blob: c189de2927ab6b55ab47baf199d347d65481e906 [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 Nagarc7572492006-07-14 00:24:40 -070028#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070029#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/cpuset.h>
31#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070032#include <linux/signal.h>
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080033#include <linux/posix-timers.h>
Matt Helsley9f460802005-11-07 00:59:16 -080034#include <linux/cn_proc.h>
Ingo Molnarde5097c2006-01-09 15:59:21 -080035#include <linux/mutex.h>
Ingo Molnar0771dfe2006-03-27 01:16:22 -080036#include <linux/futex.h>
Ingo Molnar34f192c2006-03-27 01:16:24 -080037#include <linux/compat.h>
Jens Axboeb92ce552006-04-11 13:52:07 +020038#include <linux/pipe_fs_i.h>
Al Virofa84cb92006-03-29 20:30:19 -050039#include <linux/audit.h> /* for audit_free() */
Adrian Bunk83cc5ed2006-06-25 05:47:41 -070040#include <linux/resource.h>
David Howells0d67a462006-08-29 19:05:56 +010041#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#include <asm/uaccess.h>
44#include <asm/unistd.h>
45#include <asm/pgtable.h>
46#include <asm/mmu_context.h>
47
48extern void sem_exit (void);
49extern struct task_struct *child_reaper;
50
Adrian Bunk408b6642005-05-01 08:59:29 -070051static void exit_mm(struct task_struct * tsk);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static void __unhash_process(struct task_struct *p)
54{
55 nr_threads--;
56 detach_pid(p, PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 if (thread_group_leader(p)) {
58 detach_pid(p, PIDTYPE_PGID);
59 detach_pid(p, PIDTYPE_SID);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080060
Eric W. Biederman5e85d4a2006-04-18 22:20:16 -070061 list_del_rcu(&p->tasks);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -080062 __get_cpu_var(process_counts)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 }
Oleg Nesterov47e65322006-03-28 16:11:25 -080064 list_del_rcu(&p->thread_group);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080065 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066}
67
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080068/*
69 * This function expects the tasklist_lock write-locked.
70 */
71static void __exit_signal(struct task_struct *tsk)
72{
73 struct signal_struct *sig = tsk->signal;
74 struct sighand_struct *sighand;
75
76 BUG_ON(!sig);
77 BUG_ON(!atomic_read(&sig->count));
78
79 rcu_read_lock();
80 sighand = rcu_dereference(tsk->sighand);
81 spin_lock(&sighand->siglock);
82
83 posix_cpu_timers_exit(tsk);
84 if (atomic_dec_and_test(&sig->count))
85 posix_cpu_timers_exit_group(tsk);
86 else {
87 /*
88 * If there is any task waiting for the group exit
89 * then notify it:
90 */
91 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
92 wake_up_process(sig->group_exit_task);
93 sig->group_exit_task = NULL;
94 }
95 if (tsk == sig->curr_target)
96 sig->curr_target = next_thread(tsk);
97 /*
98 * Accumulate here the counters for all threads but the
99 * group leader as they die, so they can be added into
100 * the process-wide totals when those are taken.
101 * The group leader stays around as a zombie as long
102 * as there are other threads. When it gets reaped,
103 * the exit.c code will add its counts into these totals.
104 * We won't ever get here for the group leader, since it
105 * will have been the last reference on the signal_struct.
106 */
107 sig->utime = cputime_add(sig->utime, tsk->utime);
108 sig->stime = cputime_add(sig->stime, tsk->stime);
109 sig->min_flt += tsk->min_flt;
110 sig->maj_flt += tsk->maj_flt;
111 sig->nvcsw += tsk->nvcsw;
112 sig->nivcsw += tsk->nivcsw;
113 sig->sched_time += tsk->sched_time;
114 sig = NULL; /* Marker for below. */
115 }
116
Oleg Nesterov58767002006-03-28 16:11:20 -0800117 __unhash_process(tsk);
118
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800119 tsk->signal = NULL;
Oleg Nesterova7e53282006-03-28 16:11:27 -0800120 tsk->sighand = NULL;
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800121 spin_unlock(&sighand->siglock);
122 rcu_read_unlock();
123
Oleg Nesterova7e53282006-03-28 16:11:27 -0800124 __cleanup_sighand(sighand);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800125 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
126 flush_sigqueue(&tsk->pending);
127 if (sig) {
128 flush_sigqueue(&sig->shared_pending);
129 __cleanup_signal(sig);
130 }
131}
132
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800133static void delayed_put_task_struct(struct rcu_head *rhp)
134{
135 put_task_struct(container_of(rhp, struct task_struct, rcu));
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138void release_task(struct task_struct * p)
139{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700140 struct task_struct *leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 int zap_leader;
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800142repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 atomic_dec(&p->user->processes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 write_lock_irq(&tasklist_lock);
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800145 ptrace_unlink(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
147 __exit_signal(p);
Oleg Nesterov35f5cad2006-03-28 16:11:19 -0800148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 /*
150 * If we are the last non-leader member of the thread
151 * group, and the leader is zombie, then notify the
152 * group leader's parent process. (if it wants notification.)
153 */
154 zap_leader = 0;
155 leader = p->group_leader;
156 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
157 BUG_ON(leader->exit_signal == -1);
158 do_notify_parent(leader, leader->exit_signal);
159 /*
160 * If we were the last child thread and the leader has
161 * exited already, and the leader's parent ignores SIGCHLD,
162 * then we are the one who should release the leader.
163 *
164 * do_notify_parent() will have marked it self-reaping in
165 * that case.
166 */
167 zap_leader = (leader->exit_signal == -1);
168 }
169
170 sched_exit(p);
171 write_unlock_irq(&tasklist_lock);
Eric W. Biederman48e64842006-06-26 00:25:48 -0700172 proc_flush_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 release_thread(p);
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800174 call_rcu(&p->rcu, delayed_put_task_struct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 p = leader;
177 if (unlikely(zap_leader))
178 goto repeat;
179}
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181/*
182 * This checks not only the pgrp, but falls back on the pid if no
183 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
184 * without this...
185 */
186int session_of_pgrp(int pgrp)
187{
188 struct task_struct *p;
189 int sid = -1;
190
191 read_lock(&tasklist_lock);
192 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
193 if (p->signal->session > 0) {
194 sid = p->signal->session;
195 goto out;
196 }
197 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
198 p = find_task_by_pid(pgrp);
199 if (p)
200 sid = p->signal->session;
201out:
202 read_unlock(&tasklist_lock);
203
204 return sid;
205}
206
207/*
208 * Determine if a process group is "orphaned", according to the POSIX
209 * definition in 2.2.2.52. Orphaned process groups are not to be affected
210 * by terminal-generated stop signals. Newly orphaned process groups are
211 * to receive a SIGHUP and a SIGCONT.
212 *
213 * "I ask you, have you ever known what it is to be an orphan?"
214 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700215static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
217 struct task_struct *p;
218 int ret = 1;
219
220 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
221 if (p == ignored_task
222 || p->exit_state
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700223 || is_init(p->real_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 continue;
225 if (process_group(p->real_parent) != pgrp
226 && p->real_parent->signal->session == p->signal->session) {
227 ret = 0;
228 break;
229 }
230 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
231 return ret; /* (sighing) "Often!" */
232}
233
234int is_orphaned_pgrp(int pgrp)
235{
236 int retval;
237
238 read_lock(&tasklist_lock);
239 retval = will_become_orphaned_pgrp(pgrp, NULL);
240 read_unlock(&tasklist_lock);
241
242 return retval;
243}
244
Arjan van de Ven858119e2006-01-14 13:20:43 -0800245static int has_stopped_jobs(int pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 int retval = 0;
248 struct task_struct *p;
249
250 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
251 if (p->state != TASK_STOPPED)
252 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 retval = 1;
254 break;
255 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
256 return retval;
257}
258
259/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700260 * reparent_to_init - Reparent the calling kernel thread to the init task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * If a kernel thread is launched as a result of a system call, or if
263 * it ever exits, it should generally reparent itself to init so that
264 * it is correctly cleaned up on exit.
265 *
266 * The various task state such as scheduling policy and priority may have
267 * been inherited from a user process, so we reset them to sane values here.
268 *
269 * NOTE that reparent_to_init() gives the caller full capabilities.
270 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800271static void reparent_to_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 write_lock_irq(&tasklist_lock);
274
275 ptrace_unlink(current);
276 /* Reparent to init */
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800277 remove_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 current->parent = child_reaper;
279 current->real_parent = child_reaper;
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800280 add_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282 /* Set the exit signal to SIGCHLD so we signal init on exit */
283 current->exit_signal = SIGCHLD;
284
Oleg Nesterov1c573af2006-09-29 02:00:51 -0700285 if (!has_rt_policy(current) && (task_nice(current) < 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 set_user_nice(current, 0);
287 /* cpus_allowed? */
288 /* rt_priority? */
289 /* signals? */
290 security_task_reparent_to_init(current);
291 memcpy(current->signal->rlim, init_task.signal->rlim,
292 sizeof(current->signal->rlim));
293 atomic_inc(&(INIT_USER->__count));
294 write_unlock_irq(&tasklist_lock);
295 switch_uid(INIT_USER);
296}
297
298void __set_special_pids(pid_t session, pid_t pgrp)
299{
Oren Laadane19f2472006-01-08 01:03:58 -0800300 struct task_struct *curr = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
302 if (curr->signal->session != session) {
303 detach_pid(curr, PIDTYPE_SID);
304 curr->signal->session = session;
305 attach_pid(curr, PIDTYPE_SID, session);
306 }
307 if (process_group(curr) != pgrp) {
308 detach_pid(curr, PIDTYPE_PGID);
309 curr->signal->pgrp = pgrp;
310 attach_pid(curr, PIDTYPE_PGID, pgrp);
311 }
312}
313
314void set_special_pids(pid_t session, pid_t pgrp)
315{
316 write_lock_irq(&tasklist_lock);
317 __set_special_pids(session, pgrp);
318 write_unlock_irq(&tasklist_lock);
319}
320
321/*
322 * Let kernel threads use this to say that they
323 * allow a certain signal (since daemonize() will
324 * have disabled all of them by default).
325 */
326int allow_signal(int sig)
327{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700328 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return -EINVAL;
330
331 spin_lock_irq(&current->sighand->siglock);
332 sigdelset(&current->blocked, sig);
333 if (!current->mm) {
334 /* Kernel threads handle their own signals.
335 Let the signal code know it'll be handled, so
336 that they don't get converted to SIGKILL or
337 just silently dropped */
338 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
339 }
340 recalc_sigpending();
341 spin_unlock_irq(&current->sighand->siglock);
342 return 0;
343}
344
345EXPORT_SYMBOL(allow_signal);
346
347int disallow_signal(int sig)
348{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700349 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return -EINVAL;
351
352 spin_lock_irq(&current->sighand->siglock);
353 sigaddset(&current->blocked, sig);
354 recalc_sigpending();
355 spin_unlock_irq(&current->sighand->siglock);
356 return 0;
357}
358
359EXPORT_SYMBOL(disallow_signal);
360
361/*
362 * Put all the gunge required to become a kernel thread without
363 * attached user resources in one place where it belongs.
364 */
365
366void daemonize(const char *name, ...)
367{
368 va_list args;
369 struct fs_struct *fs;
370 sigset_t blocked;
371
372 va_start(args, name);
373 vsnprintf(current->comm, sizeof(current->comm), name, args);
374 va_end(args);
375
376 /*
377 * If we were started as result of loading a module, close all of the
378 * user space pages. We don't need them, and if we didn't close them
379 * they would be locked into memory.
380 */
381 exit_mm(current);
382
383 set_special_pids(1, 1);
Ingo Molnar70522e12006-03-23 03:00:31 -0800384 mutex_lock(&tty_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 current->signal->tty = NULL;
Ingo Molnar70522e12006-03-23 03:00:31 -0800386 mutex_unlock(&tty_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388 /* Block and flush all signals */
389 sigfillset(&blocked);
390 sigprocmask(SIG_BLOCK, &blocked, NULL);
391 flush_signals(current);
392
393 /* Become as one with the init task */
394
395 exit_fs(current); /* current->fs->count--; */
396 fs = init_task.fs;
397 current->fs = fs;
398 atomic_inc(&fs->count);
Björn Steinbrink59148112006-02-18 18:12:43 +0100399 exit_namespace(current);
400 current->namespace = init_task.namespace;
401 get_namespace(current->namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 exit_files(current);
403 current->files = init_task.files;
404 atomic_inc(&current->files->count);
405
406 reparent_to_init();
407}
408
409EXPORT_SYMBOL(daemonize);
410
Arjan van de Ven858119e2006-01-14 13:20:43 -0800411static void close_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
413 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700414 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700417
418 /*
419 * It is safe to dereference the fd table without RCU or
420 * ->file_lock because this is the last reference to the
421 * files structure.
422 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700423 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 for (;;) {
425 unsigned long set;
426 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700427 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700429 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 while (set) {
431 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700432 struct file * file = xchg(&fdt->fd[i], NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (file)
434 filp_close(file, files);
435 }
436 i++;
437 set >>= 1;
438 }
439 }
440}
441
442struct files_struct *get_files_struct(struct task_struct *task)
443{
444 struct files_struct *files;
445
446 task_lock(task);
447 files = task->files;
448 if (files)
449 atomic_inc(&files->count);
450 task_unlock(task);
451
452 return files;
453}
454
455void fastcall put_files_struct(struct files_struct *files)
456{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700457 struct fdtable *fdt;
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (atomic_dec_and_test(&files->count)) {
460 close_files(files);
461 /*
462 * Free the fd and fdset arrays if we expanded them.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700463 * If the fdtable was embedded, pass files for freeing
464 * at the end of the RCU grace period. Otherwise,
465 * you can free files immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700467 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700468 if (fdt == &files->fdtab)
469 fdt->free_files = files;
470 else
471 kmem_cache_free(files_cachep, files);
472 free_fdtable(fdt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474}
475
476EXPORT_SYMBOL(put_files_struct);
477
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700478void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
479{
480 struct files_struct *old;
481
482 old = tsk->files;
483 task_lock(tsk);
484 tsk->files = files;
485 task_unlock(tsk);
486 put_files_struct(old);
487}
488EXPORT_SYMBOL(reset_files_struct);
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490static inline void __exit_files(struct task_struct *tsk)
491{
492 struct files_struct * files = tsk->files;
493
494 if (files) {
495 task_lock(tsk);
496 tsk->files = NULL;
497 task_unlock(tsk);
498 put_files_struct(files);
499 }
500}
501
502void exit_files(struct task_struct *tsk)
503{
504 __exit_files(tsk);
505}
506
507static inline void __put_fs_struct(struct fs_struct *fs)
508{
509 /* No need to hold fs->lock if we are killing it */
510 if (atomic_dec_and_test(&fs->count)) {
511 dput(fs->root);
512 mntput(fs->rootmnt);
513 dput(fs->pwd);
514 mntput(fs->pwdmnt);
515 if (fs->altroot) {
516 dput(fs->altroot);
517 mntput(fs->altrootmnt);
518 }
519 kmem_cache_free(fs_cachep, fs);
520 }
521}
522
523void put_fs_struct(struct fs_struct *fs)
524{
525 __put_fs_struct(fs);
526}
527
528static inline void __exit_fs(struct task_struct *tsk)
529{
530 struct fs_struct * fs = tsk->fs;
531
532 if (fs) {
533 task_lock(tsk);
534 tsk->fs = NULL;
535 task_unlock(tsk);
536 __put_fs_struct(fs);
537 }
538}
539
540void exit_fs(struct task_struct *tsk)
541{
542 __exit_fs(tsk);
543}
544
545EXPORT_SYMBOL_GPL(exit_fs);
546
547/*
548 * Turn us into a lazy TLB process if we
549 * aren't already..
550 */
Adrian Bunk408b6642005-05-01 08:59:29 -0700551static void exit_mm(struct task_struct * tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
553 struct mm_struct *mm = tsk->mm;
554
555 mm_release(tsk, mm);
556 if (!mm)
557 return;
558 /*
559 * Serialize with any possible pending coredump.
560 * We must hold mmap_sem around checking core_waiters
561 * and clearing tsk->mm. The core-inducing thread
562 * will increment core_waiters for each thread in the
563 * group with ->mm != NULL.
564 */
565 down_read(&mm->mmap_sem);
566 if (mm->core_waiters) {
567 up_read(&mm->mmap_sem);
568 down_write(&mm->mmap_sem);
569 if (!--mm->core_waiters)
570 complete(mm->core_startup_done);
571 up_write(&mm->mmap_sem);
572
573 wait_for_completion(&mm->core_done);
574 down_read(&mm->mmap_sem);
575 }
576 atomic_inc(&mm->mm_count);
Eric Sesterhenn125e1872006-06-23 02:06:06 -0700577 BUG_ON(mm != tsk->active_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 /* more a memory barrier than a real lock */
579 task_lock(tsk);
580 tsk->mm = NULL;
581 up_read(&mm->mmap_sem);
582 enter_lazy_tlb(mm, current);
583 task_unlock(tsk);
584 mmput(mm);
585}
586
Ingo Molnar36c8b582006-07-03 00:25:41 -0700587static inline void
588choose_new_parent(struct task_struct *p, struct task_struct *reaper)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
590 /*
591 * Make sure we're not reparenting to ourselves and that
592 * the parent is not a zombie.
593 */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800594 BUG_ON(p == reaper || reaper->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 p->real_parent = reaper;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
Ingo Molnar36c8b582006-07-03 00:25:41 -0700598static void
599reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 /* We don't want people slaying init. */
602 if (p->exit_signal != -1)
603 p->exit_signal = SIGCHLD;
604
605 if (p->pdeath_signal)
606 /* We already hold the tasklist_lock here. */
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800607 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 /* Move the child from its dying parent to the new one. */
610 if (unlikely(traced)) {
611 /* Preserve ptrace links if someone else is tracing this child. */
612 list_del_init(&p->ptrace_list);
613 if (p->parent != p->real_parent)
614 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
615 } else {
616 /* If this child is being traced, then we're the one tracing it
617 * anyway, so let go of it.
618 */
619 p->ptrace = 0;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800620 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 p->parent = p->real_parent;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800622 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 /* If we'd notified the old parent about this child's death,
625 * also notify the new parent.
626 */
627 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
628 thread_group_empty(p))
629 do_notify_parent(p, p->exit_signal);
630 else if (p->state == TASK_TRACED) {
631 /*
632 * If it was at a trace stop, turn it into
633 * a normal stop since it's no longer being
634 * traced.
635 */
636 ptrace_untrace(p);
637 }
638 }
639
640 /*
641 * process group orphan check
642 * Case ii: Our child is in a different pgrp
643 * than we are, and it was the only connection
644 * outside, so the child pgrp is now orphaned.
645 */
646 if ((process_group(p) != process_group(father)) &&
647 (p->signal->session == father->signal->session)) {
648 int pgrp = process_group(p);
649
650 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800651 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
652 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654 }
655}
656
657/*
658 * When we die, we re-parent all our children.
659 * Try to give them to another thread in our thread
660 * group, and if no such member exists, give it to
661 * the global child reaper process (ie "init")
662 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700663static void
664forget_original_parent(struct task_struct *father, struct list_head *to_release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 struct task_struct *p, *reaper = father;
667 struct list_head *_p, *_n;
668
669 do {
670 reaper = next_thread(reaper);
671 if (reaper == father) {
672 reaper = child_reaper;
673 break;
674 }
675 } while (reaper->exit_state);
676
677 /*
678 * There are only two places where our children can be:
679 *
680 * - in our child list
681 * - in our ptraced child list
682 *
683 * Search them and reparent children.
684 */
685 list_for_each_safe(_p, _n, &father->children) {
686 int ptrace;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700687 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
689 ptrace = p->ptrace;
690
691 /* if father isn't the real parent, then ptrace must be enabled */
692 BUG_ON(father != p->real_parent && !ptrace);
693
694 if (father == p->real_parent) {
695 /* reparent with a reaper, real father it's us */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800696 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 reparent_thread(p, father, 0);
698 } else {
699 /* reparent ptraced task to its real parent */
700 __ptrace_unlink (p);
701 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
702 thread_group_empty(p))
703 do_notify_parent(p, p->exit_signal);
704 }
705
706 /*
707 * if the ptraced child is a zombie with exit_signal == -1
708 * we must collect it before we exit, or it will remain
709 * zombie forever since we prevented it from self-reap itself
710 * while it was being traced by us, to be able to see it in wait4.
711 */
712 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
713 list_add(&p->ptrace_list, to_release);
714 }
715 list_for_each_safe(_p, _n, &father->ptrace_children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -0700716 p = list_entry(_p, struct task_struct, ptrace_list);
Oleg Nesterovd799f032006-03-28 16:11:04 -0800717 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 reparent_thread(p, father, 1);
719 }
720}
721
722/*
723 * Send signals to all our closest relatives so that they know
724 * to properly mourn us..
725 */
726static void exit_notify(struct task_struct *tsk)
727{
728 int state;
729 struct task_struct *t;
730 struct list_head ptrace_dead, *_p, *_n;
731
732 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
733 && !thread_group_empty(tsk)) {
734 /*
735 * This occurs when there was a race between our exit
736 * syscall and a group signal choosing us as the one to
737 * wake up. It could be that we are the only thread
738 * alerted to check for pending signals, but another thread
739 * should be woken now to take the signal since we will not.
740 * Now we'll wake all the threads in the group just to make
741 * sure someone gets all the pending signals.
742 */
743 read_lock(&tasklist_lock);
744 spin_lock_irq(&tsk->sighand->siglock);
745 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
746 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
747 recalc_sigpending_tsk(t);
748 if (signal_pending(t))
749 signal_wake_up(t, 0);
750 }
751 spin_unlock_irq(&tsk->sighand->siglock);
752 read_unlock(&tasklist_lock);
753 }
754
755 write_lock_irq(&tasklist_lock);
756
757 /*
758 * This does two things:
759 *
760 * A. Make init inherit all the child processes
761 * B. Check to see if any process groups have become orphaned
762 * as a result of our exiting, and if they have any stopped
763 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
764 */
765
766 INIT_LIST_HEAD(&ptrace_dead);
767 forget_original_parent(tsk, &ptrace_dead);
768 BUG_ON(!list_empty(&tsk->children));
769 BUG_ON(!list_empty(&tsk->ptrace_children));
770
771 /*
772 * Check to see if any process groups have become orphaned
773 * as a result of our exiting, and if they have any stopped
774 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
775 *
776 * Case i: Our father is in a different pgrp than we are
777 * and we were the only connection outside, so our pgrp
778 * is about to become orphaned.
779 */
780
781 t = tsk->real_parent;
782
783 if ((process_group(t) != process_group(tsk)) &&
784 (t->signal->session == tsk->signal->session) &&
785 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
786 has_stopped_jobs(process_group(tsk))) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800787 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
788 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
790
791 /* Let father know we died
792 *
793 * Thread signals are configurable, but you aren't going to use
794 * that to send signals to arbitary processes.
795 * That stops right now.
796 *
797 * If the parent exec id doesn't match the exec id we saved
798 * when we started then we know the parent has changed security
799 * domain.
800 *
801 * If our self_exec id doesn't match our parent_exec_id then
802 * we have changed execution domain as these two values started
803 * the same after a fork.
804 *
805 */
806
807 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
808 ( tsk->parent_exec_id != t->self_exec_id ||
809 tsk->self_exec_id != tsk->parent_exec_id)
810 && !capable(CAP_KILL))
811 tsk->exit_signal = SIGCHLD;
812
813
814 /* If something other than our normal parent is ptracing us, then
815 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
816 * only has special meaning to our real parent.
817 */
818 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
819 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
820 do_notify_parent(tsk, signal);
821 } else if (tsk->ptrace) {
822 do_notify_parent(tsk, SIGCHLD);
823 }
824
825 state = EXIT_ZOMBIE;
826 if (tsk->exit_signal == -1 &&
827 (likely(tsk->ptrace == 0) ||
828 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
829 state = EXIT_DEAD;
830 tsk->exit_state = state;
831
832 write_unlock_irq(&tasklist_lock);
833
834 list_for_each_safe(_p, _n, &ptrace_dead) {
835 list_del_init(_p);
Ingo Molnar36c8b582006-07-03 00:25:41 -0700836 t = list_entry(_p, struct task_struct, ptrace_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 release_task(t);
838 }
839
840 /* If the process is dead, release it - nobody will wait for it */
841 if (state == EXIT_DEAD)
842 release_task(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
845fastcall NORET_TYPE void do_exit(long code)
846{
847 struct task_struct *tsk = current;
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700848 struct taskstats *tidstats;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 int group_dead;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700850 unsigned int mycpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 profile_task_exit(tsk);
853
Jens Axboe22e2c502005-06-27 10:55:12 +0200854 WARN_ON(atomic_read(&tsk->fs_excl));
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (unlikely(in_interrupt()))
857 panic("Aiee, killing interrupt handler!");
858 if (unlikely(!tsk->pid))
859 panic("Attempted to kill the idle task!");
Eric W. Biedermanfef23e72006-03-28 16:10:58 -0800860 if (unlikely(tsk == child_reaper))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 panic("Attempted to kill init!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
864 current->ptrace_message = code;
865 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
866 }
867
Alexander Nybergdf164db2005-06-23 00:09:13 -0700868 /*
869 * We're taking recursive faults here in do_exit. Safest is to just
870 * leave this task alone and wait for reboot.
871 */
872 if (unlikely(tsk->flags & PF_EXITING)) {
873 printk(KERN_ALERT
874 "Fixing recursive fault but reboot is needed!\n");
Al Viroafc847b2006-02-28 12:51:55 -0500875 if (tsk->io_context)
876 exit_io_context();
Alexander Nybergdf164db2005-06-23 00:09:13 -0700877 set_current_state(TASK_UNINTERRUPTIBLE);
878 schedule();
879 }
880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 tsk->flags |= PF_EXITING;
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 if (unlikely(in_atomic()))
884 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
885 current->comm, current->pid,
886 preempt_count());
887
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700888 taskstats_exit_alloc(&tidstats, &mycpu);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700889
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 acct_update_integrals(tsk);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700891 if (tsk->mm) {
892 update_hiwater_rss(tsk->mm);
893 update_hiwater_vm(tsk->mm);
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 group_dead = atomic_dec_and_test(&tsk->signal->live);
Andrew Mortonc3068952005-08-04 16:49:32 -0700896 if (group_dead) {
Thomas Gleixner2ff678b2006-01-09 20:52:34 -0800897 hrtimer_cancel(&tsk->signal->real_timer);
Roland McGrath25f407f2005-10-21 15:03:29 -0700898 exit_itimers(tsk->signal);
Andrew Mortonc3068952005-08-04 16:49:32 -0700899 }
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700900 acct_collect(code, group_dead);
Ingo Molnar0771dfe2006-03-27 01:16:22 -0800901 if (unlikely(tsk->robust_list))
902 exit_robust_list(tsk);
Anton Blanchard2aa92582006-06-25 05:48:44 -0700903#if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
Ingo Molnar34f192c2006-03-27 01:16:24 -0800904 if (unlikely(tsk->compat_robust_list))
905 compat_exit_robust_list(tsk);
906#endif
Al Virofa84cb92006-03-29 20:30:19 -0500907 if (unlikely(tsk->audit_context))
908 audit_free(tsk);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700909 taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700910 taskstats_exit_free(tidstats);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 exit_mm(tsk);
913
KaiGai Kohei0e464812006-06-25 05:49:24 -0700914 if (group_dead)
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700915 acct_process();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 exit_sem(tsk);
917 __exit_files(tsk);
918 __exit_fs(tsk);
919 exit_namespace(tsk);
920 exit_thread();
921 cpuset_exit(tsk);
922 exit_keys(tsk);
923
924 if (group_dead && tsk->signal->leader)
925 disassociate_ctty(1);
926
Al Viroa1261f52005-11-13 16:06:55 -0800927 module_put(task_thread_info(tsk)->exec_domain->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (tsk->binfmt)
929 module_put(tsk->binfmt->module);
930
931 tsk->exit_code = code;
Matt Helsley9f460802005-11-07 00:59:16 -0800932 proc_exit_connector(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 exit_notify(tsk);
934#ifdef CONFIG_NUMA
935 mpol_free(tsk->mempolicy);
936 tsk->mempolicy = NULL;
937#endif
Ingo Molnarde5097c2006-01-09 15:59:21 -0800938 /*
Ingo Molnarc87e2832006-06-27 02:54:58 -0700939 * This must happen late, after the PID is not
940 * hashed anymore:
941 */
942 if (unlikely(!list_empty(&tsk->pi_state_list)))
943 exit_pi_state_list(tsk);
944 if (unlikely(current->pi_state_cache))
945 kfree(current->pi_state_cache);
946 /*
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700947 * Make sure we are holding no locks:
Ingo Molnarde5097c2006-01-09 15:59:21 -0800948 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700949 debug_check_no_locks_held(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Al Viroafc847b2006-02-28 12:51:55 -0500951 if (tsk->io_context)
952 exit_io_context();
953
Jens Axboeb92ce552006-04-11 13:52:07 +0200954 if (tsk->splice_pipe)
955 __free_pipe_info(tsk->splice_pipe);
956
Coywolf Qi Hunt74072512005-10-30 15:02:47 -0800957 preempt_disable();
Oleg Nesterov55a101f2006-09-29 02:01:10 -0700958 /* causes final put_task_struct in finish_task_switch(). */
Oleg Nesterovc394cc92006-09-29 02:01:11 -0700959 tsk->state = TASK_DEAD;
Coywolf Qi Hunt74072512005-10-30 15:02:47 -0800960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 schedule();
962 BUG();
963 /* Avoid "noreturn function does return". */
Alan Cox54306cf2006-09-29 02:00:42 -0700964 for (;;)
965 cpu_relax(); /* For when BUG is null */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
Russ Anderson012914d2005-04-23 00:08:00 -0700968EXPORT_SYMBOL_GPL(do_exit);
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970NORET_TYPE void complete_and_exit(struct completion *comp, long code)
971{
972 if (comp)
973 complete(comp);
Oleg Nesterov55a101f2006-09-29 02:01:10 -0700974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 do_exit(code);
976}
977
978EXPORT_SYMBOL(complete_and_exit);
979
980asmlinkage long sys_exit(int error_code)
981{
982 do_exit((error_code&0xff)<<8);
983}
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985/*
986 * Take down every thread in the group. This is called by fatal signals
987 * as well as by sys_exit_group (below).
988 */
989NORET_TYPE void
990do_group_exit(int exit_code)
991{
992 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
993
994 if (current->signal->flags & SIGNAL_GROUP_EXIT)
995 exit_code = current->signal->group_exit_code;
996 else if (!thread_group_empty(current)) {
997 struct signal_struct *const sig = current->signal;
998 struct sighand_struct *const sighand = current->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 spin_lock_irq(&sighand->siglock);
1000 if (sig->flags & SIGNAL_GROUP_EXIT)
1001 /* Another thread got here before we took the lock. */
1002 exit_code = sig->group_exit_code;
1003 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 sig->group_exit_code = exit_code;
1005 zap_other_threads(current);
1006 }
1007 spin_unlock_irq(&sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009
1010 do_exit(exit_code);
1011 /* NOTREACHED */
1012}
1013
1014/*
1015 * this kills every thread in the thread group. Note that any externally
1016 * wait4()-ing process will get the correct exit code - even if this
1017 * thread is not the thread group leader.
1018 */
1019asmlinkage void sys_exit_group(int error_code)
1020{
1021 do_group_exit((error_code & 0xff) << 8);
1022}
1023
Ingo Molnar36c8b582006-07-03 00:25:41 -07001024static int eligible_child(pid_t pid, int options, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 if (pid > 0) {
1027 if (p->pid != pid)
1028 return 0;
1029 } else if (!pid) {
1030 if (process_group(p) != process_group(current))
1031 return 0;
1032 } else if (pid != -1) {
1033 if (process_group(p) != -pid)
1034 return 0;
1035 }
1036
1037 /*
1038 * Do not consider detached threads that are
1039 * not ptraced:
1040 */
1041 if (p->exit_signal == -1 && !p->ptrace)
1042 return 0;
1043
1044 /* Wait for all children (clone and not) if __WALL is set;
1045 * otherwise, wait for clone children *only* if __WCLONE is
1046 * set; otherwise, wait for non-clone children *only*. (Note:
1047 * A "clone" child here is one that reports to its parent
1048 * using a signal other than SIGCHLD.) */
1049 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1050 && !(options & __WALL))
1051 return 0;
1052 /*
1053 * Do not consider thread group leaders that are
1054 * in a non-empty thread group:
1055 */
Oleg Nesterov3b6362b2006-09-02 21:22:16 +04001056 if (delay_group_leader(p))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return 2;
1058
1059 if (security_task_wait(p))
1060 return 0;
1061
1062 return 1;
1063}
1064
Ingo Molnar36c8b582006-07-03 00:25:41 -07001065static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 int why, int status,
1067 struct siginfo __user *infop,
1068 struct rusage __user *rusagep)
1069{
1070 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07001071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 put_task_struct(p);
1073 if (!retval)
1074 retval = put_user(SIGCHLD, &infop->si_signo);
1075 if (!retval)
1076 retval = put_user(0, &infop->si_errno);
1077 if (!retval)
1078 retval = put_user((short)why, &infop->si_code);
1079 if (!retval)
1080 retval = put_user(pid, &infop->si_pid);
1081 if (!retval)
1082 retval = put_user(uid, &infop->si_uid);
1083 if (!retval)
1084 retval = put_user(status, &infop->si_status);
1085 if (!retval)
1086 retval = pid;
1087 return retval;
1088}
1089
1090/*
1091 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1092 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1093 * the lock and this task is uninteresting. If we return nonzero, we have
1094 * released the lock and the system call should return.
1095 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001096static int wait_task_zombie(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 struct siginfo __user *infop,
1098 int __user *stat_addr, struct rusage __user *ru)
1099{
1100 unsigned long state;
1101 int retval;
1102 int status;
1103
1104 if (unlikely(noreap)) {
1105 pid_t pid = p->pid;
1106 uid_t uid = p->uid;
1107 int exit_code = p->exit_code;
1108 int why, status;
1109
1110 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1111 return 0;
1112 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1113 return 0;
1114 get_task_struct(p);
1115 read_unlock(&tasklist_lock);
1116 if ((exit_code & 0x7f) == 0) {
1117 why = CLD_EXITED;
1118 status = exit_code >> 8;
1119 } else {
1120 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1121 status = exit_code & 0x7f;
1122 }
1123 return wait_noreap_copyout(p, pid, uid, why,
1124 status, infop, ru);
1125 }
1126
1127 /*
1128 * Try to move the task's state to DEAD
1129 * only one thread is allowed to do this:
1130 */
1131 state = xchg(&p->exit_state, EXIT_DEAD);
1132 if (state != EXIT_ZOMBIE) {
1133 BUG_ON(state != EXIT_DEAD);
1134 return 0;
1135 }
1136 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1137 /*
1138 * This can only happen in a race with a ptraced thread
1139 * dying on another processor.
1140 */
1141 return 0;
1142 }
1143
1144 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
Jesper Juhl3795e162006-01-09 20:54:39 -08001145 struct signal_struct *psig;
1146 struct signal_struct *sig;
1147
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 /*
1149 * The resource counters for the group leader are in its
1150 * own task_struct. Those for dead threads in the group
1151 * are in its signal_struct, as are those for the child
1152 * processes it has previously reaped. All these
1153 * accumulate in the parent's signal_struct c* fields.
1154 *
1155 * We don't bother to take a lock here to protect these
1156 * p->signal fields, because they are only touched by
1157 * __exit_signal, which runs with tasklist_lock
1158 * write-locked anyway, and so is excluded here. We do
1159 * need to protect the access to p->parent->signal fields,
1160 * as other threads in the parent group can be right
1161 * here reaping other children at the same time.
1162 */
1163 spin_lock_irq(&p->parent->sighand->siglock);
Jesper Juhl3795e162006-01-09 20:54:39 -08001164 psig = p->parent->signal;
1165 sig = p->signal;
1166 psig->cutime =
1167 cputime_add(psig->cutime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 cputime_add(p->utime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001169 cputime_add(sig->utime,
1170 sig->cutime)));
1171 psig->cstime =
1172 cputime_add(psig->cstime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 cputime_add(p->stime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001174 cputime_add(sig->stime,
1175 sig->cstime)));
1176 psig->cmin_flt +=
1177 p->min_flt + sig->min_flt + sig->cmin_flt;
1178 psig->cmaj_flt +=
1179 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1180 psig->cnvcsw +=
1181 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1182 psig->cnivcsw +=
1183 p->nivcsw + sig->nivcsw + sig->cnivcsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 spin_unlock_irq(&p->parent->sighand->siglock);
1185 }
1186
1187 /*
1188 * Now we are sure this task is interesting, and no other
1189 * thread can reap it because we set its state to EXIT_DEAD.
1190 */
1191 read_unlock(&tasklist_lock);
1192
1193 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1194 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1195 ? p->signal->group_exit_code : p->exit_code;
1196 if (!retval && stat_addr)
1197 retval = put_user(status, stat_addr);
1198 if (!retval && infop)
1199 retval = put_user(SIGCHLD, &infop->si_signo);
1200 if (!retval && infop)
1201 retval = put_user(0, &infop->si_errno);
1202 if (!retval && infop) {
1203 int why;
1204
1205 if ((status & 0x7f) == 0) {
1206 why = CLD_EXITED;
1207 status >>= 8;
1208 } else {
1209 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1210 status &= 0x7f;
1211 }
1212 retval = put_user((short)why, &infop->si_code);
1213 if (!retval)
1214 retval = put_user(status, &infop->si_status);
1215 }
1216 if (!retval && infop)
1217 retval = put_user(p->pid, &infop->si_pid);
1218 if (!retval && infop)
1219 retval = put_user(p->uid, &infop->si_uid);
1220 if (retval) {
1221 // TODO: is this safe?
1222 p->exit_state = EXIT_ZOMBIE;
1223 return retval;
1224 }
1225 retval = p->pid;
1226 if (p->real_parent != p->parent) {
1227 write_lock_irq(&tasklist_lock);
1228 /* Double-check with lock held. */
1229 if (p->real_parent != p->parent) {
1230 __ptrace_unlink(p);
1231 // TODO: is this safe?
1232 p->exit_state = EXIT_ZOMBIE;
1233 /*
1234 * If this is not a detached task, notify the parent.
1235 * If it's still not detached after that, don't release
1236 * it now.
1237 */
1238 if (p->exit_signal != -1) {
1239 do_notify_parent(p, p->exit_signal);
1240 if (p->exit_signal != -1)
1241 p = NULL;
1242 }
1243 }
1244 write_unlock_irq(&tasklist_lock);
1245 }
1246 if (p != NULL)
1247 release_task(p);
1248 BUG_ON(!retval);
1249 return retval;
1250}
1251
1252/*
1253 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1254 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1255 * the lock and this task is uninteresting. If we return nonzero, we have
1256 * released the lock and the system call should return.
1257 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001258static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1259 int noreap, struct siginfo __user *infop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 int __user *stat_addr, struct rusage __user *ru)
1261{
1262 int retval, exit_code;
1263
1264 if (!p->exit_code)
1265 return 0;
1266 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1267 p->signal && p->signal->group_stop_count > 0)
1268 /*
1269 * A group stop is in progress and this is the group leader.
1270 * We won't report until all threads have stopped.
1271 */
1272 return 0;
1273
1274 /*
1275 * Now we are pretty sure this task is interesting.
1276 * Make sure it doesn't get reaped out from under us while we
1277 * give up the lock and then examine it below. We don't want to
1278 * keep holding onto the tasklist_lock while we call getrusage and
1279 * possibly take page faults for user memory.
1280 */
1281 get_task_struct(p);
1282 read_unlock(&tasklist_lock);
1283
1284 if (unlikely(noreap)) {
1285 pid_t pid = p->pid;
1286 uid_t uid = p->uid;
1287 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1288
1289 exit_code = p->exit_code;
1290 if (unlikely(!exit_code) ||
Linus Torvalds14bf01b2005-10-01 11:04:18 -07001291 unlikely(p->state & TASK_TRACED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 goto bail_ref;
1293 return wait_noreap_copyout(p, pid, uid,
1294 why, (exit_code << 8) | 0x7f,
1295 infop, ru);
1296 }
1297
1298 write_lock_irq(&tasklist_lock);
1299
1300 /*
1301 * This uses xchg to be atomic with the thread resuming and setting
1302 * it. It must also be done with the write lock held to prevent a
1303 * race with the EXIT_ZOMBIE case.
1304 */
1305 exit_code = xchg(&p->exit_code, 0);
1306 if (unlikely(p->exit_state)) {
1307 /*
1308 * The task resumed and then died. Let the next iteration
1309 * catch it in EXIT_ZOMBIE. Note that exit_code might
1310 * already be zero here if it resumed and did _exit(0).
1311 * The task itself is dead and won't touch exit_code again;
1312 * other processors in this function are locked out.
1313 */
1314 p->exit_code = exit_code;
1315 exit_code = 0;
1316 }
1317 if (unlikely(exit_code == 0)) {
1318 /*
1319 * Another thread in this function got to it first, or it
1320 * resumed, or it resumed and then died.
1321 */
1322 write_unlock_irq(&tasklist_lock);
1323bail_ref:
1324 put_task_struct(p);
1325 /*
1326 * We are returning to the wait loop without having successfully
1327 * removed the process and having released the lock. We cannot
1328 * continue, since the "p" task pointer is potentially stale.
1329 *
1330 * Return -EAGAIN, and do_wait() will restart the loop from the
1331 * beginning. Do _not_ re-acquire the lock.
1332 */
1333 return -EAGAIN;
1334 }
1335
1336 /* move to end of parent's list to avoid starvation */
1337 remove_parent(p);
Oleg Nesterov8fafabd2006-03-28 16:11:05 -08001338 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 write_unlock_irq(&tasklist_lock);
1341
1342 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1343 if (!retval && stat_addr)
1344 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1345 if (!retval && infop)
1346 retval = put_user(SIGCHLD, &infop->si_signo);
1347 if (!retval && infop)
1348 retval = put_user(0, &infop->si_errno);
1349 if (!retval && infop)
1350 retval = put_user((short)((p->ptrace & PT_PTRACED)
1351 ? CLD_TRAPPED : CLD_STOPPED),
1352 &infop->si_code);
1353 if (!retval && infop)
1354 retval = put_user(exit_code, &infop->si_status);
1355 if (!retval && infop)
1356 retval = put_user(p->pid, &infop->si_pid);
1357 if (!retval && infop)
1358 retval = put_user(p->uid, &infop->si_uid);
1359 if (!retval)
1360 retval = p->pid;
1361 put_task_struct(p);
1362
1363 BUG_ON(!retval);
1364 return retval;
1365}
1366
1367/*
1368 * Handle do_wait work for one task in a live, non-stopped state.
1369 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1370 * the lock and this task is uninteresting. If we return nonzero, we have
1371 * released the lock and the system call should return.
1372 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001373static int wait_task_continued(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 struct siginfo __user *infop,
1375 int __user *stat_addr, struct rusage __user *ru)
1376{
1377 int retval;
1378 pid_t pid;
1379 uid_t uid;
1380
1381 if (unlikely(!p->signal))
1382 return 0;
1383
1384 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1385 return 0;
1386
1387 spin_lock_irq(&p->sighand->siglock);
1388 /* Re-check with the lock held. */
1389 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1390 spin_unlock_irq(&p->sighand->siglock);
1391 return 0;
1392 }
1393 if (!noreap)
1394 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1395 spin_unlock_irq(&p->sighand->siglock);
1396
1397 pid = p->pid;
1398 uid = p->uid;
1399 get_task_struct(p);
1400 read_unlock(&tasklist_lock);
1401
1402 if (!infop) {
1403 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1404 put_task_struct(p);
1405 if (!retval && stat_addr)
1406 retval = put_user(0xffff, stat_addr);
1407 if (!retval)
1408 retval = p->pid;
1409 } else {
1410 retval = wait_noreap_copyout(p, pid, uid,
1411 CLD_CONTINUED, SIGCONT,
1412 infop, ru);
1413 BUG_ON(retval == 0);
1414 }
1415
1416 return retval;
1417}
1418
1419
1420static inline int my_ptrace_child(struct task_struct *p)
1421{
1422 if (!(p->ptrace & PT_PTRACED))
1423 return 0;
1424 if (!(p->ptrace & PT_ATTACHED))
1425 return 1;
1426 /*
1427 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1428 * we are the attacher. If we are the real parent, this is a race
1429 * inside ptrace_attach. It is waiting for the tasklist_lock,
1430 * which we have to switch the parent links, but has already set
1431 * the flags in p->ptrace.
1432 */
1433 return (p->parent != p->real_parent);
1434}
1435
1436static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1437 int __user *stat_addr, struct rusage __user *ru)
1438{
1439 DECLARE_WAITQUEUE(wait, current);
1440 struct task_struct *tsk;
1441 int flag, retval;
1442
1443 add_wait_queue(&current->signal->wait_chldexit,&wait);
1444repeat:
1445 /*
1446 * We will set this flag if we see any child that might later
1447 * match our criteria, even if we are not able to reap it yet.
1448 */
1449 flag = 0;
1450 current->state = TASK_INTERRUPTIBLE;
1451 read_lock(&tasklist_lock);
1452 tsk = current;
1453 do {
1454 struct task_struct *p;
1455 struct list_head *_p;
1456 int ret;
1457
1458 list_for_each(_p,&tsk->children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -07001459 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
1461 ret = eligible_child(pid, options, p);
1462 if (!ret)
1463 continue;
1464
1465 switch (p->state) {
1466 case TASK_TRACED:
Roland McGrath7f2a5252005-10-30 15:02:50 -08001467 /*
1468 * When we hit the race with PTRACE_ATTACH,
1469 * we will not report this child. But the
1470 * race means it has not yet been moved to
1471 * our ptrace_children list, so we need to
1472 * set the flag here to avoid a spurious ECHILD
1473 * when the race happens with the only child.
1474 */
1475 flag = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (!my_ptrace_child(p))
1477 continue;
1478 /*FALLTHROUGH*/
1479 case TASK_STOPPED:
1480 /*
1481 * It's stopped now, so it might later
1482 * continue, exit, or stop again.
1483 */
1484 flag = 1;
1485 if (!(options & WUNTRACED) &&
1486 !my_ptrace_child(p))
1487 continue;
1488 retval = wait_task_stopped(p, ret == 2,
1489 (options & WNOWAIT),
1490 infop,
1491 stat_addr, ru);
1492 if (retval == -EAGAIN)
1493 goto repeat;
1494 if (retval != 0) /* He released the lock. */
1495 goto end;
1496 break;
1497 default:
1498 // case EXIT_DEAD:
1499 if (p->exit_state == EXIT_DEAD)
1500 continue;
1501 // case EXIT_ZOMBIE:
1502 if (p->exit_state == EXIT_ZOMBIE) {
1503 /*
1504 * Eligible but we cannot release
1505 * it yet:
1506 */
1507 if (ret == 2)
1508 goto check_continued;
1509 if (!likely(options & WEXITED))
1510 continue;
1511 retval = wait_task_zombie(
1512 p, (options & WNOWAIT),
1513 infop, stat_addr, ru);
1514 /* He released the lock. */
1515 if (retval != 0)
1516 goto end;
1517 break;
1518 }
1519check_continued:
1520 /*
1521 * It's running now, so it might later
1522 * exit, stop, or stop and then continue.
1523 */
1524 flag = 1;
1525 if (!unlikely(options & WCONTINUED))
1526 continue;
1527 retval = wait_task_continued(
1528 p, (options & WNOWAIT),
1529 infop, stat_addr, ru);
1530 if (retval != 0) /* He released the lock. */
1531 goto end;
1532 break;
1533 }
1534 }
1535 if (!flag) {
1536 list_for_each(_p, &tsk->ptrace_children) {
1537 p = list_entry(_p, struct task_struct,
1538 ptrace_list);
1539 if (!eligible_child(pid, options, p))
1540 continue;
1541 flag = 1;
1542 break;
1543 }
1544 }
1545 if (options & __WNOTHREAD)
1546 break;
1547 tsk = next_thread(tsk);
Eric Sesterhenn125e1872006-06-23 02:06:06 -07001548 BUG_ON(tsk->signal != current->signal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 } while (tsk != current);
1550
1551 read_unlock(&tasklist_lock);
1552 if (flag) {
1553 retval = 0;
1554 if (options & WNOHANG)
1555 goto end;
1556 retval = -ERESTARTSYS;
1557 if (signal_pending(current))
1558 goto end;
1559 schedule();
1560 goto repeat;
1561 }
1562 retval = -ECHILD;
1563end:
1564 current->state = TASK_RUNNING;
1565 remove_wait_queue(&current->signal->wait_chldexit,&wait);
1566 if (infop) {
1567 if (retval > 0)
1568 retval = 0;
1569 else {
1570 /*
1571 * For a WNOHANG return, clear out all the fields
1572 * we would set so the user can easily tell the
1573 * difference.
1574 */
1575 if (!retval)
1576 retval = put_user(0, &infop->si_signo);
1577 if (!retval)
1578 retval = put_user(0, &infop->si_errno);
1579 if (!retval)
1580 retval = put_user(0, &infop->si_code);
1581 if (!retval)
1582 retval = put_user(0, &infop->si_pid);
1583 if (!retval)
1584 retval = put_user(0, &infop->si_uid);
1585 if (!retval)
1586 retval = put_user(0, &infop->si_status);
1587 }
1588 }
1589 return retval;
1590}
1591
1592asmlinkage long sys_waitid(int which, pid_t pid,
1593 struct siginfo __user *infop, int options,
1594 struct rusage __user *ru)
1595{
1596 long ret;
1597
1598 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1599 return -EINVAL;
1600 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1601 return -EINVAL;
1602
1603 switch (which) {
1604 case P_ALL:
1605 pid = -1;
1606 break;
1607 case P_PID:
1608 if (pid <= 0)
1609 return -EINVAL;
1610 break;
1611 case P_PGID:
1612 if (pid <= 0)
1613 return -EINVAL;
1614 pid = -pid;
1615 break;
1616 default:
1617 return -EINVAL;
1618 }
1619
1620 ret = do_wait(pid, options, infop, NULL, ru);
1621
1622 /* avoid REGPARM breakage on x86: */
1623 prevent_tail_call(ret);
1624 return ret;
1625}
1626
1627asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1628 int options, struct rusage __user *ru)
1629{
1630 long ret;
1631
1632 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1633 __WNOTHREAD|__WCLONE|__WALL))
1634 return -EINVAL;
1635 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1636
1637 /* avoid REGPARM breakage on x86: */
1638 prevent_tail_call(ret);
1639 return ret;
1640}
1641
1642#ifdef __ARCH_WANT_SYS_WAITPID
1643
1644/*
1645 * sys_waitpid() remains for compatibility. waitpid() should be
1646 * implemented by calling sys_wait4() from libc.a.
1647 */
1648asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1649{
1650 return sys_wait4(pid, stat_addr, options, NULL);
1651}
1652
1653#endif