blob: 4741376c8decc439c6133007536211667b8b4bb5 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/module.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080011#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080015#include <linux/mnt_namespace.h>
Jens Axboeda9cbc82008-06-30 20:42:08 +020016#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/key.h>
18#include <linux/security.h>
19#include <linux/cpu.h>
20#include <linux/acct.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070021#include <linux/tsacct_kern.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040023#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/binfmts.h>
Serge E. Hallynab516012006-10-02 02:18:06 -070025#include <linux/nsproxy.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080026#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/ptrace.h>
28#include <linux/profile.h>
29#include <linux/mount.h>
30#include <linux/proc_fs.h>
Eric W. Biederman49d769d2007-05-09 02:34:33 -070031#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/mempolicy.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070033#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070034#include <linux/delayacct.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070035#include <linux/freezer.h>
Paul Menageb4f48b62007-10-18 23:39:33 -070036#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070038#include <linux/signal.h>
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080039#include <linux/posix-timers.h>
Matt Helsley9f460802005-11-07 00:59:16 -080040#include <linux/cn_proc.h>
Ingo Molnarde5097c2006-01-09 15:59:21 -080041#include <linux/mutex.h>
Ingo Molnar0771dfe2006-03-27 01:16:22 -080042#include <linux/futex.h>
Jens Axboeb92ce552006-04-11 13:52:07 +020043#include <linux/pipe_fs_i.h>
Al Virofa84cb92006-03-29 20:30:19 -050044#include <linux/audit.h> /* for audit_free() */
Adrian Bunk83cc5ed2006-06-25 05:47:41 -070045#include <linux/resource.h>
David Howells0d67a462006-08-29 19:05:56 +010046#include <linux/blkdev.h>
Eric Dumazet6eaeeab2007-05-10 22:22:37 -070047#include <linux/task_io_accounting_ops.h>
Roland McGrath30199f52008-07-25 19:45:46 -070048#include <linux/tracehook.h>
Al Viro5ad4e532009-03-29 19:50:06 -040049#include <linux/fs_struct.h>
David Howellsd84f4f92008-11-14 10:39:23 +110050#include <linux/init_task.h>
Mathieu Desnoyers0a16b602008-07-18 12:16:17 -040051#include <trace/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#include <asm/uaccess.h>
54#include <asm/unistd.h>
55#include <asm/pgtable.h>
56#include <asm/mmu_context.h>
David Howellsd84f4f92008-11-14 10:39:23 +110057#include "cred-internals.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Mathieu Desnoyers7e066fb2008-11-14 17:47:47 -050059DEFINE_TRACE(sched_process_free);
60DEFINE_TRACE(sched_process_exit);
61DEFINE_TRACE(sched_process_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Adrian Bunk408b6642005-05-01 08:59:29 -070063static void exit_mm(struct task_struct * tsk);
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065static void __unhash_process(struct task_struct *p)
66{
67 nr_threads--;
68 detach_pid(p, PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 if (thread_group_leader(p)) {
70 detach_pid(p, PIDTYPE_PGID);
71 detach_pid(p, PIDTYPE_SID);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080072
Eric W. Biederman5e85d4a2006-04-18 22:20:16 -070073 list_del_rcu(&p->tasks);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -080074 __get_cpu_var(process_counts)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 }
Oleg Nesterov47e65322006-03-28 16:11:25 -080076 list_del_rcu(&p->thread_group);
Roland McGrathf4700212008-03-24 18:36:23 -070077 list_del_init(&p->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078}
79
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080080/*
81 * This function expects the tasklist_lock write-locked.
82 */
83static void __exit_signal(struct task_struct *tsk)
84{
85 struct signal_struct *sig = tsk->signal;
86 struct sighand_struct *sighand;
87
88 BUG_ON(!sig);
89 BUG_ON(!atomic_read(&sig->count));
90
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080091 sighand = rcu_dereference(tsk->sighand);
92 spin_lock(&sighand->siglock);
93
94 posix_cpu_timers_exit(tsk);
95 if (atomic_dec_and_test(&sig->count))
96 posix_cpu_timers_exit_group(tsk);
97 else {
98 /*
99 * If there is any task waiting for the group exit
100 * then notify it:
101 */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700102 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800103 wake_up_process(sig->group_exit_task);
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700104
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800105 if (tsk == sig->curr_target)
106 sig->curr_target = next_thread(tsk);
107 /*
108 * Accumulate here the counters for all threads but the
109 * group leader as they die, so they can be added into
110 * the process-wide totals when those are taken.
111 * The group leader stays around as a zombie as long
112 * as there are other threads. When it gets reaped,
113 * the exit.c code will add its counts into these totals.
114 * We won't ever get here for the group leader, since it
115 * will have been the last reference on the signal_struct.
116 */
Peter Zijlstra32bd6712009-02-05 12:24:15 +0100117 sig->utime = cputime_add(sig->utime, task_utime(tsk));
118 sig->stime = cputime_add(sig->stime, task_stime(tsk));
Balbir Singh49048622008-09-05 18:12:23 +0200119 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800120 sig->min_flt += tsk->min_flt;
121 sig->maj_flt += tsk->maj_flt;
122 sig->nvcsw += tsk->nvcsw;
123 sig->nivcsw += tsk->nivcsw;
Eric Dumazet6eaeeab2007-05-10 22:22:37 -0700124 sig->inblock += task_io_get_inblock(tsk);
125 sig->oublock += task_io_get_oublock(tsk);
Andrea Righi59954772008-07-27 17:29:15 +0200126 task_io_accounting_add(&sig->ioac, &tsk->ioac);
Peter Zijlstra32bd6712009-02-05 12:24:15 +0100127 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800128 sig = NULL; /* Marker for below. */
129 }
130
Oleg Nesterov58767002006-03-28 16:11:20 -0800131 __unhash_process(tsk);
132
Oleg Nesterovda7978b2008-05-23 13:04:41 -0700133 /*
134 * Do this under ->siglock, we can race with another thread
135 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136 */
137 flush_sigqueue(&tsk->pending);
138
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800139 tsk->signal = NULL;
Oleg Nesterova7e53282006-03-28 16:11:27 -0800140 tsk->sighand = NULL;
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800141 spin_unlock(&sighand->siglock);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800142
Oleg Nesterova7e53282006-03-28 16:11:27 -0800143 __cleanup_sighand(sighand);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800144 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800145 if (sig) {
146 flush_sigqueue(&sig->shared_pending);
Oleg Nesterov093a8e82006-10-28 10:38:51 -0700147 taskstats_tgid_free(sig);
Oleg Nesterovad474ca2008-11-10 15:39:30 +0100148 /*
149 * Make sure ->signal can't go away under rq->lock,
150 * see account_group_exec_runtime().
151 */
152 task_rq_unlock_wait(tsk);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800153 __cleanup_signal(sig);
154 }
155}
156
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800157static void delayed_put_task_struct(struct rcu_head *rhp)
158{
Mathieu Desnoyers0a16b602008-07-18 12:16:17 -0400159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160
Ingo Molnareef6cbf2008-12-19 10:20:42 +0100161#ifdef CONFIG_PERF_COUNTERS
162 WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list));
163#endif
Mathieu Desnoyers0a16b602008-07-18 12:16:17 -0400164 trace_sched_process_free(tsk);
165 put_task_struct(tsk);
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800166}
167
Roland McGrathf4700212008-03-24 18:36:23 -0700168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169void release_task(struct task_struct * p)
170{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700171 struct task_struct *leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 int zap_leader;
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800173repeat:
Roland McGrathdae33572008-07-25 19:45:48 -0700174 tracehook_prepare_release_task(p);
David Howellsc69e8d92008-11-14 10:39:19 +1100175 /* don't need to get the RCU readlock here - the process is dead and
176 * can't be modifying its own credentials */
177 atomic_dec(&__task_cred(p)->user->processes);
178
Pavel Emelyanov60347f62007-10-18 23:40:03 -0700179 proc_flush_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 write_lock_irq(&tasklist_lock);
Roland McGrathdae33572008-07-25 19:45:48 -0700181 tracehook_finish_release_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 __exit_signal(p);
Oleg Nesterov35f5cad2006-03-28 16:11:19 -0800183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 /*
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
188 */
189 zap_leader = 0;
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
Oleg Nesterovd839fd42008-04-30 00:53:11 -0700192 BUG_ON(task_detached(leader));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 do_notify_parent(leader, leader->exit_signal);
194 /*
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
198 *
199 * do_notify_parent() will have marked it self-reaping in
200 * that case.
201 */
Oleg Nesterovd839fd42008-04-30 00:53:11 -0700202 zap_leader = task_detached(leader);
Roland McGrathdae33572008-07-25 19:45:48 -0700203
204 /*
205 * This maintains the invariant that release_task()
206 * only runs on a task in EXIT_DEAD, just for sanity.
207 */
208 if (zap_leader)
209 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 write_unlock_irq(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 release_thread(p);
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800214 call_rcu(&p->rcu, delayed_put_task_struct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216 p = leader;
217 if (unlikely(zap_leader))
218 goto repeat;
219}
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221/*
222 * This checks not only the pgrp, but falls back on the pid if no
223 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
224 * without this...
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800225 *
226 * The caller must hold rcu lock or the tasklist lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800228struct pid *session_of_pgrp(struct pid *pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 struct task_struct *p;
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800231 struct pid *sid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800233 p = pid_task(pgrp, PIDTYPE_PGID);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800234 if (p == NULL)
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800235 p = pid_task(pgrp, PIDTYPE_PID);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800236 if (p != NULL)
Eric W. Biederman04a2e6a2007-02-12 00:52:56 -0800237 sid = task_session(p);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return sid;
240}
241
242/*
243 * Determine if a process group is "orphaned", according to the POSIX
244 * definition in 2.2.2.52. Orphaned process groups are not to be affected
245 * by terminal-generated stop signals. Newly orphaned process groups are
246 * to receive a SIGHUP and a SIGCONT.
247 *
248 * "I ask you, have you ever known what it is to be an orphan?"
249 */
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800250static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
252 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800254 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
Oleg Nesterov05e83df2008-03-02 21:44:42 +0300255 if ((p == ignored_task) ||
256 (p->exit_state && thread_group_empty(p)) ||
257 is_global_init(p->real_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 continue;
Oleg Nesterov05e83df2008-03-02 21:44:42 +0300259
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800260 if (task_pgrp(p->real_parent) != pgrp &&
Oleg Nesterov05e83df2008-03-02 21:44:42 +0300261 task_session(p->real_parent) == task_session(p))
262 return 0;
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800263 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
Oleg Nesterov05e83df2008-03-02 21:44:42 +0300264
265 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
Eric W. Biederman3e7cd6c2007-02-12 00:52:58 -0800268int is_current_pgrp_orphaned(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 int retval;
271
272 read_lock(&tasklist_lock);
Eric W. Biederman3e7cd6c2007-02-12 00:52:58 -0800273 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 read_unlock(&tasklist_lock);
275
276 return retval;
277}
278
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800279static int has_stopped_jobs(struct pid *pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 int retval = 0;
282 struct task_struct *p;
283
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800284 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
Matthew Wilcox338077e2007-12-06 11:09:35 -0500285 if (!task_is_stopped(p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 retval = 1;
288 break;
Eric W. Biederman0475ac02007-02-12 00:52:57 -0800289 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 return retval;
291}
292
Oleg Nesterovf49ee502008-03-02 21:44:40 +0300293/*
294 * Check to see if any process groups have become orphaned as
295 * a result of our exiting, and if they have any stopped jobs,
296 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
297 */
298static void
299kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
300{
301 struct pid *pgrp = task_pgrp(tsk);
302 struct task_struct *ignored_task = tsk;
303
304 if (!parent)
305 /* exit: our father is in a different pgrp than
306 * we are and we were the only connection outside.
307 */
308 parent = tsk->real_parent;
309 else
310 /* reparent: our child is in a different pgrp than
311 * we are, and it was the only connection outside.
312 */
313 ignored_task = NULL;
314
315 if (task_pgrp(parent) != pgrp &&
316 task_session(parent) == task_session(tsk) &&
317 will_become_orphaned_pgrp(pgrp, ignored_task) &&
318 has_stopped_jobs(pgrp)) {
319 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
320 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
321 }
322}
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324/**
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700325 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 *
327 * If a kernel thread is launched as a result of a system call, or if
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700328 * it ever exits, it should generally reparent itself to kthreadd so it
329 * isn't in the way of other processes and is correctly cleaned up on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 *
331 * The various task state such as scheduling policy and priority may have
332 * been inherited from a user process, so we reset them to sane values here.
333 *
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700334 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700336static void reparent_to_kthreadd(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 write_lock_irq(&tasklist_lock);
339
340 ptrace_unlink(current);
341 /* Reparent to init */
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700342 current->real_parent = current->parent = kthreadd_task;
Roland McGrathf4700212008-03-24 18:36:23 -0700343 list_move_tail(&current->sibling, &current->real_parent->children);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 /* Set the exit signal to SIGCHLD so we signal init on exit */
346 current->exit_signal = SIGCHLD;
347
Ingo Molnare05606d2007-07-09 18:51:59 +0200348 if (task_nice(current) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 set_user_nice(current, 0);
350 /* cpus_allowed? */
351 /* rt_priority? */
352 /* signals? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 memcpy(current->signal->rlim, init_task.signal->rlim,
354 sizeof(current->signal->rlim));
David Howellsd84f4f92008-11-14 10:39:23 +1100355
356 atomic_inc(&init_cred.usage);
357 commit_creds(&init_cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 write_unlock_irq(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360
Oleg Nesterov8520d7c2008-02-08 04:19:09 -0800361void __set_special_pids(struct pid *pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Oren Laadane19f2472006-01-08 01:03:58 -0800363 struct task_struct *curr = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Oleg Nesterov1b0f7ff2009-04-02 16:58:39 -0700365 if (task_session(curr) != pid)
Oleg Nesterov7d8da092008-04-30 00:54:27 -0700366 change_pid(curr, PIDTYPE_SID, pid);
Oleg Nesterov1b0f7ff2009-04-02 16:58:39 -0700367
368 if (task_pgrp(curr) != pid)
Oleg Nesterov7d8da092008-04-30 00:54:27 -0700369 change_pid(curr, PIDTYPE_PGID, pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
Oleg Nesterov8520d7c2008-02-08 04:19:09 -0800372static void set_special_pids(struct pid *pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
374 write_lock_irq(&tasklist_lock);
Oleg Nesterov8520d7c2008-02-08 04:19:09 -0800375 __set_special_pids(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 write_unlock_irq(&tasklist_lock);
377}
378
379/*
380 * Let kernel threads use this to say that they
381 * allow a certain signal (since daemonize() will
382 * have disabled all of them by default).
383 */
384int allow_signal(int sig)
385{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700386 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return -EINVAL;
388
389 spin_lock_irq(&current->sighand->siglock);
390 sigdelset(&current->blocked, sig);
391 if (!current->mm) {
392 /* Kernel threads handle their own signals.
393 Let the signal code know it'll be handled, so
394 that they don't get converted to SIGKILL or
395 just silently dropped */
396 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
397 }
398 recalc_sigpending();
399 spin_unlock_irq(&current->sighand->siglock);
400 return 0;
401}
402
403EXPORT_SYMBOL(allow_signal);
404
405int disallow_signal(int sig)
406{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700407 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 return -EINVAL;
409
410 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterov10ab8252007-05-09 02:34:37 -0700411 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 recalc_sigpending();
413 spin_unlock_irq(&current->sighand->siglock);
414 return 0;
415}
416
417EXPORT_SYMBOL(disallow_signal);
418
419/*
420 * Put all the gunge required to become a kernel thread without
421 * attached user resources in one place where it belongs.
422 */
423
424void daemonize(const char *name, ...)
425{
426 va_list args;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 sigset_t blocked;
428
429 va_start(args, name);
430 vsnprintf(current->comm, sizeof(current->comm), name, args);
431 va_end(args);
432
433 /*
434 * If we were started as result of loading a module, close all of the
435 * user space pages. We don't need them, and if we didn't close them
436 * they would be locked into memory.
437 */
438 exit_mm(current);
Rafael J. Wysocki83144182007-07-17 04:03:35 -0700439 /*
440 * We don't want to have TIF_FREEZE set if the system-wide hibernation
441 * or suspend transition begins right now.
442 */
Oleg Nesterov7b34e422008-07-25 01:47:37 -0700443 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Oleg Nesterov8520d7c2008-02-08 04:19:09 -0800445 if (current->nsproxy != &init_nsproxy) {
446 get_nsproxy(&init_nsproxy);
447 switch_task_namespaces(current, &init_nsproxy);
448 }
Oleg Nesterov297bd422008-02-08 04:19:10 -0800449 set_special_pids(&init_struct_pid);
Peter Zijlstra24ec8392006-12-08 02:36:04 -0800450 proc_clear_tty(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 /* Block and flush all signals */
453 sigfillset(&blocked);
454 sigprocmask(SIG_BLOCK, &blocked, NULL);
455 flush_signals(current);
456
457 /* Become as one with the init task */
458
Al Viro3e93cd62009-03-29 19:00:13 -0400459 daemonize_fs_struct();
Daniel Walkerd4c5e412007-10-18 23:39:59 -0700460 exit_files(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 current->files = init_task.files;
462 atomic_inc(&current->files->count);
463
Eric W. Biederman49d769d2007-05-09 02:34:33 -0700464 reparent_to_kthreadd();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467EXPORT_SYMBOL(daemonize);
468
Arjan van de Ven858119e2006-01-14 13:20:43 -0800469static void close_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
471 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700472 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700475
476 /*
477 * It is safe to dereference the fd table without RCU or
478 * ->file_lock because this is the last reference to the
479 * files structure.
480 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700481 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 for (;;) {
483 unsigned long set;
484 i = j * __NFDBITS;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800485 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700487 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 while (set) {
489 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700490 struct file * file = xchg(&fdt->fd[i], NULL);
Ingo Molnar944be0b2007-02-12 00:52:26 -0800491 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 filp_close(file, files);
Ingo Molnar944be0b2007-02-12 00:52:26 -0800493 cond_resched();
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496 i++;
497 set >>= 1;
498 }
499 }
500}
501
502struct files_struct *get_files_struct(struct task_struct *task)
503{
504 struct files_struct *files;
505
506 task_lock(task);
507 files = task->files;
508 if (files)
509 atomic_inc(&files->count);
510 task_unlock(task);
511
512 return files;
513}
514
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800515void put_files_struct(struct files_struct *files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700517 struct fdtable *fdt;
518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (atomic_dec_and_test(&files->count)) {
520 close_files(files);
521 /*
522 * Free the fd and fdset arrays if we expanded them.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700523 * If the fdtable was embedded, pass files for freeing
524 * at the end of the RCU grace period. Otherwise,
525 * you can free files immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700527 fdt = files_fdtable(files);
Vadim Lobanov4fd45812006-12-10 02:21:17 -0800528 if (fdt != &files->fdtab)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700529 kmem_cache_free(files_cachep, files);
Vadim Lobanov01b2d932006-12-22 01:10:43 -0800530 free_fdtable(fdt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
532}
533
Al Viro3b125382008-04-22 05:31:30 -0400534void reset_files_struct(struct files_struct *files)
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700535{
Al Viro3b125382008-04-22 05:31:30 -0400536 struct task_struct *tsk = current;
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700537 struct files_struct *old;
538
539 old = tsk->files;
540 task_lock(tsk);
541 tsk->files = files;
542 task_unlock(tsk);
543 put_files_struct(old);
544}
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700545
Al Viro1ec7f1d2008-04-22 05:35:42 -0400546void exit_files(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
548 struct files_struct * files = tsk->files;
549
550 if (files) {
551 task_lock(tsk);
552 tsk->files = NULL;
553 task_unlock(tsk);
554 put_files_struct(files);
555 }
556}
557
Balbir Singhcf475ad2008-04-29 01:00:16 -0700558#ifdef CONFIG_MM_OWNER
559/*
560 * Task p is exiting and it owned mm, lets find a new owner for it
561 */
562static inline int
563mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
564{
565 /*
566 * If there are other users of the mm and the owner (us) is exiting
567 * we need to find a new owner to take on the responsibility.
568 */
Balbir Singhcf475ad2008-04-29 01:00:16 -0700569 if (atomic_read(&mm->mm_users) <= 1)
570 return 0;
571 if (mm->owner != p)
572 return 0;
573 return 1;
574}
575
576void mm_update_next_owner(struct mm_struct *mm)
577{
578 struct task_struct *c, *g, *p = current;
579
580retry:
581 if (!mm_need_new_owner(mm, p))
582 return;
583
584 read_lock(&tasklist_lock);
585 /*
586 * Search in the children
587 */
588 list_for_each_entry(c, &p->children, sibling) {
589 if (c->mm == mm)
590 goto assign_new_owner;
591 }
592
593 /*
594 * Search in the siblings
595 */
596 list_for_each_entry(c, &p->parent->children, sibling) {
597 if (c->mm == mm)
598 goto assign_new_owner;
599 }
600
601 /*
602 * Search through everything else. We should not get
603 * here often
604 */
605 do_each_thread(g, c) {
606 if (c->mm == mm)
607 goto assign_new_owner;
608 } while_each_thread(g, c);
609
610 read_unlock(&tasklist_lock);
Balbir Singh31a78f22008-09-28 23:09:31 +0100611 /*
612 * We found no owner yet mm_users > 1: this implies that we are
613 * most likely racing with swapoff (try_to_unuse()) or /proc or
Hugh Dickinse5991372009-01-06 14:39:22 -0800614 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
Balbir Singh31a78f22008-09-28 23:09:31 +0100615 */
Balbir Singh31a78f22008-09-28 23:09:31 +0100616 mm->owner = NULL;
Balbir Singhcf475ad2008-04-29 01:00:16 -0700617 return;
618
619assign_new_owner:
620 BUG_ON(c == p);
621 get_task_struct(c);
622 /*
623 * The task_lock protects c->mm from changing.
624 * We always want mm->owner->mm == mm
625 */
626 task_lock(c);
Hugh Dickinse5991372009-01-06 14:39:22 -0800627 /*
628 * Delay read_unlock() till we have the task_lock()
629 * to ensure that c does not slip away underneath us
630 */
631 read_unlock(&tasklist_lock);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700632 if (c->mm != mm) {
633 task_unlock(c);
634 put_task_struct(c);
635 goto retry;
636 }
Balbir Singhcf475ad2008-04-29 01:00:16 -0700637 mm->owner = c;
638 task_unlock(c);
639 put_task_struct(c);
640}
641#endif /* CONFIG_MM_OWNER */
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643/*
644 * Turn us into a lazy TLB process if we
645 * aren't already..
646 */
Adrian Bunk408b6642005-05-01 08:59:29 -0700647static void exit_mm(struct task_struct * tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 struct mm_struct *mm = tsk->mm;
Oleg Nesterovb564daf2008-07-25 01:47:44 -0700650 struct core_state *core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652 mm_release(tsk, mm);
653 if (!mm)
654 return;
655 /*
656 * Serialize with any possible pending coredump.
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700657 * We must hold mmap_sem around checking core_state
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 * and clearing tsk->mm. The core-inducing thread
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700659 * will increment ->nr_threads for each thread in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 * group with ->mm != NULL.
661 */
662 down_read(&mm->mmap_sem);
Oleg Nesterovb564daf2008-07-25 01:47:44 -0700663 core_state = mm->core_state;
664 if (core_state) {
665 struct core_thread self;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 up_read(&mm->mmap_sem);
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -0700667
Oleg Nesterovb564daf2008-07-25 01:47:44 -0700668 self.task = tsk;
669 self.next = xchg(&core_state->dumper.next, &self);
670 /*
671 * Implies mb(), the result of xchg() must be visible
672 * to core_state->dumper.
673 */
674 if (atomic_dec_and_test(&core_state->nr_threads))
675 complete(&core_state->startup);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Oleg Nesterova94e2d42008-07-25 01:47:46 -0700677 for (;;) {
678 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
679 if (!self.task) /* see coredump_finish() */
680 break;
681 schedule();
682 }
683 __set_task_state(tsk, TASK_RUNNING);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 down_read(&mm->mmap_sem);
685 }
686 atomic_inc(&mm->mm_count);
Eric Sesterhenn125e1872006-06-23 02:06:06 -0700687 BUG_ON(mm != tsk->active_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 /* more a memory barrier than a real lock */
689 task_lock(tsk);
690 tsk->mm = NULL;
691 up_read(&mm->mmap_sem);
692 enter_lazy_tlb(mm, current);
Rafael J. Wysocki0c1eecf2007-07-19 01:47:33 -0700693 /* We don't want this task to be frozen prematurely */
694 clear_freeze_flag(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 task_unlock(tsk);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700696 mm_update_next_owner(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 mmput(mm);
698}
699
Roland McGrathf4700212008-03-24 18:36:23 -0700700/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 * When we die, we re-parent all our children.
702 * Try to give them to another thread in our thread
703 * group, and if no such member exists, give it to
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800704 * the child reaper process (ie "init") in our pid
705 * space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 */
Oleg Nesterov950bbab2008-09-02 14:35:49 -0700707static struct task_struct *find_new_reaper(struct task_struct *father)
708{
709 struct pid_namespace *pid_ns = task_active_pid_ns(father);
710 struct task_struct *thread;
711
712 thread = father;
713 while_each_thread(father, thread) {
714 if (thread->flags & PF_EXITING)
715 continue;
716 if (unlikely(pid_ns->child_reaper == father))
717 pid_ns->child_reaper = thread;
718 return thread;
719 }
720
721 if (unlikely(pid_ns->child_reaper == father)) {
722 write_unlock_irq(&tasklist_lock);
723 if (unlikely(pid_ns == &init_pid_ns))
724 panic("Attempted to kill init!");
725
726 zap_pid_ns_processes(pid_ns);
727 write_lock_irq(&tasklist_lock);
728 /*
729 * We can not clear ->child_reaper or leave it alone.
730 * There may by stealth EXIT_DEAD tasks on ->children,
731 * forget_original_parent() must move them somewhere.
732 */
733 pid_ns->child_reaper = init_pid_ns.child_reaper;
734 }
735
736 return pid_ns->child_reaper;
737}
738
Oleg Nesterov5dfc80b2009-04-02 16:58:19 -0700739/*
740* Any that need to be release_task'd are put on the @dead list.
741 */
742static void reparent_thread(struct task_struct *father, struct task_struct *p,
743 struct list_head *dead)
744{
745 if (p->pdeath_signal)
746 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
747
748 list_move_tail(&p->sibling, &p->real_parent->children);
749
750 if (task_detached(p))
751 return;
752 /*
753 * If this is a threaded reparent there is no need to
754 * notify anyone anything has happened.
755 */
756 if (same_thread_group(p->real_parent, father))
757 return;
758
759 /* We don't want people slaying init. */
760 p->exit_signal = SIGCHLD;
761
762 /* If it has exited notify the new parent about this child's death. */
763 if (!p->ptrace &&
764 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
765 do_notify_parent(p, p->exit_signal);
766 if (task_detached(p)) {
767 p->exit_state = EXIT_DEAD;
768 list_move_tail(&p->sibling, dead);
769 }
770 }
771
772 kill_orphaned_pgrp(p, father);
773}
774
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700775static void forget_original_parent(struct task_struct *father)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Oleg Nesterov950bbab2008-09-02 14:35:49 -0700777 struct task_struct *p, *n, *reaper;
Oleg Nesterov5dfc80b2009-04-02 16:58:19 -0700778 LIST_HEAD(dead_children);
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700779
Oleg Nesterov39c626a2009-04-02 16:58:18 -0700780 exit_ptrace(father);
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700781
782 write_lock_irq(&tasklist_lock);
Oleg Nesterov950bbab2008-09-02 14:35:49 -0700783 reaper = find_new_reaper(father);
Roland McGrathf4700212008-03-24 18:36:23 -0700784
Matthias Kaehlcke03ff1792007-10-18 23:39:57 -0700785 list_for_each_entry_safe(p, n, &father->children, sibling) {
Oleg Nesterov84eb6462007-10-16 23:26:49 -0700786 p->real_parent = reaper;
Roland McGrathf4700212008-03-24 18:36:23 -0700787 if (p->parent == father) {
788 BUG_ON(p->ptrace);
789 p->parent = p->real_parent;
790 }
Oleg Nesterov5dfc80b2009-04-02 16:58:19 -0700791 reparent_thread(father, p, &dead_children);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700793 write_unlock_irq(&tasklist_lock);
Oleg Nesterov5dfc80b2009-04-02 16:58:19 -0700794
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700795 BUG_ON(!list_empty(&father->children));
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700796
Oleg Nesterov5dfc80b2009-04-02 16:58:19 -0700797 list_for_each_entry_safe(p, n, &dead_children, sibling) {
798 list_del_init(&p->sibling);
Oleg Nesterov39c626a2009-04-02 16:58:18 -0700799 release_task(p);
800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
803/*
804 * Send signals to all our closest relatives so that they know
805 * to properly mourn us..
806 */
Oleg Nesterov821c7de2008-03-02 21:44:44 +0300807static void exit_notify(struct task_struct *tsk, int group_dead)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
Roland McGrath2b2a1ff2008-07-25 19:45:54 -0700809 int signal;
810 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 /*
813 * This does two things:
814 *
815 * A. Make init inherit all the child processes
816 * B. Check to see if any process groups have become orphaned
817 * as a result of our exiting, and if they have any stopped
818 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
819 */
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700820 forget_original_parent(tsk);
Pavel Emelyanov2e4a7072007-10-18 23:40:01 -0700821 exit_task_namespaces(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Oleg Nesterov762a24b2007-10-18 23:40:00 -0700823 write_lock_irq(&tasklist_lock);
Oleg Nesterov821c7de2008-03-02 21:44:44 +0300824 if (group_dead)
825 kill_orphaned_pgrp(tsk->group_leader, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Oleg Nesterov24728442007-08-04 01:04:41 +0400827 /* Let father know we died
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 *
829 * Thread signals are configurable, but you aren't going to use
Daniel Walkerd4c5e412007-10-18 23:39:59 -0700830 * that to send signals to arbitary processes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 * That stops right now.
832 *
833 * If the parent exec id doesn't match the exec id we saved
834 * when we started then we know the parent has changed security
835 * domain.
836 *
837 * If our self_exec id doesn't match our parent_exec_id then
838 * we have changed execution domain as these two values started
839 * the same after a fork.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 */
Oleg Nesterovd839fd42008-04-30 00:53:11 -0700841 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
Oleg Nesterovf49ee502008-03-02 21:44:40 +0300842 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
Oleg Nesterov432870d2009-04-06 16:16:02 +0200843 tsk->self_exec_id != tsk->parent_exec_id))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 tsk->exit_signal = SIGCHLD;
845
Roland McGrath2b2a1ff2008-07-25 19:45:54 -0700846 signal = tracehook_notify_death(tsk, &cookie, group_dead);
Roland McGrath5c7edcd2008-07-31 02:04:09 -0700847 if (signal >= 0)
Roland McGrath2b2a1ff2008-07-25 19:45:54 -0700848 signal = do_notify_parent(tsk, signal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Roland McGrath5c7edcd2008-07-31 02:04:09 -0700850 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700852 /* mt-exec, de_thread() is waiting for us */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700853 if (thread_group_leader(tsk) &&
Steve VanDeBogart2633f0e2008-08-26 15:14:36 -0700854 tsk->signal->group_exit_task &&
855 tsk->signal->notify_count < 0)
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700856 wake_up_process(tsk->signal->group_exit_task);
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 write_unlock_irq(&tasklist_lock);
859
Roland McGrath2b2a1ff2008-07-25 19:45:54 -0700860 tracehook_report_death(tsk, signal, cookie, group_dead);
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 /* If the process is dead, release it - nobody will wait for it */
Roland McGrath5c7edcd2008-07-31 02:04:09 -0700863 if (signal == DEATH_REAP)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 release_task(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
Jeff Dikee18eecb2007-07-15 23:38:48 -0700867#ifdef CONFIG_DEBUG_STACK_USAGE
868static void check_stack_usage(void)
869{
870 static DEFINE_SPINLOCK(low_water_lock);
871 static int lowest_to_date = THREAD_SIZE;
Jeff Dikee18eecb2007-07-15 23:38:48 -0700872 unsigned long free;
873
Eric Sandeen7c9f8862008-04-22 16:38:23 -0500874 free = stack_not_used(current);
Jeff Dikee18eecb2007-07-15 23:38:48 -0700875
876 if (free >= lowest_to_date)
877 return;
878
879 spin_lock(&low_water_lock);
880 if (free < lowest_to_date) {
881 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
882 "left\n",
883 current->comm, free);
884 lowest_to_date = free;
885 }
886 spin_unlock(&low_water_lock);
887}
888#else
889static inline void check_stack_usage(void) {}
890#endif
891
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800892NORET_TYPE void do_exit(long code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
894 struct task_struct *tsk = current;
895 int group_dead;
896
897 profile_task_exit(tsk);
898
Jens Axboe22e2c502005-06-27 10:55:12 +0200899 WARN_ON(atomic_read(&tsk->fs_excl));
900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 if (unlikely(in_interrupt()))
902 panic("Aiee, killing interrupt handler!");
903 if (unlikely(!tsk->pid))
904 panic("Attempted to kill the idle task!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Roland McGrath30199f52008-07-25 19:45:46 -0700906 tracehook_report_exit(&code);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Alexander Nybergdf164db2005-06-23 00:09:13 -0700908 /*
909 * We're taking recursive faults here in do_exit. Safest is to just
910 * leave this task alone and wait for reboot.
911 */
912 if (unlikely(tsk->flags & PF_EXITING)) {
913 printk(KERN_ALERT
914 "Fixing recursive fault but reboot is needed!\n");
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700915 /*
916 * We can do this unlocked here. The futex code uses
917 * this flag just to verify whether the pi state
918 * cleanup has been done or not. In the worst case it
919 * loops once more. We pretend that the cleanup was
920 * done as there is no way to return. Either the
921 * OWNER_DIED bit is set by now or we push the blocked
922 * task into the wait for ever nirwana as well.
923 */
924 tsk->flags |= PF_EXITPIDONE;
Alexander Nybergdf164db2005-06-23 00:09:13 -0700925 set_current_state(TASK_UNINTERRUPTIBLE);
926 schedule();
927 }
928
Thomas Gleixner3aa551c2009-03-23 18:28:15 +0100929 exit_irq_thread();
930
Oleg Nesterovd12619b2008-02-08 04:19:12 -0800931 exit_signals(tsk); /* sets PF_EXITING */
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700932 /*
933 * tsk->flags are checked in the futex code to protect against
934 * an exiting task cleaning up the robust pi futexes.
935 */
Oleg Nesterovd2ee7192007-10-16 23:26:47 -0700936 smp_mb();
937 spin_unlock_wait(&tsk->pi_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (unlikely(in_atomic()))
940 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700941 current->comm, task_pid_nr(current),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 preempt_count());
943
944 acct_update_integrals(tsk);
Oleg Nesterov901608d2009-01-06 14:40:29 -0800945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 group_dead = atomic_dec_and_test(&tsk->signal->live);
Andrew Mortonc3068952005-08-04 16:49:32 -0700947 if (group_dead) {
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700948 hrtimer_cancel(&tsk->signal->real_timer);
Roland McGrath25f407f2005-10-21 15:03:29 -0700949 exit_itimers(tsk->signal);
Andrew Mortonc3068952005-08-04 16:49:32 -0700950 }
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700951 acct_collect(code, group_dead);
Miloslav Trmac522ed772007-07-15 23:40:56 -0700952 if (group_dead)
953 tty_audit_exit();
Al Virofa84cb92006-03-29 20:30:19 -0500954 if (unlikely(tsk->audit_context))
955 audit_free(tsk);
Oleg Nesterov115085e2006-12-06 20:36:51 -0800956
Jonathan Limf2ab6d82007-08-30 23:56:23 -0700957 tsk->exit_code = code;
Oleg Nesterov115085e2006-12-06 20:36:51 -0800958 taskstats_exit(tsk, group_dead);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 exit_mm(tsk);
961
KaiGai Kohei0e464812006-06-25 05:49:24 -0700962 if (group_dead)
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700963 acct_process();
Mathieu Desnoyers0a16b602008-07-18 12:16:17 -0400964 trace_sched_process_exit(tsk);
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 exit_sem(tsk);
Al Viro1ec7f1d2008-04-22 05:35:42 -0400967 exit_files(tsk);
968 exit_fs(tsk);
Jeff Dikee18eecb2007-07-15 23:38:48 -0700969 check_stack_usage();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 exit_thread();
Paul Menageb4f48b62007-10-18 23:39:33 -0700971 cgroup_exit(tsk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 if (group_dead && tsk->signal->leader)
974 disassociate_ctty(1);
975
Al Viroa1261f52005-11-13 16:06:55 -0800976 module_put(task_thread_info(tsk)->exec_domain->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (tsk->binfmt)
978 module_put(tsk->binfmt->module);
979
Matt Helsley9f460802005-11-07 00:59:16 -0800980 proc_exit_connector(tsk);
Oleg Nesterov821c7de2008-03-02 21:44:44 +0300981 exit_notify(tsk, group_dead);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982#ifdef CONFIG_NUMA
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700983 mpol_put(tsk->mempolicy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 tsk->mempolicy = NULL;
985#endif
Ingo Molnar9b51f662008-12-12 13:49:45 +0100986#ifdef CONFIG_FUTEX
Ingo Molnarc87e2832006-06-27 02:54:58 -0700987 if (unlikely(!list_empty(&tsk->pi_state_list)))
988 exit_pi_state_list(tsk);
989 if (unlikely(current->pi_state_cache))
990 kfree(current->pi_state_cache);
Alexey Dobriyan42b2dd02007-10-16 23:27:30 -0700991#endif
Ingo Molnarc87e2832006-06-27 02:54:58 -0700992 /*
Ingo Molnar9a11b492006-07-03 00:24:33 -0700993 * Make sure we are holding no locks:
Ingo Molnarde5097c2006-01-09 15:59:21 -0800994 */
Ingo Molnar9a11b492006-07-03 00:24:33 -0700995 debug_check_no_locks_held(tsk);
Alexey Kuznetsov778e9a92007-06-08 13:47:00 -0700996 /*
997 * We can do this unlocked here. The futex code uses this flag
998 * just to verify whether the pi state cleanup has been done
999 * or not. In the worst case it loops once more.
1000 */
1001 tsk->flags |= PF_EXITPIDONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Al Viroafc847b2006-02-28 12:51:55 -05001003 if (tsk->io_context)
1004 exit_io_context();
1005
Jens Axboeb92ce552006-04-11 13:52:07 +02001006 if (tsk->splice_pipe)
1007 __free_pipe_info(tsk->splice_pipe);
1008
Coywolf Qi Hunt74072512005-10-30 15:02:47 -08001009 preempt_disable();
Oleg Nesterov55a101f2006-09-29 02:01:10 -07001010 /* causes final put_task_struct in finish_task_switch(). */
Oleg Nesterovc394cc92006-09-29 02:01:11 -07001011 tsk->state = TASK_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 schedule();
1013 BUG();
1014 /* Avoid "noreturn function does return". */
Alan Cox54306cf2006-09-29 02:00:42 -07001015 for (;;)
1016 cpu_relax(); /* For when BUG is null */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
Russ Anderson012914d2005-04-23 00:08:00 -07001019EXPORT_SYMBOL_GPL(do_exit);
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1022{
1023 if (comp)
1024 complete(comp);
Oleg Nesterov55a101f2006-09-29 02:01:10 -07001025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 do_exit(code);
1027}
1028
1029EXPORT_SYMBOL(complete_and_exit);
1030
Heiko Carstens754fe8d2009-01-14 14:14:09 +01001031SYSCALL_DEFINE1(exit, int, error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
1033 do_exit((error_code&0xff)<<8);
1034}
1035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036/*
1037 * Take down every thread in the group. This is called by fatal signals
1038 * as well as by sys_exit_group (below).
1039 */
1040NORET_TYPE void
1041do_group_exit(int exit_code)
1042{
Oleg Nesterovbfc4b082008-04-30 00:52:36 -07001043 struct signal_struct *sig = current->signal;
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1046
Oleg Nesterovbfc4b082008-04-30 00:52:36 -07001047 if (signal_group_exit(sig))
1048 exit_code = sig->group_exit_code;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 else if (!thread_group_empty(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 struct sighand_struct *const sighand = current->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 spin_lock_irq(&sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001052 if (signal_group_exit(sig))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 /* Another thread got here before we took the lock. */
1054 exit_code = sig->group_exit_code;
1055 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 sig->group_exit_code = exit_code;
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001057 sig->flags = SIGNAL_GROUP_EXIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 zap_other_threads(current);
1059 }
1060 spin_unlock_irq(&sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062
1063 do_exit(exit_code);
1064 /* NOTREACHED */
1065}
1066
1067/*
1068 * this kills every thread in the thread group. Note that any externally
1069 * wait4()-ing process will get the correct exit code - even if this
1070 * thread is not the thread group leader.
1071 */
Heiko Carstens754fe8d2009-01-14 14:14:09 +01001072SYSCALL_DEFINE1(exit_group, int, error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 do_group_exit((error_code & 0xff) << 8);
Heiko Carstens2ed7c032009-01-14 14:13:54 +01001075 /* NOTREACHED */
1076 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077}
1078
Eric W. Biederman161550d2008-02-08 04:19:14 -08001079static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1080{
1081 struct pid *pid = NULL;
1082 if (type == PIDTYPE_PID)
1083 pid = task->pids[type].pid;
1084 else if (type < PIDTYPE_MAX)
1085 pid = task->group_leader->pids[type].pid;
1086 return pid;
1087}
1088
1089static int eligible_child(enum pid_type type, struct pid *pid, int options,
1090 struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
Roland McGrath73243282007-05-06 14:50:20 -07001092 int err;
1093
Eric W. Biederman161550d2008-02-08 04:19:14 -08001094 if (type < PIDTYPE_MAX) {
1095 if (task_pid_type(p, type) != pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return 0;
1097 }
1098
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 /* Wait for all children (clone and not) if __WALL is set;
1100 * otherwise, wait for clone children *only* if __WCLONE is
1101 * set; otherwise, wait for non-clone children *only*. (Note:
1102 * A "clone" child here is one that reports to its parent
1103 * using a signal other than SIGCHLD.) */
1104 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1105 && !(options & __WALL))
1106 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Roland McGrath73243282007-05-06 14:50:20 -07001108 err = security_task_wait(p);
Roland McGrath14dd0b82008-03-30 18:41:25 -07001109 if (err)
1110 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Roland McGrath14dd0b82008-03-30 18:41:25 -07001112 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113}
1114
Ingo Molnar36c8b582006-07-03 00:25:41 -07001115static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 int why, int status,
1117 struct siginfo __user *infop,
1118 struct rusage __user *rusagep)
1119{
1120 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07001121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 put_task_struct(p);
1123 if (!retval)
1124 retval = put_user(SIGCHLD, &infop->si_signo);
1125 if (!retval)
1126 retval = put_user(0, &infop->si_errno);
1127 if (!retval)
1128 retval = put_user((short)why, &infop->si_code);
1129 if (!retval)
1130 retval = put_user(pid, &infop->si_pid);
1131 if (!retval)
1132 retval = put_user(uid, &infop->si_uid);
1133 if (!retval)
1134 retval = put_user(status, &infop->si_status);
1135 if (!retval)
1136 retval = pid;
1137 return retval;
1138}
1139
1140/*
1141 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1142 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1143 * the lock and this task is uninteresting. If we return nonzero, we have
1144 * released the lock and the system call should return.
1145 */
Roland McGrath98abed02008-03-19 19:24:59 -07001146static int wait_task_zombie(struct task_struct *p, int options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 struct siginfo __user *infop,
1148 int __user *stat_addr, struct rusage __user *ru)
1149{
1150 unsigned long state;
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001151 int retval, status, traced;
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001152 pid_t pid = task_pid_vnr(p);
David Howellsc69e8d92008-11-14 10:39:19 +11001153 uid_t uid = __task_cred(p)->uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Roland McGrath98abed02008-03-19 19:24:59 -07001155 if (!likely(options & WEXITED))
1156 return 0;
1157
1158 if (unlikely(options & WNOWAIT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 int exit_code = p->exit_code;
1160 int why, status;
1161
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 get_task_struct(p);
1163 read_unlock(&tasklist_lock);
1164 if ((exit_code & 0x7f) == 0) {
1165 why = CLD_EXITED;
1166 status = exit_code >> 8;
1167 } else {
1168 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1169 status = exit_code & 0x7f;
1170 }
1171 return wait_noreap_copyout(p, pid, uid, why,
1172 status, infop, ru);
1173 }
1174
1175 /*
1176 * Try to move the task's state to DEAD
1177 * only one thread is allowed to do this:
1178 */
1179 state = xchg(&p->exit_state, EXIT_DEAD);
1180 if (state != EXIT_ZOMBIE) {
1181 BUG_ON(state != EXIT_DEAD);
1182 return 0;
1183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
Oleg Nesterov53b6f9f2008-04-30 00:53:13 -07001185 traced = ptrace_reparented(p);
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001186
1187 if (likely(!traced)) {
Jesper Juhl3795e162006-01-09 20:54:39 -08001188 struct signal_struct *psig;
1189 struct signal_struct *sig;
Frank Mayharf06febc2008-09-12 09:54:39 -07001190 struct task_cputime cputime;
Jesper Juhl3795e162006-01-09 20:54:39 -08001191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 /*
1193 * The resource counters for the group leader are in its
1194 * own task_struct. Those for dead threads in the group
1195 * are in its signal_struct, as are those for the child
1196 * processes it has previously reaped. All these
1197 * accumulate in the parent's signal_struct c* fields.
1198 *
1199 * We don't bother to take a lock here to protect these
1200 * p->signal fields, because they are only touched by
1201 * __exit_signal, which runs with tasklist_lock
1202 * write-locked anyway, and so is excluded here. We do
1203 * need to protect the access to p->parent->signal fields,
1204 * as other threads in the parent group can be right
1205 * here reaping other children at the same time.
Frank Mayharf06febc2008-09-12 09:54:39 -07001206 *
1207 * We use thread_group_cputime() to get times for the thread
1208 * group, which consolidates times for all threads in the
1209 * group including the group leader.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 */
Oleg Nesterov2b5fe6d2008-11-17 15:40:08 +01001211 thread_group_cputime(p, &cputime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 spin_lock_irq(&p->parent->sighand->siglock);
Jesper Juhl3795e162006-01-09 20:54:39 -08001213 psig = p->parent->signal;
1214 sig = p->signal;
1215 psig->cutime =
1216 cputime_add(psig->cutime,
Frank Mayharf06febc2008-09-12 09:54:39 -07001217 cputime_add(cputime.utime,
1218 sig->cutime));
Jesper Juhl3795e162006-01-09 20:54:39 -08001219 psig->cstime =
1220 cputime_add(psig->cstime,
Frank Mayharf06febc2008-09-12 09:54:39 -07001221 cputime_add(cputime.stime,
1222 sig->cstime));
Laurent Vivier9ac52312007-10-15 17:00:19 +02001223 psig->cgtime =
1224 cputime_add(psig->cgtime,
1225 cputime_add(p->gtime,
1226 cputime_add(sig->gtime,
1227 sig->cgtime)));
Jesper Juhl3795e162006-01-09 20:54:39 -08001228 psig->cmin_flt +=
1229 p->min_flt + sig->min_flt + sig->cmin_flt;
1230 psig->cmaj_flt +=
1231 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1232 psig->cnvcsw +=
1233 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1234 psig->cnivcsw +=
1235 p->nivcsw + sig->nivcsw + sig->cnivcsw;
Eric Dumazet6eaeeab2007-05-10 22:22:37 -07001236 psig->cinblock +=
1237 task_io_get_inblock(p) +
1238 sig->inblock + sig->cinblock;
1239 psig->coublock +=
1240 task_io_get_oublock(p) +
1241 sig->oublock + sig->coublock;
Andrea Righi59954772008-07-27 17:29:15 +02001242 task_io_accounting_add(&psig->ioac, &p->ioac);
1243 task_io_accounting_add(&psig->ioac, &sig->ioac);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 spin_unlock_irq(&p->parent->sighand->siglock);
1245 }
1246
1247 /*
1248 * Now we are sure this task is interesting, and no other
1249 * thread can reap it because we set its state to EXIT_DEAD.
1250 */
1251 read_unlock(&tasklist_lock);
1252
Ingo Molnareef6cbf2008-12-19 10:20:42 +01001253 /*
1254 * Flush inherited counters to the parent - before the parent
1255 * gets woken up by child-exit notifications.
1256 */
1257 perf_counter_exit_task(p);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1260 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1261 ? p->signal->group_exit_code : p->exit_code;
1262 if (!retval && stat_addr)
1263 retval = put_user(status, stat_addr);
1264 if (!retval && infop)
1265 retval = put_user(SIGCHLD, &infop->si_signo);
1266 if (!retval && infop)
1267 retval = put_user(0, &infop->si_errno);
1268 if (!retval && infop) {
1269 int why;
1270
1271 if ((status & 0x7f) == 0) {
1272 why = CLD_EXITED;
1273 status >>= 8;
1274 } else {
1275 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1276 status &= 0x7f;
1277 }
1278 retval = put_user((short)why, &infop->si_code);
1279 if (!retval)
1280 retval = put_user(status, &infop->si_status);
1281 }
1282 if (!retval && infop)
Oleg Nesterov3a515e42008-02-08 04:19:07 -08001283 retval = put_user(pid, &infop->si_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (!retval && infop)
David Howellsc69e8d92008-11-14 10:39:19 +11001285 retval = put_user(uid, &infop->si_uid);
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001286 if (!retval)
Oleg Nesterov3a515e42008-02-08 04:19:07 -08001287 retval = pid;
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001288
1289 if (traced) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 write_lock_irq(&tasklist_lock);
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001291 /* We dropped tasklist, ptracer could die and untrace */
1292 ptrace_unlink(p);
1293 /*
1294 * If this is not a detached task, notify the parent.
1295 * If it's still not detached after that, don't release
1296 * it now.
1297 */
Oleg Nesterovd839fd42008-04-30 00:53:11 -07001298 if (!task_detached(p)) {
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001299 do_notify_parent(p, p->exit_signal);
Oleg Nesterovd839fd42008-04-30 00:53:11 -07001300 if (!task_detached(p)) {
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001301 p->exit_state = EXIT_ZOMBIE;
1302 p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
1304 }
1305 write_unlock_irq(&tasklist_lock);
1306 }
1307 if (p != NULL)
1308 release_task(p);
Oleg Nesterov2f4e6e22007-10-16 23:26:58 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return retval;
1311}
1312
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001313static int *task_stopped_code(struct task_struct *p, bool ptrace)
1314{
1315 if (ptrace) {
1316 if (task_is_stopped_or_traced(p))
1317 return &p->exit_code;
1318 } else {
1319 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1320 return &p->signal->group_exit_code;
1321 }
1322 return NULL;
1323}
1324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325/*
1326 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1327 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1328 * the lock and this task is uninteresting. If we return nonzero, we have
1329 * released the lock and the system call should return.
1330 */
Roland McGrathf4700212008-03-24 18:36:23 -07001331static int wait_task_stopped(int ptrace, struct task_struct *p,
Roland McGrath98abed02008-03-19 19:24:59 -07001332 int options, struct siginfo __user *infop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 int __user *stat_addr, struct rusage __user *ru)
1334{
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001335 int retval, exit_code, *p_code, why;
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001336 uid_t uid = 0; /* unneeded, required by compiler */
Oleg Nesterovc8950782007-11-28 16:21:24 -08001337 pid_t pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Roland McGrathf4700212008-03-24 18:36:23 -07001339 if (!(options & WUNTRACED))
Roland McGrath98abed02008-03-19 19:24:59 -07001340 return 0;
1341
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001342 exit_code = 0;
1343 spin_lock_irq(&p->sighand->siglock);
1344
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001345 p_code = task_stopped_code(p, ptrace);
1346 if (unlikely(!p_code))
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001347 goto unlock_sig;
1348
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001349 exit_code = *p_code;
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001350 if (!exit_code)
1351 goto unlock_sig;
1352
Roland McGrath98abed02008-03-19 19:24:59 -07001353 if (!unlikely(options & WNOWAIT))
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001354 *p_code = 0;
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001355
David Howellsc69e8d92008-11-14 10:39:19 +11001356 /* don't need the RCU readlock here as we're holding a spinlock */
1357 uid = __task_cred(p)->uid;
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001358unlock_sig:
1359 spin_unlock_irq(&p->sighand->siglock);
1360 if (!exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return 0;
1362
1363 /*
1364 * Now we are pretty sure this task is interesting.
1365 * Make sure it doesn't get reaped out from under us while we
1366 * give up the lock and then examine it below. We don't want to
1367 * keep holding onto the tasklist_lock while we call getrusage and
1368 * possibly take page faults for user memory.
1369 */
1370 get_task_struct(p);
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001371 pid = task_pid_vnr(p);
Roland McGrathf4700212008-03-24 18:36:23 -07001372 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 read_unlock(&tasklist_lock);
1374
Roland McGrath98abed02008-03-19 19:24:59 -07001375 if (unlikely(options & WNOWAIT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return wait_noreap_copyout(p, pid, uid,
Scott James Remnante6ceb322007-11-28 16:22:07 -08001377 why, exit_code,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 infop, ru);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
1380 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1381 if (!retval && stat_addr)
1382 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1383 if (!retval && infop)
1384 retval = put_user(SIGCHLD, &infop->si_signo);
1385 if (!retval && infop)
1386 retval = put_user(0, &infop->si_errno);
1387 if (!retval && infop)
Roland McGrath6efcae42008-03-08 11:41:22 -08001388 retval = put_user((short)why, &infop->si_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (!retval && infop)
1390 retval = put_user(exit_code, &infop->si_status);
1391 if (!retval && infop)
Oleg Nesterovc8950782007-11-28 16:21:24 -08001392 retval = put_user(pid, &infop->si_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 if (!retval && infop)
Oleg Nesterovee7c82d2008-02-08 04:19:01 -08001394 retval = put_user(uid, &infop->si_uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 if (!retval)
Oleg Nesterovc8950782007-11-28 16:21:24 -08001396 retval = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 put_task_struct(p);
1398
1399 BUG_ON(!retval);
1400 return retval;
1401}
1402
1403/*
1404 * Handle do_wait work for one task in a live, non-stopped state.
1405 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1406 * the lock and this task is uninteresting. If we return nonzero, we have
1407 * released the lock and the system call should return.
1408 */
Roland McGrath98abed02008-03-19 19:24:59 -07001409static int wait_task_continued(struct task_struct *p, int options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 struct siginfo __user *infop,
1411 int __user *stat_addr, struct rusage __user *ru)
1412{
1413 int retval;
1414 pid_t pid;
1415 uid_t uid;
1416
Roland McGrath98abed02008-03-19 19:24:59 -07001417 if (!unlikely(options & WCONTINUED))
1418 return 0;
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1421 return 0;
1422
1423 spin_lock_irq(&p->sighand->siglock);
1424 /* Re-check with the lock held. */
1425 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1426 spin_unlock_irq(&p->sighand->siglock);
1427 return 0;
1428 }
Roland McGrath98abed02008-03-19 19:24:59 -07001429 if (!unlikely(options & WNOWAIT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
David Howellsc69e8d92008-11-14 10:39:19 +11001431 uid = __task_cred(p)->uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 spin_unlock_irq(&p->sighand->siglock);
1433
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08001434 pid = task_pid_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 get_task_struct(p);
1436 read_unlock(&tasklist_lock);
1437
1438 if (!infop) {
1439 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1440 put_task_struct(p);
1441 if (!retval && stat_addr)
1442 retval = put_user(0xffff, stat_addr);
1443 if (!retval)
Oleg Nesterov3a515e42008-02-08 04:19:07 -08001444 retval = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 } else {
1446 retval = wait_noreap_copyout(p, pid, uid,
1447 CLD_CONTINUED, SIGCONT,
1448 infop, ru);
1449 BUG_ON(retval == 0);
1450 }
1451
1452 return retval;
1453}
1454
Roland McGrath98abed02008-03-19 19:24:59 -07001455/*
1456 * Consider @p for a wait by @parent.
1457 *
1458 * -ECHILD should be in *@notask_error before the first call.
1459 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1460 * Returns zero if the search for a child should continue;
Roland McGrath14dd0b82008-03-30 18:41:25 -07001461 * then *@notask_error is 0 if @p is an eligible child,
1462 * or another error from security_task_wait(), or still -ECHILD.
Roland McGrath98abed02008-03-19 19:24:59 -07001463 */
Roland McGrathf4700212008-03-24 18:36:23 -07001464static int wait_consider_task(struct task_struct *parent, int ptrace,
Roland McGrath98abed02008-03-19 19:24:59 -07001465 struct task_struct *p, int *notask_error,
1466 enum pid_type type, struct pid *pid, int options,
1467 struct siginfo __user *infop,
1468 int __user *stat_addr, struct rusage __user *ru)
1469{
1470 int ret = eligible_child(type, pid, options, p);
Roland McGrath14dd0b82008-03-30 18:41:25 -07001471 if (!ret)
Roland McGrath98abed02008-03-19 19:24:59 -07001472 return ret;
1473
Roland McGrath14dd0b82008-03-30 18:41:25 -07001474 if (unlikely(ret < 0)) {
1475 /*
1476 * If we have not yet seen any eligible child,
1477 * then let this error code replace -ECHILD.
1478 * A permission error will give the user a clue
1479 * to look for security policy problems, rather
1480 * than for mysterious wait bugs.
1481 */
1482 if (*notask_error)
1483 *notask_error = ret;
1484 }
1485
Roland McGrathf4700212008-03-24 18:36:23 -07001486 if (likely(!ptrace) && unlikely(p->ptrace)) {
1487 /*
1488 * This child is hidden by ptrace.
1489 * We aren't allowed to see it now, but eventually we will.
1490 */
1491 *notask_error = 0;
1492 return 0;
1493 }
1494
Roland McGrath98abed02008-03-19 19:24:59 -07001495 if (p->exit_state == EXIT_DEAD)
1496 return 0;
1497
1498 /*
1499 * We don't reap group leaders with subthreads.
1500 */
1501 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1502 return wait_task_zombie(p, options, infop, stat_addr, ru);
1503
1504 /*
1505 * It's stopped or running now, so it might
1506 * later continue, exit, or stop again.
1507 */
1508 *notask_error = 0;
1509
Oleg Nesterov90bc8d82009-04-02 16:57:58 -07001510 if (task_stopped_code(p, ptrace))
Roland McGrathf4700212008-03-24 18:36:23 -07001511 return wait_task_stopped(ptrace, p, options,
1512 infop, stat_addr, ru);
Roland McGrath98abed02008-03-19 19:24:59 -07001513
1514 return wait_task_continued(p, options, infop, stat_addr, ru);
1515}
1516
1517/*
1518 * Do the work of do_wait() for one thread in the group, @tsk.
1519 *
1520 * -ECHILD should be in *@notask_error before the first call.
1521 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1522 * Returns zero if the search for a child should continue; then
Roland McGrath14dd0b82008-03-30 18:41:25 -07001523 * *@notask_error is 0 if there were any eligible children,
1524 * or another error from security_task_wait(), or still -ECHILD.
Roland McGrath98abed02008-03-19 19:24:59 -07001525 */
1526static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1527 enum pid_type type, struct pid *pid, int options,
1528 struct siginfo __user *infop, int __user *stat_addr,
1529 struct rusage __user *ru)
1530{
1531 struct task_struct *p;
1532
1533 list_for_each_entry(p, &tsk->children, sibling) {
Roland McGrathf4700212008-03-24 18:36:23 -07001534 /*
1535 * Do not consider detached threads.
1536 */
1537 if (!task_detached(p)) {
1538 int ret = wait_consider_task(tsk, 0, p, notask_error,
1539 type, pid, options,
1540 infop, stat_addr, ru);
1541 if (ret)
1542 return ret;
1543 }
Roland McGrath98abed02008-03-19 19:24:59 -07001544 }
1545
1546 return 0;
1547}
1548
1549static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1550 enum pid_type type, struct pid *pid, int options,
1551 struct siginfo __user *infop, int __user *stat_addr,
1552 struct rusage __user *ru)
1553{
1554 struct task_struct *p;
1555
1556 /*
Roland McGrathf4700212008-03-24 18:36:23 -07001557 * Traditionally we see ptrace'd stopped tasks regardless of options.
Roland McGrath98abed02008-03-19 19:24:59 -07001558 */
Roland McGrathf4700212008-03-24 18:36:23 -07001559 options |= WUNTRACED;
Roland McGrath98abed02008-03-19 19:24:59 -07001560
Roland McGrathf4700212008-03-24 18:36:23 -07001561 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1562 int ret = wait_consider_task(tsk, 1, p, notask_error,
1563 type, pid, options,
1564 infop, stat_addr, ru);
1565 if (ret)
Roland McGrath98abed02008-03-19 19:24:59 -07001566 return ret;
Roland McGrath98abed02008-03-19 19:24:59 -07001567 }
1568
1569 return 0;
1570}
1571
Eric W. Biederman161550d2008-02-08 04:19:14 -08001572static long do_wait(enum pid_type type, struct pid *pid, int options,
1573 struct siginfo __user *infop, int __user *stat_addr,
1574 struct rusage __user *ru)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
1576 DECLARE_WAITQUEUE(wait, current);
1577 struct task_struct *tsk;
Roland McGrath98abed02008-03-19 19:24:59 -07001578 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Mathieu Desnoyers0a16b602008-07-18 12:16:17 -04001580 trace_sched_process_wait(pid);
1581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 add_wait_queue(&current->signal->wait_chldexit,&wait);
1583repeat:
Roland McGrath98abed02008-03-19 19:24:59 -07001584 /*
1585 * If there is nothing that can match our critiera just get out.
1586 * We will clear @retval to zero if we see any child that might later
1587 * match our criteria, even if we are not able to reap it yet.
1588 */
Eric W. Biederman161550d2008-02-08 04:19:14 -08001589 retval = -ECHILD;
1590 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1591 goto end;
1592
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 current->state = TASK_INTERRUPTIBLE;
1594 read_lock(&tasklist_lock);
1595 tsk = current;
1596 do {
Roland McGrath98abed02008-03-19 19:24:59 -07001597 int tsk_result = do_wait_thread(tsk, &retval,
1598 type, pid, options,
1599 infop, stat_addr, ru);
1600 if (!tsk_result)
1601 tsk_result = ptrace_do_wait(tsk, &retval,
1602 type, pid, options,
1603 infop, stat_addr, ru);
1604 if (tsk_result) {
1605 /*
1606 * tasklist_lock is unlocked and we have a final result.
1607 */
1608 retval = tsk_result;
1609 goto end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 }
Roland McGrath98abed02008-03-19 19:24:59 -07001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (options & __WNOTHREAD)
1613 break;
1614 tsk = next_thread(tsk);
Eric Sesterhenn125e1872006-06-23 02:06:06 -07001615 BUG_ON(tsk->signal != current->signal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 } while (tsk != current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 read_unlock(&tasklist_lock);
Oleg Nesterovf2cc3eb2008-02-08 04:19:06 -08001618
Roland McGrath98abed02008-03-19 19:24:59 -07001619 if (!retval && !(options & WNOHANG)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 retval = -ERESTARTSYS;
Roland McGrath98abed02008-03-19 19:24:59 -07001621 if (!signal_pending(current)) {
1622 schedule();
1623 goto repeat;
1624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
Roland McGrath98abed02008-03-19 19:24:59 -07001626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627end:
1628 current->state = TASK_RUNNING;
1629 remove_wait_queue(&current->signal->wait_chldexit,&wait);
1630 if (infop) {
1631 if (retval > 0)
Oleg Nesterov9cbab812008-02-08 04:19:02 -08001632 retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 else {
1634 /*
1635 * For a WNOHANG return, clear out all the fields
1636 * we would set so the user can easily tell the
1637 * difference.
1638 */
1639 if (!retval)
1640 retval = put_user(0, &infop->si_signo);
1641 if (!retval)
1642 retval = put_user(0, &infop->si_errno);
1643 if (!retval)
1644 retval = put_user(0, &infop->si_code);
1645 if (!retval)
1646 retval = put_user(0, &infop->si_pid);
1647 if (!retval)
1648 retval = put_user(0, &infop->si_uid);
1649 if (!retval)
1650 retval = put_user(0, &infop->si_status);
1651 }
1652 }
1653 return retval;
1654}
1655
Heiko Carstens17da2bd2009-01-14 14:14:10 +01001656SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1657 infop, int, options, struct rusage __user *, ru)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
Eric W. Biederman161550d2008-02-08 04:19:14 -08001659 struct pid *pid = NULL;
1660 enum pid_type type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 long ret;
1662
1663 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1664 return -EINVAL;
1665 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1666 return -EINVAL;
1667
1668 switch (which) {
1669 case P_ALL:
Eric W. Biederman161550d2008-02-08 04:19:14 -08001670 type = PIDTYPE_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 break;
1672 case P_PID:
Eric W. Biederman161550d2008-02-08 04:19:14 -08001673 type = PIDTYPE_PID;
1674 if (upid <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 return -EINVAL;
1676 break;
1677 case P_PGID:
Eric W. Biederman161550d2008-02-08 04:19:14 -08001678 type = PIDTYPE_PGID;
1679 if (upid <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 break;
1682 default:
1683 return -EINVAL;
1684 }
1685
Eric W. Biederman161550d2008-02-08 04:19:14 -08001686 if (type < PIDTYPE_MAX)
1687 pid = find_get_pid(upid);
1688 ret = do_wait(type, pid, options, infop, NULL, ru);
1689 put_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 /* avoid REGPARM breakage on x86: */
Roland McGrath54a01512008-04-10 15:37:38 -07001692 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 return ret;
1694}
1695
Heiko Carstens754fe8d2009-01-14 14:14:09 +01001696SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1697 int, options, struct rusage __user *, ru)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698{
Eric W. Biederman161550d2008-02-08 04:19:14 -08001699 struct pid *pid = NULL;
1700 enum pid_type type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 long ret;
1702
1703 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1704 __WNOTHREAD|__WCLONE|__WALL))
1705 return -EINVAL;
Eric W. Biederman161550d2008-02-08 04:19:14 -08001706
1707 if (upid == -1)
1708 type = PIDTYPE_MAX;
1709 else if (upid < 0) {
1710 type = PIDTYPE_PGID;
1711 pid = find_get_pid(-upid);
1712 } else if (upid == 0) {
1713 type = PIDTYPE_PGID;
Oleg Nesterov2ae448e2009-04-02 16:58:36 -07001714 pid = get_task_pid(current, PIDTYPE_PGID);
Eric W. Biederman161550d2008-02-08 04:19:14 -08001715 } else /* upid > 0 */ {
1716 type = PIDTYPE_PID;
1717 pid = find_get_pid(upid);
1718 }
1719
1720 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1721 put_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723 /* avoid REGPARM breakage on x86: */
Roland McGrath54a01512008-04-10 15:37:38 -07001724 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 return ret;
1726}
1727
1728#ifdef __ARCH_WANT_SYS_WAITPID
1729
1730/*
1731 * sys_waitpid() remains for compatibility. waitpid() should be
1732 * implemented by calling sys_wait4() from libc.a.
1733 */
Heiko Carstens17da2bd2009-01-14 14:14:10 +01001734SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735{
1736 return sys_wait4(pid, stat_addr, options, NULL);
1737}
1738
1739#endif