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