]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/exit.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
1da177e4 LT |
7 | #include <linux/mm.h> |
8 | #include <linux/slab.h> | |
4eb5aaa3 | 9 | #include <linux/sched/autogroup.h> |
6e84f315 | 10 | #include <linux/sched/mm.h> |
03441a34 | 11 | #include <linux/sched/stat.h> |
29930025 | 12 | #include <linux/sched/task.h> |
68db0cf1 | 13 | #include <linux/sched/task_stack.h> |
32ef5517 | 14 | #include <linux/sched/cputime.h> |
1da177e4 | 15 | #include <linux/interrupt.h> |
1da177e4 | 16 | #include <linux/module.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 LT |
18 | #include <linux/completion.h> |
19 | #include <linux/personality.h> | |
20 | #include <linux/tty.h> | |
da9cbc87 | 21 | #include <linux/iocontext.h> |
1da177e4 | 22 | #include <linux/key.h> |
1da177e4 LT |
23 | #include <linux/cpu.h> |
24 | #include <linux/acct.h> | |
8f0ab514 | 25 | #include <linux/tsacct_kern.h> |
1da177e4 | 26 | #include <linux/file.h> |
9f3acc31 | 27 | #include <linux/fdtable.h> |
80d26af8 | 28 | #include <linux/freezer.h> |
1da177e4 | 29 | #include <linux/binfmts.h> |
ab516013 | 30 | #include <linux/nsproxy.h> |
84d73786 | 31 | #include <linux/pid_namespace.h> |
1da177e4 LT |
32 | #include <linux/ptrace.h> |
33 | #include <linux/profile.h> | |
34 | #include <linux/mount.h> | |
35 | #include <linux/proc_fs.h> | |
49d769d5 | 36 | #include <linux/kthread.h> |
1da177e4 | 37 | #include <linux/mempolicy.h> |
c757249a | 38 | #include <linux/taskstats_kern.h> |
ca74e92b | 39 | #include <linux/delayacct.h> |
b4f48b63 | 40 | #include <linux/cgroup.h> |
1da177e4 | 41 | #include <linux/syscalls.h> |
7ed20e1a | 42 | #include <linux/signal.h> |
6a14c5c9 | 43 | #include <linux/posix-timers.h> |
9f46080c | 44 | #include <linux/cn_proc.h> |
de5097c2 | 45 | #include <linux/mutex.h> |
0771dfef | 46 | #include <linux/futex.h> |
b92ce558 | 47 | #include <linux/pipe_fs_i.h> |
fa84cb93 | 48 | #include <linux/audit.h> /* for audit_free() */ |
83cc5ed3 | 49 | #include <linux/resource.h> |
0d67a46d | 50 | #include <linux/blkdev.h> |
6eaeeaba | 51 | #include <linux/task_io_accounting_ops.h> |
30199f5a | 52 | #include <linux/tracehook.h> |
5ad4e53b | 53 | #include <linux/fs_struct.h> |
ca49ca71 | 54 | #include <linux/userfaultfd_k.h> |
d84f4f99 | 55 | #include <linux/init_task.h> |
cdd6c482 | 56 | #include <linux/perf_event.h> |
ad8d75ff | 57 | #include <trace/events/sched.h> |
24f1e32c | 58 | #include <linux/hw_breakpoint.h> |
3d5992d2 | 59 | #include <linux/oom.h> |
54848d73 | 60 | #include <linux/writeback.h> |
40401530 | 61 | #include <linux/shm.h> |
5c9a8750 | 62 | #include <linux/kcov.h> |
53d3eaa3 | 63 | #include <linux/random.h> |
8f95c90c | 64 | #include <linux/rcuwait.h> |
7e95a225 | 65 | #include <linux/compat.h> |
1da177e4 | 66 | |
7c0f6ba6 | 67 | #include <linux/uaccess.h> |
1da177e4 LT |
68 | #include <asm/unistd.h> |
69 | #include <asm/pgtable.h> | |
70 | #include <asm/mmu_context.h> | |
71 | ||
d40e48e0 | 72 | static void __unhash_process(struct task_struct *p, bool group_dead) |
1da177e4 LT |
73 | { |
74 | nr_threads--; | |
50d75f8d | 75 | detach_pid(p, PIDTYPE_PID); |
d40e48e0 | 76 | if (group_dead) { |
1da177e4 LT |
77 | detach_pid(p, PIDTYPE_PGID); |
78 | detach_pid(p, PIDTYPE_SID); | |
c97d9893 | 79 | |
5e85d4ab | 80 | list_del_rcu(&p->tasks); |
9cd80bbb | 81 | list_del_init(&p->sibling); |
909ea964 | 82 | __this_cpu_dec(process_counts); |
1da177e4 | 83 | } |
47e65328 | 84 | list_del_rcu(&p->thread_group); |
0c740d0a | 85 | list_del_rcu(&p->thread_node); |
1da177e4 LT |
86 | } |
87 | ||
6a14c5c9 ON |
88 | /* |
89 | * This function expects the tasklist_lock write-locked. | |
90 | */ | |
91 | static void __exit_signal(struct task_struct *tsk) | |
92 | { | |
93 | struct signal_struct *sig = tsk->signal; | |
d40e48e0 | 94 | bool group_dead = thread_group_leader(tsk); |
6a14c5c9 | 95 | struct sighand_struct *sighand; |
4ada856f | 96 | struct tty_struct *uninitialized_var(tty); |
5613fda9 | 97 | u64 utime, stime; |
6a14c5c9 | 98 | |
d11c563d | 99 | sighand = rcu_dereference_check(tsk->sighand, |
db1466b3 | 100 | lockdep_tasklist_lock_is_held()); |
6a14c5c9 ON |
101 | spin_lock(&sighand->siglock); |
102 | ||
baa73d9e | 103 | #ifdef CONFIG_POSIX_TIMERS |
6a14c5c9 | 104 | posix_cpu_timers_exit(tsk); |
d40e48e0 | 105 | if (group_dead) { |
6a14c5c9 | 106 | posix_cpu_timers_exit_group(tsk); |
4a599942 | 107 | } else { |
e0a70217 ON |
108 | /* |
109 | * This can only happen if the caller is de_thread(). | |
110 | * FIXME: this is the temporary hack, we should teach | |
111 | * posix-cpu-timers to handle this case correctly. | |
112 | */ | |
113 | if (unlikely(has_group_leader_pid(tsk))) | |
114 | posix_cpu_timers_exit_group(tsk); | |
baa73d9e NP |
115 | } |
116 | #endif | |
e0a70217 | 117 | |
baa73d9e NP |
118 | if (group_dead) { |
119 | tty = sig->tty; | |
120 | sig->tty = NULL; | |
121 | } else { | |
6a14c5c9 ON |
122 | /* |
123 | * If there is any task waiting for the group exit | |
124 | * then notify it: | |
125 | */ | |
d344193a | 126 | if (sig->notify_count > 0 && !--sig->notify_count) |
6a14c5c9 | 127 | wake_up_process(sig->group_exit_task); |
6db840fa | 128 | |
6a14c5c9 ON |
129 | if (tsk == sig->curr_target) |
130 | sig->curr_target = next_thread(tsk); | |
6a14c5c9 ON |
131 | } |
132 | ||
53d3eaa3 NP |
133 | add_device_randomness((const void*) &tsk->se.sum_exec_runtime, |
134 | sizeof(unsigned long long)); | |
135 | ||
90ed9cbe | 136 | /* |
26e75b5c ON |
137 | * Accumulate here the counters for all threads as they die. We could |
138 | * skip the group leader because it is the last user of signal_struct, | |
139 | * but we want to avoid the race with thread_group_cputime() which can | |
140 | * see the empty ->thread_head list. | |
90ed9cbe RR |
141 | */ |
142 | task_cputime(tsk, &utime, &stime); | |
e78c3496 | 143 | write_seqlock(&sig->stats_lock); |
90ed9cbe RR |
144 | sig->utime += utime; |
145 | sig->stime += stime; | |
146 | sig->gtime += task_gtime(tsk); | |
147 | sig->min_flt += tsk->min_flt; | |
148 | sig->maj_flt += tsk->maj_flt; | |
149 | sig->nvcsw += tsk->nvcsw; | |
150 | sig->nivcsw += tsk->nivcsw; | |
151 | sig->inblock += task_io_get_inblock(tsk); | |
152 | sig->oublock += task_io_get_oublock(tsk); | |
153 | task_io_accounting_add(&sig->ioac, &tsk->ioac); | |
154 | sig->sum_sched_runtime += tsk->se.sum_exec_runtime; | |
b3ac022c | 155 | sig->nr_threads--; |
d40e48e0 | 156 | __unhash_process(tsk, group_dead); |
e78c3496 | 157 | write_sequnlock(&sig->stats_lock); |
5876700c | 158 | |
da7978b0 ON |
159 | /* |
160 | * Do this under ->siglock, we can race with another thread | |
161 | * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. | |
162 | */ | |
163 | flush_sigqueue(&tsk->pending); | |
a7e5328a | 164 | tsk->sighand = NULL; |
6a14c5c9 | 165 | spin_unlock(&sighand->siglock); |
6a14c5c9 | 166 | |
a7e5328a | 167 | __cleanup_sighand(sighand); |
a0be55de | 168 | clear_tsk_thread_flag(tsk, TIF_SIGPENDING); |
d40e48e0 | 169 | if (group_dead) { |
6a14c5c9 | 170 | flush_sigqueue(&sig->shared_pending); |
4ada856f | 171 | tty_kref_put(tty); |
6a14c5c9 ON |
172 | } |
173 | } | |
174 | ||
8c7904a0 EB |
175 | static void delayed_put_task_struct(struct rcu_head *rhp) |
176 | { | |
0a16b607 MD |
177 | struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); |
178 | ||
4e231c79 | 179 | perf_event_delayed_put(tsk); |
0a16b607 MD |
180 | trace_sched_process_free(tsk); |
181 | put_task_struct(tsk); | |
8c7904a0 EB |
182 | } |
183 | ||
f470021a | 184 | |
a0be55de | 185 | void release_task(struct task_struct *p) |
1da177e4 | 186 | { |
36c8b586 | 187 | struct task_struct *leader; |
1da177e4 | 188 | int zap_leader; |
1f09f974 | 189 | repeat: |
c69e8d9c | 190 | /* don't need to get the RCU readlock here - the process is dead and |
d11c563d PM |
191 | * can't be modifying its own credentials. But shut RCU-lockdep up */ |
192 | rcu_read_lock(); | |
c69e8d9c | 193 | atomic_dec(&__task_cred(p)->user->processes); |
d11c563d | 194 | rcu_read_unlock(); |
c69e8d9c | 195 | |
60347f67 | 196 | proc_flush_task(p); |
0203026b | 197 | |
1da177e4 | 198 | write_lock_irq(&tasklist_lock); |
a288eecc | 199 | ptrace_release_task(p); |
1da177e4 | 200 | __exit_signal(p); |
35f5cad8 | 201 | |
1da177e4 LT |
202 | /* |
203 | * If we are the last non-leader member of the thread | |
204 | * group, and the leader is zombie, then notify the | |
205 | * group leader's parent process. (if it wants notification.) | |
206 | */ | |
207 | zap_leader = 0; | |
208 | leader = p->group_leader; | |
a0be55de IA |
209 | if (leader != p && thread_group_empty(leader) |
210 | && leader->exit_state == EXIT_ZOMBIE) { | |
1da177e4 LT |
211 | /* |
212 | * If we were the last child thread and the leader has | |
213 | * exited already, and the leader's parent ignores SIGCHLD, | |
214 | * then we are the one who should release the leader. | |
dae33574 | 215 | */ |
86773473 | 216 | zap_leader = do_notify_parent(leader, leader->exit_signal); |
dae33574 RM |
217 | if (zap_leader) |
218 | leader->exit_state = EXIT_DEAD; | |
1da177e4 LT |
219 | } |
220 | ||
1da177e4 | 221 | write_unlock_irq(&tasklist_lock); |
1da177e4 | 222 | release_thread(p); |
8c7904a0 | 223 | call_rcu(&p->rcu, delayed_put_task_struct); |
1da177e4 LT |
224 | |
225 | p = leader; | |
226 | if (unlikely(zap_leader)) | |
227 | goto repeat; | |
228 | } | |
229 | ||
150593bf ON |
230 | /* |
231 | * Note that if this function returns a valid task_struct pointer (!NULL) | |
232 | * task->usage must remain >0 for the duration of the RCU critical section. | |
233 | */ | |
234 | struct task_struct *task_rcu_dereference(struct task_struct **ptask) | |
235 | { | |
236 | struct sighand_struct *sighand; | |
237 | struct task_struct *task; | |
238 | ||
239 | /* | |
240 | * We need to verify that release_task() was not called and thus | |
241 | * delayed_put_task_struct() can't run and drop the last reference | |
242 | * before rcu_read_unlock(). We check task->sighand != NULL, | |
243 | * but we can read the already freed and reused memory. | |
244 | */ | |
245 | retry: | |
246 | task = rcu_dereference(*ptask); | |
247 | if (!task) | |
248 | return NULL; | |
249 | ||
250 | probe_kernel_address(&task->sighand, sighand); | |
251 | ||
252 | /* | |
253 | * Pairs with atomic_dec_and_test() in put_task_struct(). If this task | |
254 | * was already freed we can not miss the preceding update of this | |
255 | * pointer. | |
256 | */ | |
257 | smp_rmb(); | |
258 | if (unlikely(task != READ_ONCE(*ptask))) | |
259 | goto retry; | |
260 | ||
261 | /* | |
262 | * We've re-checked that "task == *ptask", now we have two different | |
263 | * cases: | |
264 | * | |
265 | * 1. This is actually the same task/task_struct. In this case | |
266 | * sighand != NULL tells us it is still alive. | |
267 | * | |
268 | * 2. This is another task which got the same memory for task_struct. | |
269 | * We can't know this of course, and we can not trust | |
270 | * sighand != NULL. | |
271 | * | |
272 | * In this case we actually return a random value, but this is | |
273 | * correct. | |
274 | * | |
275 | * If we return NULL - we can pretend that we actually noticed that | |
276 | * *ptask was updated when the previous task has exited. Or pretend | |
277 | * that probe_slab_address(&sighand) reads NULL. | |
278 | * | |
279 | * If we return the new task (because sighand is not NULL for any | |
280 | * reason) - this is fine too. This (new) task can't go away before | |
281 | * another gp pass. | |
282 | * | |
283 | * And note: We could even eliminate the false positive if re-read | |
284 | * task->sighand once again to avoid the falsely NULL. But this case | |
285 | * is very unlikely so we don't care. | |
286 | */ | |
287 | if (!sighand) | |
288 | return NULL; | |
289 | ||
290 | return task; | |
291 | } | |
292 | ||
8f95c90c DB |
293 | void rcuwait_wake_up(struct rcuwait *w) |
294 | { | |
295 | struct task_struct *task; | |
296 | ||
297 | rcu_read_lock(); | |
298 | ||
299 | /* | |
300 | * Order condition vs @task, such that everything prior to the load | |
301 | * of @task is visible. This is the condition as to why the user called | |
302 | * rcuwait_trywake() in the first place. Pairs with set_current_state() | |
303 | * barrier (A) in rcuwait_wait_event(). | |
304 | * | |
305 | * WAIT WAKE | |
306 | * [S] tsk = current [S] cond = true | |
307 | * MB (A) MB (B) | |
308 | * [L] cond [L] tsk | |
309 | */ | |
310 | smp_rmb(); /* (B) */ | |
311 | ||
312 | /* | |
313 | * Avoid using task_rcu_dereference() magic as long as we are careful, | |
314 | * see comment in rcuwait_wait_event() regarding ->exit_state. | |
315 | */ | |
316 | task = rcu_dereference(w->task); | |
317 | if (task) | |
318 | wake_up_process(task); | |
319 | rcu_read_unlock(); | |
320 | } | |
321 | ||
150593bf ON |
322 | struct task_struct *try_get_task_struct(struct task_struct **ptask) |
323 | { | |
324 | struct task_struct *task; | |
325 | ||
326 | rcu_read_lock(); | |
327 | task = task_rcu_dereference(ptask); | |
328 | if (task) | |
329 | get_task_struct(task); | |
330 | rcu_read_unlock(); | |
331 | ||
332 | return task; | |
333 | } | |
334 | ||
1da177e4 LT |
335 | /* |
336 | * Determine if a process group is "orphaned", according to the POSIX | |
337 | * definition in 2.2.2.52. Orphaned process groups are not to be affected | |
338 | * by terminal-generated stop signals. Newly orphaned process groups are | |
339 | * to receive a SIGHUP and a SIGCONT. | |
340 | * | |
341 | * "I ask you, have you ever known what it is to be an orphan?" | |
342 | */ | |
a0be55de IA |
343 | static int will_become_orphaned_pgrp(struct pid *pgrp, |
344 | struct task_struct *ignored_task) | |
1da177e4 LT |
345 | { |
346 | struct task_struct *p; | |
1da177e4 | 347 | |
0475ac08 | 348 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { |
05e83df6 ON |
349 | if ((p == ignored_task) || |
350 | (p->exit_state && thread_group_empty(p)) || | |
351 | is_global_init(p->real_parent)) | |
1da177e4 | 352 | continue; |
05e83df6 | 353 | |
0475ac08 | 354 | if (task_pgrp(p->real_parent) != pgrp && |
05e83df6 ON |
355 | task_session(p->real_parent) == task_session(p)) |
356 | return 0; | |
0475ac08 | 357 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); |
05e83df6 ON |
358 | |
359 | return 1; | |
1da177e4 LT |
360 | } |
361 | ||
3e7cd6c4 | 362 | int is_current_pgrp_orphaned(void) |
1da177e4 LT |
363 | { |
364 | int retval; | |
365 | ||
366 | read_lock(&tasklist_lock); | |
3e7cd6c4 | 367 | retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); |
1da177e4 LT |
368 | read_unlock(&tasklist_lock); |
369 | ||
370 | return retval; | |
371 | } | |
372 | ||
961c4675 | 373 | static bool has_stopped_jobs(struct pid *pgrp) |
1da177e4 | 374 | { |
1da177e4 LT |
375 | struct task_struct *p; |
376 | ||
0475ac08 | 377 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { |
961c4675 ON |
378 | if (p->signal->flags & SIGNAL_STOP_STOPPED) |
379 | return true; | |
0475ac08 | 380 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); |
961c4675 ON |
381 | |
382 | return false; | |
1da177e4 LT |
383 | } |
384 | ||
f49ee505 ON |
385 | /* |
386 | * Check to see if any process groups have become orphaned as | |
387 | * a result of our exiting, and if they have any stopped jobs, | |
388 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
389 | */ | |
390 | static void | |
391 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) | |
392 | { | |
393 | struct pid *pgrp = task_pgrp(tsk); | |
394 | struct task_struct *ignored_task = tsk; | |
395 | ||
396 | if (!parent) | |
a0be55de IA |
397 | /* exit: our father is in a different pgrp than |
398 | * we are and we were the only connection outside. | |
399 | */ | |
f49ee505 ON |
400 | parent = tsk->real_parent; |
401 | else | |
402 | /* reparent: our child is in a different pgrp than | |
403 | * we are, and it was the only connection outside. | |
404 | */ | |
405 | ignored_task = NULL; | |
406 | ||
407 | if (task_pgrp(parent) != pgrp && | |
408 | task_session(parent) == task_session(tsk) && | |
409 | will_become_orphaned_pgrp(pgrp, ignored_task) && | |
410 | has_stopped_jobs(pgrp)) { | |
411 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); | |
412 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); | |
413 | } | |
414 | } | |
415 | ||
f98bafa0 | 416 | #ifdef CONFIG_MEMCG |
cf475ad2 | 417 | /* |
733eda7a | 418 | * A task is exiting. If it owned this mm, find a new owner for the mm. |
cf475ad2 | 419 | */ |
cf475ad2 BS |
420 | void mm_update_next_owner(struct mm_struct *mm) |
421 | { | |
422 | struct task_struct *c, *g, *p = current; | |
423 | ||
424 | retry: | |
733eda7a KH |
425 | /* |
426 | * If the exiting or execing task is not the owner, it's | |
427 | * someone else's problem. | |
428 | */ | |
429 | if (mm->owner != p) | |
cf475ad2 | 430 | return; |
733eda7a KH |
431 | /* |
432 | * The current owner is exiting/execing and there are no other | |
433 | * candidates. Do not leave the mm pointing to a possibly | |
434 | * freed task structure. | |
435 | */ | |
436 | if (atomic_read(&mm->mm_users) <= 1) { | |
437 | mm->owner = NULL; | |
438 | return; | |
439 | } | |
cf475ad2 BS |
440 | |
441 | read_lock(&tasklist_lock); | |
442 | /* | |
443 | * Search in the children | |
444 | */ | |
445 | list_for_each_entry(c, &p->children, sibling) { | |
446 | if (c->mm == mm) | |
447 | goto assign_new_owner; | |
448 | } | |
449 | ||
450 | /* | |
451 | * Search in the siblings | |
452 | */ | |
dea33cfd | 453 | list_for_each_entry(c, &p->real_parent->children, sibling) { |
cf475ad2 BS |
454 | if (c->mm == mm) |
455 | goto assign_new_owner; | |
456 | } | |
457 | ||
458 | /* | |
f87fb599 | 459 | * Search through everything else, we should not get here often. |
cf475ad2 | 460 | */ |
39af1765 ON |
461 | for_each_process(g) { |
462 | if (g->flags & PF_KTHREAD) | |
463 | continue; | |
464 | for_each_thread(g, c) { | |
465 | if (c->mm == mm) | |
466 | goto assign_new_owner; | |
467 | if (c->mm) | |
468 | break; | |
469 | } | |
f87fb599 | 470 | } |
cf475ad2 | 471 | read_unlock(&tasklist_lock); |
31a78f23 BS |
472 | /* |
473 | * We found no owner yet mm_users > 1: this implies that we are | |
474 | * most likely racing with swapoff (try_to_unuse()) or /proc or | |
e5991371 | 475 | * ptrace or page migration (get_task_mm()). Mark owner as NULL. |
31a78f23 | 476 | */ |
31a78f23 | 477 | mm->owner = NULL; |
cf475ad2 BS |
478 | return; |
479 | ||
480 | assign_new_owner: | |
481 | BUG_ON(c == p); | |
482 | get_task_struct(c); | |
483 | /* | |
484 | * The task_lock protects c->mm from changing. | |
485 | * We always want mm->owner->mm == mm | |
486 | */ | |
487 | task_lock(c); | |
e5991371 HD |
488 | /* |
489 | * Delay read_unlock() till we have the task_lock() | |
490 | * to ensure that c does not slip away underneath us | |
491 | */ | |
492 | read_unlock(&tasklist_lock); | |
cf475ad2 BS |
493 | if (c->mm != mm) { |
494 | task_unlock(c); | |
495 | put_task_struct(c); | |
496 | goto retry; | |
497 | } | |
cf475ad2 BS |
498 | mm->owner = c; |
499 | task_unlock(c); | |
500 | put_task_struct(c); | |
501 | } | |
f98bafa0 | 502 | #endif /* CONFIG_MEMCG */ |
cf475ad2 | 503 | |
1da177e4 LT |
504 | /* |
505 | * Turn us into a lazy TLB process if we | |
506 | * aren't already.. | |
507 | */ | |
0039962a | 508 | static void exit_mm(void) |
1da177e4 | 509 | { |
0039962a | 510 | struct mm_struct *mm = current->mm; |
b564daf8 | 511 | struct core_state *core_state; |
1da177e4 | 512 | |
0039962a | 513 | mm_release(current, mm); |
1da177e4 LT |
514 | if (!mm) |
515 | return; | |
4fe7efdb | 516 | sync_mm_rss(mm); |
1da177e4 LT |
517 | /* |
518 | * Serialize with any possible pending coredump. | |
999d9fc1 | 519 | * We must hold mmap_sem around checking core_state |
1da177e4 | 520 | * and clearing tsk->mm. The core-inducing thread |
999d9fc1 | 521 | * will increment ->nr_threads for each thread in the |
1da177e4 LT |
522 | * group with ->mm != NULL. |
523 | */ | |
524 | down_read(&mm->mmap_sem); | |
b564daf8 ON |
525 | core_state = mm->core_state; |
526 | if (core_state) { | |
527 | struct core_thread self; | |
a0be55de | 528 | |
1da177e4 | 529 | up_read(&mm->mmap_sem); |
1da177e4 | 530 | |
0039962a | 531 | self.task = current; |
b564daf8 ON |
532 | self.next = xchg(&core_state->dumper.next, &self); |
533 | /* | |
534 | * Implies mb(), the result of xchg() must be visible | |
535 | * to core_state->dumper. | |
536 | */ | |
537 | if (atomic_dec_and_test(&core_state->nr_threads)) | |
538 | complete(&core_state->startup); | |
1da177e4 | 539 | |
a94e2d40 | 540 | for (;;) { |
642fa448 | 541 | set_current_state(TASK_UNINTERRUPTIBLE); |
a94e2d40 ON |
542 | if (!self.task) /* see coredump_finish() */ |
543 | break; | |
80d26af8 | 544 | freezable_schedule(); |
a94e2d40 | 545 | } |
642fa448 | 546 | __set_current_state(TASK_RUNNING); |
1da177e4 LT |
547 | down_read(&mm->mmap_sem); |
548 | } | |
f1f10076 | 549 | mmgrab(mm); |
0039962a | 550 | BUG_ON(mm != current->active_mm); |
1da177e4 | 551 | /* more a memory barrier than a real lock */ |
0039962a DB |
552 | task_lock(current); |
553 | current->mm = NULL; | |
1da177e4 LT |
554 | up_read(&mm->mmap_sem); |
555 | enter_lazy_tlb(mm, current); | |
0039962a | 556 | task_unlock(current); |
cf475ad2 | 557 | mm_update_next_owner(mm); |
1da177e4 | 558 | mmput(mm); |
c32b3cbe | 559 | if (test_thread_flag(TIF_MEMDIE)) |
38531201 | 560 | exit_oom_victim(); |
1da177e4 LT |
561 | } |
562 | ||
c9dc05bf ON |
563 | static struct task_struct *find_alive_thread(struct task_struct *p) |
564 | { | |
565 | struct task_struct *t; | |
566 | ||
567 | for_each_thread(p, t) { | |
568 | if (!(t->flags & PF_EXITING)) | |
569 | return t; | |
570 | } | |
571 | return NULL; | |
572 | } | |
573 | ||
1109909c ON |
574 | static struct task_struct *find_child_reaper(struct task_struct *father) |
575 | __releases(&tasklist_lock) | |
576 | __acquires(&tasklist_lock) | |
577 | { | |
578 | struct pid_namespace *pid_ns = task_active_pid_ns(father); | |
579 | struct task_struct *reaper = pid_ns->child_reaper; | |
580 | ||
581 | if (likely(reaper != father)) | |
582 | return reaper; | |
583 | ||
c9dc05bf ON |
584 | reaper = find_alive_thread(father); |
585 | if (reaper) { | |
1109909c ON |
586 | pid_ns->child_reaper = reaper; |
587 | return reaper; | |
588 | } | |
589 | ||
590 | write_unlock_irq(&tasklist_lock); | |
591 | if (unlikely(pid_ns == &init_pid_ns)) { | |
592 | panic("Attempted to kill init! exitcode=0x%08x\n", | |
593 | father->signal->group_exit_code ?: father->exit_code); | |
594 | } | |
595 | zap_pid_ns_processes(pid_ns); | |
596 | write_lock_irq(&tasklist_lock); | |
597 | ||
598 | return father; | |
599 | } | |
600 | ||
1da177e4 | 601 | /* |
ebec18a6 LP |
602 | * When we die, we re-parent all our children, and try to: |
603 | * 1. give them to another thread in our thread group, if such a member exists | |
604 | * 2. give it to the first ancestor process which prctl'd itself as a | |
605 | * child_subreaper for its children (like a service manager) | |
606 | * 3. give it to the init process (PID 1) in our pid namespace | |
1da177e4 | 607 | */ |
1109909c ON |
608 | static struct task_struct *find_new_reaper(struct task_struct *father, |
609 | struct task_struct *child_reaper) | |
1da177e4 | 610 | { |
c9dc05bf | 611 | struct task_struct *thread, *reaper; |
1da177e4 | 612 | |
c9dc05bf ON |
613 | thread = find_alive_thread(father); |
614 | if (thread) | |
950bbabb | 615 | return thread; |
1da177e4 | 616 | |
7d24e2df | 617 | if (father->signal->has_child_subreaper) { |
c6c70f44 | 618 | unsigned int ns_level = task_pid(father)->level; |
ebec18a6 | 619 | /* |
175aed3f | 620 | * Find the first ->is_child_subreaper ancestor in our pid_ns. |
c6c70f44 ON |
621 | * We can't check reaper != child_reaper to ensure we do not |
622 | * cross the namespaces, the exiting parent could be injected | |
623 | * by setns() + fork(). | |
624 | * We check pid->level, this is slightly more efficient than | |
625 | * task_active_pid_ns(reaper) != task_active_pid_ns(father). | |
ebec18a6 | 626 | */ |
c6c70f44 ON |
627 | for (reaper = father->real_parent; |
628 | task_pid(reaper)->level == ns_level; | |
ebec18a6 | 629 | reaper = reaper->real_parent) { |
175aed3f | 630 | if (reaper == &init_task) |
ebec18a6 LP |
631 | break; |
632 | if (!reaper->signal->is_child_subreaper) | |
633 | continue; | |
c9dc05bf ON |
634 | thread = find_alive_thread(reaper); |
635 | if (thread) | |
636 | return thread; | |
ebec18a6 | 637 | } |
1da177e4 | 638 | } |
762a24be | 639 | |
1109909c | 640 | return child_reaper; |
950bbabb ON |
641 | } |
642 | ||
5dfc80be ON |
643 | /* |
644 | * Any that need to be release_task'd are put on the @dead list. | |
645 | */ | |
9cd80bbb | 646 | static void reparent_leader(struct task_struct *father, struct task_struct *p, |
5dfc80be ON |
647 | struct list_head *dead) |
648 | { | |
2831096e | 649 | if (unlikely(p->exit_state == EXIT_DEAD)) |
5dfc80be ON |
650 | return; |
651 | ||
abd50b39 | 652 | /* We don't want people slaying init. */ |
5dfc80be ON |
653 | p->exit_signal = SIGCHLD; |
654 | ||
655 | /* If it has exited notify the new parent about this child's death. */ | |
d21142ec | 656 | if (!p->ptrace && |
5dfc80be | 657 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { |
86773473 | 658 | if (do_notify_parent(p, p->exit_signal)) { |
5dfc80be | 659 | p->exit_state = EXIT_DEAD; |
dc2fd4b0 | 660 | list_add(&p->ptrace_entry, dead); |
5dfc80be ON |
661 | } |
662 | } | |
663 | ||
664 | kill_orphaned_pgrp(p, father); | |
665 | } | |
666 | ||
482a3767 ON |
667 | /* |
668 | * This does two things: | |
669 | * | |
670 | * A. Make init inherit all the child processes | |
671 | * B. Check to see if any process groups have become orphaned | |
672 | * as a result of our exiting, and if they have any stopped | |
673 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
674 | */ | |
675 | static void forget_original_parent(struct task_struct *father, | |
676 | struct list_head *dead) | |
1da177e4 | 677 | { |
482a3767 | 678 | struct task_struct *p, *t, *reaper; |
762a24be | 679 | |
7c8bd232 | 680 | if (unlikely(!list_empty(&father->ptraced))) |
482a3767 | 681 | exit_ptrace(father, dead); |
f470021a | 682 | |
7c8bd232 | 683 | /* Can drop and reacquire tasklist_lock */ |
1109909c | 684 | reaper = find_child_reaper(father); |
ad9e206a | 685 | if (list_empty(&father->children)) |
482a3767 | 686 | return; |
1109909c ON |
687 | |
688 | reaper = find_new_reaper(father, reaper); | |
2831096e | 689 | list_for_each_entry(p, &father->children, sibling) { |
57a05918 | 690 | for_each_thread(p, t) { |
9cd80bbb | 691 | t->real_parent = reaper; |
57a05918 ON |
692 | BUG_ON((!t->ptrace) != (t->parent == father)); |
693 | if (likely(!t->ptrace)) | |
9cd80bbb | 694 | t->parent = t->real_parent; |
9cd80bbb ON |
695 | if (t->pdeath_signal) |
696 | group_send_sig_info(t->pdeath_signal, | |
697 | SEND_SIG_NOINFO, t); | |
57a05918 | 698 | } |
2831096e ON |
699 | /* |
700 | * If this is a threaded reparent there is no need to | |
701 | * notify anyone anything has happened. | |
702 | */ | |
703 | if (!same_thread_group(reaper, father)) | |
482a3767 | 704 | reparent_leader(father, p, dead); |
1da177e4 | 705 | } |
2831096e | 706 | list_splice_tail_init(&father->children, &reaper->children); |
1da177e4 LT |
707 | } |
708 | ||
709 | /* | |
710 | * Send signals to all our closest relatives so that they know | |
711 | * to properly mourn us.. | |
712 | */ | |
821c7de7 | 713 | static void exit_notify(struct task_struct *tsk, int group_dead) |
1da177e4 | 714 | { |
53c8f9f1 | 715 | bool autoreap; |
482a3767 ON |
716 | struct task_struct *p, *n; |
717 | LIST_HEAD(dead); | |
1da177e4 | 718 | |
762a24be | 719 | write_lock_irq(&tasklist_lock); |
482a3767 ON |
720 | forget_original_parent(tsk, &dead); |
721 | ||
821c7de7 ON |
722 | if (group_dead) |
723 | kill_orphaned_pgrp(tsk->group_leader, NULL); | |
1da177e4 | 724 | |
45cdf5cc ON |
725 | if (unlikely(tsk->ptrace)) { |
726 | int sig = thread_group_leader(tsk) && | |
727 | thread_group_empty(tsk) && | |
728 | !ptrace_reparented(tsk) ? | |
729 | tsk->exit_signal : SIGCHLD; | |
730 | autoreap = do_notify_parent(tsk, sig); | |
731 | } else if (thread_group_leader(tsk)) { | |
732 | autoreap = thread_group_empty(tsk) && | |
733 | do_notify_parent(tsk, tsk->exit_signal); | |
734 | } else { | |
735 | autoreap = true; | |
736 | } | |
1da177e4 | 737 | |
53c8f9f1 | 738 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; |
6c66e7db ON |
739 | if (tsk->exit_state == EXIT_DEAD) |
740 | list_add(&tsk->ptrace_entry, &dead); | |
1da177e4 | 741 | |
9c339168 ON |
742 | /* mt-exec, de_thread() is waiting for group leader */ |
743 | if (unlikely(tsk->signal->notify_count < 0)) | |
6db840fa | 744 | wake_up_process(tsk->signal->group_exit_task); |
1da177e4 LT |
745 | write_unlock_irq(&tasklist_lock); |
746 | ||
482a3767 ON |
747 | list_for_each_entry_safe(p, n, &dead, ptrace_entry) { |
748 | list_del_init(&p->ptrace_entry); | |
749 | release_task(p); | |
750 | } | |
1da177e4 LT |
751 | } |
752 | ||
e18eecb8 JD |
753 | #ifdef CONFIG_DEBUG_STACK_USAGE |
754 | static void check_stack_usage(void) | |
755 | { | |
756 | static DEFINE_SPINLOCK(low_water_lock); | |
757 | static int lowest_to_date = THREAD_SIZE; | |
e18eecb8 JD |
758 | unsigned long free; |
759 | ||
7c9f8861 | 760 | free = stack_not_used(current); |
e18eecb8 JD |
761 | |
762 | if (free >= lowest_to_date) | |
763 | return; | |
764 | ||
765 | spin_lock(&low_water_lock); | |
766 | if (free < lowest_to_date) { | |
627393d4 | 767 | pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", |
a0be55de | 768 | current->comm, task_pid_nr(current), free); |
e18eecb8 JD |
769 | lowest_to_date = free; |
770 | } | |
771 | spin_unlock(&low_water_lock); | |
772 | } | |
773 | #else | |
774 | static inline void check_stack_usage(void) {} | |
775 | #endif | |
776 | ||
9af6528e | 777 | void __noreturn do_exit(long code) |
1da177e4 LT |
778 | { |
779 | struct task_struct *tsk = current; | |
780 | int group_dead; | |
3f95aa81 | 781 | TASKS_RCU(int tasks_rcu_i); |
1da177e4 LT |
782 | |
783 | profile_task_exit(tsk); | |
5c9a8750 | 784 | kcov_task_exit(tsk); |
1da177e4 | 785 | |
73c10101 | 786 | WARN_ON(blk_needs_flush_plug(tsk)); |
22e2c507 | 787 | |
1da177e4 LT |
788 | if (unlikely(in_interrupt())) |
789 | panic("Aiee, killing interrupt handler!"); | |
790 | if (unlikely(!tsk->pid)) | |
791 | panic("Attempted to kill the idle task!"); | |
1da177e4 | 792 | |
33dd94ae NE |
793 | /* |
794 | * If do_exit is called because this processes oopsed, it's possible | |
795 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before | |
796 | * continuing. Amongst other possible reasons, this is to prevent | |
797 | * mm_release()->clear_child_tid() from writing to a user-controlled | |
798 | * kernel address. | |
799 | */ | |
800 | set_fs(USER_DS); | |
801 | ||
a288eecc | 802 | ptrace_event(PTRACE_EVENT_EXIT, code); |
1da177e4 | 803 | |
e0e81739 DH |
804 | validate_creds_for_do_exit(tsk); |
805 | ||
df164db5 AN |
806 | /* |
807 | * We're taking recursive faults here in do_exit. Safest is to just | |
808 | * leave this task alone and wait for reboot. | |
809 | */ | |
810 | if (unlikely(tsk->flags & PF_EXITING)) { | |
a0be55de | 811 | pr_alert("Fixing recursive fault but reboot is needed!\n"); |
778e9a9c AK |
812 | /* |
813 | * We can do this unlocked here. The futex code uses | |
814 | * this flag just to verify whether the pi state | |
815 | * cleanup has been done or not. In the worst case it | |
816 | * loops once more. We pretend that the cleanup was | |
817 | * done as there is no way to return. Either the | |
818 | * OWNER_DIED bit is set by now or we push the blocked | |
819 | * task into the wait for ever nirwana as well. | |
820 | */ | |
821 | tsk->flags |= PF_EXITPIDONE; | |
df164db5 AN |
822 | set_current_state(TASK_UNINTERRUPTIBLE); |
823 | schedule(); | |
824 | } | |
825 | ||
d12619b5 | 826 | exit_signals(tsk); /* sets PF_EXITING */ |
778e9a9c | 827 | /* |
be3e7844 PZ |
828 | * Ensure that all new tsk->pi_lock acquisitions must observe |
829 | * PF_EXITING. Serializes against futex.c:attach_to_pi_owner(). | |
778e9a9c | 830 | */ |
d2ee7198 | 831 | smp_mb(); |
be3e7844 PZ |
832 | /* |
833 | * Ensure that we must observe the pi_state in exit_mm() -> | |
834 | * mm_release() -> exit_pi_state_list(). | |
835 | */ | |
1d615482 | 836 | raw_spin_unlock_wait(&tsk->pi_lock); |
1da177e4 | 837 | |
1dc0fffc | 838 | if (unlikely(in_atomic())) { |
a0be55de IA |
839 | pr_info("note: %s[%d] exited with preempt_count %d\n", |
840 | current->comm, task_pid_nr(current), | |
841 | preempt_count()); | |
1dc0fffc PZ |
842 | preempt_count_set(PREEMPT_ENABLED); |
843 | } | |
1da177e4 | 844 | |
48d212a2 LT |
845 | /* sync mm's RSS info before statistics gathering */ |
846 | if (tsk->mm) | |
847 | sync_mm_rss(tsk->mm); | |
51229b49 | 848 | acct_update_integrals(tsk); |
1da177e4 | 849 | group_dead = atomic_dec_and_test(&tsk->signal->live); |
c3068951 | 850 | if (group_dead) { |
baa73d9e | 851 | #ifdef CONFIG_POSIX_TIMERS |
778e9a9c | 852 | hrtimer_cancel(&tsk->signal->real_timer); |
25f407f0 | 853 | exit_itimers(tsk->signal); |
baa73d9e | 854 | #endif |
1f10206c JP |
855 | if (tsk->mm) |
856 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); | |
c3068951 | 857 | } |
f6ec29a4 | 858 | acct_collect(code, group_dead); |
522ed776 MT |
859 | if (group_dead) |
860 | tty_audit_exit(); | |
a4ff8dba | 861 | audit_free(tsk); |
115085ea | 862 | |
48d212a2 | 863 | tsk->exit_code = code; |
115085ea | 864 | taskstats_exit(tsk, group_dead); |
c757249a | 865 | |
0039962a | 866 | exit_mm(); |
1da177e4 | 867 | |
0e464814 | 868 | if (group_dead) |
f6ec29a4 | 869 | acct_process(); |
0a16b607 MD |
870 | trace_sched_process_exit(tsk); |
871 | ||
1da177e4 | 872 | exit_sem(tsk); |
b34a6b1d | 873 | exit_shm(tsk); |
1ec7f1dd AV |
874 | exit_files(tsk); |
875 | exit_fs(tsk); | |
c39df5fa ON |
876 | if (group_dead) |
877 | disassociate_ctty(1); | |
8aac6270 | 878 | exit_task_namespaces(tsk); |
ed3e694d | 879 | exit_task_work(tsk); |
e6464694 | 880 | exit_thread(tsk); |
0b3fcf17 SE |
881 | |
882 | /* | |
883 | * Flush inherited counters to the parent - before the parent | |
884 | * gets woken up by child-exit notifications. | |
885 | * | |
886 | * because of cgroup mode, must be called before cgroup_exit() | |
887 | */ | |
888 | perf_event_exit_task(tsk); | |
889 | ||
8e5bfa8c | 890 | sched_autogroup_exit_task(tsk); |
1ec41830 | 891 | cgroup_exit(tsk); |
1da177e4 | 892 | |
24f1e32c FW |
893 | /* |
894 | * FIXME: do that only when needed, using sched_exit tracepoint | |
895 | */ | |
7c8df286 | 896 | flush_ptrace_hw_breakpoint(tsk); |
33b2fb30 | 897 | |
49f5903b | 898 | TASKS_RCU(preempt_disable()); |
3f95aa81 | 899 | TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu)); |
49f5903b | 900 | TASKS_RCU(preempt_enable()); |
821c7de7 | 901 | exit_notify(tsk, group_dead); |
ef982393 | 902 | proc_exit_connector(tsk); |
c11600e4 | 903 | mpol_put_task_policy(tsk); |
42b2dd0a | 904 | #ifdef CONFIG_FUTEX |
c87e2837 IM |
905 | if (unlikely(current->pi_state_cache)) |
906 | kfree(current->pi_state_cache); | |
42b2dd0a | 907 | #endif |
de5097c2 | 908 | /* |
9a11b49a | 909 | * Make sure we are holding no locks: |
de5097c2 | 910 | */ |
1b1d2fb4 | 911 | debug_check_no_locks_held(); |
778e9a9c AK |
912 | /* |
913 | * We can do this unlocked here. The futex code uses this flag | |
914 | * just to verify whether the pi state cleanup has been done | |
915 | * or not. In the worst case it loops once more. | |
916 | */ | |
917 | tsk->flags |= PF_EXITPIDONE; | |
1da177e4 | 918 | |
afc847b7 | 919 | if (tsk->io_context) |
b69f2292 | 920 | exit_io_context(tsk); |
afc847b7 | 921 | |
b92ce558 | 922 | if (tsk->splice_pipe) |
4b8a8f1e | 923 | free_pipe_info(tsk->splice_pipe); |
b92ce558 | 924 | |
5640f768 ED |
925 | if (tsk->task_frag.page) |
926 | put_page(tsk->task_frag.page); | |
927 | ||
e0e81739 DH |
928 | validate_creds_for_do_exit(tsk); |
929 | ||
4bcb8232 | 930 | check_stack_usage(); |
7407251a | 931 | preempt_disable(); |
54848d73 WF |
932 | if (tsk->nr_dirtied) |
933 | __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); | |
f41d911f | 934 | exit_rcu(); |
3f95aa81 | 935 | TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i)); |
b5740f4b | 936 | |
9af6528e | 937 | do_task_dead(); |
1da177e4 | 938 | } |
012914da RA |
939 | EXPORT_SYMBOL_GPL(do_exit); |
940 | ||
9402c95f | 941 | void complete_and_exit(struct completion *comp, long code) |
1da177e4 LT |
942 | { |
943 | if (comp) | |
944 | complete(comp); | |
55a101f8 | 945 | |
1da177e4 LT |
946 | do_exit(code); |
947 | } | |
1da177e4 LT |
948 | EXPORT_SYMBOL(complete_and_exit); |
949 | ||
754fe8d2 | 950 | SYSCALL_DEFINE1(exit, int, error_code) |
1da177e4 LT |
951 | { |
952 | do_exit((error_code&0xff)<<8); | |
953 | } | |
954 | ||
1da177e4 LT |
955 | /* |
956 | * Take down every thread in the group. This is called by fatal signals | |
957 | * as well as by sys_exit_group (below). | |
958 | */ | |
9402c95f | 959 | void |
1da177e4 LT |
960 | do_group_exit(int exit_code) |
961 | { | |
bfc4b089 ON |
962 | struct signal_struct *sig = current->signal; |
963 | ||
1da177e4 LT |
964 | BUG_ON(exit_code & 0x80); /* core dumps don't get here */ |
965 | ||
bfc4b089 ON |
966 | if (signal_group_exit(sig)) |
967 | exit_code = sig->group_exit_code; | |
1da177e4 | 968 | else if (!thread_group_empty(current)) { |
1da177e4 | 969 | struct sighand_struct *const sighand = current->sighand; |
a0be55de | 970 | |
1da177e4 | 971 | spin_lock_irq(&sighand->siglock); |
ed5d2cac | 972 | if (signal_group_exit(sig)) |
1da177e4 LT |
973 | /* Another thread got here before we took the lock. */ |
974 | exit_code = sig->group_exit_code; | |
975 | else { | |
1da177e4 | 976 | sig->group_exit_code = exit_code; |
ed5d2cac | 977 | sig->flags = SIGNAL_GROUP_EXIT; |
1da177e4 LT |
978 | zap_other_threads(current); |
979 | } | |
980 | spin_unlock_irq(&sighand->siglock); | |
1da177e4 LT |
981 | } |
982 | ||
983 | do_exit(exit_code); | |
984 | /* NOTREACHED */ | |
985 | } | |
986 | ||
987 | /* | |
988 | * this kills every thread in the thread group. Note that any externally | |
989 | * wait4()-ing process will get the correct exit code - even if this | |
990 | * thread is not the thread group leader. | |
991 | */ | |
754fe8d2 | 992 | SYSCALL_DEFINE1(exit_group, int, error_code) |
1da177e4 LT |
993 | { |
994 | do_group_exit((error_code & 0xff) << 8); | |
2ed7c03e HC |
995 | /* NOTREACHED */ |
996 | return 0; | |
1da177e4 LT |
997 | } |
998 | ||
9e8ae01d ON |
999 | struct wait_opts { |
1000 | enum pid_type wo_type; | |
9e8ae01d | 1001 | int wo_flags; |
e1eb1ebc | 1002 | struct pid *wo_pid; |
9e8ae01d ON |
1003 | |
1004 | struct siginfo __user *wo_info; | |
1005 | int __user *wo_stat; | |
ce72a16f | 1006 | struct rusage *wo_rusage; |
9e8ae01d | 1007 | |
0b7570e7 | 1008 | wait_queue_t child_wait; |
9e8ae01d ON |
1009 | int notask_error; |
1010 | }; | |
1011 | ||
989264f4 ON |
1012 | static inline |
1013 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type) | |
161550d7 | 1014 | { |
989264f4 ON |
1015 | if (type != PIDTYPE_PID) |
1016 | task = task->group_leader; | |
1017 | return task->pids[type].pid; | |
161550d7 EB |
1018 | } |
1019 | ||
989264f4 | 1020 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p) |
1da177e4 | 1021 | { |
5c01ba49 ON |
1022 | return wo->wo_type == PIDTYPE_MAX || |
1023 | task_pid_type(p, wo->wo_type) == wo->wo_pid; | |
1024 | } | |
1da177e4 | 1025 | |
bf959931 ON |
1026 | static int |
1027 | eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) | |
5c01ba49 ON |
1028 | { |
1029 | if (!eligible_pid(wo, p)) | |
1030 | return 0; | |
bf959931 ON |
1031 | |
1032 | /* | |
1033 | * Wait for all children (clone and not) if __WALL is set or | |
1034 | * if it is traced by us. | |
1035 | */ | |
1036 | if (ptrace || (wo->wo_flags & __WALL)) | |
1037 | return 1; | |
1038 | ||
1039 | /* | |
1040 | * Otherwise, wait for clone children *only* if __WCLONE is set; | |
1041 | * otherwise, wait for non-clone children *only*. | |
1042 | * | |
1043 | * Note: a "clone" child here is one that reports to its parent | |
1044 | * using a signal other than SIGCHLD, or a non-leader thread which | |
1045 | * we can only see if it is traced by us. | |
1046 | */ | |
1047 | if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) | |
1da177e4 | 1048 | return 0; |
1da177e4 | 1049 | |
14dd0b81 | 1050 | return 1; |
1da177e4 LT |
1051 | } |
1052 | ||
9e8ae01d ON |
1053 | static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p, |
1054 | pid_t pid, uid_t uid, int why, int status) | |
1da177e4 | 1055 | { |
9e8ae01d | 1056 | struct siginfo __user *infop; |
ce72a16f AV |
1057 | int retval = 0; |
1058 | ||
1059 | if (wo->wo_rusage) | |
1060 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
36c8b586 | 1061 | |
1da177e4 | 1062 | put_task_struct(p); |
9e8ae01d | 1063 | infop = wo->wo_info; |
b6fe2d11 VM |
1064 | if (infop) { |
1065 | if (!retval) | |
1066 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1067 | if (!retval) | |
1068 | retval = put_user(0, &infop->si_errno); | |
1069 | if (!retval) | |
1070 | retval = put_user((short)why, &infop->si_code); | |
1071 | if (!retval) | |
1072 | retval = put_user(pid, &infop->si_pid); | |
1073 | if (!retval) | |
1074 | retval = put_user(uid, &infop->si_uid); | |
1075 | if (!retval) | |
1076 | retval = put_user(status, &infop->si_status); | |
1077 | } | |
1da177e4 LT |
1078 | if (!retval) |
1079 | retval = pid; | |
1080 | return retval; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold | |
1085 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1086 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1087 | * released the lock and the system call should return. | |
1088 | */ | |
9e8ae01d | 1089 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) |
1da177e4 | 1090 | { |
f6507f83 | 1091 | int state, retval, status; |
6c5f3e7b | 1092 | pid_t pid = task_pid_vnr(p); |
43e13cc1 | 1093 | uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); |
9e8ae01d | 1094 | struct siginfo __user *infop; |
1da177e4 | 1095 | |
9e8ae01d | 1096 | if (!likely(wo->wo_flags & WEXITED)) |
98abed02 RM |
1097 | return 0; |
1098 | ||
9e8ae01d | 1099 | if (unlikely(wo->wo_flags & WNOWAIT)) { |
1da177e4 | 1100 | int exit_code = p->exit_code; |
f3abd4f9 | 1101 | int why; |
1da177e4 | 1102 | |
1da177e4 LT |
1103 | get_task_struct(p); |
1104 | read_unlock(&tasklist_lock); | |
1029a2b5 PZ |
1105 | sched_annotate_sleep(); |
1106 | ||
1da177e4 LT |
1107 | if ((exit_code & 0x7f) == 0) { |
1108 | why = CLD_EXITED; | |
1109 | status = exit_code >> 8; | |
1110 | } else { | |
1111 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1112 | status = exit_code & 0x7f; | |
1113 | } | |
9e8ae01d | 1114 | return wait_noreap_copyout(wo, p, pid, uid, why, status); |
1da177e4 | 1115 | } |
1da177e4 | 1116 | /* |
abd50b39 | 1117 | * Move the task's state to DEAD/TRACE, only one thread can do this. |
1da177e4 | 1118 | */ |
f6507f83 ON |
1119 | state = (ptrace_reparented(p) && thread_group_leader(p)) ? |
1120 | EXIT_TRACE : EXIT_DEAD; | |
abd50b39 | 1121 | if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) |
1da177e4 | 1122 | return 0; |
986094df ON |
1123 | /* |
1124 | * We own this thread, nobody else can reap it. | |
1125 | */ | |
1126 | read_unlock(&tasklist_lock); | |
1127 | sched_annotate_sleep(); | |
f6507f83 | 1128 | |
befca967 | 1129 | /* |
f6507f83 | 1130 | * Check thread_group_leader() to exclude the traced sub-threads. |
befca967 | 1131 | */ |
f6507f83 | 1132 | if (state == EXIT_DEAD && thread_group_leader(p)) { |
f953ccd0 ON |
1133 | struct signal_struct *sig = p->signal; |
1134 | struct signal_struct *psig = current->signal; | |
1f10206c | 1135 | unsigned long maxrss; |
5613fda9 | 1136 | u64 tgutime, tgstime; |
3795e161 | 1137 | |
1da177e4 LT |
1138 | /* |
1139 | * The resource counters for the group leader are in its | |
1140 | * own task_struct. Those for dead threads in the group | |
1141 | * are in its signal_struct, as are those for the child | |
1142 | * processes it has previously reaped. All these | |
1143 | * accumulate in the parent's signal_struct c* fields. | |
1144 | * | |
1145 | * We don't bother to take a lock here to protect these | |
f953ccd0 ON |
1146 | * p->signal fields because the whole thread group is dead |
1147 | * and nobody can change them. | |
1148 | * | |
1149 | * psig->stats_lock also protects us from our sub-theads | |
1150 | * which can reap other children at the same time. Until | |
1151 | * we change k_getrusage()-like users to rely on this lock | |
1152 | * we have to take ->siglock as well. | |
0cf55e1e | 1153 | * |
a0be55de IA |
1154 | * We use thread_group_cputime_adjusted() to get times for |
1155 | * the thread group, which consolidates times for all threads | |
1156 | * in the group including the group leader. | |
1da177e4 | 1157 | */ |
e80d0a1a | 1158 | thread_group_cputime_adjusted(p, &tgutime, &tgstime); |
f953ccd0 | 1159 | spin_lock_irq(¤t->sighand->siglock); |
e78c3496 | 1160 | write_seqlock(&psig->stats_lock); |
64861634 MS |
1161 | psig->cutime += tgutime + sig->cutime; |
1162 | psig->cstime += tgstime + sig->cstime; | |
6fac4829 | 1163 | psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; |
3795e161 JJ |
1164 | psig->cmin_flt += |
1165 | p->min_flt + sig->min_flt + sig->cmin_flt; | |
1166 | psig->cmaj_flt += | |
1167 | p->maj_flt + sig->maj_flt + sig->cmaj_flt; | |
1168 | psig->cnvcsw += | |
1169 | p->nvcsw + sig->nvcsw + sig->cnvcsw; | |
1170 | psig->cnivcsw += | |
1171 | p->nivcsw + sig->nivcsw + sig->cnivcsw; | |
6eaeeaba ED |
1172 | psig->cinblock += |
1173 | task_io_get_inblock(p) + | |
1174 | sig->inblock + sig->cinblock; | |
1175 | psig->coublock += | |
1176 | task_io_get_oublock(p) + | |
1177 | sig->oublock + sig->coublock; | |
1f10206c JP |
1178 | maxrss = max(sig->maxrss, sig->cmaxrss); |
1179 | if (psig->cmaxrss < maxrss) | |
1180 | psig->cmaxrss = maxrss; | |
5995477a AR |
1181 | task_io_accounting_add(&psig->ioac, &p->ioac); |
1182 | task_io_accounting_add(&psig->ioac, &sig->ioac); | |
e78c3496 | 1183 | write_sequnlock(&psig->stats_lock); |
f953ccd0 | 1184 | spin_unlock_irq(¤t->sighand->siglock); |
1da177e4 LT |
1185 | } |
1186 | ||
ce72a16f AV |
1187 | if (wo->wo_rusage) |
1188 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1189 | retval = 0; | |
1da177e4 LT |
1190 | status = (p->signal->flags & SIGNAL_GROUP_EXIT) |
1191 | ? p->signal->group_exit_code : p->exit_code; | |
9e8ae01d ON |
1192 | if (!retval && wo->wo_stat) |
1193 | retval = put_user(status, wo->wo_stat); | |
1194 | ||
1195 | infop = wo->wo_info; | |
1da177e4 LT |
1196 | if (!retval && infop) |
1197 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1198 | if (!retval && infop) | |
1199 | retval = put_user(0, &infop->si_errno); | |
1200 | if (!retval && infop) { | |
1201 | int why; | |
1202 | ||
1203 | if ((status & 0x7f) == 0) { | |
1204 | why = CLD_EXITED; | |
1205 | status >>= 8; | |
1206 | } else { | |
1207 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1208 | status &= 0x7f; | |
1209 | } | |
1210 | retval = put_user((short)why, &infop->si_code); | |
1211 | if (!retval) | |
1212 | retval = put_user(status, &infop->si_status); | |
1213 | } | |
1214 | if (!retval && infop) | |
3a515e4a | 1215 | retval = put_user(pid, &infop->si_pid); |
1da177e4 | 1216 | if (!retval && infop) |
c69e8d9c | 1217 | retval = put_user(uid, &infop->si_uid); |
2f4e6e2a | 1218 | if (!retval) |
3a515e4a | 1219 | retval = pid; |
2f4e6e2a | 1220 | |
b4360690 | 1221 | if (state == EXIT_TRACE) { |
1da177e4 | 1222 | write_lock_irq(&tasklist_lock); |
2f4e6e2a ON |
1223 | /* We dropped tasklist, ptracer could die and untrace */ |
1224 | ptrace_unlink(p); | |
b4360690 ON |
1225 | |
1226 | /* If parent wants a zombie, don't release it now */ | |
1227 | state = EXIT_ZOMBIE; | |
1228 | if (do_notify_parent(p, p->exit_signal)) | |
1229 | state = EXIT_DEAD; | |
abd50b39 | 1230 | p->exit_state = state; |
1da177e4 LT |
1231 | write_unlock_irq(&tasklist_lock); |
1232 | } | |
abd50b39 | 1233 | if (state == EXIT_DEAD) |
1da177e4 | 1234 | release_task(p); |
2f4e6e2a | 1235 | |
1da177e4 LT |
1236 | return retval; |
1237 | } | |
1238 | ||
90bc8d8b ON |
1239 | static int *task_stopped_code(struct task_struct *p, bool ptrace) |
1240 | { | |
1241 | if (ptrace) { | |
570ac933 | 1242 | if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) |
90bc8d8b ON |
1243 | return &p->exit_code; |
1244 | } else { | |
1245 | if (p->signal->flags & SIGNAL_STOP_STOPPED) | |
1246 | return &p->signal->group_exit_code; | |
1247 | } | |
1248 | return NULL; | |
1249 | } | |
1250 | ||
19e27463 TH |
1251 | /** |
1252 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED | |
1253 | * @wo: wait options | |
1254 | * @ptrace: is the wait for ptrace | |
1255 | * @p: task to wait for | |
1256 | * | |
1257 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. | |
1258 | * | |
1259 | * CONTEXT: | |
1260 | * read_lock(&tasklist_lock), which is released if return value is | |
1261 | * non-zero. Also, grabs and releases @p->sighand->siglock. | |
1262 | * | |
1263 | * RETURNS: | |
1264 | * 0 if wait condition didn't exist and search for other wait conditions | |
1265 | * should continue. Non-zero return, -errno on failure and @p's pid on | |
1266 | * success, implies that tasklist_lock is released and wait condition | |
1267 | * search should terminate. | |
1da177e4 | 1268 | */ |
9e8ae01d ON |
1269 | static int wait_task_stopped(struct wait_opts *wo, |
1270 | int ptrace, struct task_struct *p) | |
1da177e4 | 1271 | { |
9e8ae01d | 1272 | struct siginfo __user *infop; |
90bc8d8b | 1273 | int retval, exit_code, *p_code, why; |
ee7c82da | 1274 | uid_t uid = 0; /* unneeded, required by compiler */ |
c8950783 | 1275 | pid_t pid; |
1da177e4 | 1276 | |
47918025 ON |
1277 | /* |
1278 | * Traditionally we see ptrace'd stopped tasks regardless of options. | |
1279 | */ | |
9e8ae01d | 1280 | if (!ptrace && !(wo->wo_flags & WUNTRACED)) |
98abed02 RM |
1281 | return 0; |
1282 | ||
19e27463 TH |
1283 | if (!task_stopped_code(p, ptrace)) |
1284 | return 0; | |
1285 | ||
ee7c82da ON |
1286 | exit_code = 0; |
1287 | spin_lock_irq(&p->sighand->siglock); | |
1288 | ||
90bc8d8b ON |
1289 | p_code = task_stopped_code(p, ptrace); |
1290 | if (unlikely(!p_code)) | |
ee7c82da ON |
1291 | goto unlock_sig; |
1292 | ||
90bc8d8b | 1293 | exit_code = *p_code; |
ee7c82da ON |
1294 | if (!exit_code) |
1295 | goto unlock_sig; | |
1296 | ||
9e8ae01d | 1297 | if (!unlikely(wo->wo_flags & WNOWAIT)) |
90bc8d8b | 1298 | *p_code = 0; |
ee7c82da | 1299 | |
8ca937a6 | 1300 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); |
ee7c82da ON |
1301 | unlock_sig: |
1302 | spin_unlock_irq(&p->sighand->siglock); | |
1303 | if (!exit_code) | |
1da177e4 LT |
1304 | return 0; |
1305 | ||
1306 | /* | |
1307 | * Now we are pretty sure this task is interesting. | |
1308 | * Make sure it doesn't get reaped out from under us while we | |
1309 | * give up the lock and then examine it below. We don't want to | |
1310 | * keep holding onto the tasklist_lock while we call getrusage and | |
1311 | * possibly take page faults for user memory. | |
1312 | */ | |
1313 | get_task_struct(p); | |
6c5f3e7b | 1314 | pid = task_pid_vnr(p); |
f470021a | 1315 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED; |
1da177e4 | 1316 | read_unlock(&tasklist_lock); |
1029a2b5 | 1317 | sched_annotate_sleep(); |
1da177e4 | 1318 | |
9e8ae01d ON |
1319 | if (unlikely(wo->wo_flags & WNOWAIT)) |
1320 | return wait_noreap_copyout(wo, p, pid, uid, why, exit_code); | |
1321 | ||
ce72a16f AV |
1322 | if (wo->wo_rusage) |
1323 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1324 | retval = 0; | |
9e8ae01d ON |
1325 | if (!retval && wo->wo_stat) |
1326 | retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat); | |
1da177e4 | 1327 | |
9e8ae01d | 1328 | infop = wo->wo_info; |
1da177e4 LT |
1329 | if (!retval && infop) |
1330 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1331 | if (!retval && infop) | |
1332 | retval = put_user(0, &infop->si_errno); | |
1333 | if (!retval && infop) | |
6efcae46 | 1334 | retval = put_user((short)why, &infop->si_code); |
1da177e4 LT |
1335 | if (!retval && infop) |
1336 | retval = put_user(exit_code, &infop->si_status); | |
1337 | if (!retval && infop) | |
c8950783 | 1338 | retval = put_user(pid, &infop->si_pid); |
1da177e4 | 1339 | if (!retval && infop) |
ee7c82da | 1340 | retval = put_user(uid, &infop->si_uid); |
1da177e4 | 1341 | if (!retval) |
c8950783 | 1342 | retval = pid; |
1da177e4 LT |
1343 | put_task_struct(p); |
1344 | ||
1345 | BUG_ON(!retval); | |
1346 | return retval; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Handle do_wait work for one task in a live, non-stopped state. | |
1351 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1352 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1353 | * released the lock and the system call should return. | |
1354 | */ | |
9e8ae01d | 1355 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) |
1da177e4 LT |
1356 | { |
1357 | int retval; | |
1358 | pid_t pid; | |
1359 | uid_t uid; | |
1360 | ||
9e8ae01d | 1361 | if (!unlikely(wo->wo_flags & WCONTINUED)) |
98abed02 RM |
1362 | return 0; |
1363 | ||
1da177e4 LT |
1364 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) |
1365 | return 0; | |
1366 | ||
1367 | spin_lock_irq(&p->sighand->siglock); | |
1368 | /* Re-check with the lock held. */ | |
1369 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { | |
1370 | spin_unlock_irq(&p->sighand->siglock); | |
1371 | return 0; | |
1372 | } | |
9e8ae01d | 1373 | if (!unlikely(wo->wo_flags & WNOWAIT)) |
1da177e4 | 1374 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED; |
8ca937a6 | 1375 | uid = from_kuid_munged(current_user_ns(), task_uid(p)); |
1da177e4 LT |
1376 | spin_unlock_irq(&p->sighand->siglock); |
1377 | ||
6c5f3e7b | 1378 | pid = task_pid_vnr(p); |
1da177e4 LT |
1379 | get_task_struct(p); |
1380 | read_unlock(&tasklist_lock); | |
1029a2b5 | 1381 | sched_annotate_sleep(); |
1da177e4 | 1382 | |
9e8ae01d | 1383 | if (!wo->wo_info) { |
ce72a16f AV |
1384 | if (wo->wo_rusage) |
1385 | getrusage(p, RUSAGE_BOTH, wo->wo_rusage); | |
1386 | retval = 0; | |
1da177e4 | 1387 | put_task_struct(p); |
9e8ae01d ON |
1388 | if (!retval && wo->wo_stat) |
1389 | retval = put_user(0xffff, wo->wo_stat); | |
1da177e4 | 1390 | if (!retval) |
3a515e4a | 1391 | retval = pid; |
1da177e4 | 1392 | } else { |
9e8ae01d ON |
1393 | retval = wait_noreap_copyout(wo, p, pid, uid, |
1394 | CLD_CONTINUED, SIGCONT); | |
1da177e4 LT |
1395 | BUG_ON(retval == 0); |
1396 | } | |
1397 | ||
1398 | return retval; | |
1399 | } | |
1400 | ||
98abed02 RM |
1401 | /* |
1402 | * Consider @p for a wait by @parent. | |
1403 | * | |
9e8ae01d | 1404 | * -ECHILD should be in ->notask_error before the first call. |
98abed02 RM |
1405 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. |
1406 | * Returns zero if the search for a child should continue; | |
9e8ae01d | 1407 | * then ->notask_error is 0 if @p is an eligible child, |
3a2f5a59 | 1408 | * or still -ECHILD. |
98abed02 | 1409 | */ |
b6e763f0 ON |
1410 | static int wait_consider_task(struct wait_opts *wo, int ptrace, |
1411 | struct task_struct *p) | |
98abed02 | 1412 | { |
3245d6ac ON |
1413 | /* |
1414 | * We can race with wait_task_zombie() from another thread. | |
1415 | * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition | |
1416 | * can't confuse the checks below. | |
1417 | */ | |
1418 | int exit_state = ACCESS_ONCE(p->exit_state); | |
b3ab0316 ON |
1419 | int ret; |
1420 | ||
3245d6ac | 1421 | if (unlikely(exit_state == EXIT_DEAD)) |
b3ab0316 ON |
1422 | return 0; |
1423 | ||
bf959931 | 1424 | ret = eligible_child(wo, ptrace, p); |
14dd0b81 | 1425 | if (!ret) |
98abed02 RM |
1426 | return ret; |
1427 | ||
3245d6ac | 1428 | if (unlikely(exit_state == EXIT_TRACE)) { |
50b8d257 | 1429 | /* |
abd50b39 ON |
1430 | * ptrace == 0 means we are the natural parent. In this case |
1431 | * we should clear notask_error, debugger will notify us. | |
50b8d257 | 1432 | */ |
abd50b39 | 1433 | if (likely(!ptrace)) |
50b8d257 | 1434 | wo->notask_error = 0; |
823b018e | 1435 | return 0; |
50b8d257 | 1436 | } |
823b018e | 1437 | |
377d75da ON |
1438 | if (likely(!ptrace) && unlikely(p->ptrace)) { |
1439 | /* | |
1440 | * If it is traced by its real parent's group, just pretend | |
1441 | * the caller is ptrace_do_wait() and reap this child if it | |
1442 | * is zombie. | |
1443 | * | |
1444 | * This also hides group stop state from real parent; otherwise | |
1445 | * a single stop can be reported twice as group and ptrace stop. | |
1446 | * If a ptracer wants to distinguish these two events for its | |
1447 | * own children it should create a separate process which takes | |
1448 | * the role of real parent. | |
1449 | */ | |
1450 | if (!ptrace_reparented(p)) | |
1451 | ptrace = 1; | |
1452 | } | |
1453 | ||
45cb24a1 | 1454 | /* slay zombie? */ |
3245d6ac | 1455 | if (exit_state == EXIT_ZOMBIE) { |
9b84cca2 | 1456 | /* we don't reap group leaders with subthreads */ |
7c733eb3 ON |
1457 | if (!delay_group_leader(p)) { |
1458 | /* | |
1459 | * A zombie ptracee is only visible to its ptracer. | |
1460 | * Notification and reaping will be cascaded to the | |
1461 | * real parent when the ptracer detaches. | |
1462 | */ | |
1463 | if (unlikely(ptrace) || likely(!p->ptrace)) | |
1464 | return wait_task_zombie(wo, p); | |
1465 | } | |
98abed02 | 1466 | |
f470021a | 1467 | /* |
9b84cca2 TH |
1468 | * Allow access to stopped/continued state via zombie by |
1469 | * falling through. Clearing of notask_error is complex. | |
1470 | * | |
1471 | * When !@ptrace: | |
1472 | * | |
1473 | * If WEXITED is set, notask_error should naturally be | |
1474 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set, | |
1475 | * so, if there are live subthreads, there are events to | |
1476 | * wait for. If all subthreads are dead, it's still safe | |
1477 | * to clear - this function will be called again in finite | |
1478 | * amount time once all the subthreads are released and | |
1479 | * will then return without clearing. | |
1480 | * | |
1481 | * When @ptrace: | |
1482 | * | |
1483 | * Stopped state is per-task and thus can't change once the | |
1484 | * target task dies. Only continued and exited can happen. | |
1485 | * Clear notask_error if WCONTINUED | WEXITED. | |
1486 | */ | |
1487 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) | |
1488 | wo->notask_error = 0; | |
1489 | } else { | |
1490 | /* | |
1491 | * @p is alive and it's gonna stop, continue or exit, so | |
1492 | * there always is something to wait for. | |
f470021a | 1493 | */ |
9e8ae01d | 1494 | wo->notask_error = 0; |
f470021a RM |
1495 | } |
1496 | ||
98abed02 | 1497 | /* |
45cb24a1 TH |
1498 | * Wait for stopped. Depending on @ptrace, different stopped state |
1499 | * is used and the two don't interact with each other. | |
98abed02 | 1500 | */ |
19e27463 TH |
1501 | ret = wait_task_stopped(wo, ptrace, p); |
1502 | if (ret) | |
1503 | return ret; | |
98abed02 RM |
1504 | |
1505 | /* | |
45cb24a1 TH |
1506 | * Wait for continued. There's only one continued state and the |
1507 | * ptracer can consume it which can confuse the real parent. Don't | |
1508 | * use WCONTINUED from ptracer. You don't need or want it. | |
98abed02 | 1509 | */ |
9e8ae01d | 1510 | return wait_task_continued(wo, p); |
98abed02 RM |
1511 | } |
1512 | ||
1513 | /* | |
1514 | * Do the work of do_wait() for one thread in the group, @tsk. | |
1515 | * | |
9e8ae01d | 1516 | * -ECHILD should be in ->notask_error before the first call. |
98abed02 RM |
1517 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. |
1518 | * Returns zero if the search for a child should continue; then | |
9e8ae01d | 1519 | * ->notask_error is 0 if there were any eligible children, |
3a2f5a59 | 1520 | * or still -ECHILD. |
98abed02 | 1521 | */ |
9e8ae01d | 1522 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) |
98abed02 RM |
1523 | { |
1524 | struct task_struct *p; | |
1525 | ||
1526 | list_for_each_entry(p, &tsk->children, sibling) { | |
9cd80bbb | 1527 | int ret = wait_consider_task(wo, 0, p); |
a0be55de | 1528 | |
9cd80bbb ON |
1529 | if (ret) |
1530 | return ret; | |
98abed02 RM |
1531 | } |
1532 | ||
1533 | return 0; | |
1534 | } | |
1535 | ||
9e8ae01d | 1536 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) |
98abed02 RM |
1537 | { |
1538 | struct task_struct *p; | |
1539 | ||
f470021a | 1540 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { |
b6e763f0 | 1541 | int ret = wait_consider_task(wo, 1, p); |
a0be55de | 1542 | |
f470021a | 1543 | if (ret) |
98abed02 | 1544 | return ret; |
98abed02 RM |
1545 | } |
1546 | ||
1547 | return 0; | |
1548 | } | |
1549 | ||
0b7570e7 ON |
1550 | static int child_wait_callback(wait_queue_t *wait, unsigned mode, |
1551 | int sync, void *key) | |
1552 | { | |
1553 | struct wait_opts *wo = container_of(wait, struct wait_opts, | |
1554 | child_wait); | |
1555 | struct task_struct *p = key; | |
1556 | ||
5c01ba49 | 1557 | if (!eligible_pid(wo, p)) |
0b7570e7 ON |
1558 | return 0; |
1559 | ||
b4fe5182 ON |
1560 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) |
1561 | return 0; | |
1562 | ||
0b7570e7 ON |
1563 | return default_wake_function(wait, mode, sync, key); |
1564 | } | |
1565 | ||
a7f0765e ON |
1566 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent) |
1567 | { | |
0b7570e7 ON |
1568 | __wake_up_sync_key(&parent->signal->wait_chldexit, |
1569 | TASK_INTERRUPTIBLE, 1, p); | |
a7f0765e ON |
1570 | } |
1571 | ||
9e8ae01d | 1572 | static long do_wait(struct wait_opts *wo) |
1da177e4 | 1573 | { |
1da177e4 | 1574 | struct task_struct *tsk; |
98abed02 | 1575 | int retval; |
1da177e4 | 1576 | |
9e8ae01d | 1577 | trace_sched_process_wait(wo->wo_pid); |
0a16b607 | 1578 | |
0b7570e7 ON |
1579 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); |
1580 | wo->child_wait.private = current; | |
1581 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); | |
1da177e4 | 1582 | repeat: |
98abed02 | 1583 | /* |
3da56d16 | 1584 | * If there is nothing that can match our criteria, just get out. |
9e8ae01d ON |
1585 | * We will clear ->notask_error to zero if we see any child that |
1586 | * might later match our criteria, even if we are not able to reap | |
1587 | * it yet. | |
98abed02 | 1588 | */ |
64a16caf | 1589 | wo->notask_error = -ECHILD; |
9e8ae01d ON |
1590 | if ((wo->wo_type < PIDTYPE_MAX) && |
1591 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) | |
64a16caf | 1592 | goto notask; |
161550d7 | 1593 | |
f95d39d1 | 1594 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 LT |
1595 | read_lock(&tasklist_lock); |
1596 | tsk = current; | |
1597 | do { | |
64a16caf ON |
1598 | retval = do_wait_thread(wo, tsk); |
1599 | if (retval) | |
1600 | goto end; | |
9e8ae01d | 1601 | |
64a16caf ON |
1602 | retval = ptrace_do_wait(wo, tsk); |
1603 | if (retval) | |
98abed02 | 1604 | goto end; |
98abed02 | 1605 | |
9e8ae01d | 1606 | if (wo->wo_flags & __WNOTHREAD) |
1da177e4 | 1607 | break; |
a3f6dfb7 | 1608 | } while_each_thread(current, tsk); |
1da177e4 | 1609 | read_unlock(&tasklist_lock); |
f2cc3eb1 | 1610 | |
64a16caf | 1611 | notask: |
9e8ae01d ON |
1612 | retval = wo->notask_error; |
1613 | if (!retval && !(wo->wo_flags & WNOHANG)) { | |
1da177e4 | 1614 | retval = -ERESTARTSYS; |
98abed02 RM |
1615 | if (!signal_pending(current)) { |
1616 | schedule(); | |
1617 | goto repeat; | |
1618 | } | |
1da177e4 | 1619 | } |
1da177e4 | 1620 | end: |
f95d39d1 | 1621 | __set_current_state(TASK_RUNNING); |
0b7570e7 | 1622 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); |
1da177e4 LT |
1623 | return retval; |
1624 | } | |
1625 | ||
ce72a16f AV |
1626 | static long kernel_waitid(int which, pid_t upid, struct siginfo __user *infop, |
1627 | int options, struct rusage *ru) | |
1da177e4 | 1628 | { |
9e8ae01d | 1629 | struct wait_opts wo; |
161550d7 EB |
1630 | struct pid *pid = NULL; |
1631 | enum pid_type type; | |
1da177e4 LT |
1632 | long ret; |
1633 | ||
91c4e8ea ON |
1634 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| |
1635 | __WNOTHREAD|__WCLONE|__WALL)) | |
1da177e4 LT |
1636 | return -EINVAL; |
1637 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) | |
1638 | return -EINVAL; | |
1639 | ||
1640 | switch (which) { | |
1641 | case P_ALL: | |
161550d7 | 1642 | type = PIDTYPE_MAX; |
1da177e4 LT |
1643 | break; |
1644 | case P_PID: | |
161550d7 EB |
1645 | type = PIDTYPE_PID; |
1646 | if (upid <= 0) | |
1da177e4 LT |
1647 | return -EINVAL; |
1648 | break; | |
1649 | case P_PGID: | |
161550d7 EB |
1650 | type = PIDTYPE_PGID; |
1651 | if (upid <= 0) | |
1da177e4 | 1652 | return -EINVAL; |
1da177e4 LT |
1653 | break; |
1654 | default: | |
1655 | return -EINVAL; | |
1656 | } | |
1657 | ||
161550d7 EB |
1658 | if (type < PIDTYPE_MAX) |
1659 | pid = find_get_pid(upid); | |
9e8ae01d ON |
1660 | |
1661 | wo.wo_type = type; | |
1662 | wo.wo_pid = pid; | |
1663 | wo.wo_flags = options; | |
1664 | wo.wo_info = infop; | |
1665 | wo.wo_stat = NULL; | |
1666 | wo.wo_rusage = ru; | |
1667 | ret = do_wait(&wo); | |
dfe16dfa VM |
1668 | |
1669 | if (ret > 0) { | |
1670 | ret = 0; | |
1671 | } else if (infop) { | |
1672 | /* | |
1673 | * For a WNOHANG return, clear out all the fields | |
1674 | * we would set so the user can easily tell the | |
1675 | * difference. | |
1676 | */ | |
1677 | if (!ret) | |
1678 | ret = put_user(0, &infop->si_signo); | |
1679 | if (!ret) | |
1680 | ret = put_user(0, &infop->si_errno); | |
1681 | if (!ret) | |
1682 | ret = put_user(0, &infop->si_code); | |
1683 | if (!ret) | |
1684 | ret = put_user(0, &infop->si_pid); | |
1685 | if (!ret) | |
1686 | ret = put_user(0, &infop->si_uid); | |
1687 | if (!ret) | |
1688 | ret = put_user(0, &infop->si_status); | |
1689 | } | |
1690 | ||
161550d7 | 1691 | put_pid(pid); |
1da177e4 LT |
1692 | return ret; |
1693 | } | |
1694 | ||
ce72a16f AV |
1695 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, |
1696 | infop, int, options, struct rusage __user *, ru) | |
1697 | { | |
1698 | struct rusage r; | |
1699 | long err = kernel_waitid(which, upid, infop, options, ru ? &r : NULL); | |
1700 | ||
1701 | if (!err) { | |
1702 | if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) | |
1703 | return -EFAULT; | |
1704 | } | |
1705 | return err; | |
1706 | } | |
1707 | ||
1708 | static long kernel_wait4(pid_t upid, int __user *stat_addr, | |
1709 | int options, struct rusage *ru) | |
1da177e4 | 1710 | { |
9e8ae01d | 1711 | struct wait_opts wo; |
161550d7 EB |
1712 | struct pid *pid = NULL; |
1713 | enum pid_type type; | |
1da177e4 LT |
1714 | long ret; |
1715 | ||
1716 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| | |
1717 | __WNOTHREAD|__WCLONE|__WALL)) | |
1718 | return -EINVAL; | |
161550d7 EB |
1719 | |
1720 | if (upid == -1) | |
1721 | type = PIDTYPE_MAX; | |
1722 | else if (upid < 0) { | |
1723 | type = PIDTYPE_PGID; | |
1724 | pid = find_get_pid(-upid); | |
1725 | } else if (upid == 0) { | |
1726 | type = PIDTYPE_PGID; | |
2ae448ef | 1727 | pid = get_task_pid(current, PIDTYPE_PGID); |
161550d7 EB |
1728 | } else /* upid > 0 */ { |
1729 | type = PIDTYPE_PID; | |
1730 | pid = find_get_pid(upid); | |
1731 | } | |
1732 | ||
9e8ae01d ON |
1733 | wo.wo_type = type; |
1734 | wo.wo_pid = pid; | |
1735 | wo.wo_flags = options | WEXITED; | |
1736 | wo.wo_info = NULL; | |
1737 | wo.wo_stat = stat_addr; | |
1738 | wo.wo_rusage = ru; | |
1739 | ret = do_wait(&wo); | |
161550d7 | 1740 | put_pid(pid); |
1da177e4 | 1741 | |
1da177e4 LT |
1742 | return ret; |
1743 | } | |
1744 | ||
ce72a16f AV |
1745 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, |
1746 | int, options, struct rusage __user *, ru) | |
1747 | { | |
1748 | struct rusage r; | |
1749 | long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); | |
1750 | ||
1751 | if (err > 0) { | |
1752 | if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) | |
1753 | return -EFAULT; | |
1754 | } | |
1755 | return err; | |
1756 | } | |
1757 | ||
1da177e4 LT |
1758 | #ifdef __ARCH_WANT_SYS_WAITPID |
1759 | ||
1760 | /* | |
1761 | * sys_waitpid() remains for compatibility. waitpid() should be | |
1762 | * implemented by calling sys_wait4() from libc.a. | |
1763 | */ | |
17da2bd9 | 1764 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) |
1da177e4 LT |
1765 | { |
1766 | return sys_wait4(pid, stat_addr, options, NULL); | |
1767 | } | |
1768 | ||
1769 | #endif | |
7e95a225 AV |
1770 | |
1771 | #ifdef CONFIG_COMPAT | |
1772 | COMPAT_SYSCALL_DEFINE4(wait4, | |
1773 | compat_pid_t, pid, | |
1774 | compat_uint_t __user *, stat_addr, | |
1775 | int, options, | |
1776 | struct compat_rusage __user *, ru) | |
1777 | { | |
ce72a16f AV |
1778 | struct rusage r; |
1779 | long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); | |
1780 | if (err > 0) { | |
1781 | if (ru && put_compat_rusage(&r, ru)) | |
1782 | return -EFAULT; | |
7e95a225 | 1783 | } |
ce72a16f | 1784 | return err; |
7e95a225 AV |
1785 | } |
1786 | ||
1787 | COMPAT_SYSCALL_DEFINE5(waitid, | |
1788 | int, which, compat_pid_t, pid, | |
1789 | struct compat_siginfo __user *, infop, int, options, | |
1790 | struct compat_rusage __user *, uru) | |
1791 | { | |
1792 | siginfo_t info; | |
1793 | struct rusage ru; | |
1794 | long ret; | |
1795 | mm_segment_t old_fs = get_fs(); | |
1796 | ||
1797 | memset(&info, 0, sizeof(info)); | |
1798 | ||
1799 | set_fs(KERNEL_DS); | |
ce72a16f AV |
1800 | ret = kernel_waitid(which, pid, (siginfo_t __user *)&info, options, |
1801 | uru ? &ru : NULL); | |
7e95a225 AV |
1802 | set_fs(old_fs); |
1803 | ||
1804 | if ((ret < 0) || (info.si_signo == 0)) | |
1805 | return ret; | |
1806 | ||
1807 | if (uru) { | |
1808 | /* sys_waitid() overwrites everything in ru */ | |
1809 | if (COMPAT_USE_64BIT_TIME) | |
1810 | ret = copy_to_user(uru, &ru, sizeof(ru)); | |
1811 | else | |
1812 | ret = put_compat_rusage(&ru, uru); | |
1813 | if (ret) | |
1814 | return -EFAULT; | |
1815 | } | |
1816 | ||
1817 | BUG_ON(info.si_code & __SI_MASK); | |
1818 | info.si_code |= __SI_CHLD; | |
1819 | return copy_siginfo_to_user32(infop, &info); | |
1820 | } | |
1821 | #endif |