blob: 14f17033f5635cf41fedbaec35c92b14f1a91933 [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>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080025#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/ptrace.h>
27#include <linux/profile.h>
28#include <linux/mount.h>
29#include <linux/proc_fs.h>
30#include <linux/mempolicy.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070031#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070032#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/cpuset.h>
34#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070035#include <linux/signal.h>
Oleg Nesterov6a14c5c2006-03-28 16:11:18 -080036#include <linux/posix-timers.h>
Matt Helsley9f460802005-11-07 00:59:16 -080037#include <linux/cn_proc.h>
Ingo Molnarde5097c2006-01-09 15:59:21 -080038#include <linux/mutex.h>
Ingo Molnar0771dfe2006-03-27 01:16:22 -080039#include <linux/futex.h>
Ingo Molnar34f192c2006-03-27 01:16:24 -080040#include <linux/compat.h>
Jens Axboeb92ce552006-04-11 13:52:07 +020041#include <linux/pipe_fs_i.h>
Al Virofa84cb92006-03-29 20:30:19 -050042#include <linux/audit.h> /* for audit_free() */
Adrian Bunk83cc5ed2006-06-25 05:47:41 -070043#include <linux/resource.h>
David Howells0d67a462006-08-29 19:05:56 +010044#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#include <asm/uaccess.h>
47#include <asm/unistd.h>
48#include <asm/pgtable.h>
49#include <asm/mmu_context.h>
50
51extern void sem_exit (void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
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;
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800192 int sid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 read_lock(&tasklist_lock);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800195
196 p = find_task_by_pid_type(PIDTYPE_PGID, pgrp);
197 if (p == NULL)
198 p = find_task_by_pid(pgrp);
199 if (p != NULL)
Cedric Le Goater937949d2006-12-08 02:37:54 -0800200 sid = process_session(p);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 read_unlock(&tasklist_lock);
Oleg Nesterov62dfb552006-12-08 02:38:03 -0800203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 return sid;
205}
206
207/*
208 * Determine if a process group is "orphaned", according to the POSIX
209 * definition in 2.2.2.52. Orphaned process groups are not to be affected
210 * by terminal-generated stop signals. Newly orphaned process groups are
211 * to receive a SIGHUP and a SIGCONT.
212 *
213 * "I ask you, have you ever known what it is to be an orphan?"
214 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700215static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
217 struct task_struct *p;
218 int ret = 1;
219
220 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
221 if (p == ignored_task
222 || p->exit_state
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700223 || is_init(p->real_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 continue;
Cedric Le Goater937949d2006-12-08 02:37:54 -0800225 if (process_group(p->real_parent) != pgrp &&
226 process_session(p->real_parent) == process_session(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 ret = 0;
228 break;
229 }
230 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
231 return ret; /* (sighing) "Often!" */
232}
233
234int is_orphaned_pgrp(int pgrp)
235{
236 int retval;
237
238 read_lock(&tasklist_lock);
239 retval = will_become_orphaned_pgrp(pgrp, NULL);
240 read_unlock(&tasklist_lock);
241
242 return retval;
243}
244
Arjan van de Ven858119e2006-01-14 13:20:43 -0800245static int has_stopped_jobs(int pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 int retval = 0;
248 struct task_struct *p;
249
250 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
251 if (p->state != TASK_STOPPED)
252 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 retval = 1;
254 break;
255 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
256 return retval;
257}
258
259/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800260 * reparent_to_init - Reparent the calling kernel thread to the init task of the pid space that the thread belongs to.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * If a kernel thread is launched as a result of a system call, or if
263 * it ever exits, it should generally reparent itself to init so that
264 * it is correctly cleaned up on exit.
265 *
266 * The various task state such as scheduling policy and priority may have
267 * been inherited from a user process, so we reset them to sane values here.
268 *
269 * NOTE that reparent_to_init() gives the caller full capabilities.
270 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800271static void reparent_to_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 write_lock_irq(&tasklist_lock);
274
275 ptrace_unlink(current);
276 /* Reparent to init */
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800277 remove_parent(current);
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800278 current->parent = child_reaper(current);
279 current->real_parent = child_reaper(current);
Oleg Nesterov9b678ec2006-03-28 16:11:05 -0800280 add_parent(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282 /* Set the exit signal to SIGCHLD so we signal init on exit */
283 current->exit_signal = SIGCHLD;
284
Oleg Nesterov1c573af2006-09-29 02:00:51 -0700285 if (!has_rt_policy(current) && (task_nice(current) < 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 set_user_nice(current, 0);
287 /* cpus_allowed? */
288 /* rt_priority? */
289 /* signals? */
290 security_task_reparent_to_init(current);
291 memcpy(current->signal->rlim, init_task.signal->rlim,
292 sizeof(current->signal->rlim));
293 atomic_inc(&(INIT_USER->__count));
294 write_unlock_irq(&tasklist_lock);
295 switch_uid(INIT_USER);
296}
297
298void __set_special_pids(pid_t session, pid_t pgrp)
299{
Oren Laadane19f2472006-01-08 01:03:58 -0800300 struct task_struct *curr = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
Cedric Le Goater937949d2006-12-08 02:37:54 -0800302 if (process_session(curr) != session) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 detach_pid(curr, PIDTYPE_SID);
Cedric Le Goater1ec320a2006-12-08 02:37:55 -0800304 set_signal_session(curr->signal, session);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 attach_pid(curr, PIDTYPE_SID, session);
306 }
307 if (process_group(curr) != pgrp) {
308 detach_pid(curr, PIDTYPE_PGID);
309 curr->signal->pgrp = pgrp;
310 attach_pid(curr, PIDTYPE_PGID, pgrp);
311 }
312}
313
Oleg Nesterovae424ae2006-12-08 02:36:08 -0800314static void set_special_pids(pid_t session, pid_t pgrp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
316 write_lock_irq(&tasklist_lock);
317 __set_special_pids(session, pgrp);
318 write_unlock_irq(&tasklist_lock);
319}
320
321/*
322 * Let kernel threads use this to say that they
323 * allow a certain signal (since daemonize() will
324 * have disabled all of them by default).
325 */
326int allow_signal(int sig)
327{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700328 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return -EINVAL;
330
331 spin_lock_irq(&current->sighand->siglock);
332 sigdelset(&current->blocked, sig);
333 if (!current->mm) {
334 /* Kernel threads handle their own signals.
335 Let the signal code know it'll be handled, so
336 that they don't get converted to SIGKILL or
337 just silently dropped */
338 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
339 }
340 recalc_sigpending();
341 spin_unlock_irq(&current->sighand->siglock);
342 return 0;
343}
344
345EXPORT_SYMBOL(allow_signal);
346
347int disallow_signal(int sig)
348{
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700349 if (!valid_signal(sig) || sig < 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return -EINVAL;
351
352 spin_lock_irq(&current->sighand->siglock);
353 sigaddset(&current->blocked, sig);
354 recalc_sigpending();
355 spin_unlock_irq(&current->sighand->siglock);
356 return 0;
357}
358
359EXPORT_SYMBOL(disallow_signal);
360
361/*
362 * Put all the gunge required to become a kernel thread without
363 * attached user resources in one place where it belongs.
364 */
365
366void daemonize(const char *name, ...)
367{
368 va_list args;
369 struct fs_struct *fs;
370 sigset_t blocked;
371
372 va_start(args, name);
373 vsnprintf(current->comm, sizeof(current->comm), name, args);
374 va_end(args);
375
376 /*
377 * If we were started as result of loading a module, close all of the
378 * user space pages. We don't need them, and if we didn't close them
379 * they would be locked into memory.
380 */
381 exit_mm(current);
382
383 set_special_pids(1, 1);
Peter Zijlstra24ec8392006-12-08 02:36:04 -0800384 proc_clear_tty(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386 /* Block and flush all signals */
387 sigfillset(&blocked);
388 sigprocmask(SIG_BLOCK, &blocked, NULL);
389 flush_signals(current);
390
391 /* Become as one with the init task */
392
393 exit_fs(current); /* current->fs->count--; */
394 fs = init_task.fs;
395 current->fs = fs;
396 atomic_inc(&fs->count);
Serge E. Hallynab516012006-10-02 02:18:06 -0700397
Linus Torvalds444f3782007-01-30 13:35:18 -0800398 exit_task_namespaces(current);
Serge E. Hallynab516012006-10-02 02:18:06 -0700399 current->nsproxy = init_task.nsproxy;
Serge E. Hallynab516012006-10-02 02:18:06 -0700400 get_task_namespaces(current);
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 exit_files(current);
403 current->files = init_task.files;
404 atomic_inc(&current->files->count);
405
406 reparent_to_init();
407}
408
409EXPORT_SYMBOL(daemonize);
410
Arjan van de Ven858119e2006-01-14 13:20:43 -0800411static void close_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
413 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700414 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700417
418 /*
419 * It is safe to dereference the fd table without RCU or
420 * ->file_lock because this is the last reference to the
421 * files structure.
422 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700423 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 for (;;) {
425 unsigned long set;
426 i = j * __NFDBITS;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800427 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700429 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 while (set) {
431 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700432 struct file * file = xchg(&fdt->fd[i], NULL);
Ingo Molnar944be0b2007-02-12 00:52:26 -0800433 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 filp_close(file, files);
Ingo Molnar944be0b2007-02-12 00:52:26 -0800435 cond_resched();
436 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 }
438 i++;
439 set >>= 1;
440 }
441 }
442}
443
444struct files_struct *get_files_struct(struct task_struct *task)
445{
446 struct files_struct *files;
447
448 task_lock(task);
449 files = task->files;
450 if (files)
451 atomic_inc(&files->count);
452 task_unlock(task);
453
454 return files;
455}
456
457void fastcall put_files_struct(struct files_struct *files)
458{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700459 struct fdtable *fdt;
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (atomic_dec_and_test(&files->count)) {
462 close_files(files);
463 /*
464 * Free the fd and fdset arrays if we expanded them.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700465 * If the fdtable was embedded, pass files for freeing
466 * at the end of the RCU grace period. Otherwise,
467 * you can free files immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700469 fdt = files_fdtable(files);
Vadim Lobanov4fd45812006-12-10 02:21:17 -0800470 if (fdt != &files->fdtab)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700471 kmem_cache_free(files_cachep, files);
Vadim Lobanov01b2d932006-12-22 01:10:43 -0800472 free_fdtable(fdt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474}
475
476EXPORT_SYMBOL(put_files_struct);
477
Kirill Korotaev3b9b8ab2006-09-29 02:00:05 -0700478void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
479{
480 struct files_struct *old;
481
482 old = tsk->files;
483 task_lock(tsk);
484 tsk->files = files;
485 task_unlock(tsk);
486 put_files_struct(old);
487}
488EXPORT_SYMBOL(reset_files_struct);
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490static inline void __exit_files(struct task_struct *tsk)
491{
492 struct files_struct * files = tsk->files;
493
494 if (files) {
495 task_lock(tsk);
496 tsk->files = NULL;
497 task_unlock(tsk);
498 put_files_struct(files);
499 }
500}
501
502void exit_files(struct task_struct *tsk)
503{
504 __exit_files(tsk);
505}
506
507static inline void __put_fs_struct(struct fs_struct *fs)
508{
509 /* No need to hold fs->lock if we are killing it */
510 if (atomic_dec_and_test(&fs->count)) {
511 dput(fs->root);
512 mntput(fs->rootmnt);
513 dput(fs->pwd);
514 mntput(fs->pwdmnt);
515 if (fs->altroot) {
516 dput(fs->altroot);
517 mntput(fs->altrootmnt);
518 }
519 kmem_cache_free(fs_cachep, fs);
520 }
521}
522
523void put_fs_struct(struct fs_struct *fs)
524{
525 __put_fs_struct(fs);
526}
527
528static inline void __exit_fs(struct task_struct *tsk)
529{
530 struct fs_struct * fs = tsk->fs;
531
532 if (fs) {
533 task_lock(tsk);
534 tsk->fs = NULL;
535 task_unlock(tsk);
536 __put_fs_struct(fs);
537 }
538}
539
540void exit_fs(struct task_struct *tsk)
541{
542 __exit_fs(tsk);
543}
544
545EXPORT_SYMBOL_GPL(exit_fs);
546
547/*
548 * Turn us into a lazy TLB process if we
549 * aren't already..
550 */
Adrian Bunk408b6642005-05-01 08:59:29 -0700551static void exit_mm(struct task_struct * tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
553 struct mm_struct *mm = tsk->mm;
554
555 mm_release(tsk, mm);
556 if (!mm)
557 return;
558 /*
559 * Serialize with any possible pending coredump.
560 * We must hold mmap_sem around checking core_waiters
561 * and clearing tsk->mm. The core-inducing thread
562 * will increment core_waiters for each thread in the
563 * group with ->mm != NULL.
564 */
565 down_read(&mm->mmap_sem);
566 if (mm->core_waiters) {
567 up_read(&mm->mmap_sem);
568 down_write(&mm->mmap_sem);
569 if (!--mm->core_waiters)
570 complete(mm->core_startup_done);
571 up_write(&mm->mmap_sem);
572
573 wait_for_completion(&mm->core_done);
574 down_read(&mm->mmap_sem);
575 }
576 atomic_inc(&mm->mm_count);
Eric Sesterhenn125e1872006-06-23 02:06:06 -0700577 BUG_ON(mm != tsk->active_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 /* more a memory barrier than a real lock */
579 task_lock(tsk);
580 tsk->mm = NULL;
581 up_read(&mm->mmap_sem);
582 enter_lazy_tlb(mm, current);
583 task_unlock(tsk);
584 mmput(mm);
585}
586
Ingo Molnar36c8b582006-07-03 00:25:41 -0700587static inline void
588choose_new_parent(struct task_struct *p, struct task_struct *reaper)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
590 /*
591 * Make sure we're not reparenting to ourselves and that
592 * the parent is not a zombie.
593 */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800594 BUG_ON(p == reaper || reaper->exit_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 p->real_parent = reaper;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
Ingo Molnar36c8b582006-07-03 00:25:41 -0700598static void
599reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Oleg Nesterov241ceee2006-12-24 23:30:44 +0300601 if (p->pdeath_signal)
602 /* We already hold the tasklist_lock here. */
603 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 /* Move the child from its dying parent to the new one. */
606 if (unlikely(traced)) {
607 /* Preserve ptrace links if someone else is tracing this child. */
608 list_del_init(&p->ptrace_list);
609 if (p->parent != p->real_parent)
610 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
611 } else {
612 /* If this child is being traced, then we're the one tracing it
613 * anyway, so let go of it.
614 */
615 p->ptrace = 0;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800616 remove_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 p->parent = p->real_parent;
Oleg Nesterov6ac781b2006-03-28 16:11:09 -0800618 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Eric W. Biedermanb2b2cbc2006-12-21 21:28:40 -0700620 if (p->state == TASK_TRACED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 /*
622 * If it was at a trace stop, turn it into
623 * a normal stop since it's no longer being
624 * traced.
625 */
626 ptrace_untrace(p);
627 }
628 }
629
Eric W. Biedermanb2b2cbc2006-12-21 21:28:40 -0700630 /* If this is a threaded reparent there is no need to
631 * notify anyone anything has happened.
632 */
633 if (p->real_parent->group_leader == father->group_leader)
634 return;
635
636 /* We don't want people slaying init. */
637 if (p->exit_signal != -1)
638 p->exit_signal = SIGCHLD;
Eric W. Biedermanb2b2cbc2006-12-21 21:28:40 -0700639
640 /* If we'd notified the old parent about this child's death,
641 * also notify the new parent.
642 */
643 if (!traced && p->exit_state == EXIT_ZOMBIE &&
644 p->exit_signal != -1 && thread_group_empty(p))
645 do_notify_parent(p, p->exit_signal);
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 /*
648 * process group orphan check
649 * Case ii: Our child is in a different pgrp
650 * than we are, and it was the only connection
651 * outside, so the child pgrp is now orphaned.
652 */
653 if ((process_group(p) != process_group(father)) &&
Cedric Le Goater937949d2006-12-08 02:37:54 -0800654 (process_session(p) == process_session(father))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 int pgrp = process_group(p);
656
Cedric Le Goater937949d2006-12-08 02:37:54 -0800657 if (will_become_orphaned_pgrp(pgrp, NULL) &&
658 has_stopped_jobs(pgrp)) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800659 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
660 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
662 }
663}
664
665/*
666 * When we die, we re-parent all our children.
667 * Try to give them to another thread in our thread
668 * group, and if no such member exists, give it to
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800669 * the child reaper process (ie "init") in our pid
670 * space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700672static void
673forget_original_parent(struct task_struct *father, struct list_head *to_release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 struct task_struct *p, *reaper = father;
676 struct list_head *_p, *_n;
677
678 do {
679 reaper = next_thread(reaper);
680 if (reaper == father) {
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800681 reaper = child_reaper(father);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 break;
683 }
684 } while (reaper->exit_state);
685
686 /*
687 * There are only two places where our children can be:
688 *
689 * - in our child list
690 * - in our ptraced child list
691 *
692 * Search them and reparent children.
693 */
694 list_for_each_safe(_p, _n, &father->children) {
695 int ptrace;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700696 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 ptrace = p->ptrace;
699
700 /* if father isn't the real parent, then ptrace must be enabled */
701 BUG_ON(father != p->real_parent && !ptrace);
702
703 if (father == p->real_parent) {
704 /* reparent with a reaper, real father it's us */
Oleg Nesterovd799f032006-03-28 16:11:04 -0800705 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 reparent_thread(p, father, 0);
707 } else {
708 /* reparent ptraced task to its real parent */
709 __ptrace_unlink (p);
710 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
711 thread_group_empty(p))
712 do_notify_parent(p, p->exit_signal);
713 }
714
715 /*
716 * if the ptraced child is a zombie with exit_signal == -1
717 * we must collect it before we exit, or it will remain
718 * zombie forever since we prevented it from self-reap itself
719 * while it was being traced by us, to be able to see it in wait4.
720 */
721 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
722 list_add(&p->ptrace_list, to_release);
723 }
724 list_for_each_safe(_p, _n, &father->ptrace_children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -0700725 p = list_entry(_p, struct task_struct, ptrace_list);
Oleg Nesterovd799f032006-03-28 16:11:04 -0800726 choose_new_parent(p, reaper);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 reparent_thread(p, father, 1);
728 }
729}
730
731/*
732 * Send signals to all our closest relatives so that they know
733 * to properly mourn us..
734 */
735static void exit_notify(struct task_struct *tsk)
736{
737 int state;
738 struct task_struct *t;
739 struct list_head ptrace_dead, *_p, *_n;
740
741 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
742 && !thread_group_empty(tsk)) {
743 /*
744 * This occurs when there was a race between our exit
745 * syscall and a group signal choosing us as the one to
746 * wake up. It could be that we are the only thread
747 * alerted to check for pending signals, but another thread
748 * should be woken now to take the signal since we will not.
749 * Now we'll wake all the threads in the group just to make
750 * sure someone gets all the pending signals.
751 */
752 read_lock(&tasklist_lock);
753 spin_lock_irq(&tsk->sighand->siglock);
754 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
755 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
756 recalc_sigpending_tsk(t);
757 if (signal_pending(t))
758 signal_wake_up(t, 0);
759 }
760 spin_unlock_irq(&tsk->sighand->siglock);
761 read_unlock(&tasklist_lock);
762 }
763
764 write_lock_irq(&tasklist_lock);
765
766 /*
767 * This does two things:
768 *
769 * A. Make init inherit all the child processes
770 * B. Check to see if any process groups have become orphaned
771 * as a result of our exiting, and if they have any stopped
772 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
773 */
774
775 INIT_LIST_HEAD(&ptrace_dead);
776 forget_original_parent(tsk, &ptrace_dead);
777 BUG_ON(!list_empty(&tsk->children));
778 BUG_ON(!list_empty(&tsk->ptrace_children));
779
780 /*
781 * Check to see if any process groups have become orphaned
782 * as a result of our exiting, and if they have any stopped
783 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
784 *
785 * Case i: Our father is in a different pgrp than we are
786 * and we were the only connection outside, so our pgrp
787 * is about to become orphaned.
788 */
789
790 t = tsk->real_parent;
791
792 if ((process_group(t) != process_group(tsk)) &&
Cedric Le Goater937949d2006-12-08 02:37:54 -0800793 (process_session(t) == process_session(tsk)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
795 has_stopped_jobs(process_group(tsk))) {
Oleg Nesterovb67a1b92005-10-30 15:03:44 -0800796 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
797 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
799
800 /* Let father know we died
801 *
802 * Thread signals are configurable, but you aren't going to use
803 * that to send signals to arbitary processes.
804 * That stops right now.
805 *
806 * If the parent exec id doesn't match the exec id we saved
807 * when we started then we know the parent has changed security
808 * domain.
809 *
810 * If our self_exec id doesn't match our parent_exec_id then
811 * we have changed execution domain as these two values started
812 * the same after a fork.
813 *
814 */
815
816 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
817 ( tsk->parent_exec_id != t->self_exec_id ||
818 tsk->self_exec_id != tsk->parent_exec_id)
819 && !capable(CAP_KILL))
820 tsk->exit_signal = SIGCHLD;
821
822
823 /* If something other than our normal parent is ptracing us, then
824 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
825 * only has special meaning to our real parent.
826 */
827 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
828 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
829 do_notify_parent(tsk, signal);
830 } else if (tsk->ptrace) {
831 do_notify_parent(tsk, SIGCHLD);
832 }
833
834 state = EXIT_ZOMBIE;
835 if (tsk->exit_signal == -1 &&
836 (likely(tsk->ptrace == 0) ||
837 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
838 state = EXIT_DEAD;
839 tsk->exit_state = state;
840
841 write_unlock_irq(&tasklist_lock);
842
843 list_for_each_safe(_p, _n, &ptrace_dead) {
844 list_del_init(_p);
Ingo Molnar36c8b582006-07-03 00:25:41 -0700845 t = list_entry(_p, struct task_struct, ptrace_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 release_task(t);
847 }
848
849 /* If the process is dead, release it - nobody will wait for it */
850 if (state == EXIT_DEAD)
851 release_task(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852}
853
854fastcall NORET_TYPE void do_exit(long code)
855{
856 struct task_struct *tsk = current;
857 int group_dead;
858
859 profile_task_exit(tsk);
860
Jens Axboe22e2c502005-06-27 10:55:12 +0200861 WARN_ON(atomic_read(&tsk->fs_excl));
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (unlikely(in_interrupt()))
864 panic("Aiee, killing interrupt handler!");
865 if (unlikely(!tsk->pid))
866 panic("Attempted to kill the idle task!");
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800867 if (unlikely(tsk == child_reaper(tsk))) {
868 if (tsk->nsproxy->pid_ns != &init_pid_ns)
869 tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
870 else
871 panic("Attempted to kill init!");
872 }
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
876 current->ptrace_message = code;
877 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
878 }
879
Alexander Nybergdf164db2005-06-23 00:09:13 -0700880 /*
881 * We're taking recursive faults here in do_exit. Safest is to just
882 * leave this task alone and wait for reboot.
883 */
884 if (unlikely(tsk->flags & PF_EXITING)) {
885 printk(KERN_ALERT
886 "Fixing recursive fault but reboot is needed!\n");
Al Viroafc847b2006-02-28 12:51:55 -0500887 if (tsk->io_context)
888 exit_io_context();
Alexander Nybergdf164db2005-06-23 00:09:13 -0700889 set_current_state(TASK_UNINTERRUPTIBLE);
890 schedule();
891 }
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 tsk->flags |= PF_EXITING;
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (unlikely(in_atomic()))
896 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
897 current->comm, current->pid,
898 preempt_count());
899
900 acct_update_integrals(tsk);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700901 if (tsk->mm) {
902 update_hiwater_rss(tsk->mm);
903 update_hiwater_vm(tsk->mm);
904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 group_dead = atomic_dec_and_test(&tsk->signal->live);
Andrew Mortonc3068952005-08-04 16:49:32 -0700906 if (group_dead) {
Thomas Gleixner2ff678b2006-01-09 20:52:34 -0800907 hrtimer_cancel(&tsk->signal->real_timer);
Roland McGrath25f407f2005-10-21 15:03:29 -0700908 exit_itimers(tsk->signal);
Andrew Mortonc3068952005-08-04 16:49:32 -0700909 }
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700910 acct_collect(code, group_dead);
Ingo Molnar0771dfe2006-03-27 01:16:22 -0800911 if (unlikely(tsk->robust_list))
912 exit_robust_list(tsk);
Anton Blanchard2aa92582006-06-25 05:48:44 -0700913#if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
Ingo Molnar34f192c2006-03-27 01:16:24 -0800914 if (unlikely(tsk->compat_robust_list))
915 compat_exit_robust_list(tsk);
916#endif
Al Virofa84cb92006-03-29 20:30:19 -0500917 if (unlikely(tsk->audit_context))
918 audit_free(tsk);
Oleg Nesterov115085e2006-12-06 20:36:51 -0800919
920 taskstats_exit(tsk, group_dead);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 exit_mm(tsk);
923
KaiGai Kohei0e464812006-06-25 05:49:24 -0700924 if (group_dead)
KaiGai Koheif6ec29a2006-06-25 05:49:25 -0700925 acct_process();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 exit_sem(tsk);
927 __exit_files(tsk);
928 __exit_fs(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 exit_thread();
930 cpuset_exit(tsk);
931 exit_keys(tsk);
932
933 if (group_dead && tsk->signal->leader)
934 disassociate_ctty(1);
935
Al Viroa1261f52005-11-13 16:06:55 -0800936 module_put(task_thread_info(tsk)->exec_domain->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (tsk->binfmt)
938 module_put(tsk->binfmt->module);
939
940 tsk->exit_code = code;
Matt Helsley9f460802005-11-07 00:59:16 -0800941 proc_exit_connector(tsk);
Linus Torvalds444f3782007-01-30 13:35:18 -0800942 exit_task_namespaces(tsk);
Serge E. Hallyn0f245282007-01-30 15:28:23 -0600943 exit_notify(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944#ifdef CONFIG_NUMA
945 mpol_free(tsk->mempolicy);
946 tsk->mempolicy = NULL;
947#endif
Ingo Molnarde5097c2006-01-09 15:59:21 -0800948 /*
Ingo Molnarc87e2832006-06-27 02:54:58 -0700949 * This must happen late, after the PID is not
950 * hashed anymore:
951 */
952 if (unlikely(!list_empty(&tsk->pi_state_list)))
953 exit_pi_state_list(tsk);
954 if (unlikely(current->pi_state_cache))
955 kfree(current->pi_state_cache);
956 /*
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700957 * Make sure we are holding no locks:
Ingo Molnarde5097c2006-01-09 15:59:21 -0800958 */
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700959 debug_check_no_locks_held(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Al Viroafc847b2006-02-28 12:51:55 -0500961 if (tsk->io_context)
962 exit_io_context();
963
Jens Axboeb92ce552006-04-11 13:52:07 +0200964 if (tsk->splice_pipe)
965 __free_pipe_info(tsk->splice_pipe);
966
Coywolf Qi Hunt74072512005-10-30 15:02:47 -0800967 preempt_disable();
Oleg Nesterov55a101f2006-09-29 02:01:10 -0700968 /* causes final put_task_struct in finish_task_switch(). */
Oleg Nesterovc394cc92006-09-29 02:01:11 -0700969 tsk->state = TASK_DEAD;
Coywolf Qi Hunt74072512005-10-30 15:02:47 -0800970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 schedule();
972 BUG();
973 /* Avoid "noreturn function does return". */
Alan Cox54306cf2006-09-29 02:00:42 -0700974 for (;;)
975 cpu_relax(); /* For when BUG is null */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976}
977
Russ Anderson012914d2005-04-23 00:08:00 -0700978EXPORT_SYMBOL_GPL(do_exit);
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980NORET_TYPE void complete_and_exit(struct completion *comp, long code)
981{
982 if (comp)
983 complete(comp);
Oleg Nesterov55a101f2006-09-29 02:01:10 -0700984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 do_exit(code);
986}
987
988EXPORT_SYMBOL(complete_and_exit);
989
990asmlinkage long sys_exit(int error_code)
991{
992 do_exit((error_code&0xff)<<8);
993}
994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995/*
996 * Take down every thread in the group. This is called by fatal signals
997 * as well as by sys_exit_group (below).
998 */
999NORET_TYPE void
1000do_group_exit(int exit_code)
1001{
1002 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1003
1004 if (current->signal->flags & SIGNAL_GROUP_EXIT)
1005 exit_code = current->signal->group_exit_code;
1006 else if (!thread_group_empty(current)) {
1007 struct signal_struct *const sig = current->signal;
1008 struct sighand_struct *const sighand = current->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 spin_lock_irq(&sighand->siglock);
1010 if (sig->flags & SIGNAL_GROUP_EXIT)
1011 /* Another thread got here before we took the lock. */
1012 exit_code = sig->group_exit_code;
1013 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 sig->group_exit_code = exit_code;
1015 zap_other_threads(current);
1016 }
1017 spin_unlock_irq(&sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
1020 do_exit(exit_code);
1021 /* NOTREACHED */
1022}
1023
1024/*
1025 * this kills every thread in the thread group. Note that any externally
1026 * wait4()-ing process will get the correct exit code - even if this
1027 * thread is not the thread group leader.
1028 */
1029asmlinkage void sys_exit_group(int error_code)
1030{
1031 do_group_exit((error_code & 0xff) << 8);
1032}
1033
Ingo Molnar36c8b582006-07-03 00:25:41 -07001034static int eligible_child(pid_t pid, int options, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 if (pid > 0) {
1037 if (p->pid != pid)
1038 return 0;
1039 } else if (!pid) {
1040 if (process_group(p) != process_group(current))
1041 return 0;
1042 } else if (pid != -1) {
1043 if (process_group(p) != -pid)
1044 return 0;
1045 }
1046
1047 /*
1048 * Do not consider detached threads that are
1049 * not ptraced:
1050 */
1051 if (p->exit_signal == -1 && !p->ptrace)
1052 return 0;
1053
1054 /* Wait for all children (clone and not) if __WALL is set;
1055 * otherwise, wait for clone children *only* if __WCLONE is
1056 * set; otherwise, wait for non-clone children *only*. (Note:
1057 * A "clone" child here is one that reports to its parent
1058 * using a signal other than SIGCHLD.) */
1059 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1060 && !(options & __WALL))
1061 return 0;
1062 /*
1063 * Do not consider thread group leaders that are
1064 * in a non-empty thread group:
1065 */
Oleg Nesterov3b6362b2006-09-02 21:22:16 +04001066 if (delay_group_leader(p))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 return 2;
1068
1069 if (security_task_wait(p))
1070 return 0;
1071
1072 return 1;
1073}
1074
Ingo Molnar36c8b582006-07-03 00:25:41 -07001075static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 int why, int status,
1077 struct siginfo __user *infop,
1078 struct rusage __user *rusagep)
1079{
1080 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
Ingo Molnar36c8b582006-07-03 00:25:41 -07001081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 put_task_struct(p);
1083 if (!retval)
1084 retval = put_user(SIGCHLD, &infop->si_signo);
1085 if (!retval)
1086 retval = put_user(0, &infop->si_errno);
1087 if (!retval)
1088 retval = put_user((short)why, &infop->si_code);
1089 if (!retval)
1090 retval = put_user(pid, &infop->si_pid);
1091 if (!retval)
1092 retval = put_user(uid, &infop->si_uid);
1093 if (!retval)
1094 retval = put_user(status, &infop->si_status);
1095 if (!retval)
1096 retval = pid;
1097 return retval;
1098}
1099
1100/*
1101 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1102 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1103 * the lock and this task is uninteresting. If we return nonzero, we have
1104 * released the lock and the system call should return.
1105 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001106static int wait_task_zombie(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 struct siginfo __user *infop,
1108 int __user *stat_addr, struct rusage __user *ru)
1109{
1110 unsigned long state;
1111 int retval;
1112 int status;
1113
1114 if (unlikely(noreap)) {
1115 pid_t pid = p->pid;
1116 uid_t uid = p->uid;
1117 int exit_code = p->exit_code;
1118 int why, status;
1119
1120 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1121 return 0;
1122 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1123 return 0;
1124 get_task_struct(p);
1125 read_unlock(&tasklist_lock);
1126 if ((exit_code & 0x7f) == 0) {
1127 why = CLD_EXITED;
1128 status = exit_code >> 8;
1129 } else {
1130 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1131 status = exit_code & 0x7f;
1132 }
1133 return wait_noreap_copyout(p, pid, uid, why,
1134 status, infop, ru);
1135 }
1136
1137 /*
1138 * Try to move the task's state to DEAD
1139 * only one thread is allowed to do this:
1140 */
1141 state = xchg(&p->exit_state, EXIT_DEAD);
1142 if (state != EXIT_ZOMBIE) {
1143 BUG_ON(state != EXIT_DEAD);
1144 return 0;
1145 }
1146 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1147 /*
1148 * This can only happen in a race with a ptraced thread
1149 * dying on another processor.
1150 */
1151 return 0;
1152 }
1153
1154 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
Jesper Juhl3795e162006-01-09 20:54:39 -08001155 struct signal_struct *psig;
1156 struct signal_struct *sig;
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 /*
1159 * The resource counters for the group leader are in its
1160 * own task_struct. Those for dead threads in the group
1161 * are in its signal_struct, as are those for the child
1162 * processes it has previously reaped. All these
1163 * accumulate in the parent's signal_struct c* fields.
1164 *
1165 * We don't bother to take a lock here to protect these
1166 * p->signal fields, because they are only touched by
1167 * __exit_signal, which runs with tasklist_lock
1168 * write-locked anyway, and so is excluded here. We do
1169 * need to protect the access to p->parent->signal fields,
1170 * as other threads in the parent group can be right
1171 * here reaping other children at the same time.
1172 */
1173 spin_lock_irq(&p->parent->sighand->siglock);
Jesper Juhl3795e162006-01-09 20:54:39 -08001174 psig = p->parent->signal;
1175 sig = p->signal;
1176 psig->cutime =
1177 cputime_add(psig->cutime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 cputime_add(p->utime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001179 cputime_add(sig->utime,
1180 sig->cutime)));
1181 psig->cstime =
1182 cputime_add(psig->cstime,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 cputime_add(p->stime,
Jesper Juhl3795e162006-01-09 20:54:39 -08001184 cputime_add(sig->stime,
1185 sig->cstime)));
1186 psig->cmin_flt +=
1187 p->min_flt + sig->min_flt + sig->cmin_flt;
1188 psig->cmaj_flt +=
1189 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1190 psig->cnvcsw +=
1191 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1192 psig->cnivcsw +=
1193 p->nivcsw + sig->nivcsw + sig->cnivcsw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 spin_unlock_irq(&p->parent->sighand->siglock);
1195 }
1196
1197 /*
1198 * Now we are sure this task is interesting, and no other
1199 * thread can reap it because we set its state to EXIT_DEAD.
1200 */
1201 read_unlock(&tasklist_lock);
1202
1203 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1204 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1205 ? p->signal->group_exit_code : p->exit_code;
1206 if (!retval && stat_addr)
1207 retval = put_user(status, stat_addr);
1208 if (!retval && infop)
1209 retval = put_user(SIGCHLD, &infop->si_signo);
1210 if (!retval && infop)
1211 retval = put_user(0, &infop->si_errno);
1212 if (!retval && infop) {
1213 int why;
1214
1215 if ((status & 0x7f) == 0) {
1216 why = CLD_EXITED;
1217 status >>= 8;
1218 } else {
1219 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1220 status &= 0x7f;
1221 }
1222 retval = put_user((short)why, &infop->si_code);
1223 if (!retval)
1224 retval = put_user(status, &infop->si_status);
1225 }
1226 if (!retval && infop)
1227 retval = put_user(p->pid, &infop->si_pid);
1228 if (!retval && infop)
1229 retval = put_user(p->uid, &infop->si_uid);
1230 if (retval) {
1231 // TODO: is this safe?
1232 p->exit_state = EXIT_ZOMBIE;
1233 return retval;
1234 }
1235 retval = p->pid;
1236 if (p->real_parent != p->parent) {
1237 write_lock_irq(&tasklist_lock);
1238 /* Double-check with lock held. */
1239 if (p->real_parent != p->parent) {
1240 __ptrace_unlink(p);
1241 // TODO: is this safe?
1242 p->exit_state = EXIT_ZOMBIE;
1243 /*
1244 * If this is not a detached task, notify the parent.
1245 * If it's still not detached after that, don't release
1246 * it now.
1247 */
1248 if (p->exit_signal != -1) {
1249 do_notify_parent(p, p->exit_signal);
1250 if (p->exit_signal != -1)
1251 p = NULL;
1252 }
1253 }
1254 write_unlock_irq(&tasklist_lock);
1255 }
1256 if (p != NULL)
1257 release_task(p);
1258 BUG_ON(!retval);
1259 return retval;
1260}
1261
1262/*
1263 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1264 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1265 * the lock and this task is uninteresting. If we return nonzero, we have
1266 * released the lock and the system call should return.
1267 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001268static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1269 int noreap, struct siginfo __user *infop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 int __user *stat_addr, struct rusage __user *ru)
1271{
1272 int retval, exit_code;
1273
1274 if (!p->exit_code)
1275 return 0;
1276 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1277 p->signal && p->signal->group_stop_count > 0)
1278 /*
1279 * A group stop is in progress and this is the group leader.
1280 * We won't report until all threads have stopped.
1281 */
1282 return 0;
1283
1284 /*
1285 * Now we are pretty sure this task is interesting.
1286 * Make sure it doesn't get reaped out from under us while we
1287 * give up the lock and then examine it below. We don't want to
1288 * keep holding onto the tasklist_lock while we call getrusage and
1289 * possibly take page faults for user memory.
1290 */
1291 get_task_struct(p);
1292 read_unlock(&tasklist_lock);
1293
1294 if (unlikely(noreap)) {
1295 pid_t pid = p->pid;
1296 uid_t uid = p->uid;
1297 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1298
1299 exit_code = p->exit_code;
1300 if (unlikely(!exit_code) ||
Linus Torvalds14bf01b2005-10-01 11:04:18 -07001301 unlikely(p->state & TASK_TRACED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 goto bail_ref;
1303 return wait_noreap_copyout(p, pid, uid,
1304 why, (exit_code << 8) | 0x7f,
1305 infop, ru);
1306 }
1307
1308 write_lock_irq(&tasklist_lock);
1309
1310 /*
1311 * This uses xchg to be atomic with the thread resuming and setting
1312 * it. It must also be done with the write lock held to prevent a
1313 * race with the EXIT_ZOMBIE case.
1314 */
1315 exit_code = xchg(&p->exit_code, 0);
1316 if (unlikely(p->exit_state)) {
1317 /*
1318 * The task resumed and then died. Let the next iteration
1319 * catch it in EXIT_ZOMBIE. Note that exit_code might
1320 * already be zero here if it resumed and did _exit(0).
1321 * The task itself is dead and won't touch exit_code again;
1322 * other processors in this function are locked out.
1323 */
1324 p->exit_code = exit_code;
1325 exit_code = 0;
1326 }
1327 if (unlikely(exit_code == 0)) {
1328 /*
1329 * Another thread in this function got to it first, or it
1330 * resumed, or it resumed and then died.
1331 */
1332 write_unlock_irq(&tasklist_lock);
1333bail_ref:
1334 put_task_struct(p);
1335 /*
1336 * We are returning to the wait loop without having successfully
1337 * removed the process and having released the lock. We cannot
1338 * continue, since the "p" task pointer is potentially stale.
1339 *
1340 * Return -EAGAIN, and do_wait() will restart the loop from the
1341 * beginning. Do _not_ re-acquire the lock.
1342 */
1343 return -EAGAIN;
1344 }
1345
1346 /* move to end of parent's list to avoid starvation */
1347 remove_parent(p);
Oleg Nesterov8fafabd2006-03-28 16:11:05 -08001348 add_parent(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 write_unlock_irq(&tasklist_lock);
1351
1352 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1353 if (!retval && stat_addr)
1354 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1355 if (!retval && infop)
1356 retval = put_user(SIGCHLD, &infop->si_signo);
1357 if (!retval && infop)
1358 retval = put_user(0, &infop->si_errno);
1359 if (!retval && infop)
1360 retval = put_user((short)((p->ptrace & PT_PTRACED)
1361 ? CLD_TRAPPED : CLD_STOPPED),
1362 &infop->si_code);
1363 if (!retval && infop)
1364 retval = put_user(exit_code, &infop->si_status);
1365 if (!retval && infop)
1366 retval = put_user(p->pid, &infop->si_pid);
1367 if (!retval && infop)
1368 retval = put_user(p->uid, &infop->si_uid);
1369 if (!retval)
1370 retval = p->pid;
1371 put_task_struct(p);
1372
1373 BUG_ON(!retval);
1374 return retval;
1375}
1376
1377/*
1378 * Handle do_wait work for one task in a live, non-stopped state.
1379 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1380 * the lock and this task is uninteresting. If we return nonzero, we have
1381 * released the lock and the system call should return.
1382 */
Ingo Molnar36c8b582006-07-03 00:25:41 -07001383static int wait_task_continued(struct task_struct *p, int noreap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 struct siginfo __user *infop,
1385 int __user *stat_addr, struct rusage __user *ru)
1386{
1387 int retval;
1388 pid_t pid;
1389 uid_t uid;
1390
1391 if (unlikely(!p->signal))
1392 return 0;
1393
1394 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1395 return 0;
1396
1397 spin_lock_irq(&p->sighand->siglock);
1398 /* Re-check with the lock held. */
1399 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1400 spin_unlock_irq(&p->sighand->siglock);
1401 return 0;
1402 }
1403 if (!noreap)
1404 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1405 spin_unlock_irq(&p->sighand->siglock);
1406
1407 pid = p->pid;
1408 uid = p->uid;
1409 get_task_struct(p);
1410 read_unlock(&tasklist_lock);
1411
1412 if (!infop) {
1413 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1414 put_task_struct(p);
1415 if (!retval && stat_addr)
1416 retval = put_user(0xffff, stat_addr);
1417 if (!retval)
1418 retval = p->pid;
1419 } else {
1420 retval = wait_noreap_copyout(p, pid, uid,
1421 CLD_CONTINUED, SIGCONT,
1422 infop, ru);
1423 BUG_ON(retval == 0);
1424 }
1425
1426 return retval;
1427}
1428
1429
1430static inline int my_ptrace_child(struct task_struct *p)
1431{
1432 if (!(p->ptrace & PT_PTRACED))
1433 return 0;
1434 if (!(p->ptrace & PT_ATTACHED))
1435 return 1;
1436 /*
1437 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1438 * we are the attacher. If we are the real parent, this is a race
1439 * inside ptrace_attach. It is waiting for the tasklist_lock,
1440 * which we have to switch the parent links, but has already set
1441 * the flags in p->ptrace.
1442 */
1443 return (p->parent != p->real_parent);
1444}
1445
1446static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1447 int __user *stat_addr, struct rusage __user *ru)
1448{
1449 DECLARE_WAITQUEUE(wait, current);
1450 struct task_struct *tsk;
1451 int flag, retval;
1452
1453 add_wait_queue(&current->signal->wait_chldexit,&wait);
1454repeat:
1455 /*
1456 * We will set this flag if we see any child that might later
1457 * match our criteria, even if we are not able to reap it yet.
1458 */
1459 flag = 0;
1460 current->state = TASK_INTERRUPTIBLE;
1461 read_lock(&tasklist_lock);
1462 tsk = current;
1463 do {
1464 struct task_struct *p;
1465 struct list_head *_p;
1466 int ret;
1467
1468 list_for_each(_p,&tsk->children) {
Ingo Molnar36c8b582006-07-03 00:25:41 -07001469 p = list_entry(_p, struct task_struct, sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
1471 ret = eligible_child(pid, options, p);
1472 if (!ret)
1473 continue;
1474
1475 switch (p->state) {
1476 case TASK_TRACED:
Roland McGrath7f2a5252005-10-30 15:02:50 -08001477 /*
1478 * When we hit the race with PTRACE_ATTACH,
1479 * we will not report this child. But the
1480 * race means it has not yet been moved to
1481 * our ptrace_children list, so we need to
1482 * set the flag here to avoid a spurious ECHILD
1483 * when the race happens with the only child.
1484 */
1485 flag = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (!my_ptrace_child(p))
1487 continue;
1488 /*FALLTHROUGH*/
1489 case TASK_STOPPED:
1490 /*
1491 * It's stopped now, so it might later
1492 * continue, exit, or stop again.
1493 */
1494 flag = 1;
1495 if (!(options & WUNTRACED) &&
1496 !my_ptrace_child(p))
1497 continue;
1498 retval = wait_task_stopped(p, ret == 2,
1499 (options & WNOWAIT),
1500 infop,
1501 stat_addr, ru);
1502 if (retval == -EAGAIN)
1503 goto repeat;
1504 if (retval != 0) /* He released the lock. */
1505 goto end;
1506 break;
1507 default:
1508 // case EXIT_DEAD:
1509 if (p->exit_state == EXIT_DEAD)
1510 continue;
1511 // case EXIT_ZOMBIE:
1512 if (p->exit_state == EXIT_ZOMBIE) {
1513 /*
1514 * Eligible but we cannot release
1515 * it yet:
1516 */
1517 if (ret == 2)
1518 goto check_continued;
1519 if (!likely(options & WEXITED))
1520 continue;
1521 retval = wait_task_zombie(
1522 p, (options & WNOWAIT),
1523 infop, stat_addr, ru);
1524 /* He released the lock. */
1525 if (retval != 0)
1526 goto end;
1527 break;
1528 }
1529check_continued:
1530 /*
1531 * It's running now, so it might later
1532 * exit, stop, or stop and then continue.
1533 */
1534 flag = 1;
1535 if (!unlikely(options & WCONTINUED))
1536 continue;
1537 retval = wait_task_continued(
1538 p, (options & WNOWAIT),
1539 infop, stat_addr, ru);
1540 if (retval != 0) /* He released the lock. */
1541 goto end;
1542 break;
1543 }
1544 }
1545 if (!flag) {
1546 list_for_each(_p, &tsk->ptrace_children) {
1547 p = list_entry(_p, struct task_struct,
1548 ptrace_list);
1549 if (!eligible_child(pid, options, p))
1550 continue;
1551 flag = 1;
1552 break;
1553 }
1554 }
1555 if (options & __WNOTHREAD)
1556 break;
1557 tsk = next_thread(tsk);
Eric Sesterhenn125e1872006-06-23 02:06:06 -07001558 BUG_ON(tsk->signal != current->signal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 } while (tsk != current);
1560
1561 read_unlock(&tasklist_lock);
1562 if (flag) {
1563 retval = 0;
1564 if (options & WNOHANG)
1565 goto end;
1566 retval = -ERESTARTSYS;
1567 if (signal_pending(current))
1568 goto end;
1569 schedule();
1570 goto repeat;
1571 }
1572 retval = -ECHILD;
1573end:
1574 current->state = TASK_RUNNING;
1575 remove_wait_queue(&current->signal->wait_chldexit,&wait);
1576 if (infop) {
1577 if (retval > 0)
1578 retval = 0;
1579 else {
1580 /*
1581 * For a WNOHANG return, clear out all the fields
1582 * we would set so the user can easily tell the
1583 * difference.
1584 */
1585 if (!retval)
1586 retval = put_user(0, &infop->si_signo);
1587 if (!retval)
1588 retval = put_user(0, &infop->si_errno);
1589 if (!retval)
1590 retval = put_user(0, &infop->si_code);
1591 if (!retval)
1592 retval = put_user(0, &infop->si_pid);
1593 if (!retval)
1594 retval = put_user(0, &infop->si_uid);
1595 if (!retval)
1596 retval = put_user(0, &infop->si_status);
1597 }
1598 }
1599 return retval;
1600}
1601
1602asmlinkage long sys_waitid(int which, pid_t pid,
1603 struct siginfo __user *infop, int options,
1604 struct rusage __user *ru)
1605{
1606 long ret;
1607
1608 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1609 return -EINVAL;
1610 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1611 return -EINVAL;
1612
1613 switch (which) {
1614 case P_ALL:
1615 pid = -1;
1616 break;
1617 case P_PID:
1618 if (pid <= 0)
1619 return -EINVAL;
1620 break;
1621 case P_PGID:
1622 if (pid <= 0)
1623 return -EINVAL;
1624 pid = -pid;
1625 break;
1626 default:
1627 return -EINVAL;
1628 }
1629
1630 ret = do_wait(pid, options, infop, NULL, ru);
1631
1632 /* avoid REGPARM breakage on x86: */
1633 prevent_tail_call(ret);
1634 return ret;
1635}
1636
1637asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1638 int options, struct rusage __user *ru)
1639{
1640 long ret;
1641
1642 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1643 __WNOTHREAD|__WCLONE|__WALL))
1644 return -EINVAL;
1645 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1646
1647 /* avoid REGPARM breakage on x86: */
1648 prevent_tail_call(ret);
1649 return ret;
1650}
1651
1652#ifdef __ARCH_WANT_SYS_WAITPID
1653
1654/*
1655 * sys_waitpid() remains for compatibility. waitpid() should be
1656 * implemented by calling sys_wait4() from libc.a.
1657 */
1658asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1659{
1660 return sys_wait4(pid, stat_addr, options, NULL);
1661}
1662
1663#endif