]>
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/sched/autogroup.h> | |
10 | #include <linux/sched/mm.h> | |
11 | #include <linux/sched/stat.h> | |
12 | #include <linux/sched/task.h> | |
13 | #include <linux/sched/task_stack.h> | |
14 | #include <linux/sched/cputime.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/capability.h> | |
18 | #include <linux/completion.h> | |
19 | #include <linux/personality.h> | |
20 | #include <linux/tty.h> | |
21 | #include <linux/iocontext.h> | |
22 | #include <linux/key.h> | |
23 | #include <linux/cpu.h> | |
24 | #include <linux/acct.h> | |
25 | #include <linux/tsacct_kern.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/fdtable.h> | |
28 | #include <linux/freezer.h> | |
29 | #include <linux/binfmts.h> | |
30 | #include <linux/nsproxy.h> | |
31 | #include <linux/pid_namespace.h> | |
32 | #include <linux/ptrace.h> | |
33 | #include <linux/profile.h> | |
34 | #include <linux/mount.h> | |
35 | #include <linux/proc_fs.h> | |
36 | #include <linux/kthread.h> | |
37 | #include <linux/mempolicy.h> | |
38 | #include <linux/taskstats_kern.h> | |
39 | #include <linux/delayacct.h> | |
40 | #include <linux/cgroup.h> | |
41 | #include <linux/syscalls.h> | |
42 | #include <linux/signal.h> | |
43 | #include <linux/posix-timers.h> | |
44 | #include <linux/cn_proc.h> | |
45 | #include <linux/mutex.h> | |
46 | #include <linux/futex.h> | |
47 | #include <linux/pipe_fs_i.h> | |
48 | #include <linux/audit.h> /* for audit_free() */ | |
49 | #include <linux/resource.h> | |
50 | #include <linux/blkdev.h> | |
51 | #include <linux/task_io_accounting_ops.h> | |
52 | #include <linux/tracehook.h> | |
53 | #include <linux/fs_struct.h> | |
54 | #include <linux/init_task.h> | |
55 | #include <linux/perf_event.h> | |
56 | #include <trace/events/sched.h> | |
57 | #include <linux/hw_breakpoint.h> | |
58 | #include <linux/oom.h> | |
59 | #include <linux/writeback.h> | |
60 | #include <linux/shm.h> | |
61 | #include <linux/kcov.h> | |
62 | #include <linux/random.h> | |
63 | #include <linux/rcuwait.h> | |
64 | #include <linux/compat.h> | |
65 | ||
66 | #include <linux/uaccess.h> | |
67 | #include <asm/unistd.h> | |
68 | #include <asm/pgtable.h> | |
69 | #include <asm/mmu_context.h> | |
70 | ||
71 | static void __unhash_process(struct task_struct *p, bool group_dead) | |
72 | { | |
73 | nr_threads--; | |
74 | detach_pid(p, PIDTYPE_PID); | |
75 | if (group_dead) { | |
76 | detach_pid(p, PIDTYPE_PGID); | |
77 | detach_pid(p, PIDTYPE_SID); | |
78 | ||
79 | list_del_rcu(&p->tasks); | |
80 | list_del_init(&p->sibling); | |
81 | __this_cpu_dec(process_counts); | |
82 | } | |
83 | list_del_rcu(&p->thread_group); | |
84 | list_del_rcu(&p->thread_node); | |
85 | } | |
86 | ||
87 | /* | |
88 | * This function expects the tasklist_lock write-locked. | |
89 | */ | |
90 | static void __exit_signal(struct task_struct *tsk) | |
91 | { | |
92 | struct signal_struct *sig = tsk->signal; | |
93 | bool group_dead = thread_group_leader(tsk); | |
94 | struct sighand_struct *sighand; | |
95 | struct tty_struct *uninitialized_var(tty); | |
96 | u64 utime, stime; | |
97 | ||
98 | sighand = rcu_dereference_check(tsk->sighand, | |
99 | lockdep_tasklist_lock_is_held()); | |
100 | spin_lock(&sighand->siglock); | |
101 | ||
102 | #ifdef CONFIG_POSIX_TIMERS | |
103 | posix_cpu_timers_exit(tsk); | |
104 | if (group_dead) { | |
105 | posix_cpu_timers_exit_group(tsk); | |
106 | } else { | |
107 | /* | |
108 | * This can only happen if the caller is de_thread(). | |
109 | * FIXME: this is the temporary hack, we should teach | |
110 | * posix-cpu-timers to handle this case correctly. | |
111 | */ | |
112 | if (unlikely(has_group_leader_pid(tsk))) | |
113 | posix_cpu_timers_exit_group(tsk); | |
114 | } | |
115 | #endif | |
116 | ||
117 | if (group_dead) { | |
118 | tty = sig->tty; | |
119 | sig->tty = NULL; | |
120 | } else { | |
121 | /* | |
122 | * If there is any task waiting for the group exit | |
123 | * then notify it: | |
124 | */ | |
125 | if (sig->notify_count > 0 && !--sig->notify_count) | |
126 | wake_up_process(sig->group_exit_task); | |
127 | ||
128 | if (tsk == sig->curr_target) | |
129 | sig->curr_target = next_thread(tsk); | |
130 | } | |
131 | ||
132 | add_device_randomness((const void*) &tsk->se.sum_exec_runtime, | |
133 | sizeof(unsigned long long)); | |
134 | ||
135 | /* | |
136 | * Accumulate here the counters for all threads as they die. We could | |
137 | * skip the group leader because it is the last user of signal_struct, | |
138 | * but we want to avoid the race with thread_group_cputime() which can | |
139 | * see the empty ->thread_head list. | |
140 | */ | |
141 | task_cputime(tsk, &utime, &stime); | |
142 | write_seqlock(&sig->stats_lock); | |
143 | sig->utime += utime; | |
144 | sig->stime += stime; | |
145 | sig->gtime += task_gtime(tsk); | |
146 | sig->min_flt += tsk->min_flt; | |
147 | sig->maj_flt += tsk->maj_flt; | |
148 | sig->nvcsw += tsk->nvcsw; | |
149 | sig->nivcsw += tsk->nivcsw; | |
150 | sig->inblock += task_io_get_inblock(tsk); | |
151 | sig->oublock += task_io_get_oublock(tsk); | |
152 | task_io_accounting_add(&sig->ioac, &tsk->ioac); | |
153 | sig->sum_sched_runtime += tsk->se.sum_exec_runtime; | |
154 | sig->nr_threads--; | |
155 | __unhash_process(tsk, group_dead); | |
156 | write_sequnlock(&sig->stats_lock); | |
157 | ||
158 | /* | |
159 | * Do this under ->siglock, we can race with another thread | |
160 | * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. | |
161 | */ | |
162 | flush_sigqueue(&tsk->pending); | |
163 | tsk->sighand = NULL; | |
164 | spin_unlock(&sighand->siglock); | |
165 | ||
166 | __cleanup_sighand(sighand); | |
167 | clear_tsk_thread_flag(tsk, TIF_SIGPENDING); | |
168 | if (group_dead) { | |
169 | flush_sigqueue(&sig->shared_pending); | |
170 | tty_kref_put(tty); | |
171 | } | |
172 | } | |
173 | ||
174 | static void delayed_put_task_struct(struct rcu_head *rhp) | |
175 | { | |
176 | struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); | |
177 | ||
178 | perf_event_delayed_put(tsk); | |
179 | trace_sched_process_free(tsk); | |
180 | put_task_struct(tsk); | |
181 | } | |
182 | ||
183 | ||
184 | void release_task(struct task_struct *p) | |
185 | { | |
186 | struct task_struct *leader; | |
187 | int zap_leader; | |
188 | repeat: | |
189 | /* don't need to get the RCU readlock here - the process is dead and | |
190 | * can't be modifying its own credentials. But shut RCU-lockdep up */ | |
191 | rcu_read_lock(); | |
192 | atomic_dec(&__task_cred(p)->user->processes); | |
193 | rcu_read_unlock(); | |
194 | ||
195 | proc_flush_task(p); | |
196 | ||
197 | write_lock_irq(&tasklist_lock); | |
198 | ptrace_release_task(p); | |
199 | __exit_signal(p); | |
200 | ||
201 | /* | |
202 | * If we are the last non-leader member of the thread | |
203 | * group, and the leader is zombie, then notify the | |
204 | * group leader's parent process. (if it wants notification.) | |
205 | */ | |
206 | zap_leader = 0; | |
207 | leader = p->group_leader; | |
208 | if (leader != p && thread_group_empty(leader) | |
209 | && leader->exit_state == EXIT_ZOMBIE) { | |
210 | /* | |
211 | * If we were the last child thread and the leader has | |
212 | * exited already, and the leader's parent ignores SIGCHLD, | |
213 | * then we are the one who should release the leader. | |
214 | */ | |
215 | zap_leader = do_notify_parent(leader, leader->exit_signal); | |
216 | if (zap_leader) | |
217 | leader->exit_state = EXIT_DEAD; | |
218 | } | |
219 | ||
220 | write_unlock_irq(&tasklist_lock); | |
221 | release_thread(p); | |
222 | call_rcu(&p->rcu, delayed_put_task_struct); | |
223 | ||
224 | p = leader; | |
225 | if (unlikely(zap_leader)) | |
226 | goto repeat; | |
227 | } | |
228 | ||
229 | /* | |
230 | * Note that if this function returns a valid task_struct pointer (!NULL) | |
231 | * task->usage must remain >0 for the duration of the RCU critical section. | |
232 | */ | |
233 | struct task_struct *task_rcu_dereference(struct task_struct **ptask) | |
234 | { | |
235 | struct sighand_struct *sighand; | |
236 | struct task_struct *task; | |
237 | ||
238 | /* | |
239 | * We need to verify that release_task() was not called and thus | |
240 | * delayed_put_task_struct() can't run and drop the last reference | |
241 | * before rcu_read_unlock(). We check task->sighand != NULL, | |
242 | * but we can read the already freed and reused memory. | |
243 | */ | |
244 | retry: | |
245 | task = rcu_dereference(*ptask); | |
246 | if (!task) | |
247 | return NULL; | |
248 | ||
249 | probe_kernel_address(&task->sighand, sighand); | |
250 | ||
251 | /* | |
252 | * Pairs with atomic_dec_and_test() in put_task_struct(). If this task | |
253 | * was already freed we can not miss the preceding update of this | |
254 | * pointer. | |
255 | */ | |
256 | smp_rmb(); | |
257 | if (unlikely(task != READ_ONCE(*ptask))) | |
258 | goto retry; | |
259 | ||
260 | /* | |
261 | * We've re-checked that "task == *ptask", now we have two different | |
262 | * cases: | |
263 | * | |
264 | * 1. This is actually the same task/task_struct. In this case | |
265 | * sighand != NULL tells us it is still alive. | |
266 | * | |
267 | * 2. This is another task which got the same memory for task_struct. | |
268 | * We can't know this of course, and we can not trust | |
269 | * sighand != NULL. | |
270 | * | |
271 | * In this case we actually return a random value, but this is | |
272 | * correct. | |
273 | * | |
274 | * If we return NULL - we can pretend that we actually noticed that | |
275 | * *ptask was updated when the previous task has exited. Or pretend | |
276 | * that probe_slab_address(&sighand) reads NULL. | |
277 | * | |
278 | * If we return the new task (because sighand is not NULL for any | |
279 | * reason) - this is fine too. This (new) task can't go away before | |
280 | * another gp pass. | |
281 | * | |
282 | * And note: We could even eliminate the false positive if re-read | |
283 | * task->sighand once again to avoid the falsely NULL. But this case | |
284 | * is very unlikely so we don't care. | |
285 | */ | |
286 | if (!sighand) | |
287 | return NULL; | |
288 | ||
289 | return task; | |
290 | } | |
291 | ||
292 | void rcuwait_wake_up(struct rcuwait *w) | |
293 | { | |
294 | struct task_struct *task; | |
295 | ||
296 | rcu_read_lock(); | |
297 | ||
298 | /* | |
299 | * Order condition vs @task, such that everything prior to the load | |
300 | * of @task is visible. This is the condition as to why the user called | |
301 | * rcuwait_trywake() in the first place. Pairs with set_current_state() | |
302 | * barrier (A) in rcuwait_wait_event(). | |
303 | * | |
304 | * WAIT WAKE | |
305 | * [S] tsk = current [S] cond = true | |
306 | * MB (A) MB (B) | |
307 | * [L] cond [L] tsk | |
308 | */ | |
309 | smp_rmb(); /* (B) */ | |
310 | ||
311 | /* | |
312 | * Avoid using task_rcu_dereference() magic as long as we are careful, | |
313 | * see comment in rcuwait_wait_event() regarding ->exit_state. | |
314 | */ | |
315 | task = rcu_dereference(w->task); | |
316 | if (task) | |
317 | wake_up_process(task); | |
318 | rcu_read_unlock(); | |
319 | } | |
320 | ||
321 | /* | |
322 | * Determine if a process group is "orphaned", according to the POSIX | |
323 | * definition in 2.2.2.52. Orphaned process groups are not to be affected | |
324 | * by terminal-generated stop signals. Newly orphaned process groups are | |
325 | * to receive a SIGHUP and a SIGCONT. | |
326 | * | |
327 | * "I ask you, have you ever known what it is to be an orphan?" | |
328 | */ | |
329 | static int will_become_orphaned_pgrp(struct pid *pgrp, | |
330 | struct task_struct *ignored_task) | |
331 | { | |
332 | struct task_struct *p; | |
333 | ||
334 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
335 | if ((p == ignored_task) || | |
336 | (p->exit_state && thread_group_empty(p)) || | |
337 | is_global_init(p->real_parent)) | |
338 | continue; | |
339 | ||
340 | if (task_pgrp(p->real_parent) != pgrp && | |
341 | task_session(p->real_parent) == task_session(p)) | |
342 | return 0; | |
343 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
344 | ||
345 | return 1; | |
346 | } | |
347 | ||
348 | int is_current_pgrp_orphaned(void) | |
349 | { | |
350 | int retval; | |
351 | ||
352 | read_lock(&tasklist_lock); | |
353 | retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); | |
354 | read_unlock(&tasklist_lock); | |
355 | ||
356 | return retval; | |
357 | } | |
358 | ||
359 | static bool has_stopped_jobs(struct pid *pgrp) | |
360 | { | |
361 | struct task_struct *p; | |
362 | ||
363 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { | |
364 | if (p->signal->flags & SIGNAL_STOP_STOPPED) | |
365 | return true; | |
366 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); | |
367 | ||
368 | return false; | |
369 | } | |
370 | ||
371 | /* | |
372 | * Check to see if any process groups have become orphaned as | |
373 | * a result of our exiting, and if they have any stopped jobs, | |
374 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
375 | */ | |
376 | static void | |
377 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) | |
378 | { | |
379 | struct pid *pgrp = task_pgrp(tsk); | |
380 | struct task_struct *ignored_task = tsk; | |
381 | ||
382 | if (!parent) | |
383 | /* exit: our father is in a different pgrp than | |
384 | * we are and we were the only connection outside. | |
385 | */ | |
386 | parent = tsk->real_parent; | |
387 | else | |
388 | /* reparent: our child is in a different pgrp than | |
389 | * we are, and it was the only connection outside. | |
390 | */ | |
391 | ignored_task = NULL; | |
392 | ||
393 | if (task_pgrp(parent) != pgrp && | |
394 | task_session(parent) == task_session(tsk) && | |
395 | will_become_orphaned_pgrp(pgrp, ignored_task) && | |
396 | has_stopped_jobs(pgrp)) { | |
397 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); | |
398 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); | |
399 | } | |
400 | } | |
401 | ||
402 | #ifdef CONFIG_MEMCG | |
403 | /* | |
404 | * A task is exiting. If it owned this mm, find a new owner for the mm. | |
405 | */ | |
406 | void mm_update_next_owner(struct mm_struct *mm) | |
407 | { | |
408 | struct task_struct *c, *g, *p = current; | |
409 | ||
410 | retry: | |
411 | /* | |
412 | * If the exiting or execing task is not the owner, it's | |
413 | * someone else's problem. | |
414 | */ | |
415 | if (mm->owner != p) | |
416 | return; | |
417 | /* | |
418 | * The current owner is exiting/execing and there are no other | |
419 | * candidates. Do not leave the mm pointing to a possibly | |
420 | * freed task structure. | |
421 | */ | |
422 | if (atomic_read(&mm->mm_users) <= 1) { | |
423 | mm->owner = NULL; | |
424 | return; | |
425 | } | |
426 | ||
427 | read_lock(&tasklist_lock); | |
428 | /* | |
429 | * Search in the children | |
430 | */ | |
431 | list_for_each_entry(c, &p->children, sibling) { | |
432 | if (c->mm == mm) | |
433 | goto assign_new_owner; | |
434 | } | |
435 | ||
436 | /* | |
437 | * Search in the siblings | |
438 | */ | |
439 | list_for_each_entry(c, &p->real_parent->children, sibling) { | |
440 | if (c->mm == mm) | |
441 | goto assign_new_owner; | |
442 | } | |
443 | ||
444 | /* | |
445 | * Search through everything else, we should not get here often. | |
446 | */ | |
447 | for_each_process(g) { | |
448 | if (g->flags & PF_KTHREAD) | |
449 | continue; | |
450 | for_each_thread(g, c) { | |
451 | if (c->mm == mm) | |
452 | goto assign_new_owner; | |
453 | if (c->mm) | |
454 | break; | |
455 | } | |
456 | } | |
457 | read_unlock(&tasklist_lock); | |
458 | /* | |
459 | * We found no owner yet mm_users > 1: this implies that we are | |
460 | * most likely racing with swapoff (try_to_unuse()) or /proc or | |
461 | * ptrace or page migration (get_task_mm()). Mark owner as NULL. | |
462 | */ | |
463 | mm->owner = NULL; | |
464 | return; | |
465 | ||
466 | assign_new_owner: | |
467 | BUG_ON(c == p); | |
468 | get_task_struct(c); | |
469 | /* | |
470 | * The task_lock protects c->mm from changing. | |
471 | * We always want mm->owner->mm == mm | |
472 | */ | |
473 | task_lock(c); | |
474 | /* | |
475 | * Delay read_unlock() till we have the task_lock() | |
476 | * to ensure that c does not slip away underneath us | |
477 | */ | |
478 | read_unlock(&tasklist_lock); | |
479 | if (c->mm != mm) { | |
480 | task_unlock(c); | |
481 | put_task_struct(c); | |
482 | goto retry; | |
483 | } | |
484 | mm->owner = c; | |
485 | task_unlock(c); | |
486 | put_task_struct(c); | |
487 | } | |
488 | #endif /* CONFIG_MEMCG */ | |
489 | ||
490 | /* | |
491 | * Turn us into a lazy TLB process if we | |
492 | * aren't already.. | |
493 | */ | |
494 | static void exit_mm(void) | |
495 | { | |
496 | struct mm_struct *mm = current->mm; | |
497 | struct core_state *core_state; | |
498 | ||
499 | mm_release(current, mm); | |
500 | if (!mm) | |
501 | return; | |
502 | sync_mm_rss(mm); | |
503 | /* | |
504 | * Serialize with any possible pending coredump. | |
505 | * We must hold mmap_sem around checking core_state | |
506 | * and clearing tsk->mm. The core-inducing thread | |
507 | * will increment ->nr_threads for each thread in the | |
508 | * group with ->mm != NULL. | |
509 | */ | |
510 | down_read(&mm->mmap_sem); | |
511 | core_state = mm->core_state; | |
512 | if (core_state) { | |
513 | struct core_thread self; | |
514 | ||
515 | up_read(&mm->mmap_sem); | |
516 | ||
517 | self.task = current; | |
518 | self.next = xchg(&core_state->dumper.next, &self); | |
519 | /* | |
520 | * Implies mb(), the result of xchg() must be visible | |
521 | * to core_state->dumper. | |
522 | */ | |
523 | if (atomic_dec_and_test(&core_state->nr_threads)) | |
524 | complete(&core_state->startup); | |
525 | ||
526 | for (;;) { | |
527 | set_current_state(TASK_UNINTERRUPTIBLE); | |
528 | if (!self.task) /* see coredump_finish() */ | |
529 | break; | |
530 | freezable_schedule(); | |
531 | } | |
532 | __set_current_state(TASK_RUNNING); | |
533 | down_read(&mm->mmap_sem); | |
534 | } | |
535 | mmgrab(mm); | |
536 | BUG_ON(mm != current->active_mm); | |
537 | /* more a memory barrier than a real lock */ | |
538 | task_lock(current); | |
539 | current->mm = NULL; | |
540 | up_read(&mm->mmap_sem); | |
541 | enter_lazy_tlb(mm, current); | |
542 | task_unlock(current); | |
543 | mm_update_next_owner(mm); | |
544 | mmput(mm); | |
545 | if (test_thread_flag(TIF_MEMDIE)) | |
546 | exit_oom_victim(); | |
547 | } | |
548 | ||
549 | static struct task_struct *find_alive_thread(struct task_struct *p) | |
550 | { | |
551 | struct task_struct *t; | |
552 | ||
553 | for_each_thread(p, t) { | |
554 | if (!(t->flags & PF_EXITING)) | |
555 | return t; | |
556 | } | |
557 | return NULL; | |
558 | } | |
559 | ||
560 | static struct task_struct *find_child_reaper(struct task_struct *father) | |
561 | __releases(&tasklist_lock) | |
562 | __acquires(&tasklist_lock) | |
563 | { | |
564 | struct pid_namespace *pid_ns = task_active_pid_ns(father); | |
565 | struct task_struct *reaper = pid_ns->child_reaper; | |
566 | ||
567 | if (likely(reaper != father)) | |
568 | return reaper; | |
569 | ||
570 | reaper = find_alive_thread(father); | |
571 | if (reaper) { | |
572 | pid_ns->child_reaper = reaper; | |
573 | return reaper; | |
574 | } | |
575 | ||
576 | write_unlock_irq(&tasklist_lock); | |
577 | if (unlikely(pid_ns == &init_pid_ns)) { | |
578 | panic("Attempted to kill init! exitcode=0x%08x\n", | |
579 | father->signal->group_exit_code ?: father->exit_code); | |
580 | } | |
581 | zap_pid_ns_processes(pid_ns); | |
582 | write_lock_irq(&tasklist_lock); | |
583 | ||
584 | return father; | |
585 | } | |
586 | ||
587 | /* | |
588 | * When we die, we re-parent all our children, and try to: | |
589 | * 1. give them to another thread in our thread group, if such a member exists | |
590 | * 2. give it to the first ancestor process which prctl'd itself as a | |
591 | * child_subreaper for its children (like a service manager) | |
592 | * 3. give it to the init process (PID 1) in our pid namespace | |
593 | */ | |
594 | static struct task_struct *find_new_reaper(struct task_struct *father, | |
595 | struct task_struct *child_reaper) | |
596 | { | |
597 | struct task_struct *thread, *reaper; | |
598 | ||
599 | thread = find_alive_thread(father); | |
600 | if (thread) | |
601 | return thread; | |
602 | ||
603 | if (father->signal->has_child_subreaper) { | |
604 | unsigned int ns_level = task_pid(father)->level; | |
605 | /* | |
606 | * Find the first ->is_child_subreaper ancestor in our pid_ns. | |
607 | * We can't check reaper != child_reaper to ensure we do not | |
608 | * cross the namespaces, the exiting parent could be injected | |
609 | * by setns() + fork(). | |
610 | * We check pid->level, this is slightly more efficient than | |
611 | * task_active_pid_ns(reaper) != task_active_pid_ns(father). | |
612 | */ | |
613 | for (reaper = father->real_parent; | |
614 | task_pid(reaper)->level == ns_level; | |
615 | reaper = reaper->real_parent) { | |
616 | if (reaper == &init_task) | |
617 | break; | |
618 | if (!reaper->signal->is_child_subreaper) | |
619 | continue; | |
620 | thread = find_alive_thread(reaper); | |
621 | if (thread) | |
622 | return thread; | |
623 | } | |
624 | } | |
625 | ||
626 | return child_reaper; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Any that need to be release_task'd are put on the @dead list. | |
631 | */ | |
632 | static void reparent_leader(struct task_struct *father, struct task_struct *p, | |
633 | struct list_head *dead) | |
634 | { | |
635 | if (unlikely(p->exit_state == EXIT_DEAD)) | |
636 | return; | |
637 | ||
638 | /* We don't want people slaying init. */ | |
639 | p->exit_signal = SIGCHLD; | |
640 | ||
641 | /* If it has exited notify the new parent about this child's death. */ | |
642 | if (!p->ptrace && | |
643 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { | |
644 | if (do_notify_parent(p, p->exit_signal)) { | |
645 | p->exit_state = EXIT_DEAD; | |
646 | list_add(&p->ptrace_entry, dead); | |
647 | } | |
648 | } | |
649 | ||
650 | kill_orphaned_pgrp(p, father); | |
651 | } | |
652 | ||
653 | /* | |
654 | * This does two things: | |
655 | * | |
656 | * A. Make init inherit all the child processes | |
657 | * B. Check to see if any process groups have become orphaned | |
658 | * as a result of our exiting, and if they have any stopped | |
659 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
660 | */ | |
661 | static void forget_original_parent(struct task_struct *father, | |
662 | struct list_head *dead) | |
663 | { | |
664 | struct task_struct *p, *t, *reaper; | |
665 | ||
666 | if (unlikely(!list_empty(&father->ptraced))) | |
667 | exit_ptrace(father, dead); | |
668 | ||
669 | /* Can drop and reacquire tasklist_lock */ | |
670 | reaper = find_child_reaper(father); | |
671 | if (list_empty(&father->children)) | |
672 | return; | |
673 | ||
674 | reaper = find_new_reaper(father, reaper); | |
675 | list_for_each_entry(p, &father->children, sibling) { | |
676 | for_each_thread(p, t) { | |
677 | t->real_parent = reaper; | |
678 | BUG_ON((!t->ptrace) != (t->parent == father)); | |
679 | if (likely(!t->ptrace)) | |
680 | t->parent = t->real_parent; | |
681 | if (t->pdeath_signal) | |
682 | group_send_sig_info(t->pdeath_signal, | |
683 | SEND_SIG_NOINFO, t); | |
684 | } | |
685 | /* | |
686 | * If this is a threaded reparent there is no need to | |
687 | * notify anyone anything has happened. | |
688 | */ | |
689 | if (!same_thread_group(reaper, father)) | |
690 | reparent_leader(father, p, dead); | |
691 | } | |
692 | list_splice_tail_init(&father->children, &reaper->children); | |
693 | } | |
694 | ||
695 | /* | |
696 | * Send signals to all our closest relatives so that they know | |
697 | * to properly mourn us.. | |
698 | */ | |
699 | static void exit_notify(struct task_struct *tsk, int group_dead) | |
700 | { | |
701 | bool autoreap; | |
702 | struct task_struct *p, *n; | |
703 | LIST_HEAD(dead); | |
704 | ||
705 | write_lock_irq(&tasklist_lock); | |
706 | forget_original_parent(tsk, &dead); | |
707 | ||
708 | if (group_dead) | |
709 | kill_orphaned_pgrp(tsk->group_leader, NULL); | |
710 | ||
711 | if (unlikely(tsk->ptrace)) { | |
712 | int sig = thread_group_leader(tsk) && | |
713 | thread_group_empty(tsk) && | |
714 | !ptrace_reparented(tsk) ? | |
715 | tsk->exit_signal : SIGCHLD; | |
716 | autoreap = do_notify_parent(tsk, sig); | |
717 | } else if (thread_group_leader(tsk)) { | |
718 | autoreap = thread_group_empty(tsk) && | |
719 | do_notify_parent(tsk, tsk->exit_signal); | |
720 | } else { | |
721 | autoreap = true; | |
722 | } | |
723 | ||
724 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; | |
725 | if (tsk->exit_state == EXIT_DEAD) | |
726 | list_add(&tsk->ptrace_entry, &dead); | |
727 | ||
728 | /* mt-exec, de_thread() is waiting for group leader */ | |
729 | if (unlikely(tsk->signal->notify_count < 0)) | |
730 | wake_up_process(tsk->signal->group_exit_task); | |
731 | write_unlock_irq(&tasklist_lock); | |
732 | ||
733 | list_for_each_entry_safe(p, n, &dead, ptrace_entry) { | |
734 | list_del_init(&p->ptrace_entry); | |
735 | release_task(p); | |
736 | } | |
737 | } | |
738 | ||
739 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
740 | static void check_stack_usage(void) | |
741 | { | |
742 | static DEFINE_SPINLOCK(low_water_lock); | |
743 | static int lowest_to_date = THREAD_SIZE; | |
744 | unsigned long free; | |
745 | ||
746 | free = stack_not_used(current); | |
747 | ||
748 | if (free >= lowest_to_date) | |
749 | return; | |
750 | ||
751 | spin_lock(&low_water_lock); | |
752 | if (free < lowest_to_date) { | |
753 | pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", | |
754 | current->comm, task_pid_nr(current), free); | |
755 | lowest_to_date = free; | |
756 | } | |
757 | spin_unlock(&low_water_lock); | |
758 | } | |
759 | #else | |
760 | static inline void check_stack_usage(void) {} | |
761 | #endif | |
762 | ||
763 | void __noreturn do_exit(long code) | |
764 | { | |
765 | struct task_struct *tsk = current; | |
766 | int group_dead; | |
767 | TASKS_RCU(int tasks_rcu_i); | |
768 | ||
769 | profile_task_exit(tsk); | |
770 | kcov_task_exit(tsk); | |
771 | ||
772 | WARN_ON(blk_needs_flush_plug(tsk)); | |
773 | ||
774 | if (unlikely(in_interrupt())) | |
775 | panic("Aiee, killing interrupt handler!"); | |
776 | if (unlikely(!tsk->pid)) | |
777 | panic("Attempted to kill the idle task!"); | |
778 | ||
779 | /* | |
780 | * If do_exit is called because this processes oopsed, it's possible | |
781 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before | |
782 | * continuing. Amongst other possible reasons, this is to prevent | |
783 | * mm_release()->clear_child_tid() from writing to a user-controlled | |
784 | * kernel address. | |
785 | */ | |
786 | set_fs(USER_DS); | |
787 | ||
788 | ptrace_event(PTRACE_EVENT_EXIT, code); | |
789 | ||
790 | validate_creds_for_do_exit(tsk); | |
791 | ||
792 | /* | |
793 | * We're taking recursive faults here in do_exit. Safest is to just | |
794 | * leave this task alone and wait for reboot. | |
795 | */ | |
796 | if (unlikely(tsk->flags & PF_EXITING)) { | |
797 | pr_alert("Fixing recursive fault but reboot is needed!\n"); | |
798 | /* | |
799 | * We can do this unlocked here. The futex code uses | |
800 | * this flag just to verify whether the pi state | |
801 | * cleanup has been done or not. In the worst case it | |
802 | * loops once more. We pretend that the cleanup was | |
803 | * done as there is no way to return. Either the | |
804 | * OWNER_DIED bit is set by now or we push the blocked | |
805 | * task into the wait for ever nirwana as well. | |
806 | */ | |
807 | tsk->flags |= PF_EXITPIDONE; | |
808 | set_current_state(TASK_UNINTERRUPTIBLE); | |
809 | schedule(); | |
810 | } | |
811 | ||
812 | exit_signals(tsk); /* sets PF_EXITING */ | |
813 | /* | |
814 | * Ensure that all new tsk->pi_lock acquisitions must observe | |
815 | * PF_EXITING. Serializes against futex.c:attach_to_pi_owner(). | |
816 | */ | |
817 | smp_mb(); | |
818 | /* | |
819 | * Ensure that we must observe the pi_state in exit_mm() -> | |
820 | * mm_release() -> exit_pi_state_list(). | |
821 | */ | |
822 | raw_spin_unlock_wait(&tsk->pi_lock); | |
823 | ||
824 | if (unlikely(in_atomic())) { | |
825 | pr_info("note: %s[%d] exited with preempt_count %d\n", | |
826 | current->comm, task_pid_nr(current), | |
827 | preempt_count()); | |
828 | preempt_count_set(PREEMPT_ENABLED); | |
829 | } | |
830 | ||
831 | /* sync mm's RSS info before statistics gathering */ | |
832 | if (tsk->mm) | |
833 | sync_mm_rss(tsk->mm); | |
834 | acct_update_integrals(tsk); | |
835 | group_dead = atomic_dec_and_test(&tsk->signal->live); | |
836 | if (group_dead) { | |
837 | #ifdef CONFIG_POSIX_TIMERS | |
838 | hrtimer_cancel(&tsk->signal->real_timer); | |
839 | exit_itimers(tsk->signal); | |
840 | #endif | |
841 | if (tsk->mm) | |
842 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); | |
843 | } | |
844 | acct_collect(code, group_dead); | |
845 | if (group_dead) | |
846 | tty_audit_exit(); | |
847 | audit_free(tsk); | |
848 | ||
849 | tsk->exit_code = code; | |
850 | taskstats_exit(tsk, group_dead); | |
851 | ||
852 | exit_mm(); | |
853 | ||
854 | if (group_dead) | |
855 | acct_process(); | |
856 | trace_sched_process_exit(tsk); | |
857 | ||
858 | exit_sem(tsk); | |
859 | exit_shm(tsk); | |
860 | exit_files(tsk); | |
861 | exit_fs(tsk); | |
862 | if (group_dead) | |
863 | disassociate_ctty(1); | |
864 | exit_task_namespaces(tsk); | |
865 | exit_task_work(tsk); | |
866 | exit_thread(tsk); | |
867 | ||
868 | /* | |
869 | * Flush inherited counters to the parent - before the parent | |
870 | * gets woken up by child-exit notifications. | |
871 | * | |
872 | * because of cgroup mode, must be called before cgroup_exit() | |
873 | */ | |
874 | perf_event_exit_task(tsk); | |
875 | ||
876 | sched_autogroup_exit_task(tsk); | |
877 | cgroup_exit(tsk); | |
878 | ||
879 | /* | |
880 | * FIXME: do that only when needed, using sched_exit tracepoint | |
881 | */ | |
882 | flush_ptrace_hw_breakpoint(tsk); | |
883 | ||
884 | TASKS_RCU(preempt_disable()); | |
885 | TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu)); | |
886 | TASKS_RCU(preempt_enable()); | |
887 | exit_notify(tsk, group_dead); | |
888 | proc_exit_connector(tsk); | |
889 | mpol_put_task_policy(tsk); | |
890 | #ifdef CONFIG_FUTEX | |
891 | if (unlikely(current->pi_state_cache)) | |
892 | kfree(current->pi_state_cache); | |
893 | #endif | |
894 | /* | |
895 | * Make sure we are holding no locks: | |
896 | */ | |
897 | debug_check_no_locks_held(); | |
898 | /* | |
899 | * We can do this unlocked here. The futex code uses this flag | |
900 | * just to verify whether the pi state cleanup has been done | |
901 | * or not. In the worst case it loops once more. | |
902 | */ | |
903 | tsk->flags |= PF_EXITPIDONE; | |
904 | ||
905 | if (tsk->io_context) | |
906 | exit_io_context(tsk); | |
907 | ||
908 | if (tsk->splice_pipe) | |
909 | free_pipe_info(tsk->splice_pipe); | |
910 | ||
911 | if (tsk->task_frag.page) | |
912 | put_page(tsk->task_frag.page); | |
913 | ||
914 | validate_creds_for_do_exit(tsk); | |
915 | ||
916 | check_stack_usage(); | |
917 | preempt_disable(); | |
918 | if (tsk->nr_dirtied) | |
919 | __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); | |
920 | exit_rcu(); | |
921 | TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i)); | |
922 | ||
923 | do_task_dead(); | |
924 | } | |
925 | EXPORT_SYMBOL_GPL(do_exit); | |
926 | ||
927 | void complete_and_exit(struct completion *comp, long code) | |
928 | { | |
929 | if (comp) | |
930 | complete(comp); | |
931 | ||
932 | do_exit(code); | |
933 | } | |
934 | EXPORT_SYMBOL(complete_and_exit); | |
935 | ||
936 | SYSCALL_DEFINE1(exit, int, error_code) | |
937 | { | |
938 | do_exit((error_code&0xff)<<8); | |
939 | } | |
940 | ||
941 | /* | |
942 | * Take down every thread in the group. This is called by fatal signals | |
943 | * as well as by sys_exit_group (below). | |
944 | */ | |
945 | void | |
946 | do_group_exit(int exit_code) | |
947 | { | |
948 | struct signal_struct *sig = current->signal; | |
949 | ||
950 | BUG_ON(exit_code & 0x80); /* core dumps don't get here */ | |
951 | ||
952 | if (signal_group_exit(sig)) | |
953 | exit_code = sig->group_exit_code; | |
954 | else if (!thread_group_empty(current)) { | |
955 | struct sighand_struct *const sighand = current->sighand; | |
956 | ||
957 | spin_lock_irq(&sighand->siglock); | |
958 | if (signal_group_exit(sig)) | |
959 | /* Another thread got here before we took the lock. */ | |
960 | exit_code = sig->group_exit_code; | |
961 | else { | |
962 | sig->group_exit_code = exit_code; | |
963 | sig->flags = SIGNAL_GROUP_EXIT; | |
964 | zap_other_threads(current); | |
965 | } | |
966 | spin_unlock_irq(&sighand->siglock); | |
967 | } | |
968 | ||
969 | do_exit(exit_code); | |
970 | /* NOTREACHED */ | |
971 | } | |
972 | ||
973 | /* | |
974 | * this kills every thread in the thread group. Note that any externally | |
975 | * wait4()-ing process will get the correct exit code - even if this | |
976 | * thread is not the thread group leader. | |
977 | */ | |
978 | SYSCALL_DEFINE1(exit_group, int, error_code) | |
979 | { | |
980 | do_group_exit((error_code & 0xff) << 8); | |
981 | /* NOTREACHED */ | |
982 | return 0; | |
983 | } | |
984 | ||
985 | struct waitid_info { | |
986 | pid_t pid; | |
987 | uid_t uid; | |
988 | int status; | |
989 | int cause; | |
990 | }; | |
991 | ||
992 | struct wait_opts { | |
993 | enum pid_type wo_type; | |
994 | int wo_flags; | |
995 | struct pid *wo_pid; | |
996 | ||
997 | struct waitid_info *wo_info; | |
998 | int wo_stat; | |
999 | struct rusage *wo_rusage; | |
1000 | ||
1001 | wait_queue_entry_t child_wait; | |
1002 | int notask_error; | |
1003 | }; | |
1004 | ||
1005 | static inline | |
1006 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type) | |
1007 | { | |
1008 | if (type != PIDTYPE_PID) | |
1009 | task = task->group_leader; | |
1010 | return task->pids[type].pid; | |
1011 | } | |
1012 | ||
1013 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p) | |
1014 | { | |
1015 | return wo->wo_type == PIDTYPE_MAX || | |
1016 | task_pid_type(p, wo->wo_type) == wo->wo_pid; | |
1017 | } | |
1018 | ||
1019 | static int | |
1020 | eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) | |
1021 | { | |
1022 | if (!eligible_pid(wo, p)) | |
1023 | return 0; | |
1024 | ||
1025 | /* | |
1026 | * Wait for all children (clone and not) if __WALL is set or | |
1027 | * if it is traced by us. | |
1028 | */ | |
1029 | if (ptrace || (wo->wo_flags & __WALL)) | |
1030 | return 1; | |
1031 | ||
1032 | /* | |
1033 | * Otherwise, wait for clone children *only* if __WCLONE is set; | |
1034 | * otherwise, wait for non-clone children *only*. | |
1035 | * | |
1036 | * Note: a "clone" child here is one that reports to its parent | |
1037 | * using a signal other than SIGCHLD, or a non-leader thread which | |
1038 | * we can only see if it is traced by us. | |
1039 | */ | |
1040 | if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) | |
1041 | return 0; | |
1042 | ||
1043 | return 1; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold | |
1048 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1049 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1050 | * released the lock and the system call should return. | |
1051 | */ | |
1052 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) | |
1053 | { | |
1054 | int state, status; | |
1055 | pid_t pid = task_pid_vnr(p); | |
1056 | uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1057 | struct waitid_info *infop; | |
1058 | ||
1059 | if (!likely(wo->wo_flags & WEXITED)) | |
1060 | return 0; | |
1061 | ||
1062 | if (unlikely(wo->wo_flags & WNOWAIT)) { | |
1063 | status = p->exit_code; | |
1064 | get_task_struct(p); | |
1065 | read_unlock(&tasklist_lock); | |
1066 | sched_annotate_sleep(); | |
1067 | if (wo->wo_rusage) | |
1068 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1069 | put_task_struct(p); | |
1070 | goto out_info; | |
1071 | } | |
1072 | /* | |
1073 | * Move the task's state to DEAD/TRACE, only one thread can do this. | |
1074 | */ | |
1075 | state = (ptrace_reparented(p) && thread_group_leader(p)) ? | |
1076 | EXIT_TRACE : EXIT_DEAD; | |
1077 | if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) | |
1078 | return 0; | |
1079 | /* | |
1080 | * We own this thread, nobody else can reap it. | |
1081 | */ | |
1082 | read_unlock(&tasklist_lock); | |
1083 | sched_annotate_sleep(); | |
1084 | ||
1085 | /* | |
1086 | * Check thread_group_leader() to exclude the traced sub-threads. | |
1087 | */ | |
1088 | if (state == EXIT_DEAD && thread_group_leader(p)) { | |
1089 | struct signal_struct *sig = p->signal; | |
1090 | struct signal_struct *psig = current->signal; | |
1091 | unsigned long maxrss; | |
1092 | u64 tgutime, tgstime; | |
1093 | ||
1094 | /* | |
1095 | * The resource counters for the group leader are in its | |
1096 | * own task_struct. Those for dead threads in the group | |
1097 | * are in its signal_struct, as are those for the child | |
1098 | * processes it has previously reaped. All these | |
1099 | * accumulate in the parent's signal_struct c* fields. | |
1100 | * | |
1101 | * We don't bother to take a lock here to protect these | |
1102 | * p->signal fields because the whole thread group is dead | |
1103 | * and nobody can change them. | |
1104 | * | |
1105 | * psig->stats_lock also protects us from our sub-theads | |
1106 | * which can reap other children at the same time. Until | |
1107 | * we change k_getrusage()-like users to rely on this lock | |
1108 | * we have to take ->siglock as well. | |
1109 | * | |
1110 | * We use thread_group_cputime_adjusted() to get times for | |
1111 | * the thread group, which consolidates times for all threads | |
1112 | * in the group including the group leader. | |
1113 | */ | |
1114 | thread_group_cputime_adjusted(p, &tgutime, &tgstime); | |
1115 | spin_lock_irq(¤t->sighand->siglock); | |
1116 | write_seqlock(&psig->stats_lock); | |
1117 | psig->cutime += tgutime + sig->cutime; | |
1118 | psig->cstime += tgstime + sig->cstime; | |
1119 | psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; | |
1120 | psig->cmin_flt += | |
1121 | p->min_flt + sig->min_flt + sig->cmin_flt; | |
1122 | psig->cmaj_flt += | |
1123 | p->maj_flt + sig->maj_flt + sig->cmaj_flt; | |
1124 | psig->cnvcsw += | |
1125 | p->nvcsw + sig->nvcsw + sig->cnvcsw; | |
1126 | psig->cnivcsw += | |
1127 | p->nivcsw + sig->nivcsw + sig->cnivcsw; | |
1128 | psig->cinblock += | |
1129 | task_io_get_inblock(p) + | |
1130 | sig->inblock + sig->cinblock; | |
1131 | psig->coublock += | |
1132 | task_io_get_oublock(p) + | |
1133 | sig->oublock + sig->coublock; | |
1134 | maxrss = max(sig->maxrss, sig->cmaxrss); | |
1135 | if (psig->cmaxrss < maxrss) | |
1136 | psig->cmaxrss = maxrss; | |
1137 | task_io_accounting_add(&psig->ioac, &p->ioac); | |
1138 | task_io_accounting_add(&psig->ioac, &sig->ioac); | |
1139 | write_sequnlock(&psig->stats_lock); | |
1140 | spin_unlock_irq(¤t->sighand->siglock); | |
1141 | } | |
1142 | ||
1143 | if (wo->wo_rusage) | |
1144 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1145 | status = (p->signal->flags & SIGNAL_GROUP_EXIT) | |
1146 | ? p->signal->group_exit_code : p->exit_code; | |
1147 | wo->wo_stat = status; | |
1148 | ||
1149 | if (state == EXIT_TRACE) { | |
1150 | write_lock_irq(&tasklist_lock); | |
1151 | /* We dropped tasklist, ptracer could die and untrace */ | |
1152 | ptrace_unlink(p); | |
1153 | ||
1154 | /* If parent wants a zombie, don't release it now */ | |
1155 | state = EXIT_ZOMBIE; | |
1156 | if (do_notify_parent(p, p->exit_signal)) | |
1157 | state = EXIT_DEAD; | |
1158 | p->exit_state = state; | |
1159 | write_unlock_irq(&tasklist_lock); | |
1160 | } | |
1161 | if (state == EXIT_DEAD) | |
1162 | release_task(p); | |
1163 | ||
1164 | out_info: | |
1165 | infop = wo->wo_info; | |
1166 | if (infop) { | |
1167 | if ((status & 0x7f) == 0) { | |
1168 | infop->cause = CLD_EXITED; | |
1169 | infop->status = status >> 8; | |
1170 | } else { | |
1171 | infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1172 | infop->status = status & 0x7f; | |
1173 | } | |
1174 | infop->pid = pid; | |
1175 | infop->uid = uid; | |
1176 | } | |
1177 | ||
1178 | return pid; | |
1179 | } | |
1180 | ||
1181 | static int *task_stopped_code(struct task_struct *p, bool ptrace) | |
1182 | { | |
1183 | if (ptrace) { | |
1184 | if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) | |
1185 | return &p->exit_code; | |
1186 | } else { | |
1187 | if (p->signal->flags & SIGNAL_STOP_STOPPED) | |
1188 | return &p->signal->group_exit_code; | |
1189 | } | |
1190 | return NULL; | |
1191 | } | |
1192 | ||
1193 | /** | |
1194 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED | |
1195 | * @wo: wait options | |
1196 | * @ptrace: is the wait for ptrace | |
1197 | * @p: task to wait for | |
1198 | * | |
1199 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. | |
1200 | * | |
1201 | * CONTEXT: | |
1202 | * read_lock(&tasklist_lock), which is released if return value is | |
1203 | * non-zero. Also, grabs and releases @p->sighand->siglock. | |
1204 | * | |
1205 | * RETURNS: | |
1206 | * 0 if wait condition didn't exist and search for other wait conditions | |
1207 | * should continue. Non-zero return, -errno on failure and @p's pid on | |
1208 | * success, implies that tasklist_lock is released and wait condition | |
1209 | * search should terminate. | |
1210 | */ | |
1211 | static int wait_task_stopped(struct wait_opts *wo, | |
1212 | int ptrace, struct task_struct *p) | |
1213 | { | |
1214 | struct waitid_info *infop; | |
1215 | int exit_code, *p_code, why; | |
1216 | uid_t uid = 0; /* unneeded, required by compiler */ | |
1217 | pid_t pid; | |
1218 | ||
1219 | /* | |
1220 | * Traditionally we see ptrace'd stopped tasks regardless of options. | |
1221 | */ | |
1222 | if (!ptrace && !(wo->wo_flags & WUNTRACED)) | |
1223 | return 0; | |
1224 | ||
1225 | if (!task_stopped_code(p, ptrace)) | |
1226 | return 0; | |
1227 | ||
1228 | exit_code = 0; | |
1229 | spin_lock_irq(&p->sighand->siglock); | |
1230 | ||
1231 | p_code = task_stopped_code(p, ptrace); | |
1232 | if (unlikely(!p_code)) | |
1233 | goto unlock_sig; | |
1234 | ||
1235 | exit_code = *p_code; | |
1236 | if (!exit_code) | |
1237 | goto unlock_sig; | |
1238 | ||
1239 | if (!unlikely(wo->wo_flags & WNOWAIT)) | |
1240 | *p_code = 0; | |
1241 | ||
1242 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1243 | unlock_sig: | |
1244 | spin_unlock_irq(&p->sighand->siglock); | |
1245 | if (!exit_code) | |
1246 | return 0; | |
1247 | ||
1248 | /* | |
1249 | * Now we are pretty sure this task is interesting. | |
1250 | * Make sure it doesn't get reaped out from under us while we | |
1251 | * give up the lock and then examine it below. We don't want to | |
1252 | * keep holding onto the tasklist_lock while we call getrusage and | |
1253 | * possibly take page faults for user memory. | |
1254 | */ | |
1255 | get_task_struct(p); | |
1256 | pid = task_pid_vnr(p); | |
1257 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED; | |
1258 | read_unlock(&tasklist_lock); | |
1259 | sched_annotate_sleep(); | |
1260 | if (wo->wo_rusage) | |
1261 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1262 | put_task_struct(p); | |
1263 | ||
1264 | if (likely(!(wo->wo_flags & WNOWAIT))) | |
1265 | wo->wo_stat = (exit_code << 8) | 0x7f; | |
1266 | ||
1267 | infop = wo->wo_info; | |
1268 | if (infop) { | |
1269 | infop->cause = why; | |
1270 | infop->status = exit_code; | |
1271 | infop->pid = pid; | |
1272 | infop->uid = uid; | |
1273 | } | |
1274 | return pid; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Handle do_wait work for one task in a live, non-stopped state. | |
1279 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1280 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1281 | * released the lock and the system call should return. | |
1282 | */ | |
1283 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) | |
1284 | { | |
1285 | struct waitid_info *infop; | |
1286 | pid_t pid; | |
1287 | uid_t uid; | |
1288 | ||
1289 | if (!unlikely(wo->wo_flags & WCONTINUED)) | |
1290 | return 0; | |
1291 | ||
1292 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) | |
1293 | return 0; | |
1294 | ||
1295 | spin_lock_irq(&p->sighand->siglock); | |
1296 | /* Re-check with the lock held. */ | |
1297 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { | |
1298 | spin_unlock_irq(&p->sighand->siglock); | |
1299 | return 0; | |
1300 | } | |
1301 | if (!unlikely(wo->wo_flags & WNOWAIT)) | |
1302 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED; | |
1303 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); | |
1304 | spin_unlock_irq(&p->sighand->siglock); | |
1305 | ||
1306 | pid = task_pid_vnr(p); | |
1307 | get_task_struct(p); | |
1308 | read_unlock(&tasklist_lock); | |
1309 | sched_annotate_sleep(); | |
1310 | if (wo->wo_rusage) | |
1311 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1312 | put_task_struct(p); | |
1313 | ||
1314 | infop = wo->wo_info; | |
1315 | if (!infop) { | |
1316 | wo->wo_stat = 0xffff; | |
1317 | } else { | |
1318 | infop->cause = CLD_CONTINUED; | |
1319 | infop->pid = pid; | |
1320 | infop->uid = uid; | |
1321 | infop->status = SIGCONT; | |
1322 | } | |
1323 | return pid; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Consider @p for a wait by @parent. | |
1328 | * | |
1329 | * -ECHILD should be in ->notask_error before the first call. | |
1330 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. | |
1331 | * Returns zero if the search for a child should continue; | |
1332 | * then ->notask_error is 0 if @p is an eligible child, | |
1333 | * or still -ECHILD. | |
1334 | */ | |
1335 | static int wait_consider_task(struct wait_opts *wo, int ptrace, | |
1336 | struct task_struct *p) | |
1337 | { | |
1338 | /* | |
1339 | * We can race with wait_task_zombie() from another thread. | |
1340 | * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition | |
1341 | * can't confuse the checks below. | |
1342 | */ | |
1343 | int exit_state = ACCESS_ONCE(p->exit_state); | |
1344 | int ret; | |
1345 | ||
1346 | if (unlikely(exit_state == EXIT_DEAD)) | |
1347 | return 0; | |
1348 | ||
1349 | ret = eligible_child(wo, ptrace, p); | |
1350 | if (!ret) | |
1351 | return ret; | |
1352 | ||
1353 | if (unlikely(exit_state == EXIT_TRACE)) { | |
1354 | /* | |
1355 | * ptrace == 0 means we are the natural parent. In this case | |
1356 | * we should clear notask_error, debugger will notify us. | |
1357 | */ | |
1358 | if (likely(!ptrace)) | |
1359 | wo->notask_error = 0; | |
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | if (likely(!ptrace) && unlikely(p->ptrace)) { | |
1364 | /* | |
1365 | * If it is traced by its real parent's group, just pretend | |
1366 | * the caller is ptrace_do_wait() and reap this child if it | |
1367 | * is zombie. | |
1368 | * | |
1369 | * This also hides group stop state from real parent; otherwise | |
1370 | * a single stop can be reported twice as group and ptrace stop. | |
1371 | * If a ptracer wants to distinguish these two events for its | |
1372 | * own children it should create a separate process which takes | |
1373 | * the role of real parent. | |
1374 | */ | |
1375 | if (!ptrace_reparented(p)) | |
1376 | ptrace = 1; | |
1377 | } | |
1378 | ||
1379 | /* slay zombie? */ | |
1380 | if (exit_state == EXIT_ZOMBIE) { | |
1381 | /* we don't reap group leaders with subthreads */ | |
1382 | if (!delay_group_leader(p)) { | |
1383 | /* | |
1384 | * A zombie ptracee is only visible to its ptracer. | |
1385 | * Notification and reaping will be cascaded to the | |
1386 | * real parent when the ptracer detaches. | |
1387 | */ | |
1388 | if (unlikely(ptrace) || likely(!p->ptrace)) | |
1389 | return wait_task_zombie(wo, p); | |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * Allow access to stopped/continued state via zombie by | |
1394 | * falling through. Clearing of notask_error is complex. | |
1395 | * | |
1396 | * When !@ptrace: | |
1397 | * | |
1398 | * If WEXITED is set, notask_error should naturally be | |
1399 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set, | |
1400 | * so, if there are live subthreads, there are events to | |
1401 | * wait for. If all subthreads are dead, it's still safe | |
1402 | * to clear - this function will be called again in finite | |
1403 | * amount time once all the subthreads are released and | |
1404 | * will then return without clearing. | |
1405 | * | |
1406 | * When @ptrace: | |
1407 | * | |
1408 | * Stopped state is per-task and thus can't change once the | |
1409 | * target task dies. Only continued and exited can happen. | |
1410 | * Clear notask_error if WCONTINUED | WEXITED. | |
1411 | */ | |
1412 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) | |
1413 | wo->notask_error = 0; | |
1414 | } else { | |
1415 | /* | |
1416 | * @p is alive and it's gonna stop, continue or exit, so | |
1417 | * there always is something to wait for. | |
1418 | */ | |
1419 | wo->notask_error = 0; | |
1420 | } | |
1421 | ||
1422 | /* | |
1423 | * Wait for stopped. Depending on @ptrace, different stopped state | |
1424 | * is used and the two don't interact with each other. | |
1425 | */ | |
1426 | ret = wait_task_stopped(wo, ptrace, p); | |
1427 | if (ret) | |
1428 | return ret; | |
1429 | ||
1430 | /* | |
1431 | * Wait for continued. There's only one continued state and the | |
1432 | * ptracer can consume it which can confuse the real parent. Don't | |
1433 | * use WCONTINUED from ptracer. You don't need or want it. | |
1434 | */ | |
1435 | return wait_task_continued(wo, p); | |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * Do the work of do_wait() for one thread in the group, @tsk. | |
1440 | * | |
1441 | * -ECHILD should be in ->notask_error before the first call. | |
1442 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. | |
1443 | * Returns zero if the search for a child should continue; then | |
1444 | * ->notask_error is 0 if there were any eligible children, | |
1445 | * or still -ECHILD. | |
1446 | */ | |
1447 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) | |
1448 | { | |
1449 | struct task_struct *p; | |
1450 | ||
1451 | list_for_each_entry(p, &tsk->children, sibling) { | |
1452 | int ret = wait_consider_task(wo, 0, p); | |
1453 | ||
1454 | if (ret) | |
1455 | return ret; | |
1456 | } | |
1457 | ||
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) | |
1462 | { | |
1463 | struct task_struct *p; | |
1464 | ||
1465 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { | |
1466 | int ret = wait_consider_task(wo, 1, p); | |
1467 | ||
1468 | if (ret) | |
1469 | return ret; | |
1470 | } | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode, | |
1476 | int sync, void *key) | |
1477 | { | |
1478 | struct wait_opts *wo = container_of(wait, struct wait_opts, | |
1479 | child_wait); | |
1480 | struct task_struct *p = key; | |
1481 | ||
1482 | if (!eligible_pid(wo, p)) | |
1483 | return 0; | |
1484 | ||
1485 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) | |
1486 | return 0; | |
1487 | ||
1488 | return default_wake_function(wait, mode, sync, key); | |
1489 | } | |
1490 | ||
1491 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent) | |
1492 | { | |
1493 | __wake_up_sync_key(&parent->signal->wait_chldexit, | |
1494 | TASK_INTERRUPTIBLE, 1, p); | |
1495 | } | |
1496 | ||
1497 | static long do_wait(struct wait_opts *wo) | |
1498 | { | |
1499 | struct task_struct *tsk; | |
1500 | int retval; | |
1501 | ||
1502 | trace_sched_process_wait(wo->wo_pid); | |
1503 | ||
1504 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); | |
1505 | wo->child_wait.private = current; | |
1506 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); | |
1507 | repeat: | |
1508 | /* | |
1509 | * If there is nothing that can match our criteria, just get out. | |
1510 | * We will clear ->notask_error to zero if we see any child that | |
1511 | * might later match our criteria, even if we are not able to reap | |
1512 | * it yet. | |
1513 | */ | |
1514 | wo->notask_error = -ECHILD; | |
1515 | if ((wo->wo_type < PIDTYPE_MAX) && | |
1516 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) | |
1517 | goto notask; | |
1518 | ||
1519 | set_current_state(TASK_INTERRUPTIBLE); | |
1520 | read_lock(&tasklist_lock); | |
1521 | tsk = current; | |
1522 | do { | |
1523 | retval = do_wait_thread(wo, tsk); | |
1524 | if (retval) | |
1525 | goto end; | |
1526 | ||
1527 | retval = ptrace_do_wait(wo, tsk); | |
1528 | if (retval) | |
1529 | goto end; | |
1530 | ||
1531 | if (wo->wo_flags & __WNOTHREAD) | |
1532 | break; | |
1533 | } while_each_thread(current, tsk); | |
1534 | read_unlock(&tasklist_lock); | |
1535 | ||
1536 | notask: | |
1537 | retval = wo->notask_error; | |
1538 | if (!retval && !(wo->wo_flags & WNOHANG)) { | |
1539 | retval = -ERESTARTSYS; | |
1540 | if (!signal_pending(current)) { | |
1541 | schedule(); | |
1542 | goto repeat; | |
1543 | } | |
1544 | } | |
1545 | end: | |
1546 | __set_current_state(TASK_RUNNING); | |
1547 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); | |
1548 | return retval; | |
1549 | } | |
1550 | ||
1551 | static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop, | |
1552 | int options, struct rusage *ru) | |
1553 | { | |
1554 | struct wait_opts wo; | |
1555 | struct pid *pid = NULL; | |
1556 | enum pid_type type; | |
1557 | long ret; | |
1558 | ||
1559 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| | |
1560 | __WNOTHREAD|__WCLONE|__WALL)) | |
1561 | return -EINVAL; | |
1562 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) | |
1563 | return -EINVAL; | |
1564 | ||
1565 | switch (which) { | |
1566 | case P_ALL: | |
1567 | type = PIDTYPE_MAX; | |
1568 | break; | |
1569 | case P_PID: | |
1570 | type = PIDTYPE_PID; | |
1571 | if (upid <= 0) | |
1572 | return -EINVAL; | |
1573 | break; | |
1574 | case P_PGID: | |
1575 | type = PIDTYPE_PGID; | |
1576 | if (upid <= 0) | |
1577 | return -EINVAL; | |
1578 | break; | |
1579 | default: | |
1580 | return -EINVAL; | |
1581 | } | |
1582 | ||
1583 | if (type < PIDTYPE_MAX) | |
1584 | pid = find_get_pid(upid); | |
1585 | ||
1586 | wo.wo_type = type; | |
1587 | wo.wo_pid = pid; | |
1588 | wo.wo_flags = options; | |
1589 | wo.wo_info = infop; | |
1590 | wo.wo_rusage = ru; | |
1591 | ret = do_wait(&wo); | |
1592 | ||
1593 | put_pid(pid); | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, | |
1598 | infop, int, options, struct rusage __user *, ru) | |
1599 | { | |
1600 | struct rusage r; | |
1601 | struct waitid_info info = {.status = 0}; | |
1602 | long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL); | |
1603 | int signo = 0; | |
1604 | ||
1605 | if (err > 0) { | |
1606 | signo = SIGCHLD; | |
1607 | err = 0; | |
1608 | if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) | |
1609 | return -EFAULT; | |
1610 | } | |
1611 | if (!infop) | |
1612 | return err; | |
1613 | ||
1614 | if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop))) | |
1615 | goto Efault; | |
1616 | ||
1617 | user_access_begin(); | |
1618 | unsafe_put_user(signo, &infop->si_signo, Efault); | |
1619 | unsafe_put_user(0, &infop->si_errno, Efault); | |
1620 | unsafe_put_user((short)info.cause, &infop->si_code, Efault); | |
1621 | unsafe_put_user(info.pid, &infop->si_pid, Efault); | |
1622 | unsafe_put_user(info.uid, &infop->si_uid, Efault); | |
1623 | unsafe_put_user(info.status, &infop->si_status, Efault); | |
1624 | user_access_end(); | |
1625 | return err; | |
1626 | Efault: | |
1627 | user_access_end(); | |
1628 | return -EFAULT; | |
1629 | } | |
1630 | ||
1631 | long kernel_wait4(pid_t upid, int __user *stat_addr, int options, | |
1632 | struct rusage *ru) | |
1633 | { | |
1634 | struct wait_opts wo; | |
1635 | struct pid *pid = NULL; | |
1636 | enum pid_type type; | |
1637 | long ret; | |
1638 | ||
1639 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| | |
1640 | __WNOTHREAD|__WCLONE|__WALL)) | |
1641 | return -EINVAL; | |
1642 | ||
1643 | /* -INT_MIN is not defined */ | |
1644 | if (upid == INT_MIN) | |
1645 | return -ESRCH; | |
1646 | ||
1647 | if (upid == -1) | |
1648 | type = PIDTYPE_MAX; | |
1649 | else if (upid < 0) { | |
1650 | type = PIDTYPE_PGID; | |
1651 | pid = find_get_pid(-upid); | |
1652 | } else if (upid == 0) { | |
1653 | type = PIDTYPE_PGID; | |
1654 | pid = get_task_pid(current, PIDTYPE_PGID); | |
1655 | } else /* upid > 0 */ { | |
1656 | type = PIDTYPE_PID; | |
1657 | pid = find_get_pid(upid); | |
1658 | } | |
1659 | ||
1660 | wo.wo_type = type; | |
1661 | wo.wo_pid = pid; | |
1662 | wo.wo_flags = options | WEXITED; | |
1663 | wo.wo_info = NULL; | |
1664 | wo.wo_stat = 0; | |
1665 | wo.wo_rusage = ru; | |
1666 | ret = do_wait(&wo); | |
1667 | put_pid(pid); | |
1668 | if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr)) | |
1669 | ret = -EFAULT; | |
1670 | ||
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, | |
1675 | int, options, struct rusage __user *, ru) | |
1676 | { | |
1677 | struct rusage r; | |
1678 | long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); | |
1679 | ||
1680 | if (err > 0) { | |
1681 | if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) | |
1682 | return -EFAULT; | |
1683 | } | |
1684 | return err; | |
1685 | } | |
1686 | ||
1687 | #ifdef __ARCH_WANT_SYS_WAITPID | |
1688 | ||
1689 | /* | |
1690 | * sys_waitpid() remains for compatibility. waitpid() should be | |
1691 | * implemented by calling sys_wait4() from libc.a. | |
1692 | */ | |
1693 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) | |
1694 | { | |
1695 | return sys_wait4(pid, stat_addr, options, NULL); | |
1696 | } | |
1697 | ||
1698 | #endif | |
1699 | ||
1700 | #ifdef CONFIG_COMPAT | |
1701 | COMPAT_SYSCALL_DEFINE4(wait4, | |
1702 | compat_pid_t, pid, | |
1703 | compat_uint_t __user *, stat_addr, | |
1704 | int, options, | |
1705 | struct compat_rusage __user *, ru) | |
1706 | { | |
1707 | struct rusage r; | |
1708 | long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); | |
1709 | if (err > 0) { | |
1710 | if (ru && put_compat_rusage(&r, ru)) | |
1711 | return -EFAULT; | |
1712 | } | |
1713 | return err; | |
1714 | } | |
1715 | ||
1716 | COMPAT_SYSCALL_DEFINE5(waitid, | |
1717 | int, which, compat_pid_t, pid, | |
1718 | struct compat_siginfo __user *, infop, int, options, | |
1719 | struct compat_rusage __user *, uru) | |
1720 | { | |
1721 | struct rusage ru; | |
1722 | struct waitid_info info = {.status = 0}; | |
1723 | long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL); | |
1724 | int signo = 0; | |
1725 | if (err > 0) { | |
1726 | signo = SIGCHLD; | |
1727 | err = 0; | |
1728 | if (uru) { | |
1729 | /* kernel_waitid() overwrites everything in ru */ | |
1730 | if (COMPAT_USE_64BIT_TIME) | |
1731 | err = copy_to_user(uru, &ru, sizeof(ru)); | |
1732 | else | |
1733 | err = put_compat_rusage(&ru, uru); | |
1734 | if (err) | |
1735 | return -EFAULT; | |
1736 | } | |
1737 | } | |
1738 | ||
1739 | if (!infop) | |
1740 | return err; | |
1741 | ||
1742 | if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop))) | |
1743 | goto Efault; | |
1744 | ||
1745 | user_access_begin(); | |
1746 | unsafe_put_user(signo, &infop->si_signo, Efault); | |
1747 | unsafe_put_user(0, &infop->si_errno, Efault); | |
1748 | unsafe_put_user((short)info.cause, &infop->si_code, Efault); | |
1749 | unsafe_put_user(info.pid, &infop->si_pid, Efault); | |
1750 | unsafe_put_user(info.uid, &infop->si_uid, Efault); | |
1751 | unsafe_put_user(info.status, &infop->si_status, Efault); | |
1752 | user_access_end(); | |
1753 | return err; | |
1754 | Efault: | |
1755 | user_access_end(); | |
1756 | return -EFAULT; | |
1757 | } | |
1758 | #endif |