]>
Commit | Line | Data |
---|---|---|
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/freezer.h> | |
24 | #include <linux/binfmts.h> | |
25 | #include <linux/nsproxy.h> | |
26 | #include <linux/pid_namespace.h> | |
27 | #include <linux/ptrace.h> | |
28 | #include <linux/profile.h> | |
29 | #include <linux/mount.h> | |
30 | #include <linux/proc_fs.h> | |
31 | #include <linux/kthread.h> | |
32 | #include <linux/mempolicy.h> | |
33 | #include <linux/taskstats_kern.h> | |
34 | #include <linux/delayacct.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 | #include <linux/oom.h> | |
54 | #include <linux/writeback.h> | |
55 | #include <linux/shm.h> | |
56 | #include <linux/kcov.h> | |
57 | ||
58 | #include <asm/uaccess.h> | |
59 | #include <asm/unistd.h> | |
60 | #include <asm/pgtable.h> | |
61 | #include <asm/mmu_context.h> | |
62 | ||
63 | static void __unhash_process(struct task_struct *p, bool group_dead) | |
64 | { | |
65 | nr_threads--; | |
66 | detach_pid(p, PIDTYPE_PID); | |
67 | if (group_dead) { | |
68 | detach_pid(p, PIDTYPE_PGID); | |
69 | detach_pid(p, PIDTYPE_SID); | |
70 | ||
71 | list_del_rcu(&p->tasks); | |
72 | list_del_init(&p->sibling); | |
73 | __this_cpu_dec(process_counts); | |
74 | } | |
75 | list_del_rcu(&p->thread_group); | |
76 | list_del_rcu(&p->thread_node); | |
77 | } | |
78 | ||
79 | /* | |
80 | * This function expects the tasklist_lock write-locked. | |
81 | */ | |
82 | static void __exit_signal(struct task_struct *tsk) | |
83 | { | |
84 | struct signal_struct *sig = tsk->signal; | |
85 | bool group_dead = thread_group_leader(tsk); | |
86 | struct sighand_struct *sighand; | |
87 | struct tty_struct *uninitialized_var(tty); | |
88 | cputime_t utime, stime; | |
89 | ||
90 | sighand = rcu_dereference_check(tsk->sighand, | |
91 | lockdep_tasklist_lock_is_held()); | |
92 | spin_lock(&sighand->siglock); | |
93 | ||
94 | posix_cpu_timers_exit(tsk); | |
95 | if (group_dead) { | |
96 | posix_cpu_timers_exit_group(tsk); | |
97 | tty = sig->tty; | |
98 | sig->tty = NULL; | |
99 | } else { | |
100 | /* | |
101 | * This can only happen if the caller is de_thread(). | |
102 | * FIXME: this is the temporary hack, we should teach | |
103 | * posix-cpu-timers to handle this case correctly. | |
104 | */ | |
105 | if (unlikely(has_group_leader_pid(tsk))) | |
106 | posix_cpu_timers_exit_group(tsk); | |
107 | ||
108 | /* | |
109 | * If there is any task waiting for the group exit | |
110 | * then notify it: | |
111 | */ | |
112 | if (sig->notify_count > 0 && !--sig->notify_count) | |
113 | wake_up_process(sig->group_exit_task); | |
114 | ||
115 | if (tsk == sig->curr_target) | |
116 | sig->curr_target = next_thread(tsk); | |
117 | } | |
118 | ||
119 | /* | |
120 | * Accumulate here the counters for all threads as they die. We could | |
121 | * skip the group leader because it is the last user of signal_struct, | |
122 | * but we want to avoid the race with thread_group_cputime() which can | |
123 | * see the empty ->thread_head list. | |
124 | */ | |
125 | task_cputime(tsk, &utime, &stime); | |
126 | write_seqlock(&sig->stats_lock); | |
127 | sig->utime += utime; | |
128 | sig->stime += stime; | |
129 | sig->gtime += task_gtime(tsk); | |
130 | sig->min_flt += tsk->min_flt; | |
131 | sig->maj_flt += tsk->maj_flt; | |
132 | sig->nvcsw += tsk->nvcsw; | |
133 | sig->nivcsw += tsk->nivcsw; | |
134 | sig->inblock += task_io_get_inblock(tsk); | |
135 | sig->oublock += task_io_get_oublock(tsk); | |
136 | task_io_accounting_add(&sig->ioac, &tsk->ioac); | |
137 | sig->sum_sched_runtime += tsk->se.sum_exec_runtime; | |
138 | sig->nr_threads--; | |
139 | __unhash_process(tsk, group_dead); | |
140 | write_sequnlock(&sig->stats_lock); | |
141 | ||
142 | /* | |
143 | * Do this under ->siglock, we can race with another thread | |
144 | * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. | |
145 | */ | |
146 | flush_sigqueue(&tsk->pending); | |
147 | tsk->sighand = NULL; | |
148 | spin_unlock(&sighand->siglock); | |
149 | ||
150 | __cleanup_sighand(sighand); | |
151 | clear_tsk_thread_flag(tsk, TIF_SIGPENDING); | |
152 | if (group_dead) { | |
153 | flush_sigqueue(&sig->shared_pending); | |
154 | tty_kref_put(tty); | |
155 | } | |
156 | } | |
157 | ||
158 | static void delayed_put_task_struct(struct rcu_head *rhp) | |
159 | { | |
160 | struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); | |
161 | ||
162 | perf_event_delayed_put(tsk); | |
163 | trace_sched_process_free(tsk); | |
164 | put_task_struct(tsk); | |
165 | } | |
166 | ||
167 | ||
168 | void release_task(struct task_struct *p) | |
169 | { | |
170 | struct task_struct *leader; | |
171 | int zap_leader; | |
172 | repeat: | |
173 | /* don't need to get the RCU readlock here - the process is dead and | |
174 | * can't be modifying its own credentials. But shut RCU-lockdep up */ | |
175 | rcu_read_lock(); | |
176 | atomic_dec(&__task_cred(p)->user->processes); | |
177 | rcu_read_unlock(); | |
178 | ||
179 | proc_flush_task(p); | |
180 | ||
181 | write_lock_irq(&tasklist_lock); | |
182 | ptrace_release_task(p); | |
183 | __exit_signal(p); | |
184 | ||
185 | /* | |
186 | * If we are the last non-leader member of the thread | |
187 | * group, and the leader is zombie, then notify the | |
188 | * group leader's parent process. (if it wants notification.) | |
189 | */ | |
190 | zap_leader = 0; | |
191 | leader = p->group_leader; | |
192 | if (leader != p && thread_group_empty(leader) | |
193 | && leader->exit_state == EXIT_ZOMBIE) { | |
194 | /* | |
195 | * If we were the last child thread and the leader has | |
196 | * exited already, and the leader's parent ignores SIGCHLD, | |
197 | * then we are the one who should release the leader. | |
198 | */ | |
199 | zap_leader = do_notify_parent(leader, leader->exit_signal); | |
200 | if (zap_leader) | |
201 | leader->exit_state = EXIT_DEAD; | |
202 | } | |
203 | ||
204 | write_unlock_irq(&tasklist_lock); | |
205 | release_thread(p); | |
206 | call_rcu(&p->rcu, delayed_put_task_struct); | |
207 | ||
208 | p = leader; | |
209 | if (unlikely(zap_leader)) | |
210 | goto repeat; | |
211 | } | |
212 | ||
213 | /* | |
214 | * Note that if this function returns a valid task_struct pointer (!NULL) | |
215 | * task->usage must remain >0 for the duration of the RCU critical section. | |
216 | */ | |
217 | struct task_struct *task_rcu_dereference(struct task_struct **ptask) | |
218 | { | |
219 | struct sighand_struct *sighand; | |
220 | struct task_struct *task; | |
221 | ||
222 | /* | |
223 | * We need to verify that release_task() was not called and thus | |
224 | * delayed_put_task_struct() can't run and drop the last reference | |
225 | * before rcu_read_unlock(). We check task->sighand != NULL, | |
226 | * but we can read the already freed and reused memory. | |
227 | */ | |
228 | retry: | |
229 | task = rcu_dereference(*ptask); | |
230 | if (!task) | |
231 | return NULL; | |
232 | ||
233 | probe_kernel_address(&task->sighand, sighand); | |
234 | ||
235 | /* | |
236 | * Pairs with atomic_dec_and_test() in put_task_struct(). If this task | |
237 | * was already freed we can not miss the preceding update of this | |
238 | * pointer. | |
239 | */ | |
240 | smp_rmb(); | |
241 | if (unlikely(task != READ_ONCE(*ptask))) | |
242 | goto retry; | |
243 | ||
244 | /* | |
245 | * We've re-checked that "task == *ptask", now we have two different | |
246 | * cases: | |
247 | * | |
248 | * 1. This is actually the same task/task_struct. In this case | |
249 | * sighand != NULL tells us it is still alive. | |
250 | * | |
251 | * 2. This is another task which got the same memory for task_struct. | |
252 | * We can't know this of course, and we can not trust | |
253 | * sighand != NULL. | |
254 | * | |
255 | * In this case we actually return a random value, but this is | |
256 | * correct. | |
257 | * | |
258 | * If we return NULL - we can pretend that we actually noticed that | |
259 | * *ptask was updated when the previous task has exited. Or pretend | |
260 | * that probe_slab_address(&sighand) reads NULL. | |
261 | * | |
262 | * If we return the new task (because sighand is not NULL for any | |
263 | * reason) - this is fine too. This (new) task can't go away before | |
264 | * another gp pass. | |
265 | * | |
266 | * And note: We could even eliminate the false positive if re-read | |
267 | * task->sighand once again to avoid the falsely NULL. But this case | |
268 | * is very unlikely so we don't care. | |
269 | */ | |
270 | if (!sighand) | |
271 | return NULL; | |
272 | ||
273 | return task; | |
274 | } | |
275 | ||
276 | struct task_struct *try_get_task_struct(struct task_struct **ptask) | |
277 | { | |
278 | struct task_struct *task; | |
279 | ||
280 | rcu_read_lock(); | |
281 | task = task_rcu_dereference(ptask); | |
282 | if (task) | |
283 | get_task_struct(task); | |
284 | rcu_read_unlock(); | |
285 | ||
286 | return task; | |
287 | } | |
288 | ||
289 | /* | |
290 | * Determine if a process group is "orphaned", according to the POSIX | |
291 | * definition in 2.2.2.52. Orphaned process groups are not to be affected | |
292 | * by terminal-generated stop signals. Newly orphaned process groups are | |
293 | * to receive a SIGHUP and a SIGCONT. | |
294 | * | |
295 | * "I ask you, have you ever known what it is to be an orphan?" | |
296 | */ | |
297 | static int will_become_orphaned_pgrp(struct pid *pgrp, | |
298 | struct task_struct *ignored_task) | |
299 | { | |
300 | struct task_struct *p; | |
301 | ||
302 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
303 | if ((p == ignored_task) || | |
304 | (p->exit_state && thread_group_empty(p)) || | |
305 | is_global_init(p->real_parent)) | |
306 | continue; | |
307 | ||
308 | if (task_pgrp(p->real_parent) != pgrp && | |
309 | task_session(p->real_parent) == task_session(p)) | |
310 | return 0; | |
311 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
312 | ||
313 | return 1; | |
314 | } | |
315 | ||
316 | int is_current_pgrp_orphaned(void) | |
317 | { | |
318 | int retval; | |
319 | ||
320 | read_lock(&tasklist_lock); | |
321 | retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); | |
322 | read_unlock(&tasklist_lock); | |
323 | ||
324 | return retval; | |
325 | } | |
326 | ||
327 | static bool has_stopped_jobs(struct pid *pgrp) | |
328 | { | |
329 | struct task_struct *p; | |
330 | ||
331 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
332 | if (p->signal->flags & SIGNAL_STOP_STOPPED) | |
333 | return true; | |
334 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
335 | ||
336 | return false; | |
337 | } | |
338 | ||
339 | /* | |
340 | * Check to see if any process groups have become orphaned as | |
341 | * a result of our exiting, and if they have any stopped jobs, | |
342 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
343 | */ | |
344 | static void | |
345 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) | |
346 | { | |
347 | struct pid *pgrp = task_pgrp(tsk); | |
348 | struct task_struct *ignored_task = tsk; | |
349 | ||
350 | if (!parent) | |
351 | /* exit: our father is in a different pgrp than | |
352 | * we are and we were the only connection outside. | |
353 | */ | |
354 | parent = tsk->real_parent; | |
355 | else | |
356 | /* reparent: our child is in a different pgrp than | |
357 | * we are, and it was the only connection outside. | |
358 | */ | |
359 | ignored_task = NULL; | |
360 | ||
361 | if (task_pgrp(parent) != pgrp && | |
362 | task_session(parent) == task_session(tsk) && | |
363 | will_become_orphaned_pgrp(pgrp, ignored_task) && | |
364 | has_stopped_jobs(pgrp)) { | |
365 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); | |
366 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); | |
367 | } | |
368 | } | |
369 | ||
370 | #ifdef CONFIG_MEMCG | |
371 | /* | |
372 | * A task is exiting. If it owned this mm, find a new owner for the mm. | |
373 | */ | |
374 | void mm_update_next_owner(struct mm_struct *mm) | |
375 | { | |
376 | struct task_struct *c, *g, *p = current; | |
377 | ||
378 | retry: | |
379 | /* | |
380 | * If the exiting or execing task is not the owner, it's | |
381 | * someone else's problem. | |
382 | */ | |
383 | if (mm->owner != p) | |
384 | return; | |
385 | /* | |
386 | * The current owner is exiting/execing and there are no other | |
387 | * candidates. Do not leave the mm pointing to a possibly | |
388 | * freed task structure. | |
389 | */ | |
390 | if (atomic_read(&mm->mm_users) <= 1) { | |
391 | mm->owner = NULL; | |
392 | return; | |
393 | } | |
394 | ||
395 | read_lock(&tasklist_lock); | |
396 | /* | |
397 | * Search in the children | |
398 | */ | |
399 | list_for_each_entry(c, &p->children, sibling) { | |
400 | if (c->mm == mm) | |
401 | goto assign_new_owner; | |
402 | } | |
403 | ||
404 | /* | |
405 | * Search in the siblings | |
406 | */ | |
407 | list_for_each_entry(c, &p->real_parent->children, sibling) { | |
408 | if (c->mm == mm) | |
409 | goto assign_new_owner; | |
410 | } | |
411 | ||
412 | /* | |
413 | * Search through everything else, we should not get here often. | |
414 | */ | |
415 | for_each_process(g) { | |
416 | if (g->flags & PF_KTHREAD) | |
417 | continue; | |
418 | for_each_thread(g, c) { | |
419 | if (c->mm == mm) | |
420 | goto assign_new_owner; | |
421 | if (c->mm) | |
422 | break; | |
423 | } | |
424 | } | |
425 | read_unlock(&tasklist_lock); | |
426 | /* | |
427 | * We found no owner yet mm_users > 1: this implies that we are | |
428 | * most likely racing with swapoff (try_to_unuse()) or /proc or | |
429 | * ptrace or page migration (get_task_mm()). Mark owner as NULL. | |
430 | */ | |
431 | mm->owner = NULL; | |
432 | return; | |
433 | ||
434 | assign_new_owner: | |
435 | BUG_ON(c == p); | |
436 | get_task_struct(c); | |
437 | /* | |
438 | * The task_lock protects c->mm from changing. | |
439 | * We always want mm->owner->mm == mm | |
440 | */ | |
441 | task_lock(c); | |
442 | /* | |
443 | * Delay read_unlock() till we have the task_lock() | |
444 | * to ensure that c does not slip away underneath us | |
445 | */ | |
446 | read_unlock(&tasklist_lock); | |
447 | if (c->mm != mm) { | |
448 | task_unlock(c); | |
449 | put_task_struct(c); | |
450 | goto retry; | |
451 | } | |
452 | mm->owner = c; | |
453 | task_unlock(c); | |
454 | put_task_struct(c); | |
455 | } | |
456 | #endif /* CONFIG_MEMCG */ | |
457 | ||
458 | /* | |
459 | * Turn us into a lazy TLB process if we | |
460 | * aren't already.. | |
461 | */ | |
462 | static void exit_mm(struct task_struct *tsk) | |
463 | { | |
464 | struct mm_struct *mm = tsk->mm; | |
465 | struct core_state *core_state; | |
466 | ||
467 | mm_release(tsk, mm); | |
468 | if (!mm) | |
469 | return; | |
470 | sync_mm_rss(mm); | |
471 | /* | |
472 | * Serialize with any possible pending coredump. | |
473 | * We must hold mmap_sem around checking core_state | |
474 | * and clearing tsk->mm. The core-inducing thread | |
475 | * will increment ->nr_threads for each thread in the | |
476 | * group with ->mm != NULL. | |
477 | */ | |
478 | down_read(&mm->mmap_sem); | |
479 | core_state = mm->core_state; | |
480 | if (core_state) { | |
481 | struct core_thread self; | |
482 | ||
483 | up_read(&mm->mmap_sem); | |
484 | ||
485 | self.task = tsk; | |
486 | self.next = xchg(&core_state->dumper.next, &self); | |
487 | /* | |
488 | * Implies mb(), the result of xchg() must be visible | |
489 | * to core_state->dumper. | |
490 | */ | |
491 | if (atomic_dec_and_test(&core_state->nr_threads)) | |
492 | complete(&core_state->startup); | |
493 | ||
494 | for (;;) { | |
495 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
496 | if (!self.task) /* see coredump_finish() */ | |
497 | break; | |
498 | freezable_schedule(); | |
499 | } | |
500 | __set_task_state(tsk, TASK_RUNNING); | |
501 | down_read(&mm->mmap_sem); | |
502 | } | |
503 | atomic_inc(&mm->mm_count); | |
504 | BUG_ON(mm != tsk->active_mm); | |
505 | /* more a memory barrier than a real lock */ | |
506 | task_lock(tsk); | |
507 | tsk->mm = NULL; | |
508 | up_read(&mm->mmap_sem); | |
509 | enter_lazy_tlb(mm, current); | |
510 | task_unlock(tsk); | |
511 | mm_update_next_owner(mm); | |
512 | mmput(mm); | |
513 | if (test_thread_flag(TIF_MEMDIE)) | |
514 | exit_oom_victim(tsk); | |
515 | } | |
516 | ||
517 | static struct task_struct *find_alive_thread(struct task_struct *p) | |
518 | { | |
519 | struct task_struct *t; | |
520 | ||
521 | for_each_thread(p, t) { | |
522 | if (!(t->flags & PF_EXITING)) | |
523 | return t; | |
524 | } | |
525 | return NULL; | |
526 | } | |
527 | ||
528 | static struct task_struct *find_child_reaper(struct task_struct *father) | |
529 | __releases(&tasklist_lock) | |
530 | __acquires(&tasklist_lock) | |
531 | { | |
532 | struct pid_namespace *pid_ns = task_active_pid_ns(father); | |
533 | struct task_struct *reaper = pid_ns->child_reaper; | |
534 | ||
535 | if (likely(reaper != father)) | |
536 | return reaper; | |
537 | ||
538 | reaper = find_alive_thread(father); | |
539 | if (reaper) { | |
540 | pid_ns->child_reaper = reaper; | |
541 | return reaper; | |
542 | } | |
543 | ||
544 | write_unlock_irq(&tasklist_lock); | |
545 | if (unlikely(pid_ns == &init_pid_ns)) { | |
546 | panic("Attempted to kill init! exitcode=0x%08x\n", | |
547 | father->signal->group_exit_code ?: father->exit_code); | |
548 | } | |
549 | zap_pid_ns_processes(pid_ns); | |
550 | write_lock_irq(&tasklist_lock); | |
551 | ||
552 | return father; | |
553 | } | |
554 | ||
555 | /* | |
556 | * When we die, we re-parent all our children, and try to: | |
557 | * 1. give them to another thread in our thread group, if such a member exists | |
558 | * 2. give it to the first ancestor process which prctl'd itself as a | |
559 | * child_subreaper for its children (like a service manager) | |
560 | * 3. give it to the init process (PID 1) in our pid namespace | |
561 | */ | |
562 | static struct task_struct *find_new_reaper(struct task_struct *father, | |
563 | struct task_struct *child_reaper) | |
564 | { | |
565 | struct task_struct *thread, *reaper; | |
566 | ||
567 | thread = find_alive_thread(father); | |
568 | if (thread) | |
569 | return thread; | |
570 | ||
571 | if (father->signal->has_child_subreaper) { | |
572 | /* | |
573 | * Find the first ->is_child_subreaper ancestor in our pid_ns. | |
574 | * We start from father to ensure we can not look into another | |
575 | * namespace, this is safe because all its threads are dead. | |
576 | */ | |
577 | for (reaper = father; | |
578 | !same_thread_group(reaper, child_reaper); | |
579 | reaper = reaper->real_parent) { | |
580 | /* call_usermodehelper() descendants need this check */ | |
581 | if (reaper == &init_task) | |
582 | break; | |
583 | if (!reaper->signal->is_child_subreaper) | |
584 | continue; | |
585 | thread = find_alive_thread(reaper); | |
586 | if (thread) | |
587 | return thread; | |
588 | } | |
589 | } | |
590 | ||
591 | return child_reaper; | |
592 | } | |
593 | ||
594 | /* | |
595 | * Any that need to be release_task'd are put on the @dead list. | |
596 | */ | |
597 | static void reparent_leader(struct task_struct *father, struct task_struct *p, | |
598 | struct list_head *dead) | |
599 | { | |
600 | if (unlikely(p->exit_state == EXIT_DEAD)) | |
601 | return; | |
602 | ||
603 | /* We don't want people slaying init. */ | |
604 | p->exit_signal = SIGCHLD; | |
605 | ||
606 | /* If it has exited notify the new parent about this child's death. */ | |
607 | if (!p->ptrace && | |
608 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { | |
609 | if (do_notify_parent(p, p->exit_signal)) { | |
610 | p->exit_state = EXIT_DEAD; | |
611 | list_add(&p->ptrace_entry, dead); | |
612 | } | |
613 | } | |
614 | ||
615 | kill_orphaned_pgrp(p, father); | |
616 | } | |
617 | ||
618 | /* | |
619 | * This does two things: | |
620 | * | |
621 | * A. Make init inherit all the child processes | |
622 | * B. Check to see if any process groups have become orphaned | |
623 | * as a result of our exiting, and if they have any stopped | |
624 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
625 | */ | |
626 | static void forget_original_parent(struct task_struct *father, | |
627 | struct list_head *dead) | |
628 | { | |
629 | struct task_struct *p, *t, *reaper; | |
630 | ||
631 | if (unlikely(!list_empty(&father->ptraced))) | |
632 | exit_ptrace(father, dead); | |
633 | ||
634 | /* Can drop and reacquire tasklist_lock */ | |
635 | reaper = find_child_reaper(father); | |
636 | if (list_empty(&father->children)) | |
637 | return; | |
638 | ||
639 | reaper = find_new_reaper(father, reaper); | |
640 | list_for_each_entry(p, &father->children, sibling) { | |
641 | for_each_thread(p, t) { | |
642 | t->real_parent = reaper; | |
643 | BUG_ON((!t->ptrace) != (t->parent == father)); | |
644 | if (likely(!t->ptrace)) | |
645 | t->parent = t->real_parent; | |
646 | if (t->pdeath_signal) | |
647 | group_send_sig_info(t->pdeath_signal, | |
648 | SEND_SIG_NOINFO, t); | |
649 | } | |
650 | /* | |
651 | * If this is a threaded reparent there is no need to | |
652 | * notify anyone anything has happened. | |
653 | */ | |
654 | if (!same_thread_group(reaper, father)) | |
655 | reparent_leader(father, p, dead); | |
656 | } | |
657 | list_splice_tail_init(&father->children, &reaper->children); | |
658 | } | |
659 | ||
660 | /* | |
661 | * Send signals to all our closest relatives so that they know | |
662 | * to properly mourn us.. | |
663 | */ | |
664 | static void exit_notify(struct task_struct *tsk, int group_dead) | |
665 | { | |
666 | bool autoreap; | |
667 | struct task_struct *p, *n; | |
668 | LIST_HEAD(dead); | |
669 | ||
670 | write_lock_irq(&tasklist_lock); | |
671 | forget_original_parent(tsk, &dead); | |
672 | ||
673 | if (group_dead) | |
674 | kill_orphaned_pgrp(tsk->group_leader, NULL); | |
675 | ||
676 | if (unlikely(tsk->ptrace)) { | |
677 | int sig = thread_group_leader(tsk) && | |
678 | thread_group_empty(tsk) && | |
679 | !ptrace_reparented(tsk) ? | |
680 | tsk->exit_signal : SIGCHLD; | |
681 | autoreap = do_notify_parent(tsk, sig); | |
682 | } else if (thread_group_leader(tsk)) { | |
683 | autoreap = thread_group_empty(tsk) && | |
684 | do_notify_parent(tsk, tsk->exit_signal); | |
685 | } else { | |
686 | autoreap = true; | |
687 | } | |
688 | ||
689 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; | |
690 | if (tsk->exit_state == EXIT_DEAD) | |
691 | list_add(&tsk->ptrace_entry, &dead); | |
692 | ||
693 | /* mt-exec, de_thread() is waiting for group leader */ | |
694 | if (unlikely(tsk->signal->notify_count < 0)) | |
695 | wake_up_process(tsk->signal->group_exit_task); | |
696 | write_unlock_irq(&tasklist_lock); | |
697 | ||
698 | list_for_each_entry_safe(p, n, &dead, ptrace_entry) { | |
699 | list_del_init(&p->ptrace_entry); | |
700 | release_task(p); | |
701 | } | |
702 | } | |
703 | ||
704 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
705 | static void check_stack_usage(void) | |
706 | { | |
707 | static DEFINE_SPINLOCK(low_water_lock); | |
708 | static int lowest_to_date = THREAD_SIZE; | |
709 | unsigned long free; | |
710 | ||
711 | free = stack_not_used(current); | |
712 | ||
713 | if (free >= lowest_to_date) | |
714 | return; | |
715 | ||
716 | spin_lock(&low_water_lock); | |
717 | if (free < lowest_to_date) { | |
718 | pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", | |
719 | current->comm, task_pid_nr(current), free); | |
720 | lowest_to_date = free; | |
721 | } | |
722 | spin_unlock(&low_water_lock); | |
723 | } | |
724 | #else | |
725 | static inline void check_stack_usage(void) {} | |
726 | #endif | |
727 | ||
728 | void do_exit(long code) | |
729 | { | |
730 | struct task_struct *tsk = current; | |
731 | int group_dead; | |
732 | TASKS_RCU(int tasks_rcu_i); | |
733 | ||
734 | profile_task_exit(tsk); | |
735 | kcov_task_exit(tsk); | |
736 | ||
737 | WARN_ON(blk_needs_flush_plug(tsk)); | |
738 | ||
739 | if (unlikely(in_interrupt())) | |
740 | panic("Aiee, killing interrupt handler!"); | |
741 | if (unlikely(!tsk->pid)) | |
742 | panic("Attempted to kill the idle task!"); | |
743 | ||
744 | /* | |
745 | * If do_exit is called because this processes oopsed, it's possible | |
746 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before | |
747 | * continuing. Amongst other possible reasons, this is to prevent | |
748 | * mm_release()->clear_child_tid() from writing to a user-controlled | |
749 | * kernel address. | |
750 | */ | |
751 | set_fs(USER_DS); | |
752 | ||
753 | ptrace_event(PTRACE_EVENT_EXIT, code); | |
754 | ||
755 | validate_creds_for_do_exit(tsk); | |
756 | ||
757 | /* | |
758 | * We're taking recursive faults here in do_exit. Safest is to just | |
759 | * leave this task alone and wait for reboot. | |
760 | */ | |
761 | if (unlikely(tsk->flags & PF_EXITING)) { | |
762 | pr_alert("Fixing recursive fault but reboot is needed!\n"); | |
763 | /* | |
764 | * We can do this unlocked here. The futex code uses | |
765 | * this flag just to verify whether the pi state | |
766 | * cleanup has been done or not. In the worst case it | |
767 | * loops once more. We pretend that the cleanup was | |
768 | * done as there is no way to return. Either the | |
769 | * OWNER_DIED bit is set by now or we push the blocked | |
770 | * task into the wait for ever nirwana as well. | |
771 | */ | |
772 | tsk->flags |= PF_EXITPIDONE; | |
773 | set_current_state(TASK_UNINTERRUPTIBLE); | |
774 | schedule(); | |
775 | } | |
776 | ||
777 | exit_signals(tsk); /* sets PF_EXITING */ | |
778 | /* | |
779 | * Ensure that all new tsk->pi_lock acquisitions must observe | |
780 | * PF_EXITING. Serializes against futex.c:attach_to_pi_owner(). | |
781 | */ | |
782 | smp_mb(); | |
783 | /* | |
784 | * Ensure that we must observe the pi_state in exit_mm() -> | |
785 | * mm_release() -> exit_pi_state_list(). | |
786 | */ | |
787 | raw_spin_unlock_wait(&tsk->pi_lock); | |
788 | ||
789 | if (unlikely(in_atomic())) { | |
790 | pr_info("note: %s[%d] exited with preempt_count %d\n", | |
791 | current->comm, task_pid_nr(current), | |
792 | preempt_count()); | |
793 | preempt_count_set(PREEMPT_ENABLED); | |
794 | } | |
795 | ||
796 | /* sync mm's RSS info before statistics gathering */ | |
797 | if (tsk->mm) | |
798 | sync_mm_rss(tsk->mm); | |
799 | acct_update_integrals(tsk); | |
800 | group_dead = atomic_dec_and_test(&tsk->signal->live); | |
801 | if (group_dead) { | |
802 | hrtimer_cancel(&tsk->signal->real_timer); | |
803 | exit_itimers(tsk->signal); | |
804 | if (tsk->mm) | |
805 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); | |
806 | } | |
807 | acct_collect(code, group_dead); | |
808 | if (group_dead) | |
809 | tty_audit_exit(); | |
810 | audit_free(tsk); | |
811 | ||
812 | tsk->exit_code = code; | |
813 | taskstats_exit(tsk, group_dead); | |
814 | ||
815 | exit_mm(tsk); | |
816 | ||
817 | if (group_dead) | |
818 | acct_process(); | |
819 | trace_sched_process_exit(tsk); | |
820 | ||
821 | exit_sem(tsk); | |
822 | exit_shm(tsk); | |
823 | exit_files(tsk); | |
824 | exit_fs(tsk); | |
825 | if (group_dead) | |
826 | disassociate_ctty(1); | |
827 | exit_task_namespaces(tsk); | |
828 | exit_task_work(tsk); | |
829 | exit_thread(tsk); | |
830 | ||
831 | /* | |
832 | * Flush inherited counters to the parent - before the parent | |
833 | * gets woken up by child-exit notifications. | |
834 | * | |
835 | * because of cgroup mode, must be called before cgroup_exit() | |
836 | */ | |
837 | perf_event_exit_task(tsk); | |
838 | ||
839 | cgroup_exit(tsk); | |
840 | ||
841 | /* | |
842 | * FIXME: do that only when needed, using sched_exit tracepoint | |
843 | */ | |
844 | flush_ptrace_hw_breakpoint(tsk); | |
845 | ||
846 | TASKS_RCU(preempt_disable()); | |
847 | TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu)); | |
848 | TASKS_RCU(preempt_enable()); | |
849 | exit_notify(tsk, group_dead); | |
850 | proc_exit_connector(tsk); | |
851 | #ifdef CONFIG_NUMA | |
852 | task_lock(tsk); | |
853 | mpol_put(tsk->mempolicy); | |
854 | tsk->mempolicy = NULL; | |
855 | task_unlock(tsk); | |
856 | #endif | |
857 | #ifdef CONFIG_FUTEX | |
858 | if (unlikely(current->pi_state_cache)) | |
859 | kfree(current->pi_state_cache); | |
860 | #endif | |
861 | /* | |
862 | * Make sure we are holding no locks: | |
863 | */ | |
864 | debug_check_no_locks_held(); | |
865 | /* | |
866 | * We can do this unlocked here. The futex code uses this flag | |
867 | * just to verify whether the pi state cleanup has been done | |
868 | * or not. In the worst case it loops once more. | |
869 | */ | |
870 | tsk->flags |= PF_EXITPIDONE; | |
871 | ||
872 | if (tsk->io_context) | |
873 | exit_io_context(tsk); | |
874 | ||
875 | if (tsk->splice_pipe) | |
876 | free_pipe_info(tsk->splice_pipe); | |
877 | ||
878 | if (tsk->task_frag.page) | |
879 | put_page(tsk->task_frag.page); | |
880 | ||
881 | validate_creds_for_do_exit(tsk); | |
882 | ||
883 | check_stack_usage(); | |
884 | preempt_disable(); | |
885 | if (tsk->nr_dirtied) | |
886 | __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); | |
887 | exit_rcu(); | |
888 | TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i)); | |
889 | ||
890 | /* | |
891 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed | |
892 | * when the following two conditions become true. | |
893 | * - There is race condition of mmap_sem (It is acquired by | |
894 | * exit_mm()), and | |
895 | * - SMI occurs before setting TASK_RUNINNG. | |
896 | * (or hypervisor of virtual machine switches to other guest) | |
897 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD | |
898 | * | |
899 | * To avoid it, we have to wait for releasing tsk->pi_lock which | |
900 | * is held by try_to_wake_up() | |
901 | */ | |
902 | smp_mb(); | |
903 | raw_spin_unlock_wait(&tsk->pi_lock); | |
904 | ||
905 | /* causes final put_task_struct in finish_task_switch(). */ | |
906 | tsk->state = TASK_DEAD; | |
907 | tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */ | |
908 | schedule(); | |
909 | BUG(); | |
910 | /* Avoid "noreturn function does return". */ | |
911 | for (;;) | |
912 | cpu_relax(); /* For when BUG is null */ | |
913 | } | |
914 | EXPORT_SYMBOL_GPL(do_exit); | |
915 | ||
916 | void complete_and_exit(struct completion *comp, long code) | |
917 | { | |
918 | if (comp) | |
919 | complete(comp); | |
920 | ||
921 | do_exit(code); | |
922 | } | |
923 | EXPORT_SYMBOL(complete_and_exit); | |
924 | ||
925 | SYSCALL_DEFINE1(exit, int, error_code) | |
926 | { | |
927 | do_exit((error_code&0xff)<<8); | |
928 | } | |
929 | ||
930 | /* | |
931 | * Take down every thread in the group. This is called by fatal signals | |
932 | * as well as by sys_exit_group (below). | |
933 | */ | |
934 | void | |
935 | do_group_exit(int exit_code) | |
936 | { | |
937 | struct signal_struct *sig = current->signal; | |
938 | ||
939 | BUG_ON(exit_code & 0x80); /* core dumps don't get here */ | |
940 | ||
941 | if (signal_group_exit(sig)) | |
942 | exit_code = sig->group_exit_code; | |
943 | else if (!thread_group_empty(current)) { | |
944 | struct sighand_struct *const sighand = current->sighand; | |
945 | ||
946 | spin_lock_irq(&sighand->siglock); | |
947 | if (signal_group_exit(sig)) | |
948 | /* Another thread got here before we took the lock. */ | |
949 | exit_code = sig->group_exit_code; | |
950 | else { | |
951 | sig->group_exit_code = exit_code; | |
952 | sig->flags = SIGNAL_GROUP_EXIT; | |
953 | zap_other_threads(current); | |
954 | } | |
955 | spin_unlock_irq(&sighand->siglock); | |
956 | } | |
957 | ||
958 | do_exit(exit_code); | |
959 | /* NOTREACHED */ | |
960 | } | |
961 | ||
962 | /* | |
963 | * this kills every thread in the thread group. Note that any externally | |
964 | * wait4()-ing process will get the correct exit code - even if this | |
965 | * thread is not the thread group leader. | |
966 | */ | |
967 | SYSCALL_DEFINE1(exit_group, int, error_code) | |
968 | { | |
969 | do_group_exit((error_code & 0xff) << 8); | |
970 | /* NOTREACHED */ | |
971 | return 0; | |
972 | } | |
973 | ||
974 | struct wait_opts { | |
975 | enum pid_type wo_type; | |
976 | int wo_flags; | |
977 | struct pid *wo_pid; | |
978 | ||
979 | struct siginfo __user *wo_info; | |
980 | int __user *wo_stat; | |
981 | struct rusage __user *wo_rusage; | |
982 | ||
983 | wait_queue_t child_wait; | |
984 | int notask_error; | |
985 | }; | |
986 | ||
987 | static inline | |
988 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type) | |
989 | { | |
990 | if (type != PIDTYPE_PID) | |
991 | task = task->group_leader; | |
992 | return task->pids[type].pid; | |
993 | } | |
994 | ||
995 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p) | |
996 | { | |
997 | return wo->wo_type == PIDTYPE_MAX || | |
998 | task_pid_type(p, wo->wo_type) == wo->wo_pid; | |
999 | } | |
1000 | ||
1001 | static int | |
1002 | eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) | |
1003 | { | |
1004 | if (!eligible_pid(wo, p)) | |
1005 | return 0; | |
1006 | ||
1007 | /* | |
1008 | * Wait for all children (clone and not) if __WALL is set or | |
1009 | * if it is traced by us. | |
1010 | */ | |
1011 | if (ptrace || (wo->wo_flags & __WALL)) | |
1012 | return 1; | |
1013 | ||
1014 | /* | |
1015 | * Otherwise, wait for clone children *only* if __WCLONE is set; | |
1016 | * otherwise, wait for non-clone children *only*. | |
1017 | * | |
1018 | * Note: a "clone" child here is one that reports to its parent | |
1019 | * using a signal other than SIGCHLD, or a non-leader thread which | |
1020 | * we can only see if it is traced by us. | |
1021 | */ | |
1022 | if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) | |
1023 | return 0; | |
1024 | ||
1025 | return 1; | |
1026 | } | |
1027 | ||
1028 | static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p, | |
1029 | pid_t pid, uid_t uid, int why, int status) | |
1030 | { | |
1031 | struct siginfo __user *infop; | |
1032 | int retval = wo->wo_rusage | |
1033 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; | |
1034 | ||
1035 | put_task_struct(p); | |
1036 | infop = wo->wo_info; | |
1037 | if (infop) { | |
1038 | if (!retval) | |
1039 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1040 | if (!retval) | |
1041 | retval = put_user(0, &infop->si_errno); | |
1042 | if (!retval) | |
1043 | retval = put_user((short)why, &infop->si_code); | |
1044 | if (!retval) | |
1045 | retval = put_user(pid, &infop->si_pid); | |
1046 | if (!retval) | |
1047 | retval = put_user(uid, &infop->si_uid); | |
1048 | if (!retval) | |
1049 | retval = put_user(status, &infop->si_status); | |
1050 | } | |
1051 | if (!retval) | |
1052 | retval = pid; | |
1053 | return retval; | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold | |
1058 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1059 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1060 | * released the lock and the system call should return. | |
1061 | */ | |
1062 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) | |
1063 | { | |
1064 | int state, retval, status; | |
1065 | pid_t pid = task_pid_vnr(p); | |
1066 | uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1067 | struct siginfo __user *infop; | |
1068 | ||
1069 | if (!likely(wo->wo_flags & WEXITED)) | |
1070 | return 0; | |
1071 | ||
1072 | if (unlikely(wo->wo_flags & WNOWAIT)) { | |
1073 | int exit_code = p->exit_code; | |
1074 | int why; | |
1075 | ||
1076 | get_task_struct(p); | |
1077 | read_unlock(&tasklist_lock); | |
1078 | sched_annotate_sleep(); | |
1079 | ||
1080 | if ((exit_code & 0x7f) == 0) { | |
1081 | why = CLD_EXITED; | |
1082 | status = exit_code >> 8; | |
1083 | } else { | |
1084 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1085 | status = exit_code & 0x7f; | |
1086 | } | |
1087 | return wait_noreap_copyout(wo, p, pid, uid, why, status); | |
1088 | } | |
1089 | /* | |
1090 | * Move the task's state to DEAD/TRACE, only one thread can do this. | |
1091 | */ | |
1092 | state = (ptrace_reparented(p) && thread_group_leader(p)) ? | |
1093 | EXIT_TRACE : EXIT_DEAD; | |
1094 | if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) | |
1095 | return 0; | |
1096 | /* | |
1097 | * We own this thread, nobody else can reap it. | |
1098 | */ | |
1099 | read_unlock(&tasklist_lock); | |
1100 | sched_annotate_sleep(); | |
1101 | ||
1102 | /* | |
1103 | * Check thread_group_leader() to exclude the traced sub-threads. | |
1104 | */ | |
1105 | if (state == EXIT_DEAD && thread_group_leader(p)) { | |
1106 | struct signal_struct *sig = p->signal; | |
1107 | struct signal_struct *psig = current->signal; | |
1108 | unsigned long maxrss; | |
1109 | cputime_t tgutime, tgstime; | |
1110 | ||
1111 | /* | |
1112 | * The resource counters for the group leader are in its | |
1113 | * own task_struct. Those for dead threads in the group | |
1114 | * are in its signal_struct, as are those for the child | |
1115 | * processes it has previously reaped. All these | |
1116 | * accumulate in the parent's signal_struct c* fields. | |
1117 | * | |
1118 | * We don't bother to take a lock here to protect these | |
1119 | * p->signal fields because the whole thread group is dead | |
1120 | * and nobody can change them. | |
1121 | * | |
1122 | * psig->stats_lock also protects us from our sub-theads | |
1123 | * which can reap other children at the same time. Until | |
1124 | * we change k_getrusage()-like users to rely on this lock | |
1125 | * we have to take ->siglock as well. | |
1126 | * | |
1127 | * We use thread_group_cputime_adjusted() to get times for | |
1128 | * the thread group, which consolidates times for all threads | |
1129 | * in the group including the group leader. | |
1130 | */ | |
1131 | thread_group_cputime_adjusted(p, &tgutime, &tgstime); | |
1132 | spin_lock_irq(¤t->sighand->siglock); | |
1133 | write_seqlock(&psig->stats_lock); | |
1134 | psig->cutime += tgutime + sig->cutime; | |
1135 | psig->cstime += tgstime + sig->cstime; | |
1136 | psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; | |
1137 | psig->cmin_flt += | |
1138 | p->min_flt + sig->min_flt + sig->cmin_flt; | |
1139 | psig->cmaj_flt += | |
1140 | p->maj_flt + sig->maj_flt + sig->cmaj_flt; | |
1141 | psig->cnvcsw += | |
1142 | p->nvcsw + sig->nvcsw + sig->cnvcsw; | |
1143 | psig->cnivcsw += | |
1144 | p->nivcsw + sig->nivcsw + sig->cnivcsw; | |
1145 | psig->cinblock += | |
1146 | task_io_get_inblock(p) + | |
1147 | sig->inblock + sig->cinblock; | |
1148 | psig->coublock += | |
1149 | task_io_get_oublock(p) + | |
1150 | sig->oublock + sig->coublock; | |
1151 | maxrss = max(sig->maxrss, sig->cmaxrss); | |
1152 | if (psig->cmaxrss < maxrss) | |
1153 | psig->cmaxrss = maxrss; | |
1154 | task_io_accounting_add(&psig->ioac, &p->ioac); | |
1155 | task_io_accounting_add(&psig->ioac, &sig->ioac); | |
1156 | write_sequnlock(&psig->stats_lock); | |
1157 | spin_unlock_irq(¤t->sighand->siglock); | |
1158 | } | |
1159 | ||
1160 | retval = wo->wo_rusage | |
1161 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; | |
1162 | status = (p->signal->flags & SIGNAL_GROUP_EXIT) | |
1163 | ? p->signal->group_exit_code : p->exit_code; | |
1164 | if (!retval && wo->wo_stat) | |
1165 | retval = put_user(status, wo->wo_stat); | |
1166 | ||
1167 | infop = wo->wo_info; | |
1168 | if (!retval && infop) | |
1169 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1170 | if (!retval && infop) | |
1171 | retval = put_user(0, &infop->si_errno); | |
1172 | if (!retval && infop) { | |
1173 | int why; | |
1174 | ||
1175 | if ((status & 0x7f) == 0) { | |
1176 | why = CLD_EXITED; | |
1177 | status >>= 8; | |
1178 | } else { | |
1179 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1180 | status &= 0x7f; | |
1181 | } | |
1182 | retval = put_user((short)why, &infop->si_code); | |
1183 | if (!retval) | |
1184 | retval = put_user(status, &infop->si_status); | |
1185 | } | |
1186 | if (!retval && infop) | |
1187 | retval = put_user(pid, &infop->si_pid); | |
1188 | if (!retval && infop) | |
1189 | retval = put_user(uid, &infop->si_uid); | |
1190 | if (!retval) | |
1191 | retval = pid; | |
1192 | ||
1193 | if (state == EXIT_TRACE) { | |
1194 | write_lock_irq(&tasklist_lock); | |
1195 | /* We dropped tasklist, ptracer could die and untrace */ | |
1196 | ptrace_unlink(p); | |
1197 | ||
1198 | /* If parent wants a zombie, don't release it now */ | |
1199 | state = EXIT_ZOMBIE; | |
1200 | if (do_notify_parent(p, p->exit_signal)) | |
1201 | state = EXIT_DEAD; | |
1202 | p->exit_state = state; | |
1203 | write_unlock_irq(&tasklist_lock); | |
1204 | } | |
1205 | if (state == EXIT_DEAD) | |
1206 | release_task(p); | |
1207 | ||
1208 | return retval; | |
1209 | } | |
1210 | ||
1211 | static int *task_stopped_code(struct task_struct *p, bool ptrace) | |
1212 | { | |
1213 | if (ptrace) { | |
1214 | if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) | |
1215 | return &p->exit_code; | |
1216 | } else { | |
1217 | if (p->signal->flags & SIGNAL_STOP_STOPPED) | |
1218 | return &p->signal->group_exit_code; | |
1219 | } | |
1220 | return NULL; | |
1221 | } | |
1222 | ||
1223 | /** | |
1224 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED | |
1225 | * @wo: wait options | |
1226 | * @ptrace: is the wait for ptrace | |
1227 | * @p: task to wait for | |
1228 | * | |
1229 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. | |
1230 | * | |
1231 | * CONTEXT: | |
1232 | * read_lock(&tasklist_lock), which is released if return value is | |
1233 | * non-zero. Also, grabs and releases @p->sighand->siglock. | |
1234 | * | |
1235 | * RETURNS: | |
1236 | * 0 if wait condition didn't exist and search for other wait conditions | |
1237 | * should continue. Non-zero return, -errno on failure and @p's pid on | |
1238 | * success, implies that tasklist_lock is released and wait condition | |
1239 | * search should terminate. | |
1240 | */ | |
1241 | static int wait_task_stopped(struct wait_opts *wo, | |
1242 | int ptrace, struct task_struct *p) | |
1243 | { | |
1244 | struct siginfo __user *infop; | |
1245 | int retval, exit_code, *p_code, why; | |
1246 | uid_t uid = 0; /* unneeded, required by compiler */ | |
1247 | pid_t pid; | |
1248 | ||
1249 | /* | |
1250 | * Traditionally we see ptrace'd stopped tasks regardless of options. | |
1251 | */ | |
1252 | if (!ptrace && !(wo->wo_flags & WUNTRACED)) | |
1253 | return 0; | |
1254 | ||
1255 | if (!task_stopped_code(p, ptrace)) | |
1256 | return 0; | |
1257 | ||
1258 | exit_code = 0; | |
1259 | spin_lock_irq(&p->sighand->siglock); | |
1260 | ||
1261 | p_code = task_stopped_code(p, ptrace); | |
1262 | if (unlikely(!p_code)) | |
1263 | goto unlock_sig; | |
1264 | ||
1265 | exit_code = *p_code; | |
1266 | if (!exit_code) | |
1267 | goto unlock_sig; | |
1268 | ||
1269 | if (!unlikely(wo->wo_flags & WNOWAIT)) | |
1270 | *p_code = 0; | |
1271 | ||
1272 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1273 | unlock_sig: | |
1274 | spin_unlock_irq(&p->sighand->siglock); | |
1275 | if (!exit_code) | |
1276 | return 0; | |
1277 | ||
1278 | /* | |
1279 | * Now we are pretty sure this task is interesting. | |
1280 | * Make sure it doesn't get reaped out from under us while we | |
1281 | * give up the lock and then examine it below. We don't want to | |
1282 | * keep holding onto the tasklist_lock while we call getrusage and | |
1283 | * possibly take page faults for user memory. | |
1284 | */ | |
1285 | get_task_struct(p); | |
1286 | pid = task_pid_vnr(p); | |
1287 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED; | |
1288 | read_unlock(&tasklist_lock); | |
1289 | sched_annotate_sleep(); | |
1290 | ||
1291 | if (unlikely(wo->wo_flags & WNOWAIT)) | |
1292 | return wait_noreap_copyout(wo, p, pid, uid, why, exit_code); | |
1293 | ||
1294 | retval = wo->wo_rusage | |
1295 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; | |
1296 | if (!retval && wo->wo_stat) | |
1297 | retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat); | |
1298 | ||
1299 | infop = wo->wo_info; | |
1300 | if (!retval && infop) | |
1301 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1302 | if (!retval && infop) | |
1303 | retval = put_user(0, &infop->si_errno); | |
1304 | if (!retval && infop) | |
1305 | retval = put_user((short)why, &infop->si_code); | |
1306 | if (!retval && infop) | |
1307 | retval = put_user(exit_code, &infop->si_status); | |
1308 | if (!retval && infop) | |
1309 | retval = put_user(pid, &infop->si_pid); | |
1310 | if (!retval && infop) | |
1311 | retval = put_user(uid, &infop->si_uid); | |
1312 | if (!retval) | |
1313 | retval = pid; | |
1314 | put_task_struct(p); | |
1315 | ||
1316 | BUG_ON(!retval); | |
1317 | return retval; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Handle do_wait work for one task in a live, non-stopped state. | |
1322 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1323 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1324 | * released the lock and the system call should return. | |
1325 | */ | |
1326 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) | |
1327 | { | |
1328 | int retval; | |
1329 | pid_t pid; | |
1330 | uid_t uid; | |
1331 | ||
1332 | if (!unlikely(wo->wo_flags & WCONTINUED)) | |
1333 | return 0; | |
1334 | ||
1335 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) | |
1336 | return 0; | |
1337 | ||
1338 | spin_lock_irq(&p->sighand->siglock); | |
1339 | /* Re-check with the lock held. */ | |
1340 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { | |
1341 | spin_unlock_irq(&p->sighand->siglock); | |
1342 | return 0; | |
1343 | } | |
1344 | if (!unlikely(wo->wo_flags & WNOWAIT)) | |
1345 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED; | |
1346 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1347 | spin_unlock_irq(&p->sighand->siglock); | |
1348 | ||
1349 | pid = task_pid_vnr(p); | |
1350 | get_task_struct(p); | |
1351 | read_unlock(&tasklist_lock); | |
1352 | sched_annotate_sleep(); | |
1353 | ||
1354 | if (!wo->wo_info) { | |
1355 | retval = wo->wo_rusage | |
1356 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; | |
1357 | put_task_struct(p); | |
1358 | if (!retval && wo->wo_stat) | |
1359 | retval = put_user(0xffff, wo->wo_stat); | |
1360 | if (!retval) | |
1361 | retval = pid; | |
1362 | } else { | |
1363 | retval = wait_noreap_copyout(wo, p, pid, uid, | |
1364 | CLD_CONTINUED, SIGCONT); | |
1365 | BUG_ON(retval == 0); | |
1366 | } | |
1367 | ||
1368 | return retval; | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * Consider @p for a wait by @parent. | |
1373 | * | |
1374 | * -ECHILD should be in ->notask_error before the first call. | |
1375 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. | |
1376 | * Returns zero if the search for a child should continue; | |
1377 | * then ->notask_error is 0 if @p is an eligible child, | |
1378 | * or another error from security_task_wait(), or still -ECHILD. | |
1379 | */ | |
1380 | static int wait_consider_task(struct wait_opts *wo, int ptrace, | |
1381 | struct task_struct *p) | |
1382 | { | |
1383 | /* | |
1384 | * We can race with wait_task_zombie() from another thread. | |
1385 | * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition | |
1386 | * can't confuse the checks below. | |
1387 | */ | |
1388 | int exit_state = ACCESS_ONCE(p->exit_state); | |
1389 | int ret; | |
1390 | ||
1391 | if (unlikely(exit_state == EXIT_DEAD)) | |
1392 | return 0; | |
1393 | ||
1394 | ret = eligible_child(wo, ptrace, p); | |
1395 | if (!ret) | |
1396 | return ret; | |
1397 | ||
1398 | ret = security_task_wait(p); | |
1399 | if (unlikely(ret < 0)) { | |
1400 | /* | |
1401 | * If we have not yet seen any eligible child, | |
1402 | * then let this error code replace -ECHILD. | |
1403 | * A permission error will give the user a clue | |
1404 | * to look for security policy problems, rather | |
1405 | * than for mysterious wait bugs. | |
1406 | */ | |
1407 | if (wo->notask_error) | |
1408 | wo->notask_error = ret; | |
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | if (unlikely(exit_state == EXIT_TRACE)) { | |
1413 | /* | |
1414 | * ptrace == 0 means we are the natural parent. In this case | |
1415 | * we should clear notask_error, debugger will notify us. | |
1416 | */ | |
1417 | if (likely(!ptrace)) | |
1418 | wo->notask_error = 0; | |
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | if (likely(!ptrace) && unlikely(p->ptrace)) { | |
1423 | /* | |
1424 | * If it is traced by its real parent's group, just pretend | |
1425 | * the caller is ptrace_do_wait() and reap this child if it | |
1426 | * is zombie. | |
1427 | * | |
1428 | * This also hides group stop state from real parent; otherwise | |
1429 | * a single stop can be reported twice as group and ptrace stop. | |
1430 | * If a ptracer wants to distinguish these two events for its | |
1431 | * own children it should create a separate process which takes | |
1432 | * the role of real parent. | |
1433 | */ | |
1434 | if (!ptrace_reparented(p)) | |
1435 | ptrace = 1; | |
1436 | } | |
1437 | ||
1438 | /* slay zombie? */ | |
1439 | if (exit_state == EXIT_ZOMBIE) { | |
1440 | /* we don't reap group leaders with subthreads */ | |
1441 | if (!delay_group_leader(p)) { | |
1442 | /* | |
1443 | * A zombie ptracee is only visible to its ptracer. | |
1444 | * Notification and reaping will be cascaded to the | |
1445 | * real parent when the ptracer detaches. | |
1446 | */ | |
1447 | if (unlikely(ptrace) || likely(!p->ptrace)) | |
1448 | return wait_task_zombie(wo, p); | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * Allow access to stopped/continued state via zombie by | |
1453 | * falling through. Clearing of notask_error is complex. | |
1454 | * | |
1455 | * When !@ptrace: | |
1456 | * | |
1457 | * If WEXITED is set, notask_error should naturally be | |
1458 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set, | |
1459 | * so, if there are live subthreads, there are events to | |
1460 | * wait for. If all subthreads are dead, it's still safe | |
1461 | * to clear - this function will be called again in finite | |
1462 | * amount time once all the subthreads are released and | |
1463 | * will then return without clearing. | |
1464 | * | |
1465 | * When @ptrace: | |
1466 | * | |
1467 | * Stopped state is per-task and thus can't change once the | |
1468 | * target task dies. Only continued and exited can happen. | |
1469 | * Clear notask_error if WCONTINUED | WEXITED. | |
1470 | */ | |
1471 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) | |
1472 | wo->notask_error = 0; | |
1473 | } else { | |
1474 | /* | |
1475 | * @p is alive and it's gonna stop, continue or exit, so | |
1476 | * there always is something to wait for. | |
1477 | */ | |
1478 | wo->notask_error = 0; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * Wait for stopped. Depending on @ptrace, different stopped state | |
1483 | * is used and the two don't interact with each other. | |
1484 | */ | |
1485 | ret = wait_task_stopped(wo, ptrace, p); | |
1486 | if (ret) | |
1487 | return ret; | |
1488 | ||
1489 | /* | |
1490 | * Wait for continued. There's only one continued state and the | |
1491 | * ptracer can consume it which can confuse the real parent. Don't | |
1492 | * use WCONTINUED from ptracer. You don't need or want it. | |
1493 | */ | |
1494 | return wait_task_continued(wo, p); | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Do the work of do_wait() for one thread in the group, @tsk. | |
1499 | * | |
1500 | * -ECHILD should be in ->notask_error before the first call. | |
1501 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. | |
1502 | * Returns zero if the search for a child should continue; then | |
1503 | * ->notask_error is 0 if there were any eligible children, | |
1504 | * or another error from security_task_wait(), or still -ECHILD. | |
1505 | */ | |
1506 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) | |
1507 | { | |
1508 | struct task_struct *p; | |
1509 | ||
1510 | list_for_each_entry(p, &tsk->children, sibling) { | |
1511 | int ret = wait_consider_task(wo, 0, p); | |
1512 | ||
1513 | if (ret) | |
1514 | return ret; | |
1515 | } | |
1516 | ||
1517 | return 0; | |
1518 | } | |
1519 | ||
1520 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) | |
1521 | { | |
1522 | struct task_struct *p; | |
1523 | ||
1524 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { | |
1525 | int ret = wait_consider_task(wo, 1, p); | |
1526 | ||
1527 | if (ret) | |
1528 | return ret; | |
1529 | } | |
1530 | ||
1531 | return 0; | |
1532 | } | |
1533 | ||
1534 | static int child_wait_callback(wait_queue_t *wait, unsigned mode, | |
1535 | int sync, void *key) | |
1536 | { | |
1537 | struct wait_opts *wo = container_of(wait, struct wait_opts, | |
1538 | child_wait); | |
1539 | struct task_struct *p = key; | |
1540 | ||
1541 | if (!eligible_pid(wo, p)) | |
1542 | return 0; | |
1543 | ||
1544 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) | |
1545 | return 0; | |
1546 | ||
1547 | return default_wake_function(wait, mode, sync, key); | |
1548 | } | |
1549 | ||
1550 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent) | |
1551 | { | |
1552 | __wake_up_sync_key(&parent->signal->wait_chldexit, | |
1553 | TASK_INTERRUPTIBLE, 1, p); | |
1554 | } | |
1555 | ||
1556 | static long do_wait(struct wait_opts *wo) | |
1557 | { | |
1558 | struct task_struct *tsk; | |
1559 | int retval; | |
1560 | ||
1561 | trace_sched_process_wait(wo->wo_pid); | |
1562 | ||
1563 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); | |
1564 | wo->child_wait.private = current; | |
1565 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); | |
1566 | repeat: | |
1567 | /* | |
1568 | * If there is nothing that can match our criteria, just get out. | |
1569 | * We will clear ->notask_error to zero if we see any child that | |
1570 | * might later match our criteria, even if we are not able to reap | |
1571 | * it yet. | |
1572 | */ | |
1573 | wo->notask_error = -ECHILD; | |
1574 | if ((wo->wo_type < PIDTYPE_MAX) && | |
1575 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) | |
1576 | goto notask; | |
1577 | ||
1578 | set_current_state(TASK_INTERRUPTIBLE); | |
1579 | read_lock(&tasklist_lock); | |
1580 | tsk = current; | |
1581 | do { | |
1582 | retval = do_wait_thread(wo, tsk); | |
1583 | if (retval) | |
1584 | goto end; | |
1585 | ||
1586 | retval = ptrace_do_wait(wo, tsk); | |
1587 | if (retval) | |
1588 | goto end; | |
1589 | ||
1590 | if (wo->wo_flags & __WNOTHREAD) | |
1591 | break; | |
1592 | } while_each_thread(current, tsk); | |
1593 | read_unlock(&tasklist_lock); | |
1594 | ||
1595 | notask: | |
1596 | retval = wo->notask_error; | |
1597 | if (!retval && !(wo->wo_flags & WNOHANG)) { | |
1598 | retval = -ERESTARTSYS; | |
1599 | if (!signal_pending(current)) { | |
1600 | schedule(); | |
1601 | goto repeat; | |
1602 | } | |
1603 | } | |
1604 | end: | |
1605 | __set_current_state(TASK_RUNNING); | |
1606 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); | |
1607 | return retval; | |
1608 | } | |
1609 | ||
1610 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, | |
1611 | infop, int, options, struct rusage __user *, ru) | |
1612 | { | |
1613 | struct wait_opts wo; | |
1614 | struct pid *pid = NULL; | |
1615 | enum pid_type type; | |
1616 | long ret; | |
1617 | ||
1618 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| | |
1619 | __WNOTHREAD|__WCLONE|__WALL)) | |
1620 | return -EINVAL; | |
1621 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) | |
1622 | return -EINVAL; | |
1623 | ||
1624 | switch (which) { | |
1625 | case P_ALL: | |
1626 | type = PIDTYPE_MAX; | |
1627 | break; | |
1628 | case P_PID: | |
1629 | type = PIDTYPE_PID; | |
1630 | if (upid <= 0) | |
1631 | return -EINVAL; | |
1632 | break; | |
1633 | case P_PGID: | |
1634 | type = PIDTYPE_PGID; | |
1635 | if (upid <= 0) | |
1636 | return -EINVAL; | |
1637 | break; | |
1638 | default: | |
1639 | return -EINVAL; | |
1640 | } | |
1641 | ||
1642 | if (type < PIDTYPE_MAX) | |
1643 | pid = find_get_pid(upid); | |
1644 | ||
1645 | wo.wo_type = type; | |
1646 | wo.wo_pid = pid; | |
1647 | wo.wo_flags = options; | |
1648 | wo.wo_info = infop; | |
1649 | wo.wo_stat = NULL; | |
1650 | wo.wo_rusage = ru; | |
1651 | ret = do_wait(&wo); | |
1652 | ||
1653 | if (ret > 0) { | |
1654 | ret = 0; | |
1655 | } else if (infop) { | |
1656 | /* | |
1657 | * For a WNOHANG return, clear out all the fields | |
1658 | * we would set so the user can easily tell the | |
1659 | * difference. | |
1660 | */ | |
1661 | if (!ret) | |
1662 | ret = put_user(0, &infop->si_signo); | |
1663 | if (!ret) | |
1664 | ret = put_user(0, &infop->si_errno); | |
1665 | if (!ret) | |
1666 | ret = put_user(0, &infop->si_code); | |
1667 | if (!ret) | |
1668 | ret = put_user(0, &infop->si_pid); | |
1669 | if (!ret) | |
1670 | ret = put_user(0, &infop->si_uid); | |
1671 | if (!ret) | |
1672 | ret = put_user(0, &infop->si_status); | |
1673 | } | |
1674 | ||
1675 | put_pid(pid); | |
1676 | return ret; | |
1677 | } | |
1678 | ||
1679 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, | |
1680 | int, options, struct rusage __user *, ru) | |
1681 | { | |
1682 | struct wait_opts wo; | |
1683 | struct pid *pid = NULL; | |
1684 | enum pid_type type; | |
1685 | long ret; | |
1686 | ||
1687 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| | |
1688 | __WNOTHREAD|__WCLONE|__WALL)) | |
1689 | return -EINVAL; | |
1690 | ||
1691 | if (upid == -1) | |
1692 | type = PIDTYPE_MAX; | |
1693 | else if (upid < 0) { | |
1694 | type = PIDTYPE_PGID; | |
1695 | pid = find_get_pid(-upid); | |
1696 | } else if (upid == 0) { | |
1697 | type = PIDTYPE_PGID; | |
1698 | pid = get_task_pid(current, PIDTYPE_PGID); | |
1699 | } else /* upid > 0 */ { | |
1700 | type = PIDTYPE_PID; | |
1701 | pid = find_get_pid(upid); | |
1702 | } | |
1703 | ||
1704 | wo.wo_type = type; | |
1705 | wo.wo_pid = pid; | |
1706 | wo.wo_flags = options | WEXITED; | |
1707 | wo.wo_info = NULL; | |
1708 | wo.wo_stat = stat_addr; | |
1709 | wo.wo_rusage = ru; | |
1710 | ret = do_wait(&wo); | |
1711 | put_pid(pid); | |
1712 | ||
1713 | return ret; | |
1714 | } | |
1715 | ||
1716 | #ifdef __ARCH_WANT_SYS_WAITPID | |
1717 | ||
1718 | /* | |
1719 | * sys_waitpid() remains for compatibility. waitpid() should be | |
1720 | * implemented by calling sys_wait4() from libc.a. | |
1721 | */ | |
1722 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) | |
1723 | { | |
1724 | return sys_wait4(pid, stat_addr, options, NULL); | |
1725 | } | |
1726 | ||
1727 | #endif |