blob: 3c2cf91defa764842939405aac8e1cfae29c191b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/interrupt.h>
10#include <linux/smp_lock.h>
11#include <linux/module.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080012#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/completion.h>
14#include <linux/personality.h>
15#include <linux/tty.h>
16#include <linux/namespace.h>
17#include <linux/key.h>
18#include <linux/security.h>
19#include <linux/cpu.h>
20#include <linux/acct.h>
21#include <linux/file.h>
22#include <linux/binfmts.h>
23#include <linux/ptrace.h>
24#include <linux/profile.h>
25#include <linux/mount.h>
26#include <linux/proc_fs.h>
27#include <linux/mempolicy.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070028#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/cpuset.h>
30#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070031#include <linux/signal.h>
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080032#include <linux/posix-timers.h>
Matt Helsley9f460802005-11-07 00:59:16 -080033#include <linux/cn_proc.h>
Ingo Molnarde5097c2006-01-09 15:59:21 -080034#include <linux/mutex.h>
Ingo Molnar0771dfe2006-03-27 01:16:22 -080035#include <linux/futex.h>
Ingo Molnar34f192c2006-03-27 01:16:24 -080036#include <linux/compat.h>
Jens Axboeb92ce552006-04-11 13:52:07 +020037#include <linux/pipe_fs_i.h>
Al Virofa84cb92006-03-29 20:30:19 -050038#include <linux/audit.h> /* for audit_free() */
Adrian Bunk83cc5ed2006-06-25 05:47:41 -070039#include <linux/resource.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/uaccess.h>
42#include <asm/unistd.h>
43#include <asm/pgtable.h>
44#include <asm/mmu_context.h>
45
46extern void sem_exit (void);
47extern struct task_struct *child_reaper;
48
Adrian Bunk408b6642005-05-01 08:59:29 -070049static void exit_mm(struct task_struct * tsk);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static void __unhash_process(struct task_struct *p)
52{
53 nr_threads--;
54 detach_pid(p, PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 if (thread_group_leader(p)) {
56 detach_pid(p, PIDTYPE_PGID);
57 detach_pid(p, PIDTYPE_SID);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080058
Eric W. Biederman5e85d4a2006-04-18 22:20:16 -070059 list_del_rcu(&p->tasks);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -080060 __get_cpu_var(process_counts)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 }
Oleg Nesterov47e65322006-03-28 16:11:25 -080062 list_del_rcu(&p->thread_group);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080063 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064}
65
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080066/*
67 * This function expects the tasklist_lock write-locked.
68 */
69static void __exit_signal(struct task_struct *tsk)
70{
71 struct signal_struct *sig = tsk->signal;
72 struct sighand_struct *sighand;
73
74 BUG_ON(!sig);
75 BUG_ON(!atomic_read(&sig->count));
76
77 rcu_read_lock();
78 sighand = rcu_dereference(tsk->sighand);
79 spin_lock(&sighand->siglock);
80
81 posix_cpu_timers_exit(tsk);
82 if (atomic_dec_and_test(&sig->count))
83 posix_cpu_timers_exit_group(tsk);
84 else {
85 /*
86 * If there is any task waiting for the group exit
87 * then notify it:
88 */
89 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
90 wake_up_process(sig->group_exit_task);
91 sig->group_exit_task = NULL;
92 }
93 if (tsk == sig->curr_target)
94 sig->curr_target = next_thread(tsk);
95 /*
96 * Accumulate here the counters for all threads but the
97 * group leader as they die, so they can be added into
98 * the process-wide totals when those are taken.
99 * The group leader stays around as a zombie as long
100 * as there are other threads. When it gets reaped,
101 * the exit.c code will add its counts into these totals.
102 * We won't ever get here for the group leader, since it
103 * will have been the last reference on the signal_struct.
104 */
105 sig->utime = cputime_add(sig->utime, tsk->utime);
106 sig->stime = cputime_add(sig->stime, tsk->stime);
107 sig->min_flt += tsk->min_flt;
108 sig->maj_flt += tsk->maj_flt;
109 sig->nvcsw += tsk->nvcsw;
110 sig->nivcsw += tsk->nivcsw;
111 sig->sched_time += tsk->sched_time;
112 sig = NULL; /* Marker for below. */
113 }
114
Oleg Nesterov58767002006-03-28 16:11:20 -0800115 __unhash_process(tsk);
116
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800117 tsk->signal = NULL;
Oleg Nesterova7e53282006-03-28 16:11:27 -0800118 tsk->sighand = NULL;
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800119 spin_unlock(&sighand->siglock);
120 rcu_read_unlock();
121
Oleg Nesterova7e53282006-03-28 16:11:27 -0800122 __cleanup_sighand(sighand);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800123 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
124 flush_sigqueue(&tsk->pending);
125 if (sig) {
126 flush_sigqueue(&sig->shared_pending);
127 __cleanup_signal(sig);
128 }
129}
130
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800131static void delayed_put_task_struct(struct rcu_head *rhp)
132{
133 put_task_struct(container_of(rhp, struct task_struct, rcu));
134}
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136void release_task(struct task_struct * p)
137{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700138 struct task_struct *leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 int zap_leader;
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800140repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 atomic_dec(&p->user->processes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 write_lock_irq(&tasklist_lock);
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800143 ptrace_unlink(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
145 __exit_signal(p);
Oleg Nesterov35f5cad2006-03-28 16:11:19 -0800146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 /*
148 * If we are the last non-leader member of the thread
149 * group, and the leader is zombie, then notify the
150 * group leader's parent process. (if it wants notification.)
151 */
152 zap_leader = 0;
153 leader = p->group_leader;
154 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
155 BUG_ON(leader->exit_signal == -1);
156 do_notify_parent(leader, leader->exit_signal);
157 /*
158 * If we were the last child thread and the leader has
159 * exited already, and the leader's parent ignores SIGCHLD,
160 * then we are the one who should release the leader.
161 *
162 * do_notify_parent() will have marked it self-reaping in
163 * that case.
164 */
165 zap_leader = (leader->exit_signal == -1);
166 }
167
168 sched_exit(p);
169 write_unlock_irq(&tasklist_lock);
Eric W. Biederman48e64842006-06-26 00:25:48 -0700170 proc_flush_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 release_thread(p);
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800172 call_rcu(&p->rcu, delayed_put_task_struct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 p = leader;
175 if (unlikely(zap_leader))
176 goto repeat;
177}
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179/*
180 * This checks not only the pgrp, but falls back on the pid if no
181 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
182 * without this...
183 */
184int session_of_pgrp(int pgrp)
185{
186 struct task_struct *p;
187 int sid = -1;
188
189 read_lock(&tasklist_lock);
190 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
191 if (p->signal->session > 0) {
192 sid = p->signal->session;
193 goto out;
194 }
195 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
196 p = find_task_by_pid(pgrp);
197 if (p)
198 sid = p->signal->session;
199out:
200 read_unlock(&tasklist_lock);
201
202 return sid;
203}
204
205/*
206 * Determine if a process group is "orphaned", according to the POSIX
207 * definition in 2.2.2.52. Orphaned process groups are not to be affected
208 * by terminal-generated stop signals. Newly orphaned process groups are
209 * to receive a SIGHUP and a SIGCONT.
210 *
211 * "I ask you, have you ever known what it is to be an orphan?"
212 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700213static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct task_struct *p;
216 int ret = 1;
217
218 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
219 if (p == ignored_task
220 || p->exit_state
221 || p->real_parent->pid == 1)
222 continue;
223 if (process_group(p->real_parent) != pgrp
224 && p->real_parent->signal->session == p->signal->session) {
225 ret = 0;
226 break;
227 }
228 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
229 return ret; /* (sighing) "Often!" */
230}
231
232int is_orphaned_pgrp(int pgrp)
233{
234 int retval;
235
236 read_lock(&tasklist_lock);
237 retval = will_become_orphaned_pgrp(pgrp, NULL);
238 read_unlock(&tasklist_lock);
239
240 return retval;
241}
242
Arjan van de Ven858119e2006-01-14 13:20:43 -0800243static int has_stopped_jobs(int pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 int retval = 0;
246 struct task_struct *p;
247
248 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
249 if (p->state != TASK_STOPPED)
250 continue;
251
252 /* If p is stopped by a debugger on a signal that won't
253 stop it, then don't count p as stopped. This isn't
254 perfect but it's a good approximation. */
255 if (unlikely (p->ptrace)
256 && p->exit_code != SIGSTOP
257 && p->exit_code != SIGTSTP
258 && p->exit_code != SIGTTOU
259 && p->exit_code != SIGTTIN)
260 continue;
261
262 retval = 1;
263 break;
264 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
265 return retval;
266}
267
268/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700269 * reparent_to_init - Reparent the calling kernel thread to the init task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 *
271 * If a kernel thread is launched as a result of a system call, or if
272 * it ever exits, it should generally reparent itself to init so that
273 * it is correctly cleaned up on exit.
274 *
275 * The various task state such as scheduling policy and priority may have
276 * been inherited from a user process, so we reset them to sane values here.
277 *
278 * NOTE that reparent_to_init() gives the caller full capabilities.
279 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800280static void reparent_to_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
282 write_lock_irq(&tasklist_lock);
283
284 ptrace_unlink(current);
285 /* Reparent to init */
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800286 remove_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 current->parent = child_reaper;
288 current->real_parent = child_reaper;
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800289 add_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 /* Set the exit signal to SIGCHLD so we signal init on exit */
292 current->exit_signal = SIGCHLD;
293
Ingo Molnarb0a94992006-01-14 13:20:41 -0800294 if ((current->policy == SCHED_NORMAL ||
295 current->policy == SCHED_BATCH)
296 && (task_nice(current) < 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 set_user_nice(current, 0);
298 /* cpus_allowed? */
299 /* rt_priority? */
300 /* signals? */
301 security_task_reparent_to_init(current);
302 memcpy(current->signal->rlim, init_task.signal->rlim,
303 sizeof(current->signal->rlim));
304 atomic_inc(&(INIT_USER->__count));
305 write_unlock_irq(&tasklist_lock);
306 switch_uid(INIT_USER);
307}
308
309void __set_special_pids(pid_t session, pid_t pgrp)
310{
Oren Laadane19f2472006-01-08 01:03:58 -0800311 struct task_struct *curr = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 if (curr->signal->session != session) {
314 detach_pid(curr, PIDTYPE_SID);
315 curr->signal->session = session;
316 attach_pid(curr, PIDTYPE_SID, session);
317 }
318 if (process_group(curr) != pgrp) {
319 detach_pid(curr, PIDTYPE_PGID);
320 curr->signal->pgrp = pgrp;
321 attach_pid(curr, PIDTYPE_PGID, pgrp);
322 }
323}
324
325void set_special_pids(pid_t session, pid_t pgrp)
326{
327 write_lock_irq(&tasklist_lock);
328 __set_special_pids(session, pgrp);
329 write_unlock_irq(&tasklist_lock);
330}
331
332/*
333 * Let kernel threads use this to say that they
334 * allow a certain signal (since daemonize() will
335 * have disabled all of them by default).
336 */
337int allow_signal(int sig)
338{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700339 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return -EINVAL;
341
342 spin_lock_irq(&current->sighand->siglock);
343 sigdelset(&current->blocked, sig);
344 if (!current->mm) {
345 /* Kernel threads handle their own signals.
346 Let the signal code know it'll be handled, so
347 that they don't get converted to SIGKILL or
348 just silently dropped */
349 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
350 }
351 recalc_sigpending();
352 spin_unlock_irq(&current->sighand->siglock);
353 return 0;
354}
355
356EXPORT_SYMBOL(allow_signal);
357
358int disallow_signal(int sig)
359{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700360 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return -EINVAL;
362
363 spin_lock_irq(&current->sighand->siglock);
364 sigaddset(&current->blocked, sig);
365 recalc_sigpending();
366 spin_unlock_irq(&current->sighand->siglock);
367 return 0;
368}
369
370EXPORT_SYMBOL(disallow_signal);
371
372/*
373 * Put all the gunge required to become a kernel thread without
374 * attached user resources in one place where it belongs.
375 */
376
377void daemonize(const char *name, ...)
378{
379 va_list args;
380 struct fs_struct *fs;
381 sigset_t blocked;
382
383 va_start(args, name);
384 vsnprintf(current->comm, sizeof(current->comm), name, args);
385 va_end(args);
386
387 /*
388 * If we were started as result of loading a module, close all of the
389 * user space pages. We don't need them, and if we didn't close them
390 * they would be locked into memory.
391 */
392 exit_mm(current);
393
394 set_special_pids(1, 1);
Ingo Molnar70522e12006-03-23 03:00:31 -0800395 mutex_lock(&tty_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 current->signal->tty = NULL;
Ingo Molnar70522e12006-03-23 03:00:31 -0800397 mutex_unlock(&tty_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 /* Block and flush all signals */
400 sigfillset(&blocked);
401 sigprocmask(SIG_BLOCK, &blocked, NULL);
402 flush_signals(current);
403
404 /* Become as one with the init task */
405
406 exit_fs(current); /* current->fs->count--; */
407 fs = init_task.fs;
408 current->fs = fs;
409 atomic_inc(&fs->count);
Björn Steinbrink59148112006-02-18 18:12:43 +0100410 exit_namespace(current);
411 current->namespace = init_task.namespace;
412 get_namespace(current->namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 exit_files(current);
414 current->files = init_task.files;
415 atomic_inc(&current->files->count);
416
417 reparent_to_init();
418}
419
420EXPORT_SYMBOL(daemonize);
421
Arjan van de Ven858119e2006-01-14 13:20:43 -0800422static void close_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
424 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700425 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700428
429 /*
430 * It is safe to dereference the fd table without RCU or
431 * ->file_lock because this is the last reference to the
432 * files structure.
433 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700434 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 for (;;) {
436 unsigned long set;
437 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700438 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700440 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 while (set) {
442 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700443 struct file * file = xchg(&fdt->fd[i], NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 if (file)
445 filp_close(file, files);
446 }
447 i++;
448 set >>= 1;
449 }
450 }
451}
452
453struct files_struct *get_files_struct(struct task_struct *task)
454{
455 struct files_struct *files;
456
457 task_lock(task);
458 files = task->files;
459 if (files)
460 atomic_inc(&files->count);
461 task_unlock(task);
462
463 return files;
464}
465
466void fastcall put_files_struct(struct files_struct *files)
467{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700468 struct fdtable *fdt;
469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (atomic_dec_and_test(&files->count)) {
471 close_files(files);
472 /*
473 * Free the fd and fdset arrays if we expanded them.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700474 * If the fdtable was embedded, pass files for freeing
475 * at the end of the RCU grace period. Otherwise,
476 * you can free files immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700478 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700479 if (fdt == &files->fdtab)
480 fdt->free_files = files;
481 else
482 kmem_cache_free(files_cachep, files);
483 free_fdtable(fdt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485}
486
487EXPORT_SYMBOL(put_files_struct);
488
489static inline void __exit_files(struct task_struct *tsk)
490{
491 struct files_struct * files = tsk->files;
492
493 if (files) {
494 task_lock(tsk);
495 tsk->files = NULL;
496 task_unlock(tsk);
497 put_files_struct(files);
498 }
499}
500
501void exit_files(struct task_struct *tsk)
502{
503 __exit_files(tsk);
504}
505
506static inline void __put_fs_struct(struct fs_struct *fs)
507{
508 /* No need to hold fs->lock if we are killing it */
509 if (atomic_dec_and_test(&fs->count)) {
510 dput(fs->root);
511 mntput(fs->rootmnt);
512 dput(fs->pwd);
513 mntput(fs->pwdmnt);
514 if (fs->altroot) {
515 dput(fs->altroot);
516 mntput(fs->altrootmnt);
517 }
518 kmem_cache_free(fs_cachep, fs);
519 }
520}
521
522void put_fs_struct(struct fs_struct *fs)
523{
524 __put_fs_struct(fs);
525}
526
527static inline void __exit_fs(struct task_struct *tsk)
528{
529 struct fs_struct * fs = tsk->fs;
530
531 if (fs) {
532 task_lock(tsk);
533 tsk->fs = NULL;
534 task_unlock(tsk);
535 __put_fs_struct(fs);
536 }
537}
538
539void exit_fs(struct task_struct *tsk)
540{
541 __exit_fs(tsk);
542}
543
544EXPORT_SYMBOL_GPL(exit_fs);
545
546/*
547 * Turn us into a lazy TLB process if we
548 * aren't already..
549 */
Adrian Bunk408b6642005-05-01 08:59:29 -0700550static void exit_mm(struct task_struct * tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
552 struct mm_struct *mm = tsk->mm;
553
554 mm_release(tsk, mm);
555 if (!mm)
556 return;
557 /*
558 * Serialize with any possible pending coredump.
559 * We must hold mmap_sem around checking core_waiters
560 * and clearing tsk->mm. The core-inducing thread
561 * will increment core_waiters for each thread in the
562 * group with ->mm != NULL.
563 */
564 down_read(&mm->mmap_sem);
565 if (mm->core_waiters) {
566 up_read(&mm->mmap_sem);
567 down_write(&mm->mmap_sem);
568 if (!--mm->core_waiters)
569 complete(mm->core_startup_done);
570 up_write(&mm->mmap_sem);
571
572 wait_for_completion(&mm->core_done);
573 down_read(&mm->mmap_sem);
574 }
575 atomic_inc(&mm->mm_count);
Eric Sesterhenn125e1872006-06-23 02:06:06 -0700576 BUG_ON(mm != tsk->active_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 /* more a memory barrier than a real lock */
578 task_lock(tsk);
579 tsk->mm = NULL;
580 up_read(&mm->mmap_sem);
581 enter_lazy_tlb(mm, current);
582 task_unlock(tsk);
583 mmput(mm);
584}
585
Ingo Molnar36c8b582006-07-03 00:25:41 -0700586static inline void
587choose_new_parent(struct task_struct *p, struct task_struct *reaper)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
589 /*
590 * Make sure we're not reparenting to ourselves and that
591 * the parent is not a zombie.
592 */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800593 BUG_ON(p == reaper || reaper->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 p->real_parent = reaper;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Ingo Molnar36c8b582006-07-03 00:25:41 -0700597static void
598reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
600 /* We don't want people slaying init. */
601 if (p->exit_signal != -1)
602 p->exit_signal = SIGCHLD;
603
604 if (p->pdeath_signal)
605 /* We already hold the tasklist_lock here. */
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800606 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 /* Move the child from its dying parent to the new one. */
609 if (unlikely(traced)) {
610 /* Preserve ptrace links if someone else is tracing this child. */
611 list_del_init(&p->ptrace_list);
612 if (p->parent != p->real_parent)
613 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
614 } else {
615 /* If this child is being traced, then we're the one tracing it
616 * anyway, so let go of it.
617 */
618 p->ptrace = 0;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800619 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 p->parent = p->real_parent;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800621 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
623 /* If we'd notified the old parent about this child's death,
624 * also notify the new parent.
625 */
626 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
627 thread_group_empty(p))
628 do_notify_parent(p, p->exit_signal);
629 else if (p->state == TASK_TRACED) {
630 /*
631 * If it was at a trace stop, turn it into
632 * a normal stop since it's no longer being
633 * traced.
634 */
635 ptrace_untrace(p);
636 }
637 }
638
639 /*
640 * process group orphan check
641 * Case ii: Our child is in a different pgrp
642 * than we are, and it was the only connection
643 * outside, so the child pgrp is now orphaned.
644 */
645 if ((process_group(p) != process_group(father)) &&
646 (p->signal->session == father->signal->session)) {
647 int pgrp = process_group(p);
648
649 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800650 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
651 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
653 }
654}
655
656/*
657 * When we die, we re-parent all our children.
658 * Try to give them to another thread in our thread
659 * group, and if no such member exists, give it to
660 * the global child reaper process (ie "init")
661 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700662static void
663forget_original_parent(struct task_struct *father, struct list_head *to_release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
665 struct task_struct *p, *reaper = father;
666 struct list_head *_p, *_n;
667
668 do {
669 reaper = next_thread(reaper);
670 if (reaper == father) {
671 reaper = child_reaper;
672 break;
673 }
674 } while (reaper->exit_state);
675
676 /*
677 * There are only two places where our children can be:
678 *
679 * - in our child list
680 * - in our ptraced child list
681 *
682 * Search them and reparent children.
683 */
684 list_for_each_safe(_p, _n, &father->children) {
685 int ptrace;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700686 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
688 ptrace = p->ptrace;
689
690 /* if father isn't the real parent, then ptrace must be enabled */
691 BUG_ON(father != p->real_parent && !ptrace);
692
693 if (father == p->real_parent) {
694 /* reparent with a reaper, real father it's us */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800695 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 reparent_thread(p, father, 0);
697 } else {
698 /* reparent ptraced task to its real parent */
699 __ptrace_unlink (p);
700 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
701 thread_group_empty(p))
702 do_notify_parent(p, p->exit_signal);
703 }
704
705 /*
706 * if the ptraced child is a zombie with exit_signal == -1
707 * we must collect it before we exit, or it will remain
708 * zombie forever since we prevented it from self-reap itself
709 * while it was being traced by us, to be able to see it in wait4.
710 */
711 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
712 list_add(&p->ptrace_list, to_release);
713 }
714 list_for_each_safe(_p, _n, &father->ptrace_children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -0700715 p = list_entry(_p, struct task_struct, ptrace_list);
Oleg Nesterovd799f032006-03-28 16:11:04 -0800716 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 reparent_thread(p, father, 1);
718 }
719}
720
721/*
722 * Send signals to all our closest relatives so that they know
723 * to properly mourn us..
724 */
725static void exit_notify(struct task_struct *tsk)
726{
727 int state;
728 struct task_struct *t;
729 struct list_head ptrace_dead, *_p, *_n;
730
731 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
732 && !thread_group_empty(tsk)) {
733 /*
734 * This occurs when there was a race between our exit
735 * syscall and a group signal choosing us as the one to
736 * wake up. It could be that we are the only thread
737 * alerted to check for pending signals, but another thread
738 * should be woken now to take the signal since we will not.
739 * Now we'll wake all the threads in the group just to make
740 * sure someone gets all the pending signals.
741 */
742 read_lock(&tasklist_lock);
743 spin_lock_irq(&tsk->sighand->siglock);
744 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
745 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
746 recalc_sigpending_tsk(t);
747 if (signal_pending(t))
748 signal_wake_up(t, 0);
749 }
750 spin_unlock_irq(&tsk->sighand->siglock);
751 read_unlock(&tasklist_lock);
752 }
753
754 write_lock_irq(&tasklist_lock);
755
756 /*
757 * This does two things:
758 *
759 * A. Make init inherit all the child processes
760 * B. Check to see if any process groups have become orphaned
761 * as a result of our exiting, and if they have any stopped
762 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
763 */
764
765 INIT_LIST_HEAD(&ptrace_dead);
766 forget_original_parent(tsk, &ptrace_dead);
767 BUG_ON(!list_empty(&tsk->children));
768 BUG_ON(!list_empty(&tsk->ptrace_children));
769
770 /*
771 * Check to see if any process groups have become orphaned
772 * as a result of our exiting, and if they have any stopped
773 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
774 *
775 * Case i: Our father is in a different pgrp than we are
776 * and we were the only connection outside, so our pgrp
777 * is about to become orphaned.
778 */
779
780 t = tsk->real_parent;
781
782 if ((process_group(t) != process_group(tsk)) &&
783 (t->signal->session == tsk->signal->session) &&
784 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
785 has_stopped_jobs(process_group(tsk))) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800786 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
787 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789
790 /* Let father know we died
791 *
792 * Thread signals are configurable, but you aren't going to use
793 * that to send signals to arbitary processes.
794 * That stops right now.
795 *
796 * If the parent exec id doesn't match the exec id we saved
797 * when we started then we know the parent has changed security
798 * domain.
799 *
800 * If our self_exec id doesn't match our parent_exec_id then
801 * we have changed execution domain as these two values started
802 * the same after a fork.
803 *
804 */
805
806 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
807 ( tsk->parent_exec_id != t->self_exec_id ||
808 tsk->self_exec_id != tsk->parent_exec_id)
809 && !capable(CAP_KILL))
810 tsk->exit_signal = SIGCHLD;
811
812
813 /* If something other than our normal parent is ptracing us, then
814 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
815 * only has special meaning to our real parent.
816 */
817 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
818 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
819 do_notify_parent(tsk, signal);
820 } else if (tsk->ptrace) {
821 do_notify_parent(tsk, SIGCHLD);
822 }
823
824 state = EXIT_ZOMBIE;
825 if (tsk->exit_signal == -1 &&
826 (likely(tsk->ptrace == 0) ||
827 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
828 state = EXIT_DEAD;
829 tsk->exit_state = state;
830
831 write_unlock_irq(&tasklist_lock);
832
833 list_for_each_safe(_p, _n, &ptrace_dead) {
834 list_del_init(_p);
Ingo Molnar36c8b582006-07-03 00:25:41 -0700835 t = list_entry(_p, struct task_struct, ptrace_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 release_task(t);
837 }
838
839 /* If the process is dead, release it - nobody will wait for it */
840 if (state == EXIT_DEAD)
841 release_task(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
844fastcall NORET_TYPE void do_exit(long code)
845{
846 struct task_struct *tsk = current;
847 int group_dead;
848
849 profile_task_exit(tsk);
850
Jens Axboe22e2c502005-06-27 10:55:12 +0200851 WARN_ON(atomic_read(&tsk->fs_excl));
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 if (unlikely(in_interrupt()))
854 panic("Aiee, killing interrupt handler!");
855 if (unlikely(!tsk->pid))
856 panic("Attempted to kill the idle task!");
Eric W. Biedermanfef23e72006-03-28 16:10:58 -0800857 if (unlikely(tsk == child_reaper))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 panic("Attempted to kill init!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
860 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
861 current->ptrace_message = code;
862 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
863 }
864
Alexander Nybergdf164db2005-06-23 00:09:13 -0700865 /*
866 * We're taking recursive faults here in do_exit. Safest is to just
867 * leave this task alone and wait for reboot.
868 */
869 if (unlikely(tsk->flags & PF_EXITING)) {
870 printk(KERN_ALERT
871 "Fixing recursive fault but reboot is needed!\n");
Al Viroafc847b2006-02-28 12:51:55 -0500872 if (tsk->io_context)
873 exit_io_context();
Alexander Nybergdf164db2005-06-23 00:09:13 -0700874 set_current_state(TASK_UNINTERRUPTIBLE);
875 schedule();
876 }
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 tsk->flags |= PF_EXITING;
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 if (unlikely(in_atomic()))
881 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
882 current->comm, current->pid,
883 preempt_count());
884
885 acct_update_integrals(tsk);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700886 if (tsk->mm) {
887 update_hiwater_rss(tsk->mm);
888 update_hiwater_vm(tsk->mm);
889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 group_dead = atomic_dec_and_test(&tsk->signal->live);
Andrew Mortonc3068952005-08-04 16:49:32 -0700891 if (group_dead) {
Thomas Gleixner2ff678b2006-01-09 20:52:34 -0800892 hrtimer_cancel(&tsk->signal->real_timer);
Roland McGrath25f407f2005-10-21 15:03:29 -0700893 exit_itimers(tsk->signal);
Andrew Mortonc3068952005-08-04 16:49:32 -0700894 }
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700895 acct_collect(code, group_dead);
Ingo Molnar0771dfe2006-03-27 01:16:22 -0800896 if (unlikely(tsk->robust_list))
897 exit_robust_list(tsk);
Anton Blanchard2aa92582006-06-25 05:48:44 -0700898#if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
Ingo Molnar34f192c2006-03-27 01:16:24 -0800899 if (unlikely(tsk->compat_robust_list))
900 compat_exit_robust_list(tsk);
901#endif
Al Virofa84cb92006-03-29 20:30:19 -0500902 if (unlikely(tsk->audit_context))
903 audit_free(tsk);
Shailabh Nagarca74e922006-07-14 00:24:36 -0700904 delayacct_tsk_exit(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 exit_mm(tsk);
906
KaiGai Kohei0e464812006-06-25 05:49:24 -0700907 if (group_dead)
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700908 acct_process();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 exit_sem(tsk);
910 __exit_files(tsk);
911 __exit_fs(tsk);
912 exit_namespace(tsk);
913 exit_thread();
914 cpuset_exit(tsk);
915 exit_keys(tsk);
916
917 if (group_dead && tsk->signal->leader)
918 disassociate_ctty(1);
919
Al Viroa1261f52005-11-13 16:06:55 -0800920 module_put(task_thread_info(tsk)->exec_domain->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (tsk->binfmt)
922 module_put(tsk->binfmt->module);
923
924 tsk->exit_code = code;
Matt Helsley9f460802005-11-07 00:59:16 -0800925 proc_exit_connector(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 exit_notify(tsk);
927#ifdef CONFIG_NUMA
928 mpol_free(tsk->mempolicy);
929 tsk->mempolicy = NULL;
930#endif
Ingo Molnarde5097c2006-01-09 15:59:21 -0800931 /*
Ingo Molnarc87e2832006-06-27 02:54:58 -0700932 * This must happen late, after the PID is not
933 * hashed anymore:
934 */
935 if (unlikely(!list_empty(&tsk->pi_state_list)))
936 exit_pi_state_list(tsk);
937 if (unlikely(current->pi_state_cache))
938 kfree(current->pi_state_cache);
939 /*
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700940 * Make sure we are holding no locks:
Ingo Molnarde5097c2006-01-09 15:59:21 -0800941 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700942 debug_check_no_locks_held(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Al Viroafc847b2006-02-28 12:51:55 -0500944 if (tsk->io_context)
945 exit_io_context();
946
Jens Axboeb92ce552006-04-11 13:52:07 +0200947 if (tsk->splice_pipe)
948 __free_pipe_info(tsk->splice_pipe);
949
Coywolf Qi Hunt74072512005-10-30 15:02:47 -0800950 /* PF_DEAD causes final put_task_struct after we schedule. */
951 preempt_disable();
952 BUG_ON(tsk->flags & PF_DEAD);
953 tsk->flags |= PF_DEAD;
954
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 schedule();
956 BUG();
957 /* Avoid "noreturn function does return". */
958 for (;;) ;
959}
960
Russ Anderson012914d2005-04-23 00:08:00 -0700961EXPORT_SYMBOL_GPL(do_exit);
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963NORET_TYPE void complete_and_exit(struct completion *comp, long code)
964{
965 if (comp)
966 complete(comp);
967
968 do_exit(code);
969}
970
971EXPORT_SYMBOL(complete_and_exit);
972
973asmlinkage long sys_exit(int error_code)
974{
975 do_exit((error_code&0xff)<<8);
976}
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978/*
979 * Take down every thread in the group. This is called by fatal signals
980 * as well as by sys_exit_group (below).
981 */
982NORET_TYPE void
983do_group_exit(int exit_code)
984{
985 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
986
987 if (current->signal->flags & SIGNAL_GROUP_EXIT)
988 exit_code = current->signal->group_exit_code;
989 else if (!thread_group_empty(current)) {
990 struct signal_struct *const sig = current->signal;
991 struct sighand_struct *const sighand = current->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 spin_lock_irq(&sighand->siglock);
993 if (sig->flags & SIGNAL_GROUP_EXIT)
994 /* Another thread got here before we took the lock. */
995 exit_code = sig->group_exit_code;
996 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 sig->group_exit_code = exit_code;
998 zap_other_threads(current);
999 }
1000 spin_unlock_irq(&sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
1002
1003 do_exit(exit_code);
1004 /* NOTREACHED */
1005}
1006
1007/*
1008 * this kills every thread in the thread group. Note that any externally
1009 * wait4()-ing process will get the correct exit code - even if this
1010 * thread is not the thread group leader.
1011 */
1012asmlinkage void sys_exit_group(int error_code)
1013{
1014 do_group_exit((error_code & 0xff) << 8);
1015}
1016
Ingo Molnar36c8b582006-07-03 00:25:41 -07001017static int eligible_child(pid_t pid, int options, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
1019 if (pid > 0) {
1020 if (p->pid != pid)
1021 return 0;
1022 } else if (!pid) {
1023 if (process_group(p) != process_group(current))
1024 return 0;
1025 } else if (pid != -1) {
1026 if (process_group(p) != -pid)
1027 return 0;
1028 }
1029
1030 /*
1031 * Do not consider detached threads that are
1032 * not ptraced:
1033 */
1034 if (p->exit_signal == -1 && !p->ptrace)
1035 return 0;
1036
1037 /* Wait for all children (clone and not) if __WALL is set;
1038 * otherwise, wait for clone children *only* if __WCLONE is
1039 * set; otherwise, wait for non-clone children *only*. (Note:
1040 * A "clone" child here is one that reports to its parent
1041 * using a signal other than SIGCHLD.) */
1042 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1043 && !(options & __WALL))
1044 return 0;
1045 /*
1046 * Do not consider thread group leaders that are
1047 * in a non-empty thread group:
1048 */
1049 if (current->tgid != p->tgid && delay_group_leader(p))
1050 return 2;
1051
1052 if (security_task_wait(p))
1053 return 0;
1054
1055 return 1;
1056}
1057
Ingo Molnar36c8b582006-07-03 00:25:41 -07001058static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 int why, int status,
1060 struct siginfo __user *infop,
1061 struct rusage __user *rusagep)
1062{
1063 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07001064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 put_task_struct(p);
1066 if (!retval)
1067 retval = put_user(SIGCHLD, &infop->si_signo);
1068 if (!retval)
1069 retval = put_user(0, &infop->si_errno);
1070 if (!retval)
1071 retval = put_user((short)why, &infop->si_code);
1072 if (!retval)
1073 retval = put_user(pid, &infop->si_pid);
1074 if (!retval)
1075 retval = put_user(uid, &infop->si_uid);
1076 if (!retval)
1077 retval = put_user(status, &infop->si_status);
1078 if (!retval)
1079 retval = pid;
1080 return retval;
1081}
1082
1083/*
1084 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1085 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1086 * the lock and this task is uninteresting. If we return nonzero, we have
1087 * released the lock and the system call should return.
1088 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001089static int wait_task_zombie(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct siginfo __user *infop,
1091 int __user *stat_addr, struct rusage __user *ru)
1092{
1093 unsigned long state;
1094 int retval;
1095 int status;
1096
1097 if (unlikely(noreap)) {
1098 pid_t pid = p->pid;
1099 uid_t uid = p->uid;
1100 int exit_code = p->exit_code;
1101 int why, status;
1102
1103 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1104 return 0;
1105 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1106 return 0;
1107 get_task_struct(p);
1108 read_unlock(&tasklist_lock);
1109 if ((exit_code & 0x7f) == 0) {
1110 why = CLD_EXITED;
1111 status = exit_code >> 8;
1112 } else {
1113 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1114 status = exit_code & 0x7f;
1115 }
1116 return wait_noreap_copyout(p, pid, uid, why,
1117 status, infop, ru);
1118 }
1119
1120 /*
1121 * Try to move the task's state to DEAD
1122 * only one thread is allowed to do this:
1123 */
1124 state = xchg(&p->exit_state, EXIT_DEAD);
1125 if (state != EXIT_ZOMBIE) {
1126 BUG_ON(state != EXIT_DEAD);
1127 return 0;
1128 }
1129 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1130 /*
1131 * This can only happen in a race with a ptraced thread
1132 * dying on another processor.
1133 */
1134 return 0;
1135 }
1136
1137 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
Jesper Juhl3795e162006-01-09 20:54:39 -08001138 struct signal_struct *psig;
1139 struct signal_struct *sig;
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 /*
1142 * The resource counters for the group leader are in its
1143 * own task_struct. Those for dead threads in the group
1144 * are in its signal_struct, as are those for the child
1145 * processes it has previously reaped. All these
1146 * accumulate in the parent's signal_struct c* fields.
1147 *
1148 * We don't bother to take a lock here to protect these
1149 * p->signal fields, because they are only touched by
1150 * __exit_signal, which runs with tasklist_lock
1151 * write-locked anyway, and so is excluded here. We do
1152 * need to protect the access to p->parent->signal fields,
1153 * as other threads in the parent group can be right
1154 * here reaping other children at the same time.
1155 */
1156 spin_lock_irq(&p->parent->sighand->siglock);
Jesper Juhl3795e162006-01-09 20:54:39 -08001157 psig = p->parent->signal;
1158 sig = p->signal;
1159 psig->cutime =
1160 cputime_add(psig->cutime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 cputime_add(p->utime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001162 cputime_add(sig->utime,
1163 sig->cutime)));
1164 psig->cstime =
1165 cputime_add(psig->cstime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 cputime_add(p->stime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001167 cputime_add(sig->stime,
1168 sig->cstime)));
1169 psig->cmin_flt +=
1170 p->min_flt + sig->min_flt + sig->cmin_flt;
1171 psig->cmaj_flt +=
1172 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1173 psig->cnvcsw +=
1174 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1175 psig->cnivcsw +=
1176 p->nivcsw + sig->nivcsw + sig->cnivcsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 spin_unlock_irq(&p->parent->sighand->siglock);
1178 }
1179
1180 /*
1181 * Now we are sure this task is interesting, and no other
1182 * thread can reap it because we set its state to EXIT_DEAD.
1183 */
1184 read_unlock(&tasklist_lock);
1185
1186 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1187 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1188 ? p->signal->group_exit_code : p->exit_code;
1189 if (!retval && stat_addr)
1190 retval = put_user(status, stat_addr);
1191 if (!retval && infop)
1192 retval = put_user(SIGCHLD, &infop->si_signo);
1193 if (!retval && infop)
1194 retval = put_user(0, &infop->si_errno);
1195 if (!retval && infop) {
1196 int why;
1197
1198 if ((status & 0x7f) == 0) {
1199 why = CLD_EXITED;
1200 status >>= 8;
1201 } else {
1202 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1203 status &= 0x7f;
1204 }
1205 retval = put_user((short)why, &infop->si_code);
1206 if (!retval)
1207 retval = put_user(status, &infop->si_status);
1208 }
1209 if (!retval && infop)
1210 retval = put_user(p->pid, &infop->si_pid);
1211 if (!retval && infop)
1212 retval = put_user(p->uid, &infop->si_uid);
1213 if (retval) {
1214 // TODO: is this safe?
1215 p->exit_state = EXIT_ZOMBIE;
1216 return retval;
1217 }
1218 retval = p->pid;
1219 if (p->real_parent != p->parent) {
1220 write_lock_irq(&tasklist_lock);
1221 /* Double-check with lock held. */
1222 if (p->real_parent != p->parent) {
1223 __ptrace_unlink(p);
1224 // TODO: is this safe?
1225 p->exit_state = EXIT_ZOMBIE;
1226 /*
1227 * If this is not a detached task, notify the parent.
1228 * If it's still not detached after that, don't release
1229 * it now.
1230 */
1231 if (p->exit_signal != -1) {
1232 do_notify_parent(p, p->exit_signal);
1233 if (p->exit_signal != -1)
1234 p = NULL;
1235 }
1236 }
1237 write_unlock_irq(&tasklist_lock);
1238 }
1239 if (p != NULL)
1240 release_task(p);
1241 BUG_ON(!retval);
1242 return retval;
1243}
1244
1245/*
1246 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1247 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1248 * the lock and this task is uninteresting. If we return nonzero, we have
1249 * released the lock and the system call should return.
1250 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001251static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1252 int noreap, struct siginfo __user *infop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 int __user *stat_addr, struct rusage __user *ru)
1254{
1255 int retval, exit_code;
1256
1257 if (!p->exit_code)
1258 return 0;
1259 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1260 p->signal && p->signal->group_stop_count > 0)
1261 /*
1262 * A group stop is in progress and this is the group leader.
1263 * We won't report until all threads have stopped.
1264 */
1265 return 0;
1266
1267 /*
1268 * Now we are pretty sure this task is interesting.
1269 * Make sure it doesn't get reaped out from under us while we
1270 * give up the lock and then examine it below. We don't want to
1271 * keep holding onto the tasklist_lock while we call getrusage and
1272 * possibly take page faults for user memory.
1273 */
1274 get_task_struct(p);
1275 read_unlock(&tasklist_lock);
1276
1277 if (unlikely(noreap)) {
1278 pid_t pid = p->pid;
1279 uid_t uid = p->uid;
1280 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1281
1282 exit_code = p->exit_code;
1283 if (unlikely(!exit_code) ||
Linus Torvalds14bf01b2005-10-01 11:04:18 -07001284 unlikely(p->state & TASK_TRACED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 goto bail_ref;
1286 return wait_noreap_copyout(p, pid, uid,
1287 why, (exit_code << 8) | 0x7f,
1288 infop, ru);
1289 }
1290
1291 write_lock_irq(&tasklist_lock);
1292
1293 /*
1294 * This uses xchg to be atomic with the thread resuming and setting
1295 * it. It must also be done with the write lock held to prevent a
1296 * race with the EXIT_ZOMBIE case.
1297 */
1298 exit_code = xchg(&p->exit_code, 0);
1299 if (unlikely(p->exit_state)) {
1300 /*
1301 * The task resumed and then died. Let the next iteration
1302 * catch it in EXIT_ZOMBIE. Note that exit_code might
1303 * already be zero here if it resumed and did _exit(0).
1304 * The task itself is dead and won't touch exit_code again;
1305 * other processors in this function are locked out.
1306 */
1307 p->exit_code = exit_code;
1308 exit_code = 0;
1309 }
1310 if (unlikely(exit_code == 0)) {
1311 /*
1312 * Another thread in this function got to it first, or it
1313 * resumed, or it resumed and then died.
1314 */
1315 write_unlock_irq(&tasklist_lock);
1316bail_ref:
1317 put_task_struct(p);
1318 /*
1319 * We are returning to the wait loop without having successfully
1320 * removed the process and having released the lock. We cannot
1321 * continue, since the "p" task pointer is potentially stale.
1322 *
1323 * Return -EAGAIN, and do_wait() will restart the loop from the
1324 * beginning. Do _not_ re-acquire the lock.
1325 */
1326 return -EAGAIN;
1327 }
1328
1329 /* move to end of parent's list to avoid starvation */
1330 remove_parent(p);
Oleg Nesterov8fafabd2006-03-28 16:11:05 -08001331 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 write_unlock_irq(&tasklist_lock);
1334
1335 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1336 if (!retval && stat_addr)
1337 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1338 if (!retval && infop)
1339 retval = put_user(SIGCHLD, &infop->si_signo);
1340 if (!retval && infop)
1341 retval = put_user(0, &infop->si_errno);
1342 if (!retval && infop)
1343 retval = put_user((short)((p->ptrace & PT_PTRACED)
1344 ? CLD_TRAPPED : CLD_STOPPED),
1345 &infop->si_code);
1346 if (!retval && infop)
1347 retval = put_user(exit_code, &infop->si_status);
1348 if (!retval && infop)
1349 retval = put_user(p->pid, &infop->si_pid);
1350 if (!retval && infop)
1351 retval = put_user(p->uid, &infop->si_uid);
1352 if (!retval)
1353 retval = p->pid;
1354 put_task_struct(p);
1355
1356 BUG_ON(!retval);
1357 return retval;
1358}
1359
1360/*
1361 * Handle do_wait work for one task in a live, non-stopped state.
1362 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1363 * the lock and this task is uninteresting. If we return nonzero, we have
1364 * released the lock and the system call should return.
1365 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001366static int wait_task_continued(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 struct siginfo __user *infop,
1368 int __user *stat_addr, struct rusage __user *ru)
1369{
1370 int retval;
1371 pid_t pid;
1372 uid_t uid;
1373
1374 if (unlikely(!p->signal))
1375 return 0;
1376
1377 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1378 return 0;
1379
1380 spin_lock_irq(&p->sighand->siglock);
1381 /* Re-check with the lock held. */
1382 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1383 spin_unlock_irq(&p->sighand->siglock);
1384 return 0;
1385 }
1386 if (!noreap)
1387 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1388 spin_unlock_irq(&p->sighand->siglock);
1389
1390 pid = p->pid;
1391 uid = p->uid;
1392 get_task_struct(p);
1393 read_unlock(&tasklist_lock);
1394
1395 if (!infop) {
1396 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1397 put_task_struct(p);
1398 if (!retval && stat_addr)
1399 retval = put_user(0xffff, stat_addr);
1400 if (!retval)
1401 retval = p->pid;
1402 } else {
1403 retval = wait_noreap_copyout(p, pid, uid,
1404 CLD_CONTINUED, SIGCONT,
1405 infop, ru);
1406 BUG_ON(retval == 0);
1407 }
1408
1409 return retval;
1410}
1411
1412
1413static inline int my_ptrace_child(struct task_struct *p)
1414{
1415 if (!(p->ptrace & PT_PTRACED))
1416 return 0;
1417 if (!(p->ptrace & PT_ATTACHED))
1418 return 1;
1419 /*
1420 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1421 * we are the attacher. If we are the real parent, this is a race
1422 * inside ptrace_attach. It is waiting for the tasklist_lock,
1423 * which we have to switch the parent links, but has already set
1424 * the flags in p->ptrace.
1425 */
1426 return (p->parent != p->real_parent);
1427}
1428
1429static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1430 int __user *stat_addr, struct rusage __user *ru)
1431{
1432 DECLARE_WAITQUEUE(wait, current);
1433 struct task_struct *tsk;
1434 int flag, retval;
1435
1436 add_wait_queue(&current->signal->wait_chldexit,&wait);
1437repeat:
1438 /*
1439 * We will set this flag if we see any child that might later
1440 * match our criteria, even if we are not able to reap it yet.
1441 */
1442 flag = 0;
1443 current->state = TASK_INTERRUPTIBLE;
1444 read_lock(&tasklist_lock);
1445 tsk = current;
1446 do {
1447 struct task_struct *p;
1448 struct list_head *_p;
1449 int ret;
1450
1451 list_for_each(_p,&tsk->children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -07001452 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
1454 ret = eligible_child(pid, options, p);
1455 if (!ret)
1456 continue;
1457
1458 switch (p->state) {
1459 case TASK_TRACED:
Roland McGrath7f2a5252005-10-30 15:02:50 -08001460 /*
1461 * When we hit the race with PTRACE_ATTACH,
1462 * we will not report this child. But the
1463 * race means it has not yet been moved to
1464 * our ptrace_children list, so we need to
1465 * set the flag here to avoid a spurious ECHILD
1466 * when the race happens with the only child.
1467 */
1468 flag = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (!my_ptrace_child(p))
1470 continue;
1471 /*FALLTHROUGH*/
1472 case TASK_STOPPED:
1473 /*
1474 * It's stopped now, so it might later
1475 * continue, exit, or stop again.
1476 */
1477 flag = 1;
1478 if (!(options & WUNTRACED) &&
1479 !my_ptrace_child(p))
1480 continue;
1481 retval = wait_task_stopped(p, ret == 2,
1482 (options & WNOWAIT),
1483 infop,
1484 stat_addr, ru);
1485 if (retval == -EAGAIN)
1486 goto repeat;
1487 if (retval != 0) /* He released the lock. */
1488 goto end;
1489 break;
1490 default:
1491 // case EXIT_DEAD:
1492 if (p->exit_state == EXIT_DEAD)
1493 continue;
1494 // case EXIT_ZOMBIE:
1495 if (p->exit_state == EXIT_ZOMBIE) {
1496 /*
1497 * Eligible but we cannot release
1498 * it yet:
1499 */
1500 if (ret == 2)
1501 goto check_continued;
1502 if (!likely(options & WEXITED))
1503 continue;
1504 retval = wait_task_zombie(
1505 p, (options & WNOWAIT),
1506 infop, stat_addr, ru);
1507 /* He released the lock. */
1508 if (retval != 0)
1509 goto end;
1510 break;
1511 }
1512check_continued:
1513 /*
1514 * It's running now, so it might later
1515 * exit, stop, or stop and then continue.
1516 */
1517 flag = 1;
1518 if (!unlikely(options & WCONTINUED))
1519 continue;
1520 retval = wait_task_continued(
1521 p, (options & WNOWAIT),
1522 infop, stat_addr, ru);
1523 if (retval != 0) /* He released the lock. */
1524 goto end;
1525 break;
1526 }
1527 }
1528 if (!flag) {
1529 list_for_each(_p, &tsk->ptrace_children) {
1530 p = list_entry(_p, struct task_struct,
1531 ptrace_list);
1532 if (!eligible_child(pid, options, p))
1533 continue;
1534 flag = 1;
1535 break;
1536 }
1537 }
1538 if (options & __WNOTHREAD)
1539 break;
1540 tsk = next_thread(tsk);
Eric Sesterhenn125e1872006-06-23 02:06:06 -07001541 BUG_ON(tsk->signal != current->signal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 } while (tsk != current);
1543
1544 read_unlock(&tasklist_lock);
1545 if (flag) {
1546 retval = 0;
1547 if (options & WNOHANG)
1548 goto end;
1549 retval = -ERESTARTSYS;
1550 if (signal_pending(current))
1551 goto end;
1552 schedule();
1553 goto repeat;
1554 }
1555 retval = -ECHILD;
1556end:
1557 current->state = TASK_RUNNING;
1558 remove_wait_queue(&current->signal->wait_chldexit,&wait);
1559 if (infop) {
1560 if (retval > 0)
1561 retval = 0;
1562 else {
1563 /*
1564 * For a WNOHANG return, clear out all the fields
1565 * we would set so the user can easily tell the
1566 * difference.
1567 */
1568 if (!retval)
1569 retval = put_user(0, &infop->si_signo);
1570 if (!retval)
1571 retval = put_user(0, &infop->si_errno);
1572 if (!retval)
1573 retval = put_user(0, &infop->si_code);
1574 if (!retval)
1575 retval = put_user(0, &infop->si_pid);
1576 if (!retval)
1577 retval = put_user(0, &infop->si_uid);
1578 if (!retval)
1579 retval = put_user(0, &infop->si_status);
1580 }
1581 }
1582 return retval;
1583}
1584
1585asmlinkage long sys_waitid(int which, pid_t pid,
1586 struct siginfo __user *infop, int options,
1587 struct rusage __user *ru)
1588{
1589 long ret;
1590
1591 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1592 return -EINVAL;
1593 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1594 return -EINVAL;
1595
1596 switch (which) {
1597 case P_ALL:
1598 pid = -1;
1599 break;
1600 case P_PID:
1601 if (pid <= 0)
1602 return -EINVAL;
1603 break;
1604 case P_PGID:
1605 if (pid <= 0)
1606 return -EINVAL;
1607 pid = -pid;
1608 break;
1609 default:
1610 return -EINVAL;
1611 }
1612
1613 ret = do_wait(pid, options, infop, NULL, ru);
1614
1615 /* avoid REGPARM breakage on x86: */
1616 prevent_tail_call(ret);
1617 return ret;
1618}
1619
1620asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1621 int options, struct rusage __user *ru)
1622{
1623 long ret;
1624
1625 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1626 __WNOTHREAD|__WCLONE|__WALL))
1627 return -EINVAL;
1628 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1629
1630 /* avoid REGPARM breakage on x86: */
1631 prevent_tail_call(ret);
1632 return ret;
1633}
1634
1635#ifdef __ARCH_WANT_SYS_WAITPID
1636
1637/*
1638 * sys_waitpid() remains for compatibility. waitpid() should be
1639 * implemented by calling sys_wait4() from libc.a.
1640 */
1641asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1642{
1643 return sys_wait4(pid, stat_addr, options, NULL);
1644}
1645
1646#endif