blob: 28d9feedfd27a4633f68e172f8ab83c78e827116 [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>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080016#include <linux/mnt_namespace.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>
23#include <linux/binfmts.h>
Serge E. Hallynab516012006-10-02 02:18:06 -070024#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/ptrace.h>
26#include <linux/profile.h>
27#include <linux/mount.h>
28#include <linux/proc_fs.h>
29#include <linux/mempolicy.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070030#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070031#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/cpuset.h>
33#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070034#include <linux/signal.h>
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080035#include <linux/posix-timers.h>
Matt Helsley9f460802005-11-07 00:59:16 -080036#include <linux/cn_proc.h>
Ingo Molnarde5097c2006-01-09 15:59:21 -080037#include <linux/mutex.h>
Ingo Molnar0771dfe2006-03-27 01:16:22 -080038#include <linux/futex.h>
Ingo Molnar34f192c2006-03-27 01:16:24 -080039#include <linux/compat.h>
Jens Axboeb92ce552006-04-11 13:52:07 +020040#include <linux/pipe_fs_i.h>
Al Virofa84cb92006-03-29 20:30:19 -050041#include <linux/audit.h> /* for audit_free() */
Adrian Bunk83cc5ed2006-06-25 05:47:41 -070042#include <linux/resource.h>
David Howells0d67a462006-08-29 19:05:56 +010043#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <asm/uaccess.h>
46#include <asm/unistd.h>
47#include <asm/pgtable.h>
48#include <asm/mmu_context.h>
49
50extern void sem_exit (void);
51extern struct task_struct *child_reaper;
52
Adrian Bunk408b6642005-05-01 08:59:29 -070053static void exit_mm(struct task_struct * tsk);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static void __unhash_process(struct task_struct *p)
56{
57 nr_threads--;
58 detach_pid(p, PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 if (thread_group_leader(p)) {
60 detach_pid(p, PIDTYPE_PGID);
61 detach_pid(p, PIDTYPE_SID);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080062
Eric W. Biederman5e85d4a2006-04-18 22:20:16 -070063 list_del_rcu(&p->tasks);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -080064 __get_cpu_var(process_counts)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 }
Oleg Nesterov47e65322006-03-28 16:11:25 -080066 list_del_rcu(&p->thread_group);
Oleg Nesterovc97d9892006-03-28 16:11:06 -080067 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068}
69
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080070/*
71 * This function expects the tasklist_lock write-locked.
72 */
73static void __exit_signal(struct task_struct *tsk)
74{
75 struct signal_struct *sig = tsk->signal;
76 struct sighand_struct *sighand;
77
78 BUG_ON(!sig);
79 BUG_ON(!atomic_read(&sig->count));
80
81 rcu_read_lock();
82 sighand = rcu_dereference(tsk->sighand);
83 spin_lock(&sighand->siglock);
84
85 posix_cpu_timers_exit(tsk);
86 if (atomic_dec_and_test(&sig->count))
87 posix_cpu_timers_exit_group(tsk);
88 else {
89 /*
90 * If there is any task waiting for the group exit
91 * then notify it:
92 */
93 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
94 wake_up_process(sig->group_exit_task);
95 sig->group_exit_task = NULL;
96 }
97 if (tsk == sig->curr_target)
98 sig->curr_target = next_thread(tsk);
99 /*
100 * Accumulate here the counters for all threads but the
101 * group leader as they die, so they can be added into
102 * the process-wide totals when those are taken.
103 * The group leader stays around as a zombie as long
104 * as there are other threads. When it gets reaped,
105 * the exit.c code will add its counts into these totals.
106 * We won't ever get here for the group leader, since it
107 * will have been the last reference on the signal_struct.
108 */
109 sig->utime = cputime_add(sig->utime, tsk->utime);
110 sig->stime = cputime_add(sig->stime, tsk->stime);
111 sig->min_flt += tsk->min_flt;
112 sig->maj_flt += tsk->maj_flt;
113 sig->nvcsw += tsk->nvcsw;
114 sig->nivcsw += tsk->nivcsw;
115 sig->sched_time += tsk->sched_time;
116 sig = NULL; /* Marker for below. */
117 }
118
Oleg Nesterov58767002006-03-28 16:11:20 -0800119 __unhash_process(tsk);
120
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800121 tsk->signal = NULL;
Oleg Nesterova7e53282006-03-28 16:11:27 -0800122 tsk->sighand = NULL;
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800123 spin_unlock(&sighand->siglock);
124 rcu_read_unlock();
125
Oleg Nesterova7e53282006-03-28 16:11:27 -0800126 __cleanup_sighand(sighand);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800127 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
128 flush_sigqueue(&tsk->pending);
129 if (sig) {
130 flush_sigqueue(&sig->shared_pending);
Oleg Nesterov093a8e82006-10-28 10:38:51 -0700131 taskstats_tgid_free(sig);
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -0800132 __cleanup_signal(sig);
133 }
134}
135
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800136static void delayed_put_task_struct(struct rcu_head *rhp)
137{
138 put_task_struct(container_of(rhp, struct task_struct, rcu));
139}
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141void release_task(struct task_struct * p)
142{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700143 struct task_struct *leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 int zap_leader;
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800145repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 atomic_dec(&p->user->processes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 write_lock_irq(&tasklist_lock);
Oleg Nesterov1f09f972006-03-28 16:11:11 -0800148 ptrace_unlink(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
150 __exit_signal(p);
Oleg Nesterov35f5cad2006-03-28 16:11:19 -0800151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 /*
153 * If we are the last non-leader member of the thread
154 * group, and the leader is zombie, then notify the
155 * group leader's parent process. (if it wants notification.)
156 */
157 zap_leader = 0;
158 leader = p->group_leader;
159 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
160 BUG_ON(leader->exit_signal == -1);
161 do_notify_parent(leader, leader->exit_signal);
162 /*
163 * If we were the last child thread and the leader has
164 * exited already, and the leader's parent ignores SIGCHLD,
165 * then we are the one who should release the leader.
166 *
167 * do_notify_parent() will have marked it self-reaping in
168 * that case.
169 */
170 zap_leader = (leader->exit_signal == -1);
171 }
172
173 sched_exit(p);
174 write_unlock_irq(&tasklist_lock);
Eric W. Biederman48e64842006-06-26 00:25:48 -0700175 proc_flush_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 release_thread(p);
Eric W. Biederman8c7904a2006-03-31 02:31:37 -0800177 call_rcu(&p->rcu, delayed_put_task_struct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 p = leader;
180 if (unlikely(zap_leader))
181 goto repeat;
182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/*
185 * This checks not only the pgrp, but falls back on the pid if no
186 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
187 * without this...
188 */
189int session_of_pgrp(int pgrp)
190{
191 struct task_struct *p;
192 int sid = -1;
193
194 read_lock(&tasklist_lock);
195 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
Cedric Le Goater937949d2006-12-08 02:37:54 -0800196 if (process_session(p) > 0) {
197 sid = process_session(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 goto out;
199 }
200 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
201 p = find_task_by_pid(pgrp);
202 if (p)
Cedric Le Goater937949d2006-12-08 02:37:54 -0800203 sid = process_session(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204out:
205 read_unlock(&tasklist_lock);
206
207 return sid;
208}
209
210/*
211 * Determine if a process group is "orphaned", according to the POSIX
212 * definition in 2.2.2.52. Orphaned process groups are not to be affected
213 * by terminal-generated stop signals. Newly orphaned process groups are
214 * to receive a SIGHUP and a SIGCONT.
215 *
216 * "I ask you, have you ever known what it is to be an orphan?"
217 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700218static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 struct task_struct *p;
221 int ret = 1;
222
223 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
224 if (p == ignored_task
225 || p->exit_state
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700226 || is_init(p->real_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 continue;
Cedric Le Goater937949d2006-12-08 02:37:54 -0800228 if (process_group(p->real_parent) != pgrp &&
229 process_session(p->real_parent) == process_session(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 ret = 0;
231 break;
232 }
233 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
234 return ret; /* (sighing) "Often!" */
235}
236
237int is_orphaned_pgrp(int pgrp)
238{
239 int retval;
240
241 read_lock(&tasklist_lock);
242 retval = will_become_orphaned_pgrp(pgrp, NULL);
243 read_unlock(&tasklist_lock);
244
245 return retval;
246}
247
Arjan van de Ven858119e2006-01-14 13:20:43 -0800248static int has_stopped_jobs(int pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
250 int retval = 0;
251 struct task_struct *p;
252
253 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
254 if (p->state != TASK_STOPPED)
255 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 retval = 1;
257 break;
258 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
259 return retval;
260}
261
262/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700263 * reparent_to_init - Reparent the calling kernel thread to the init task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 *
265 * If a kernel thread is launched as a result of a system call, or if
266 * it ever exits, it should generally reparent itself to init so that
267 * it is correctly cleaned up on exit.
268 *
269 * The various task state such as scheduling policy and priority may have
270 * been inherited from a user process, so we reset them to sane values here.
271 *
272 * NOTE that reparent_to_init() gives the caller full capabilities.
273 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800274static void reparent_to_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 write_lock_irq(&tasklist_lock);
277
278 ptrace_unlink(current);
279 /* Reparent to init */
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800280 remove_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 current->parent = child_reaper;
282 current->real_parent = child_reaper;
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800283 add_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 /* Set the exit signal to SIGCHLD so we signal init on exit */
286 current->exit_signal = SIGCHLD;
287
Oleg Nesterov1c573af2006-09-29 02:00:51 -0700288 if (!has_rt_policy(current) && (task_nice(current) < 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 set_user_nice(current, 0);
290 /* cpus_allowed? */
291 /* rt_priority? */
292 /* signals? */
293 security_task_reparent_to_init(current);
294 memcpy(current->signal->rlim, init_task.signal->rlim,
295 sizeof(current->signal->rlim));
296 atomic_inc(&(INIT_USER->__count));
297 write_unlock_irq(&tasklist_lock);
298 switch_uid(INIT_USER);
299}
300
301void __set_special_pids(pid_t session, pid_t pgrp)
302{
Oren Laadane19f2472006-01-08 01:03:58 -0800303 struct task_struct *curr = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Cedric Le Goater937949d2006-12-08 02:37:54 -0800305 if (process_session(curr) != session) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 detach_pid(curr, PIDTYPE_SID);
Cedric Le Goater1ec320a2006-12-08 02:37:55 -0800307 set_signal_session(curr->signal, session);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 attach_pid(curr, PIDTYPE_SID, session);
309 }
310 if (process_group(curr) != pgrp) {
311 detach_pid(curr, PIDTYPE_PGID);
312 curr->signal->pgrp = pgrp;
313 attach_pid(curr, PIDTYPE_PGID, pgrp);
314 }
315}
316
Oleg Nesterovae424ae2006-12-08 02:36:08 -0800317static void set_special_pids(pid_t session, pid_t pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 write_lock_irq(&tasklist_lock);
320 __set_special_pids(session, pgrp);
321 write_unlock_irq(&tasklist_lock);
322}
323
324/*
325 * Let kernel threads use this to say that they
326 * allow a certain signal (since daemonize() will
327 * have disabled all of them by default).
328 */
329int allow_signal(int sig)
330{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700331 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return -EINVAL;
333
334 spin_lock_irq(&current->sighand->siglock);
335 sigdelset(&current->blocked, sig);
336 if (!current->mm) {
337 /* Kernel threads handle their own signals.
338 Let the signal code know it'll be handled, so
339 that they don't get converted to SIGKILL or
340 just silently dropped */
341 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
342 }
343 recalc_sigpending();
344 spin_unlock_irq(&current->sighand->siglock);
345 return 0;
346}
347
348EXPORT_SYMBOL(allow_signal);
349
350int disallow_signal(int sig)
351{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700352 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return -EINVAL;
354
355 spin_lock_irq(&current->sighand->siglock);
356 sigaddset(&current->blocked, sig);
357 recalc_sigpending();
358 spin_unlock_irq(&current->sighand->siglock);
359 return 0;
360}
361
362EXPORT_SYMBOL(disallow_signal);
363
364/*
365 * Put all the gunge required to become a kernel thread without
366 * attached user resources in one place where it belongs.
367 */
368
369void daemonize(const char *name, ...)
370{
371 va_list args;
372 struct fs_struct *fs;
373 sigset_t blocked;
374
375 va_start(args, name);
376 vsnprintf(current->comm, sizeof(current->comm), name, args);
377 va_end(args);
378
379 /*
380 * If we were started as result of loading a module, close all of the
381 * user space pages. We don't need them, and if we didn't close them
382 * they would be locked into memory.
383 */
384 exit_mm(current);
385
386 set_special_pids(1, 1);
Peter Zijlstra24ec8392006-12-08 02:36:04 -0800387 proc_clear_tty(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 /* Block and flush all signals */
390 sigfillset(&blocked);
391 sigprocmask(SIG_BLOCK, &blocked, NULL);
392 flush_signals(current);
393
394 /* Become as one with the init task */
395
396 exit_fs(current); /* current->fs->count--; */
397 fs = init_task.fs;
398 current->fs = fs;
399 atomic_inc(&fs->count);
Serge E. Hallynab516012006-10-02 02:18:06 -0700400
Serge E. Hallynab516012006-10-02 02:18:06 -0700401 exit_task_namespaces(current);
Serge E. Hallynab516012006-10-02 02:18:06 -0700402 current->nsproxy = init_task.nsproxy;
Serge E. Hallynab516012006-10-02 02:18:06 -0700403 get_task_namespaces(current);
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 exit_files(current);
406 current->files = init_task.files;
407 atomic_inc(&current->files->count);
408
409 reparent_to_init();
410}
411
412EXPORT_SYMBOL(daemonize);
413
Arjan van de Ven858119e2006-01-14 13:20:43 -0800414static void close_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700417 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700420
421 /*
422 * It is safe to dereference the fd table without RCU or
423 * ->file_lock because this is the last reference to the
424 * files structure.
425 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700426 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 for (;;) {
428 unsigned long set;
429 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700430 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700432 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 while (set) {
434 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700435 struct file * file = xchg(&fdt->fd[i], NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 if (file)
437 filp_close(file, files);
438 }
439 i++;
440 set >>= 1;
441 }
442 }
443}
444
445struct files_struct *get_files_struct(struct task_struct *task)
446{
447 struct files_struct *files;
448
449 task_lock(task);
450 files = task->files;
451 if (files)
452 atomic_inc(&files->count);
453 task_unlock(task);
454
455 return files;
456}
457
458void fastcall put_files_struct(struct files_struct *files)
459{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700460 struct fdtable *fdt;
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (atomic_dec_and_test(&files->count)) {
463 close_files(files);
464 /*
465 * Free the fd and fdset arrays if we expanded them.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700466 * If the fdtable was embedded, pass files for freeing
467 * at the end of the RCU grace period. Otherwise,
468 * you can free files immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700470 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700471 if (fdt == &files->fdtab)
472 fdt->free_files = files;
473 else
474 kmem_cache_free(files_cachep, files);
475 free_fdtable(fdt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
477}
478
479EXPORT_SYMBOL(put_files_struct);
480
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700481void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
482{
483 struct files_struct *old;
484
485 old = tsk->files;
486 task_lock(tsk);
487 tsk->files = files;
488 task_unlock(tsk);
489 put_files_struct(old);
490}
491EXPORT_SYMBOL(reset_files_struct);
492
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493static inline void __exit_files(struct task_struct *tsk)
494{
495 struct files_struct * files = tsk->files;
496
497 if (files) {
498 task_lock(tsk);
499 tsk->files = NULL;
500 task_unlock(tsk);
501 put_files_struct(files);
502 }
503}
504
505void exit_files(struct task_struct *tsk)
506{
507 __exit_files(tsk);
508}
509
510static inline void __put_fs_struct(struct fs_struct *fs)
511{
512 /* No need to hold fs->lock if we are killing it */
513 if (atomic_dec_and_test(&fs->count)) {
514 dput(fs->root);
515 mntput(fs->rootmnt);
516 dput(fs->pwd);
517 mntput(fs->pwdmnt);
518 if (fs->altroot) {
519 dput(fs->altroot);
520 mntput(fs->altrootmnt);
521 }
522 kmem_cache_free(fs_cachep, fs);
523 }
524}
525
526void put_fs_struct(struct fs_struct *fs)
527{
528 __put_fs_struct(fs);
529}
530
531static inline void __exit_fs(struct task_struct *tsk)
532{
533 struct fs_struct * fs = tsk->fs;
534
535 if (fs) {
536 task_lock(tsk);
537 tsk->fs = NULL;
538 task_unlock(tsk);
539 __put_fs_struct(fs);
540 }
541}
542
543void exit_fs(struct task_struct *tsk)
544{
545 __exit_fs(tsk);
546}
547
548EXPORT_SYMBOL_GPL(exit_fs);
549
550/*
551 * Turn us into a lazy TLB process if we
552 * aren't already..
553 */
Adrian Bunk408b6642005-05-01 08:59:29 -0700554static void exit_mm(struct task_struct * tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 struct mm_struct *mm = tsk->mm;
557
558 mm_release(tsk, mm);
559 if (!mm)
560 return;
561 /*
562 * Serialize with any possible pending coredump.
563 * We must hold mmap_sem around checking core_waiters
564 * and clearing tsk->mm. The core-inducing thread
565 * will increment core_waiters for each thread in the
566 * group with ->mm != NULL.
567 */
568 down_read(&mm->mmap_sem);
569 if (mm->core_waiters) {
570 up_read(&mm->mmap_sem);
571 down_write(&mm->mmap_sem);
572 if (!--mm->core_waiters)
573 complete(mm->core_startup_done);
574 up_write(&mm->mmap_sem);
575
576 wait_for_completion(&mm->core_done);
577 down_read(&mm->mmap_sem);
578 }
579 atomic_inc(&mm->mm_count);
Eric Sesterhenn125e1872006-06-23 02:06:06 -0700580 BUG_ON(mm != tsk->active_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 /* more a memory barrier than a real lock */
582 task_lock(tsk);
583 tsk->mm = NULL;
584 up_read(&mm->mmap_sem);
585 enter_lazy_tlb(mm, current);
586 task_unlock(tsk);
587 mmput(mm);
588}
589
Ingo Molnar36c8b582006-07-03 00:25:41 -0700590static inline void
591choose_new_parent(struct task_struct *p, struct task_struct *reaper)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
593 /*
594 * Make sure we're not reparenting to ourselves and that
595 * the parent is not a zombie.
596 */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800597 BUG_ON(p == reaper || reaper->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 p->real_parent = reaper;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
Ingo Molnar36c8b582006-07-03 00:25:41 -0700601static void
602reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 /* We don't want people slaying init. */
605 if (p->exit_signal != -1)
606 p->exit_signal = SIGCHLD;
607
608 if (p->pdeath_signal)
609 /* We already hold the tasklist_lock here. */
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800610 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 /* Move the child from its dying parent to the new one. */
613 if (unlikely(traced)) {
614 /* Preserve ptrace links if someone else is tracing this child. */
615 list_del_init(&p->ptrace_list);
616 if (p->parent != p->real_parent)
617 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
618 } else {
619 /* If this child is being traced, then we're the one tracing it
620 * anyway, so let go of it.
621 */
622 p->ptrace = 0;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800623 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 p->parent = p->real_parent;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800625 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 /* If we'd notified the old parent about this child's death,
628 * also notify the new parent.
629 */
630 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
631 thread_group_empty(p))
632 do_notify_parent(p, p->exit_signal);
633 else if (p->state == TASK_TRACED) {
634 /*
635 * If it was at a trace stop, turn it into
636 * a normal stop since it's no longer being
637 * traced.
638 */
639 ptrace_untrace(p);
640 }
641 }
642
643 /*
644 * process group orphan check
645 * Case ii: Our child is in a different pgrp
646 * than we are, and it was the only connection
647 * outside, so the child pgrp is now orphaned.
648 */
649 if ((process_group(p) != process_group(father)) &&
Cedric Le Goater937949d2006-12-08 02:37:54 -0800650 (process_session(p) == process_session(father))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 int pgrp = process_group(p);
652
Cedric Le Goater937949d2006-12-08 02:37:54 -0800653 if (will_become_orphaned_pgrp(pgrp, NULL) &&
654 has_stopped_jobs(pgrp)) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800655 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
656 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 }
658 }
659}
660
661/*
662 * When we die, we re-parent all our children.
663 * Try to give them to another thread in our thread
664 * group, and if no such member exists, give it to
665 * the global child reaper process (ie "init")
666 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700667static void
668forget_original_parent(struct task_struct *father, struct list_head *to_release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 struct task_struct *p, *reaper = father;
671 struct list_head *_p, *_n;
672
673 do {
674 reaper = next_thread(reaper);
675 if (reaper == father) {
676 reaper = child_reaper;
677 break;
678 }
679 } while (reaper->exit_state);
680
681 /*
682 * There are only two places where our children can be:
683 *
684 * - in our child list
685 * - in our ptraced child list
686 *
687 * Search them and reparent children.
688 */
689 list_for_each_safe(_p, _n, &father->children) {
690 int ptrace;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700691 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 ptrace = p->ptrace;
694
695 /* if father isn't the real parent, then ptrace must be enabled */
696 BUG_ON(father != p->real_parent && !ptrace);
697
698 if (father == p->real_parent) {
699 /* reparent with a reaper, real father it's us */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800700 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 reparent_thread(p, father, 0);
702 } else {
703 /* reparent ptraced task to its real parent */
704 __ptrace_unlink (p);
705 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
706 thread_group_empty(p))
707 do_notify_parent(p, p->exit_signal);
708 }
709
710 /*
711 * if the ptraced child is a zombie with exit_signal == -1
712 * we must collect it before we exit, or it will remain
713 * zombie forever since we prevented it from self-reap itself
714 * while it was being traced by us, to be able to see it in wait4.
715 */
716 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
717 list_add(&p->ptrace_list, to_release);
718 }
719 list_for_each_safe(_p, _n, &father->ptrace_children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -0700720 p = list_entry(_p, struct task_struct, ptrace_list);
Oleg Nesterovd799f032006-03-28 16:11:04 -0800721 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 reparent_thread(p, father, 1);
723 }
724}
725
726/*
727 * Send signals to all our closest relatives so that they know
728 * to properly mourn us..
729 */
730static void exit_notify(struct task_struct *tsk)
731{
732 int state;
733 struct task_struct *t;
734 struct list_head ptrace_dead, *_p, *_n;
735
736 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
737 && !thread_group_empty(tsk)) {
738 /*
739 * This occurs when there was a race between our exit
740 * syscall and a group signal choosing us as the one to
741 * wake up. It could be that we are the only thread
742 * alerted to check for pending signals, but another thread
743 * should be woken now to take the signal since we will not.
744 * Now we'll wake all the threads in the group just to make
745 * sure someone gets all the pending signals.
746 */
747 read_lock(&tasklist_lock);
748 spin_lock_irq(&tsk->sighand->siglock);
749 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
750 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
751 recalc_sigpending_tsk(t);
752 if (signal_pending(t))
753 signal_wake_up(t, 0);
754 }
755 spin_unlock_irq(&tsk->sighand->siglock);
756 read_unlock(&tasklist_lock);
757 }
758
759 write_lock_irq(&tasklist_lock);
760
761 /*
762 * This does two things:
763 *
764 * A. Make init inherit all the child processes
765 * B. Check to see if any process groups have become orphaned
766 * as a result of our exiting, and if they have any stopped
767 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
768 */
769
770 INIT_LIST_HEAD(&ptrace_dead);
771 forget_original_parent(tsk, &ptrace_dead);
772 BUG_ON(!list_empty(&tsk->children));
773 BUG_ON(!list_empty(&tsk->ptrace_children));
774
775 /*
776 * Check to see if any process groups have become orphaned
777 * as a result of our exiting, and if they have any stopped
778 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
779 *
780 * Case i: Our father is in a different pgrp than we are
781 * and we were the only connection outside, so our pgrp
782 * is about to become orphaned.
783 */
784
785 t = tsk->real_parent;
786
787 if ((process_group(t) != process_group(tsk)) &&
Cedric Le Goater937949d2006-12-08 02:37:54 -0800788 (process_session(t) == process_session(tsk)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
790 has_stopped_jobs(process_group(tsk))) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800791 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
792 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 }
794
795 /* Let father know we died
796 *
797 * Thread signals are configurable, but you aren't going to use
798 * that to send signals to arbitary processes.
799 * That stops right now.
800 *
801 * If the parent exec id doesn't match the exec id we saved
802 * when we started then we know the parent has changed security
803 * domain.
804 *
805 * If our self_exec id doesn't match our parent_exec_id then
806 * we have changed execution domain as these two values started
807 * the same after a fork.
808 *
809 */
810
811 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
812 ( tsk->parent_exec_id != t->self_exec_id ||
813 tsk->self_exec_id != tsk->parent_exec_id)
814 && !capable(CAP_KILL))
815 tsk->exit_signal = SIGCHLD;
816
817
818 /* If something other than our normal parent is ptracing us, then
819 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
820 * only has special meaning to our real parent.
821 */
822 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
823 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
824 do_notify_parent(tsk, signal);
825 } else if (tsk->ptrace) {
826 do_notify_parent(tsk, SIGCHLD);
827 }
828
829 state = EXIT_ZOMBIE;
830 if (tsk->exit_signal == -1 &&
831 (likely(tsk->ptrace == 0) ||
832 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
833 state = EXIT_DEAD;
834 tsk->exit_state = state;
835
836 write_unlock_irq(&tasklist_lock);
837
838 list_for_each_safe(_p, _n, &ptrace_dead) {
839 list_del_init(_p);
Ingo Molnar36c8b582006-07-03 00:25:41 -0700840 t = list_entry(_p, struct task_struct, ptrace_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 release_task(t);
842 }
843
844 /* If the process is dead, release it - nobody will wait for it */
845 if (state == EXIT_DEAD)
846 release_task(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
848
849fastcall NORET_TYPE void do_exit(long code)
850{
851 struct task_struct *tsk = current;
852 int group_dead;
853
854 profile_task_exit(tsk);
855
Jens Axboe22e2c502005-06-27 10:55:12 +0200856 WARN_ON(atomic_read(&tsk->fs_excl));
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (unlikely(in_interrupt()))
859 panic("Aiee, killing interrupt handler!");
860 if (unlikely(!tsk->pid))
861 panic("Attempted to kill the idle task!");
Eric W. Biedermanfef23e72006-03-28 16:10:58 -0800862 if (unlikely(tsk == child_reaper))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 panic("Attempted to kill init!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
866 current->ptrace_message = code;
867 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
868 }
869
Alexander Nybergdf164db2005-06-23 00:09:13 -0700870 /*
871 * We're taking recursive faults here in do_exit. Safest is to just
872 * leave this task alone and wait for reboot.
873 */
874 if (unlikely(tsk->flags & PF_EXITING)) {
875 printk(KERN_ALERT
876 "Fixing recursive fault but reboot is needed!\n");
Al Viroafc847b2006-02-28 12:51:55 -0500877 if (tsk->io_context)
878 exit_io_context();
Alexander Nybergdf164db2005-06-23 00:09:13 -0700879 set_current_state(TASK_UNINTERRUPTIBLE);
880 schedule();
881 }
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 tsk->flags |= PF_EXITING;
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (unlikely(in_atomic()))
886 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
887 current->comm, current->pid,
888 preempt_count());
889
890 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);
Oleg Nesterov115085e2006-12-06 20:36:51 -0800909
910 taskstats_exit(tsk, group_dead);
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 exit_thread();
920 cpuset_exit(tsk);
921 exit_keys(tsk);
922
923 if (group_dead && tsk->signal->leader)
924 disassociate_ctty(1);
925
Al Viroa1261f542005-11-13 16:06:55 -0800926 module_put(task_thread_info(tsk)->exec_domain->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (tsk->binfmt)
928 module_put(tsk->binfmt->module);
929
930 tsk->exit_code = code;
Matt Helsley9f460802005-11-07 00:59:16 -0800931 proc_exit_connector(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 exit_notify(tsk);
Cedric Le Goaterfab413a2006-10-02 02:18:09 -0700933 exit_task_namespaces(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934#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