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