upid); 1754 } else if (upid == 0) { 1755 type = PIDTYPE_PGID; 1756 pid = get_task_pid(current, PIDTYPE_PGID); 1757 } else /* upid > 0 */ { 1758 type = PIDTYPE_PID; 1759 pid = find_get_pid(upid); 1760 } 1761 1762 wo.wo_type = type; 1763 wo.wo_pid = pid; 1764 wo.wo_flags = options | WEXITED; 1765 wo.wo_info = NULL; 1766 wo.wo_stat = stat_addr; 1767 wo.wo_rusage = ru; 1768 ret = do_wait(&wo); 1769 put_pid(pid); 1770 1771 /* avoid REGPARM breakage on x86: */ 1772 asmlinkage_protect(4, ret, upid, stat_addr, options, ru); 1773 return ret; 1774} 1775 1776#ifdef __ARCH_WANT_SYS_WAITPID 1777 1778/* 1779 * sys_waitpid() remains for compatibility. waitpid() should be 1780 * implemented by calling sys_wait4() from libc.a. 1781 */ 1782SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1783{ 1784 return sys_wait4(pid, stat_addr, options, NULL); 1785} 1786 1787#endif 1788