]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/exit.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/config.h> | |
8 | #include <linux/mm.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/smp_lock.h> | |
12 | #include <linux/module.h> | |
c59ede7b | 13 | #include <linux/capability.h> |
1da177e4 LT |
14 | #include <linux/completion.h> |
15 | #include <linux/personality.h> | |
16 | #include <linux/tty.h> | |
17 | #include <linux/namespace.h> | |
18 | #include <linux/key.h> | |
19 | #include <linux/security.h> | |
20 | #include <linux/cpu.h> | |
21 | #include <linux/acct.h> | |
22 | #include <linux/file.h> | |
23 | #include <linux/binfmts.h> | |
24 | #include <linux/ptrace.h> | |
25 | #include <linux/profile.h> | |
26 | #include <linux/mount.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/mempolicy.h> | |
29 | #include <linux/cpuset.h> | |
30 | #include <linux/syscalls.h> | |
7ed20e1a | 31 | #include <linux/signal.h> |
6a14c5c9 | 32 | #include <linux/posix-timers.h> |
9f46080c | 33 | #include <linux/cn_proc.h> |
de5097c2 | 34 | #include <linux/mutex.h> |
0771dfef | 35 | #include <linux/futex.h> |
34f192c6 | 36 | #include <linux/compat.h> |
b92ce558 | 37 | #include <linux/pipe_fs_i.h> |
fa84cb93 | 38 | #include <linux/audit.h> /* for audit_free() */ |
1da177e4 LT |
39 | |
40 | #include <asm/uaccess.h> | |
41 | #include <asm/unistd.h> | |
42 | #include <asm/pgtable.h> | |
43 | #include <asm/mmu_context.h> | |
44 | ||
45 | extern void sem_exit (void); | |
46 | extern struct task_struct *child_reaper; | |
47 | ||
48 | int getrusage(struct task_struct *, int, struct rusage __user *); | |
49 | ||
408b664a AB |
50 | static void exit_mm(struct task_struct * tsk); |
51 | ||
1da177e4 LT |
52 | static void __unhash_process(struct task_struct *p) |
53 | { | |
54 | nr_threads--; | |
55 | detach_pid(p, PIDTYPE_PID); | |
1da177e4 LT |
56 | if (thread_group_leader(p)) { |
57 | detach_pid(p, PIDTYPE_PGID); | |
58 | detach_pid(p, PIDTYPE_SID); | |
c97d9893 | 59 | |
5e85d4ab | 60 | list_del_rcu(&p->tasks); |
73b9ebfe | 61 | __get_cpu_var(process_counts)--; |
1da177e4 | 62 | } |
47e65328 | 63 | list_del_rcu(&p->thread_group); |
c97d9893 | 64 | remove_parent(p); |
1da177e4 LT |
65 | } |
66 | ||
6a14c5c9 ON |
67 | /* |
68 | * This function expects the tasklist_lock write-locked. | |
69 | */ | |
70 | static void __exit_signal(struct task_struct *tsk) | |
71 | { | |
72 | struct signal_struct *sig = tsk->signal; | |
73 | struct sighand_struct *sighand; | |
74 | ||
75 | BUG_ON(!sig); | |
76 | BUG_ON(!atomic_read(&sig->count)); | |
77 | ||
78 | rcu_read_lock(); | |
79 | sighand = rcu_dereference(tsk->sighand); | |
80 | spin_lock(&sighand->siglock); | |
81 | ||
82 | posix_cpu_timers_exit(tsk); | |
83 | if (atomic_dec_and_test(&sig->count)) | |
84 | posix_cpu_timers_exit_group(tsk); | |
85 | else { | |
86 | /* | |
87 | * If there is any task waiting for the group exit | |
88 | * then notify it: | |
89 | */ | |
90 | if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) { | |
91 | wake_up_process(sig->group_exit_task); | |
92 | sig->group_exit_task = NULL; | |
93 | } | |
94 | if (tsk == sig->curr_target) | |
95 | sig->curr_target = next_thread(tsk); | |
96 | /* | |
97 | * Accumulate here the counters for all threads but the | |
98 | * group leader as they die, so they can be added into | |
99 | * the process-wide totals when those are taken. | |
100 | * The group leader stays around as a zombie as long | |
101 | * as there are other threads. When it gets reaped, | |
102 | * the exit.c code will add its counts into these totals. | |
103 | * We won't ever get here for the group leader, since it | |
104 | * will have been the last reference on the signal_struct. | |
105 | */ | |
106 | sig->utime = cputime_add(sig->utime, tsk->utime); | |
107 | sig->stime = cputime_add(sig->stime, tsk->stime); | |
108 | sig->min_flt += tsk->min_flt; | |
109 | sig->maj_flt += tsk->maj_flt; | |
110 | sig->nvcsw += tsk->nvcsw; | |
111 | sig->nivcsw += tsk->nivcsw; | |
112 | sig->sched_time += tsk->sched_time; | |
113 | sig = NULL; /* Marker for below. */ | |
114 | } | |
115 | ||
5876700c ON |
116 | __unhash_process(tsk); |
117 | ||
6a14c5c9 | 118 | tsk->signal = NULL; |
a7e5328a | 119 | tsk->sighand = NULL; |
6a14c5c9 ON |
120 | spin_unlock(&sighand->siglock); |
121 | rcu_read_unlock(); | |
122 | ||
a7e5328a | 123 | __cleanup_sighand(sighand); |
6a14c5c9 ON |
124 | clear_tsk_thread_flag(tsk,TIF_SIGPENDING); |
125 | flush_sigqueue(&tsk->pending); | |
126 | if (sig) { | |
127 | flush_sigqueue(&sig->shared_pending); | |
128 | __cleanup_signal(sig); | |
129 | } | |
130 | } | |
131 | ||
8c7904a0 EB |
132 | static void delayed_put_task_struct(struct rcu_head *rhp) |
133 | { | |
134 | put_task_struct(container_of(rhp, struct task_struct, rcu)); | |
135 | } | |
136 | ||
1da177e4 LT |
137 | void release_task(struct task_struct * p) |
138 | { | |
139 | int zap_leader; | |
140 | task_t *leader; | |
141 | struct dentry *proc_dentry; | |
142 | ||
1f09f974 | 143 | repeat: |
1da177e4 LT |
144 | atomic_dec(&p->user->processes); |
145 | spin_lock(&p->proc_lock); | |
146 | proc_dentry = proc_pid_unhash(p); | |
147 | write_lock_irq(&tasklist_lock); | |
1f09f974 | 148 | ptrace_unlink(p); |
1da177e4 LT |
149 | BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children)); |
150 | __exit_signal(p); | |
35f5cad8 | 151 | |
1da177e4 LT |
152 | /* |
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); | |
175 | spin_unlock(&p->proc_lock); | |
176 | proc_pid_flush(proc_dentry); | |
177 | release_thread(p); | |
8c7904a0 | 178 | call_rcu(&p->rcu, delayed_put_task_struct); |
1da177e4 LT |
179 | |
180 | p = leader; | |
181 | if (unlikely(zap_leader)) | |
182 | goto repeat; | |
183 | } | |
184 | ||
1da177e4 LT |
185 | /* |
186 | * This checks not only the pgrp, but falls back on the pid if no | |
187 | * satisfactory pgrp is found. I dunno - gdb doesn't work correctly | |
188 | * without this... | |
189 | */ | |
190 | int session_of_pgrp(int pgrp) | |
191 | { | |
192 | struct task_struct *p; | |
193 | int sid = -1; | |
194 | ||
195 | read_lock(&tasklist_lock); | |
196 | do_each_task_pid(pgrp, PIDTYPE_PGID, p) { | |
197 | if (p->signal->session > 0) { | |
198 | sid = p->signal->session; | |
199 | goto out; | |
200 | } | |
201 | } while_each_task_pid(pgrp, PIDTYPE_PGID, p); | |
202 | p = find_task_by_pid(pgrp); | |
203 | if (p) | |
204 | sid = p->signal->session; | |
205 | out: | |
206 | read_unlock(&tasklist_lock); | |
207 | ||
208 | return sid; | |
209 | } | |
210 | ||
211 | /* | |
212 | * Determine if a process group is "orphaned", according to the POSIX | |
213 | * definition in 2.2.2.52. Orphaned process groups are not to be affected | |
214 | * by terminal-generated stop signals. Newly orphaned process groups are | |
215 | * to receive a SIGHUP and a SIGCONT. | |
216 | * | |
217 | * "I ask you, have you ever known what it is to be an orphan?" | |
218 | */ | |
219 | static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task) | |
220 | { | |
221 | struct task_struct *p; | |
222 | int ret = 1; | |
223 | ||
224 | do_each_task_pid(pgrp, PIDTYPE_PGID, p) { | |
225 | if (p == ignored_task | |
226 | || p->exit_state | |
227 | || p->real_parent->pid == 1) | |
228 | continue; | |
229 | if (process_group(p->real_parent) != pgrp | |
230 | && p->real_parent->signal->session == p->signal->session) { | |
231 | ret = 0; | |
232 | break; | |
233 | } | |
234 | } while_each_task_pid(pgrp, PIDTYPE_PGID, p); | |
235 | return ret; /* (sighing) "Often!" */ | |
236 | } | |
237 | ||
238 | int is_orphaned_pgrp(int pgrp) | |
239 | { | |
240 | int retval; | |
241 | ||
242 | read_lock(&tasklist_lock); | |
243 | retval = will_become_orphaned_pgrp(pgrp, NULL); | |
244 | read_unlock(&tasklist_lock); | |
245 | ||
246 | return retval; | |
247 | } | |
248 | ||
858119e1 | 249 | static int has_stopped_jobs(int pgrp) |
1da177e4 LT |
250 | { |
251 | int retval = 0; | |
252 | struct task_struct *p; | |
253 | ||
254 | do_each_task_pid(pgrp, PIDTYPE_PGID, p) { | |
255 | if (p->state != TASK_STOPPED) | |
256 | continue; | |
257 | ||
258 | /* If p is stopped by a debugger on a signal that won't | |
259 | stop it, then don't count p as stopped. This isn't | |
260 | perfect but it's a good approximation. */ | |
261 | if (unlikely (p->ptrace) | |
262 | && p->exit_code != SIGSTOP | |
263 | && p->exit_code != SIGTSTP | |
264 | && p->exit_code != SIGTTOU | |
265 | && p->exit_code != SIGTTIN) | |
266 | continue; | |
267 | ||
268 | retval = 1; | |
269 | break; | |
270 | } while_each_task_pid(pgrp, PIDTYPE_PGID, p); | |
271 | return retval; | |
272 | } | |
273 | ||
274 | /** | |
4dc3b16b | 275 | * reparent_to_init - Reparent the calling kernel thread to the init task. |
1da177e4 LT |
276 | * |
277 | * If a kernel thread is launched as a result of a system call, or if | |
278 | * it ever exits, it should generally reparent itself to init so that | |
279 | * it is correctly cleaned up on exit. | |
280 | * | |
281 | * The various task state such as scheduling policy and priority may have | |
282 | * been inherited from a user process, so we reset them to sane values here. | |
283 | * | |
284 | * NOTE that reparent_to_init() gives the caller full capabilities. | |
285 | */ | |
858119e1 | 286 | static void reparent_to_init(void) |
1da177e4 LT |
287 | { |
288 | write_lock_irq(&tasklist_lock); | |
289 | ||
290 | ptrace_unlink(current); | |
291 | /* Reparent to init */ | |
9b678ece | 292 | remove_parent(current); |
1da177e4 LT |
293 | current->parent = child_reaper; |
294 | current->real_parent = child_reaper; | |
9b678ece | 295 | add_parent(current); |
1da177e4 LT |
296 | |
297 | /* Set the exit signal to SIGCHLD so we signal init on exit */ | |
298 | current->exit_signal = SIGCHLD; | |
299 | ||
b0a9499c IM |
300 | if ((current->policy == SCHED_NORMAL || |
301 | current->policy == SCHED_BATCH) | |
302 | && (task_nice(current) < 0)) | |
1da177e4 LT |
303 | set_user_nice(current, 0); |
304 | /* cpus_allowed? */ | |
305 | /* rt_priority? */ | |
306 | /* signals? */ | |
307 | security_task_reparent_to_init(current); | |
308 | memcpy(current->signal->rlim, init_task.signal->rlim, | |
309 | sizeof(current->signal->rlim)); | |
310 | atomic_inc(&(INIT_USER->__count)); | |
311 | write_unlock_irq(&tasklist_lock); | |
312 | switch_uid(INIT_USER); | |
313 | } | |
314 | ||
315 | void __set_special_pids(pid_t session, pid_t pgrp) | |
316 | { | |
e19f247a | 317 | struct task_struct *curr = current->group_leader; |
1da177e4 LT |
318 | |
319 | if (curr->signal->session != session) { | |
320 | detach_pid(curr, PIDTYPE_SID); | |
321 | curr->signal->session = session; | |
322 | attach_pid(curr, PIDTYPE_SID, session); | |
323 | } | |
324 | if (process_group(curr) != pgrp) { | |
325 | detach_pid(curr, PIDTYPE_PGID); | |
326 | curr->signal->pgrp = pgrp; | |
327 | attach_pid(curr, PIDTYPE_PGID, pgrp); | |
328 | } | |
329 | } | |
330 | ||
331 | void set_special_pids(pid_t session, pid_t pgrp) | |
332 | { | |
333 | write_lock_irq(&tasklist_lock); | |
334 | __set_special_pids(session, pgrp); | |
335 | write_unlock_irq(&tasklist_lock); | |
336 | } | |
337 | ||
338 | /* | |
339 | * Let kernel threads use this to say that they | |
340 | * allow a certain signal (since daemonize() will | |
341 | * have disabled all of them by default). | |
342 | */ | |
343 | int allow_signal(int sig) | |
344 | { | |
7ed20e1a | 345 | if (!valid_signal(sig) || sig < 1) |
1da177e4 LT |
346 | return -EINVAL; |
347 | ||
348 | spin_lock_irq(¤t->sighand->siglock); | |
349 | sigdelset(¤t->blocked, sig); | |
350 | if (!current->mm) { | |
351 | /* Kernel threads handle their own signals. | |
352 | Let the signal code know it'll be handled, so | |
353 | that they don't get converted to SIGKILL or | |
354 | just silently dropped */ | |
355 | current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2; | |
356 | } | |
357 | recalc_sigpending(); | |
358 | spin_unlock_irq(¤t->sighand->siglock); | |
359 | return 0; | |
360 | } | |
361 | ||
362 | EXPORT_SYMBOL(allow_signal); | |
363 | ||
364 | int disallow_signal(int sig) | |
365 | { | |
7ed20e1a | 366 | if (!valid_signal(sig) || sig < 1) |
1da177e4 LT |
367 | return -EINVAL; |
368 | ||
369 | spin_lock_irq(¤t->sighand->siglock); | |
370 | sigaddset(¤t->blocked, sig); | |
371 | recalc_sigpending(); | |
372 | spin_unlock_irq(¤t->sighand->siglock); | |
373 | return 0; | |
374 | } | |
375 | ||
376 | EXPORT_SYMBOL(disallow_signal); | |
377 | ||
378 | /* | |
379 | * Put all the gunge required to become a kernel thread without | |
380 | * attached user resources in one place where it belongs. | |
381 | */ | |
382 | ||
383 | void daemonize(const char *name, ...) | |
384 | { | |
385 | va_list args; | |
386 | struct fs_struct *fs; | |
387 | sigset_t blocked; | |
388 | ||
389 | va_start(args, name); | |
390 | vsnprintf(current->comm, sizeof(current->comm), name, args); | |
391 | va_end(args); | |
392 | ||
393 | /* | |
394 | * If we were started as result of loading a module, close all of the | |
395 | * user space pages. We don't need them, and if we didn't close them | |
396 | * they would be locked into memory. | |
397 | */ | |
398 | exit_mm(current); | |
399 | ||
400 | set_special_pids(1, 1); | |
70522e12 | 401 | mutex_lock(&tty_mutex); |
1da177e4 | 402 | current->signal->tty = NULL; |
70522e12 | 403 | mutex_unlock(&tty_mutex); |
1da177e4 LT |
404 | |
405 | /* Block and flush all signals */ | |
406 | sigfillset(&blocked); | |
407 | sigprocmask(SIG_BLOCK, &blocked, NULL); | |
408 | flush_signals(current); | |
409 | ||
410 | /* Become as one with the init task */ | |
411 | ||
412 | exit_fs(current); /* current->fs->count--; */ | |
413 | fs = init_task.fs; | |
414 | current->fs = fs; | |
415 | atomic_inc(&fs->count); | |
5914811a BS |
416 | exit_namespace(current); |
417 | current->namespace = init_task.namespace; | |
418 | get_namespace(current->namespace); | |
1da177e4 LT |
419 | exit_files(current); |
420 | current->files = init_task.files; | |
421 | atomic_inc(¤t->files->count); | |
422 | ||
423 | reparent_to_init(); | |
424 | } | |
425 | ||
426 | EXPORT_SYMBOL(daemonize); | |
427 | ||
858119e1 | 428 | static void close_files(struct files_struct * files) |
1da177e4 LT |
429 | { |
430 | int i, j; | |
badf1662 | 431 | struct fdtable *fdt; |
1da177e4 LT |
432 | |
433 | j = 0; | |
4fb3a538 DS |
434 | |
435 | /* | |
436 | * It is safe to dereference the fd table without RCU or | |
437 | * ->file_lock because this is the last reference to the | |
438 | * files structure. | |
439 | */ | |
badf1662 | 440 | fdt = files_fdtable(files); |
1da177e4 LT |
441 | for (;;) { |
442 | unsigned long set; | |
443 | i = j * __NFDBITS; | |
badf1662 | 444 | if (i >= fdt->max_fdset || i >= fdt->max_fds) |
1da177e4 | 445 | break; |
badf1662 | 446 | set = fdt->open_fds->fds_bits[j++]; |
1da177e4 LT |
447 | while (set) { |
448 | if (set & 1) { | |
badf1662 | 449 | struct file * file = xchg(&fdt->fd[i], NULL); |
1da177e4 LT |
450 | if (file) |
451 | filp_close(file, files); | |
452 | } | |
453 | i++; | |
454 | set >>= 1; | |
455 | } | |
456 | } | |
457 | } | |
458 | ||
459 | struct files_struct *get_files_struct(struct task_struct *task) | |
460 | { | |
461 | struct files_struct *files; | |
462 | ||
463 | task_lock(task); | |
464 | files = task->files; | |
465 | if (files) | |
466 | atomic_inc(&files->count); | |
467 | task_unlock(task); | |
468 | ||
469 | return files; | |
470 | } | |
471 | ||
472 | void fastcall put_files_struct(struct files_struct *files) | |
473 | { | |
badf1662 DS |
474 | struct fdtable *fdt; |
475 | ||
1da177e4 LT |
476 | if (atomic_dec_and_test(&files->count)) { |
477 | close_files(files); | |
478 | /* | |
479 | * Free the fd and fdset arrays if we expanded them. | |
ab2af1f5 DS |
480 | * If the fdtable was embedded, pass files for freeing |
481 | * at the end of the RCU grace period. Otherwise, | |
482 | * you can free files immediately. | |
1da177e4 | 483 | */ |
badf1662 | 484 | fdt = files_fdtable(files); |
ab2af1f5 DS |
485 | if (fdt == &files->fdtab) |
486 | fdt->free_files = files; | |
487 | else | |
488 | kmem_cache_free(files_cachep, files); | |
489 | free_fdtable(fdt); | |
1da177e4 LT |
490 | } |
491 | } | |
492 | ||
493 | EXPORT_SYMBOL(put_files_struct); | |
494 | ||
495 | static inline void __exit_files(struct task_struct *tsk) | |
496 | { | |
497 | struct files_struct * files = tsk->files; | |
498 | ||
499 | if (files) { | |
500 | task_lock(tsk); | |
501 | tsk->files = NULL; | |
502 | task_unlock(tsk); | |
503 | put_files_struct(files); | |
504 | } | |
505 | } | |
506 | ||
507 | void exit_files(struct task_struct *tsk) | |
508 | { | |
509 | __exit_files(tsk); | |
510 | } | |
511 | ||
512 | static inline void __put_fs_struct(struct fs_struct *fs) | |
513 | { | |
514 | /* No need to hold fs->lock if we are killing it */ | |
515 | if (atomic_dec_and_test(&fs->count)) { | |
516 | dput(fs->root); | |
517 | mntput(fs->rootmnt); | |
518 | dput(fs->pwd); | |
519 | mntput(fs->pwdmnt); | |
520 | if (fs->altroot) { | |
521 | dput(fs->altroot); | |
522 | mntput(fs->altrootmnt); | |
523 | } | |
524 | kmem_cache_free(fs_cachep, fs); | |
525 | } | |
526 | } | |
527 | ||
528 | void put_fs_struct(struct fs_struct *fs) | |
529 | { | |
530 | __put_fs_struct(fs); | |
531 | } | |
532 | ||
533 | static inline void __exit_fs(struct task_struct *tsk) | |
534 | { | |
535 | struct fs_struct * fs = tsk->fs; | |
536 | ||
537 | if (fs) { | |
538 | task_lock(tsk); | |
539 | tsk->fs = NULL; | |
540 | task_unlock(tsk); | |
541 | __put_fs_struct(fs); | |
542 | } | |
543 | } | |
544 | ||
545 | void exit_fs(struct task_struct *tsk) | |
546 | { | |
547 | __exit_fs(tsk); | |
548 | } | |
549 | ||
550 | EXPORT_SYMBOL_GPL(exit_fs); | |
551 | ||
552 | /* | |
553 | * Turn us into a lazy TLB process if we | |
554 | * aren't already.. | |
555 | */ | |
408b664a | 556 | static void exit_mm(struct task_struct * tsk) |
1da177e4 LT |
557 | { |
558 | struct mm_struct *mm = tsk->mm; | |
559 | ||
560 | mm_release(tsk, mm); | |
561 | if (!mm) | |
562 | return; | |
563 | /* | |
564 | * Serialize with any possible pending coredump. | |
565 | * We must hold mmap_sem around checking core_waiters | |
566 | * and clearing tsk->mm. The core-inducing thread | |
567 | * will increment core_waiters for each thread in the | |
568 | * group with ->mm != NULL. | |
569 | */ | |
570 | down_read(&mm->mmap_sem); | |
571 | if (mm->core_waiters) { | |
572 | up_read(&mm->mmap_sem); | |
573 | down_write(&mm->mmap_sem); | |
574 | if (!--mm->core_waiters) | |
575 | complete(mm->core_startup_done); | |
576 | up_write(&mm->mmap_sem); | |
577 | ||
578 | wait_for_completion(&mm->core_done); | |
579 | down_read(&mm->mmap_sem); | |
580 | } | |
581 | atomic_inc(&mm->mm_count); | |
582 | if (mm != tsk->active_mm) BUG(); | |
583 | /* more a memory barrier than a real lock */ | |
584 | task_lock(tsk); | |
585 | tsk->mm = NULL; | |
586 | up_read(&mm->mmap_sem); | |
587 | enter_lazy_tlb(mm, current); | |
588 | task_unlock(tsk); | |
589 | mmput(mm); | |
590 | } | |
591 | ||
d799f035 | 592 | static inline void choose_new_parent(task_t *p, task_t *reaper) |
1da177e4 LT |
593 | { |
594 | /* | |
595 | * Make sure we're not reparenting to ourselves and that | |
596 | * the parent is not a zombie. | |
597 | */ | |
d799f035 | 598 | BUG_ON(p == reaper || reaper->exit_state); |
1da177e4 | 599 | p->real_parent = reaper; |
1da177e4 LT |
600 | } |
601 | ||
858119e1 | 602 | static void reparent_thread(task_t *p, task_t *father, int traced) |
1da177e4 LT |
603 | { |
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. */ | |
b67a1b9e | 610 | group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p); |
1da177e4 LT |
611 | |
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; | |
6ac781b1 | 623 | remove_parent(p); |
1da177e4 | 624 | p->parent = p->real_parent; |
6ac781b1 | 625 | add_parent(p); |
1da177e4 LT |
626 | |
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)) && | |
650 | (p->signal->session == father->signal->session)) { | |
651 | int pgrp = process_group(p); | |
652 | ||
653 | if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) { | |
b67a1b9e ON |
654 | __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp); |
655 | __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp); | |
1da177e4 LT |
656 | } |
657 | } | |
658 | } | |
659 | ||
660 | /* | |
661 | * When we die, we re-parent all our children. | |
662 | * Try to give them to another thread in our thread | |
663 | * group, and if no such member exists, give it to | |
664 | * the global child reaper process (ie "init") | |
665 | */ | |
858119e1 | 666 | static void forget_original_parent(struct task_struct * father, |
1da177e4 LT |
667 | struct list_head *to_release) |
668 | { | |
669 | struct task_struct *p, *reaper = father; | |
670 | struct list_head *_p, *_n; | |
671 | ||
672 | do { | |
673 | reaper = next_thread(reaper); | |
674 | if (reaper == father) { | |
675 | reaper = child_reaper; | |
676 | break; | |
677 | } | |
678 | } while (reaper->exit_state); | |
679 | ||
680 | /* | |
681 | * There are only two places where our children can be: | |
682 | * | |
683 | * - in our child list | |
684 | * - in our ptraced child list | |
685 | * | |
686 | * Search them and reparent children. | |
687 | */ | |
688 | list_for_each_safe(_p, _n, &father->children) { | |
689 | int ptrace; | |
690 | p = list_entry(_p,struct task_struct,sibling); | |
691 | ||
692 | ptrace = p->ptrace; | |
693 | ||
694 | /* if father isn't the real parent, then ptrace must be enabled */ | |
695 | BUG_ON(father != p->real_parent && !ptrace); | |
696 | ||
697 | if (father == p->real_parent) { | |
698 | /* reparent with a reaper, real father it's us */ | |
d799f035 | 699 | choose_new_parent(p, reaper); |
1da177e4 LT |
700 | reparent_thread(p, father, 0); |
701 | } else { | |
702 | /* reparent ptraced task to its real parent */ | |
703 | __ptrace_unlink (p); | |
704 | if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 && | |
705 | thread_group_empty(p)) | |
706 | do_notify_parent(p, p->exit_signal); | |
707 | } | |
708 | ||
709 | /* | |
710 | * if the ptraced child is a zombie with exit_signal == -1 | |
711 | * we must collect it before we exit, or it will remain | |
712 | * zombie forever since we prevented it from self-reap itself | |
713 | * while it was being traced by us, to be able to see it in wait4. | |
714 | */ | |
715 | if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1)) | |
716 | list_add(&p->ptrace_list, to_release); | |
717 | } | |
718 | list_for_each_safe(_p, _n, &father->ptrace_children) { | |
719 | p = list_entry(_p,struct task_struct,ptrace_list); | |
d799f035 | 720 | choose_new_parent(p, reaper); |
1da177e4 LT |
721 | reparent_thread(p, father, 1); |
722 | } | |
723 | } | |
724 | ||
725 | /* | |
726 | * Send signals to all our closest relatives so that they know | |
727 | * to properly mourn us.. | |
728 | */ | |
729 | static void exit_notify(struct task_struct *tsk) | |
730 | { | |
731 | int state; | |
732 | struct task_struct *t; | |
733 | struct list_head ptrace_dead, *_p, *_n; | |
734 | ||
735 | if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT) | |
736 | && !thread_group_empty(tsk)) { | |
737 | /* | |
738 | * This occurs when there was a race between our exit | |
739 | * syscall and a group signal choosing us as the one to | |
740 | * wake up. It could be that we are the only thread | |
741 | * alerted to check for pending signals, but another thread | |
742 | * should be woken now to take the signal since we will not. | |
743 | * Now we'll wake all the threads in the group just to make | |
744 | * sure someone gets all the pending signals. | |
745 | */ | |
746 | read_lock(&tasklist_lock); | |
747 | spin_lock_irq(&tsk->sighand->siglock); | |
748 | for (t = next_thread(tsk); t != tsk; t = next_thread(t)) | |
749 | if (!signal_pending(t) && !(t->flags & PF_EXITING)) { | |
750 | recalc_sigpending_tsk(t); | |
751 | if (signal_pending(t)) | |
752 | signal_wake_up(t, 0); | |
753 | } | |
754 | spin_unlock_irq(&tsk->sighand->siglock); | |
755 | read_unlock(&tasklist_lock); | |
756 | } | |
757 | ||
758 | write_lock_irq(&tasklist_lock); | |
759 | ||
760 | /* | |
761 | * This does two things: | |
762 | * | |
763 | * A. Make init inherit all the child processes | |
764 | * B. Check to see if any process groups have become orphaned | |
765 | * as a result of our exiting, and if they have any stopped | |
766 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
767 | */ | |
768 | ||
769 | INIT_LIST_HEAD(&ptrace_dead); | |
770 | forget_original_parent(tsk, &ptrace_dead); | |
771 | BUG_ON(!list_empty(&tsk->children)); | |
772 | BUG_ON(!list_empty(&tsk->ptrace_children)); | |
773 | ||
774 | /* | |
775 | * Check to see if any process groups have become orphaned | |
776 | * as a result of our exiting, and if they have any stopped | |
777 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) | |
778 | * | |
779 | * Case i: Our father is in a different pgrp than we are | |
780 | * and we were the only connection outside, so our pgrp | |
781 | * is about to become orphaned. | |
782 | */ | |
783 | ||
784 | t = tsk->real_parent; | |
785 | ||
786 | if ((process_group(t) != process_group(tsk)) && | |
787 | (t->signal->session == tsk->signal->session) && | |
788 | will_become_orphaned_pgrp(process_group(tsk), tsk) && | |
789 | has_stopped_jobs(process_group(tsk))) { | |
b67a1b9e ON |
790 | __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk)); |
791 | __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk)); | |
1da177e4 LT |
792 | } |
793 | ||
794 | /* Let father know we died | |
795 | * | |
796 | * Thread signals are configurable, but you aren't going to use | |
797 | * that to send signals to arbitary processes. | |
798 | * That stops right now. | |
799 | * | |
800 | * If the parent exec id doesn't match the exec id we saved | |
801 | * when we started then we know the parent has changed security | |
802 | * domain. | |
803 | * | |
804 | * If our self_exec id doesn't match our parent_exec_id then | |
805 | * we have changed execution domain as these two values started | |
806 | * the same after a fork. | |
807 | * | |
808 | */ | |
809 | ||
810 | if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 && | |
811 | ( tsk->parent_exec_id != t->self_exec_id || | |
812 | tsk->self_exec_id != tsk->parent_exec_id) | |
813 | && !capable(CAP_KILL)) | |
814 | tsk->exit_signal = SIGCHLD; | |
815 | ||
816 | ||
817 | /* If something other than our normal parent is ptracing us, then | |
818 | * send it a SIGCHLD instead of honoring exit_signal. exit_signal | |
819 | * only has special meaning to our real parent. | |
820 | */ | |
821 | if (tsk->exit_signal != -1 && thread_group_empty(tsk)) { | |
822 | int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD; | |
823 | do_notify_parent(tsk, signal); | |
824 | } else if (tsk->ptrace) { | |
825 | do_notify_parent(tsk, SIGCHLD); | |
826 | } | |
827 | ||
828 | state = EXIT_ZOMBIE; | |
829 | if (tsk->exit_signal == -1 && | |
830 | (likely(tsk->ptrace == 0) || | |
831 | unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT))) | |
832 | state = EXIT_DEAD; | |
833 | tsk->exit_state = state; | |
834 | ||
835 | write_unlock_irq(&tasklist_lock); | |
836 | ||
837 | list_for_each_safe(_p, _n, &ptrace_dead) { | |
838 | list_del_init(_p); | |
839 | t = list_entry(_p,struct task_struct,ptrace_list); | |
840 | release_task(t); | |
841 | } | |
842 | ||
843 | /* If the process is dead, release it - nobody will wait for it */ | |
844 | if (state == EXIT_DEAD) | |
845 | release_task(tsk); | |
1da177e4 LT |
846 | } |
847 | ||
848 | fastcall NORET_TYPE void do_exit(long code) | |
849 | { | |
850 | struct task_struct *tsk = current; | |
851 | int group_dead; | |
852 | ||
853 | profile_task_exit(tsk); | |
854 | ||
22e2c507 JA |
855 | WARN_ON(atomic_read(&tsk->fs_excl)); |
856 | ||
1da177e4 LT |
857 | if (unlikely(in_interrupt())) |
858 | panic("Aiee, killing interrupt handler!"); | |
859 | if (unlikely(!tsk->pid)) | |
860 | panic("Attempted to kill the idle task!"); | |
fef23e7f | 861 | if (unlikely(tsk == child_reaper)) |
1da177e4 | 862 | panic("Attempted to kill init!"); |
1da177e4 LT |
863 | |
864 | if (unlikely(current->ptrace & PT_TRACE_EXIT)) { | |
865 | current->ptrace_message = code; | |
866 | ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP); | |
867 | } | |
868 | ||
df164db5 AN |
869 | /* |
870 | * We're taking recursive faults here in do_exit. Safest is to just | |
871 | * leave this task alone and wait for reboot. | |
872 | */ | |
873 | if (unlikely(tsk->flags & PF_EXITING)) { | |
874 | printk(KERN_ALERT | |
875 | "Fixing recursive fault but reboot is needed!\n"); | |
afc847b7 AV |
876 | if (tsk->io_context) |
877 | exit_io_context(); | |
df164db5 AN |
878 | set_current_state(TASK_UNINTERRUPTIBLE); |
879 | schedule(); | |
880 | } | |
881 | ||
1da177e4 LT |
882 | tsk->flags |= PF_EXITING; |
883 | ||
a362f463 LT |
884 | /* |
885 | * Make sure we don't try to process any timer firings | |
886 | * while we are already exiting. | |
887 | */ | |
888 | tsk->it_virt_expires = cputime_zero; | |
889 | tsk->it_prof_expires = cputime_zero; | |
890 | tsk->it_sched_expires = 0; | |
891 | ||
1da177e4 LT |
892 | if (unlikely(in_atomic())) |
893 | printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n", | |
894 | current->comm, current->pid, | |
895 | preempt_count()); | |
896 | ||
897 | acct_update_integrals(tsk); | |
365e9c87 HD |
898 | if (tsk->mm) { |
899 | update_hiwater_rss(tsk->mm); | |
900 | update_hiwater_vm(tsk->mm); | |
901 | } | |
1da177e4 | 902 | group_dead = atomic_dec_and_test(&tsk->signal->live); |
c3068951 | 903 | if (group_dead) { |
2ff678b8 | 904 | hrtimer_cancel(&tsk->signal->real_timer); |
25f407f0 | 905 | exit_itimers(tsk->signal); |
1da177e4 | 906 | acct_process(code); |
c3068951 | 907 | } |
0771dfef IM |
908 | if (unlikely(tsk->robust_list)) |
909 | exit_robust_list(tsk); | |
34f192c6 IM |
910 | #ifdef CONFIG_COMPAT |
911 | if (unlikely(tsk->compat_robust_list)) | |
912 | compat_exit_robust_list(tsk); | |
913 | #endif | |
fa84cb93 AV |
914 | if (unlikely(tsk->audit_context)) |
915 | audit_free(tsk); | |
1da177e4 LT |
916 | exit_mm(tsk); |
917 | ||
918 | exit_sem(tsk); | |
919 | __exit_files(tsk); | |
920 | __exit_fs(tsk); | |
921 | exit_namespace(tsk); | |
922 | exit_thread(); | |
923 | cpuset_exit(tsk); | |
924 | exit_keys(tsk); | |
925 | ||
926 | if (group_dead && tsk->signal->leader) | |
927 | disassociate_ctty(1); | |
928 | ||
a1261f54 | 929 | module_put(task_thread_info(tsk)->exec_domain->module); |
1da177e4 LT |
930 | if (tsk->binfmt) |
931 | module_put(tsk->binfmt->module); | |
932 | ||
933 | tsk->exit_code = code; | |
9f46080c | 934 | proc_exit_connector(tsk); |
1da177e4 LT |
935 | exit_notify(tsk); |
936 | #ifdef CONFIG_NUMA | |
937 | mpol_free(tsk->mempolicy); | |
938 | tsk->mempolicy = NULL; | |
939 | #endif | |
de5097c2 IM |
940 | /* |
941 | * If DEBUG_MUTEXES is on, make sure we are holding no locks: | |
942 | */ | |
943 | mutex_debug_check_no_locks_held(tsk); | |
1da177e4 | 944 | |
afc847b7 AV |
945 | if (tsk->io_context) |
946 | exit_io_context(); | |
947 | ||
b92ce558 JA |
948 | if (tsk->splice_pipe) |
949 | __free_pipe_info(tsk->splice_pipe); | |
950 | ||
7407251a CQH |
951 | /* PF_DEAD causes final put_task_struct after we schedule. */ |
952 | preempt_disable(); | |
953 | BUG_ON(tsk->flags & PF_DEAD); | |
954 | tsk->flags |= PF_DEAD; | |
955 | ||
1da177e4 LT |
956 | schedule(); |
957 | BUG(); | |
958 | /* Avoid "noreturn function does return". */ | |
959 | for (;;) ; | |
960 | } | |
961 | ||
012914da RA |
962 | EXPORT_SYMBOL_GPL(do_exit); |
963 | ||
1da177e4 LT |
964 | NORET_TYPE void complete_and_exit(struct completion *comp, long code) |
965 | { | |
966 | if (comp) | |
967 | complete(comp); | |
968 | ||
969 | do_exit(code); | |
970 | } | |
971 | ||
972 | EXPORT_SYMBOL(complete_and_exit); | |
973 | ||
974 | asmlinkage long sys_exit(int error_code) | |
975 | { | |
976 | do_exit((error_code&0xff)<<8); | |
977 | } | |
978 | ||
1da177e4 LT |
979 | /* |
980 | * Take down every thread in the group. This is called by fatal signals | |
981 | * as well as by sys_exit_group (below). | |
982 | */ | |
983 | NORET_TYPE void | |
984 | do_group_exit(int exit_code) | |
985 | { | |
986 | BUG_ON(exit_code & 0x80); /* core dumps don't get here */ | |
987 | ||
988 | if (current->signal->flags & SIGNAL_GROUP_EXIT) | |
989 | exit_code = current->signal->group_exit_code; | |
990 | else if (!thread_group_empty(current)) { | |
991 | struct signal_struct *const sig = current->signal; | |
992 | struct sighand_struct *const sighand = current->sighand; | |
1da177e4 LT |
993 | spin_lock_irq(&sighand->siglock); |
994 | if (sig->flags & SIGNAL_GROUP_EXIT) | |
995 | /* Another thread got here before we took the lock. */ | |
996 | exit_code = sig->group_exit_code; | |
997 | else { | |
1da177e4 LT |
998 | sig->group_exit_code = exit_code; |
999 | zap_other_threads(current); | |
1000 | } | |
1001 | spin_unlock_irq(&sighand->siglock); | |
1da177e4 LT |
1002 | } |
1003 | ||
1004 | do_exit(exit_code); | |
1005 | /* NOTREACHED */ | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * this kills every thread in the thread group. Note that any externally | |
1010 | * wait4()-ing process will get the correct exit code - even if this | |
1011 | * thread is not the thread group leader. | |
1012 | */ | |
1013 | asmlinkage void sys_exit_group(int error_code) | |
1014 | { | |
1015 | do_group_exit((error_code & 0xff) << 8); | |
1016 | } | |
1017 | ||
1018 | static int eligible_child(pid_t pid, int options, task_t *p) | |
1019 | { | |
1020 | if (pid > 0) { | |
1021 | if (p->pid != pid) | |
1022 | return 0; | |
1023 | } else if (!pid) { | |
1024 | if (process_group(p) != process_group(current)) | |
1025 | return 0; | |
1026 | } else if (pid != -1) { | |
1027 | if (process_group(p) != -pid) | |
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Do not consider detached threads that are | |
1033 | * not ptraced: | |
1034 | */ | |
1035 | if (p->exit_signal == -1 && !p->ptrace) | |
1036 | return 0; | |
1037 | ||
1038 | /* Wait for all children (clone and not) if __WALL is set; | |
1039 | * otherwise, wait for clone children *only* if __WCLONE is | |
1040 | * set; otherwise, wait for non-clone children *only*. (Note: | |
1041 | * A "clone" child here is one that reports to its parent | |
1042 | * using a signal other than SIGCHLD.) */ | |
1043 | if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0)) | |
1044 | && !(options & __WALL)) | |
1045 | return 0; | |
1046 | /* | |
1047 | * Do not consider thread group leaders that are | |
1048 | * in a non-empty thread group: | |
1049 | */ | |
1050 | if (current->tgid != p->tgid && delay_group_leader(p)) | |
1051 | return 2; | |
1052 | ||
1053 | if (security_task_wait(p)) | |
1054 | return 0; | |
1055 | ||
1056 | return 1; | |
1057 | } | |
1058 | ||
1059 | static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid, | |
1060 | int why, int status, | |
1061 | struct siginfo __user *infop, | |
1062 | struct rusage __user *rusagep) | |
1063 | { | |
1064 | int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0; | |
1065 | put_task_struct(p); | |
1066 | if (!retval) | |
1067 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1068 | if (!retval) | |
1069 | retval = put_user(0, &infop->si_errno); | |
1070 | if (!retval) | |
1071 | retval = put_user((short)why, &infop->si_code); | |
1072 | if (!retval) | |
1073 | retval = put_user(pid, &infop->si_pid); | |
1074 | if (!retval) | |
1075 | retval = put_user(uid, &infop->si_uid); | |
1076 | if (!retval) | |
1077 | retval = put_user(status, &infop->si_status); | |
1078 | if (!retval) | |
1079 | retval = pid; | |
1080 | return retval; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold | |
1085 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1086 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1087 | * released the lock and the system call should return. | |
1088 | */ | |
1089 | static int wait_task_zombie(task_t *p, int noreap, | |
1090 | struct siginfo __user *infop, | |
1091 | int __user *stat_addr, struct rusage __user *ru) | |
1092 | { | |
1093 | unsigned long state; | |
1094 | int retval; | |
1095 | int status; | |
1096 | ||
1097 | if (unlikely(noreap)) { | |
1098 | pid_t pid = p->pid; | |
1099 | uid_t uid = p->uid; | |
1100 | int exit_code = p->exit_code; | |
1101 | int why, status; | |
1102 | ||
1103 | if (unlikely(p->exit_state != EXIT_ZOMBIE)) | |
1104 | return 0; | |
1105 | if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) | |
1106 | return 0; | |
1107 | get_task_struct(p); | |
1108 | read_unlock(&tasklist_lock); | |
1109 | if ((exit_code & 0x7f) == 0) { | |
1110 | why = CLD_EXITED; | |
1111 | status = exit_code >> 8; | |
1112 | } else { | |
1113 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1114 | status = exit_code & 0x7f; | |
1115 | } | |
1116 | return wait_noreap_copyout(p, pid, uid, why, | |
1117 | status, infop, ru); | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * Try to move the task's state to DEAD | |
1122 | * only one thread is allowed to do this: | |
1123 | */ | |
1124 | state = xchg(&p->exit_state, EXIT_DEAD); | |
1125 | if (state != EXIT_ZOMBIE) { | |
1126 | BUG_ON(state != EXIT_DEAD); | |
1127 | return 0; | |
1128 | } | |
1129 | if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) { | |
1130 | /* | |
1131 | * This can only happen in a race with a ptraced thread | |
1132 | * dying on another processor. | |
1133 | */ | |
1134 | return 0; | |
1135 | } | |
1136 | ||
1137 | if (likely(p->real_parent == p->parent) && likely(p->signal)) { | |
3795e161 JJ |
1138 | struct signal_struct *psig; |
1139 | struct signal_struct *sig; | |
1140 | ||
1da177e4 LT |
1141 | /* |
1142 | * The resource counters for the group leader are in its | |
1143 | * own task_struct. Those for dead threads in the group | |
1144 | * are in its signal_struct, as are those for the child | |
1145 | * processes it has previously reaped. All these | |
1146 | * accumulate in the parent's signal_struct c* fields. | |
1147 | * | |
1148 | * We don't bother to take a lock here to protect these | |
1149 | * p->signal fields, because they are only touched by | |
1150 | * __exit_signal, which runs with tasklist_lock | |
1151 | * write-locked anyway, and so is excluded here. We do | |
1152 | * need to protect the access to p->parent->signal fields, | |
1153 | * as other threads in the parent group can be right | |
1154 | * here reaping other children at the same time. | |
1155 | */ | |
1156 | spin_lock_irq(&p->parent->sighand->siglock); | |
3795e161 JJ |
1157 | psig = p->parent->signal; |
1158 | sig = p->signal; | |
1159 | psig->cutime = | |
1160 | cputime_add(psig->cutime, | |
1da177e4 | 1161 | cputime_add(p->utime, |
3795e161 JJ |
1162 | cputime_add(sig->utime, |
1163 | sig->cutime))); | |
1164 | psig->cstime = | |
1165 | cputime_add(psig->cstime, | |
1da177e4 | 1166 | cputime_add(p->stime, |
3795e161 JJ |
1167 | cputime_add(sig->stime, |
1168 | sig->cstime))); | |
1169 | psig->cmin_flt += | |
1170 | p->min_flt + sig->min_flt + sig->cmin_flt; | |
1171 | psig->cmaj_flt += | |
1172 | p->maj_flt + sig->maj_flt + sig->cmaj_flt; | |
1173 | psig->cnvcsw += | |
1174 | p->nvcsw + sig->nvcsw + sig->cnvcsw; | |
1175 | psig->cnivcsw += | |
1176 | p->nivcsw + sig->nivcsw + sig->cnivcsw; | |
1da177e4 LT |
1177 | spin_unlock_irq(&p->parent->sighand->siglock); |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * Now we are sure this task is interesting, and no other | |
1182 | * thread can reap it because we set its state to EXIT_DEAD. | |
1183 | */ | |
1184 | read_unlock(&tasklist_lock); | |
1185 | ||
1186 | retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0; | |
1187 | status = (p->signal->flags & SIGNAL_GROUP_EXIT) | |
1188 | ? p->signal->group_exit_code : p->exit_code; | |
1189 | if (!retval && stat_addr) | |
1190 | retval = put_user(status, stat_addr); | |
1191 | if (!retval && infop) | |
1192 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1193 | if (!retval && infop) | |
1194 | retval = put_user(0, &infop->si_errno); | |
1195 | if (!retval && infop) { | |
1196 | int why; | |
1197 | ||
1198 | if ((status & 0x7f) == 0) { | |
1199 | why = CLD_EXITED; | |
1200 | status >>= 8; | |
1201 | } else { | |
1202 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; | |
1203 | status &= 0x7f; | |
1204 | } | |
1205 | retval = put_user((short)why, &infop->si_code); | |
1206 | if (!retval) | |
1207 | retval = put_user(status, &infop->si_status); | |
1208 | } | |
1209 | if (!retval && infop) | |
1210 | retval = put_user(p->pid, &infop->si_pid); | |
1211 | if (!retval && infop) | |
1212 | retval = put_user(p->uid, &infop->si_uid); | |
1213 | if (retval) { | |
1214 | // TODO: is this safe? | |
1215 | p->exit_state = EXIT_ZOMBIE; | |
1216 | return retval; | |
1217 | } | |
1218 | retval = p->pid; | |
1219 | if (p->real_parent != p->parent) { | |
1220 | write_lock_irq(&tasklist_lock); | |
1221 | /* Double-check with lock held. */ | |
1222 | if (p->real_parent != p->parent) { | |
1223 | __ptrace_unlink(p); | |
1224 | // TODO: is this safe? | |
1225 | p->exit_state = EXIT_ZOMBIE; | |
1226 | /* | |
1227 | * If this is not a detached task, notify the parent. | |
1228 | * If it's still not detached after that, don't release | |
1229 | * it now. | |
1230 | */ | |
1231 | if (p->exit_signal != -1) { | |
1232 | do_notify_parent(p, p->exit_signal); | |
1233 | if (p->exit_signal != -1) | |
1234 | p = NULL; | |
1235 | } | |
1236 | } | |
1237 | write_unlock_irq(&tasklist_lock); | |
1238 | } | |
1239 | if (p != NULL) | |
1240 | release_task(p); | |
1241 | BUG_ON(!retval); | |
1242 | return retval; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold | |
1247 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1248 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1249 | * released the lock and the system call should return. | |
1250 | */ | |
1251 | static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap, | |
1252 | struct siginfo __user *infop, | |
1253 | int __user *stat_addr, struct rusage __user *ru) | |
1254 | { | |
1255 | int retval, exit_code; | |
1256 | ||
1257 | if (!p->exit_code) | |
1258 | return 0; | |
1259 | if (delayed_group_leader && !(p->ptrace & PT_PTRACED) && | |
1260 | p->signal && p->signal->group_stop_count > 0) | |
1261 | /* | |
1262 | * A group stop is in progress and this is the group leader. | |
1263 | * We won't report until all threads have stopped. | |
1264 | */ | |
1265 | return 0; | |
1266 | ||
1267 | /* | |
1268 | * Now we are pretty sure this task is interesting. | |
1269 | * Make sure it doesn't get reaped out from under us while we | |
1270 | * give up the lock and then examine it below. We don't want to | |
1271 | * keep holding onto the tasklist_lock while we call getrusage and | |
1272 | * possibly take page faults for user memory. | |
1273 | */ | |
1274 | get_task_struct(p); | |
1275 | read_unlock(&tasklist_lock); | |
1276 | ||
1277 | if (unlikely(noreap)) { | |
1278 | pid_t pid = p->pid; | |
1279 | uid_t uid = p->uid; | |
1280 | int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED; | |
1281 | ||
1282 | exit_code = p->exit_code; | |
1283 | if (unlikely(!exit_code) || | |
14bf01bb | 1284 | unlikely(p->state & TASK_TRACED)) |
1da177e4 LT |
1285 | goto bail_ref; |
1286 | return wait_noreap_copyout(p, pid, uid, | |
1287 | why, (exit_code << 8) | 0x7f, | |
1288 | infop, ru); | |
1289 | } | |
1290 | ||
1291 | write_lock_irq(&tasklist_lock); | |
1292 | ||
1293 | /* | |
1294 | * This uses xchg to be atomic with the thread resuming and setting | |
1295 | * it. It must also be done with the write lock held to prevent a | |
1296 | * race with the EXIT_ZOMBIE case. | |
1297 | */ | |
1298 | exit_code = xchg(&p->exit_code, 0); | |
1299 | if (unlikely(p->exit_state)) { | |
1300 | /* | |
1301 | * The task resumed and then died. Let the next iteration | |
1302 | * catch it in EXIT_ZOMBIE. Note that exit_code might | |
1303 | * already be zero here if it resumed and did _exit(0). | |
1304 | * The task itself is dead and won't touch exit_code again; | |
1305 | * other processors in this function are locked out. | |
1306 | */ | |
1307 | p->exit_code = exit_code; | |
1308 | exit_code = 0; | |
1309 | } | |
1310 | if (unlikely(exit_code == 0)) { | |
1311 | /* | |
1312 | * Another thread in this function got to it first, or it | |
1313 | * resumed, or it resumed and then died. | |
1314 | */ | |
1315 | write_unlock_irq(&tasklist_lock); | |
1316 | bail_ref: | |
1317 | put_task_struct(p); | |
1318 | /* | |
1319 | * We are returning to the wait loop without having successfully | |
1320 | * removed the process and having released the lock. We cannot | |
1321 | * continue, since the "p" task pointer is potentially stale. | |
1322 | * | |
1323 | * Return -EAGAIN, and do_wait() will restart the loop from the | |
1324 | * beginning. Do _not_ re-acquire the lock. | |
1325 | */ | |
1326 | return -EAGAIN; | |
1327 | } | |
1328 | ||
1329 | /* move to end of parent's list to avoid starvation */ | |
1330 | remove_parent(p); | |
8fafabd8 | 1331 | add_parent(p); |
1da177e4 LT |
1332 | |
1333 | write_unlock_irq(&tasklist_lock); | |
1334 | ||
1335 | retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0; | |
1336 | if (!retval && stat_addr) | |
1337 | retval = put_user((exit_code << 8) | 0x7f, stat_addr); | |
1338 | if (!retval && infop) | |
1339 | retval = put_user(SIGCHLD, &infop->si_signo); | |
1340 | if (!retval && infop) | |
1341 | retval = put_user(0, &infop->si_errno); | |
1342 | if (!retval && infop) | |
1343 | retval = put_user((short)((p->ptrace & PT_PTRACED) | |
1344 | ? CLD_TRAPPED : CLD_STOPPED), | |
1345 | &infop->si_code); | |
1346 | if (!retval && infop) | |
1347 | retval = put_user(exit_code, &infop->si_status); | |
1348 | if (!retval && infop) | |
1349 | retval = put_user(p->pid, &infop->si_pid); | |
1350 | if (!retval && infop) | |
1351 | retval = put_user(p->uid, &infop->si_uid); | |
1352 | if (!retval) | |
1353 | retval = p->pid; | |
1354 | put_task_struct(p); | |
1355 | ||
1356 | BUG_ON(!retval); | |
1357 | return retval; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * Handle do_wait work for one task in a live, non-stopped state. | |
1362 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold | |
1363 | * the lock and this task is uninteresting. If we return nonzero, we have | |
1364 | * released the lock and the system call should return. | |
1365 | */ | |
1366 | static int wait_task_continued(task_t *p, int noreap, | |
1367 | struct siginfo __user *infop, | |
1368 | int __user *stat_addr, struct rusage __user *ru) | |
1369 | { | |
1370 | int retval; | |
1371 | pid_t pid; | |
1372 | uid_t uid; | |
1373 | ||
1374 | if (unlikely(!p->signal)) | |
1375 | return 0; | |
1376 | ||
1377 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) | |
1378 | return 0; | |
1379 | ||
1380 | spin_lock_irq(&p->sighand->siglock); | |
1381 | /* Re-check with the lock held. */ | |
1382 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { | |
1383 | spin_unlock_irq(&p->sighand->siglock); | |
1384 | return 0; | |
1385 | } | |
1386 | if (!noreap) | |
1387 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED; | |
1388 | spin_unlock_irq(&p->sighand->siglock); | |
1389 | ||
1390 | pid = p->pid; | |
1391 | uid = p->uid; | |
1392 | get_task_struct(p); | |
1393 | read_unlock(&tasklist_lock); | |
1394 | ||
1395 | if (!infop) { | |
1396 | retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0; | |
1397 | put_task_struct(p); | |
1398 | if (!retval && stat_addr) | |
1399 | retval = put_user(0xffff, stat_addr); | |
1400 | if (!retval) | |
1401 | retval = p->pid; | |
1402 | } else { | |
1403 | retval = wait_noreap_copyout(p, pid, uid, | |
1404 | CLD_CONTINUED, SIGCONT, | |
1405 | infop, ru); | |
1406 | BUG_ON(retval == 0); | |
1407 | } | |
1408 | ||
1409 | return retval; | |
1410 | } | |
1411 | ||
1412 | ||
1413 | static inline int my_ptrace_child(struct task_struct *p) | |
1414 | { | |
1415 | if (!(p->ptrace & PT_PTRACED)) | |
1416 | return 0; | |
1417 | if (!(p->ptrace & PT_ATTACHED)) | |
1418 | return 1; | |
1419 | /* | |
1420 | * This child was PTRACE_ATTACH'd. We should be seeing it only if | |
1421 | * we are the attacher. If we are the real parent, this is a race | |
1422 | * inside ptrace_attach. It is waiting for the tasklist_lock, | |
1423 | * which we have to switch the parent links, but has already set | |
1424 | * the flags in p->ptrace. | |
1425 | */ | |
1426 | return (p->parent != p->real_parent); | |
1427 | } | |
1428 | ||
1429 | static long do_wait(pid_t pid, int options, struct siginfo __user *infop, | |
1430 | int __user *stat_addr, struct rusage __user *ru) | |
1431 | { | |
1432 | DECLARE_WAITQUEUE(wait, current); | |
1433 | struct task_struct *tsk; | |
1434 | int flag, retval; | |
1435 | ||
1436 | add_wait_queue(¤t->signal->wait_chldexit,&wait); | |
1437 | repeat: | |
1438 | /* | |
1439 | * We will set this flag if we see any child that might later | |
1440 | * match our criteria, even if we are not able to reap it yet. | |
1441 | */ | |
1442 | flag = 0; | |
1443 | current->state = TASK_INTERRUPTIBLE; | |
1444 | read_lock(&tasklist_lock); | |
1445 | tsk = current; | |
1446 | do { | |
1447 | struct task_struct *p; | |
1448 | struct list_head *_p; | |
1449 | int ret; | |
1450 | ||
1451 | list_for_each(_p,&tsk->children) { | |
1452 | p = list_entry(_p,struct task_struct,sibling); | |
1453 | ||
1454 | ret = eligible_child(pid, options, p); | |
1455 | if (!ret) | |
1456 | continue; | |
1457 | ||
1458 | switch (p->state) { | |
1459 | case TASK_TRACED: | |
7f2a5255 RM |
1460 | /* |
1461 | * When we hit the race with PTRACE_ATTACH, | |
1462 | * we will not report this child. But the | |
1463 | * race means it has not yet been moved to | |
1464 | * our ptrace_children list, so we need to | |
1465 | * set the flag here to avoid a spurious ECHILD | |
1466 | * when the race happens with the only child. | |
1467 | */ | |
1468 | flag = 1; | |
1da177e4 LT |
1469 | if (!my_ptrace_child(p)) |
1470 | continue; | |
1471 | /*FALLTHROUGH*/ | |
1472 | case TASK_STOPPED: | |
1473 | /* | |
1474 | * It's stopped now, so it might later | |
1475 | * continue, exit, or stop again. | |
1476 | */ | |
1477 | flag = 1; | |
1478 | if (!(options & WUNTRACED) && | |
1479 | !my_ptrace_child(p)) | |
1480 | continue; | |
1481 | retval = wait_task_stopped(p, ret == 2, | |
1482 | (options & WNOWAIT), | |
1483 | infop, | |
1484 | stat_addr, ru); | |
1485 | if (retval == -EAGAIN) | |
1486 | goto repeat; | |
1487 | if (retval != 0) /* He released the lock. */ | |
1488 | goto end; | |
1489 | break; | |
1490 | default: | |
1491 | // case EXIT_DEAD: | |
1492 | if (p->exit_state == EXIT_DEAD) | |
1493 | continue; | |
1494 | // case EXIT_ZOMBIE: | |
1495 | if (p->exit_state == EXIT_ZOMBIE) { | |
1496 | /* | |
1497 | * Eligible but we cannot release | |
1498 | * it yet: | |
1499 | */ | |
1500 | if (ret == 2) | |
1501 | goto check_continued; | |
1502 | if (!likely(options & WEXITED)) | |
1503 | continue; | |
1504 | retval = wait_task_zombie( | |
1505 | p, (options & WNOWAIT), | |
1506 | infop, stat_addr, ru); | |
1507 | /* He released the lock. */ | |
1508 | if (retval != 0) | |
1509 | goto end; | |
1510 | break; | |
1511 | } | |
1512 | check_continued: | |
1513 | /* | |
1514 | * It's running now, so it might later | |
1515 | * exit, stop, or stop and then continue. | |
1516 | */ | |
1517 | flag = 1; | |
1518 | if (!unlikely(options & WCONTINUED)) | |
1519 | continue; | |
1520 | retval = wait_task_continued( | |
1521 | p, (options & WNOWAIT), | |
1522 | infop, stat_addr, ru); | |
1523 | if (retval != 0) /* He released the lock. */ | |
1524 | goto end; | |
1525 | break; | |
1526 | } | |
1527 | } | |
1528 | if (!flag) { | |
1529 | list_for_each(_p, &tsk->ptrace_children) { | |
1530 | p = list_entry(_p, struct task_struct, | |
1531 | ptrace_list); | |
1532 | if (!eligible_child(pid, options, p)) | |
1533 | continue; | |
1534 | flag = 1; | |
1535 | break; | |
1536 | } | |
1537 | } | |
1538 | if (options & __WNOTHREAD) | |
1539 | break; | |
1540 | tsk = next_thread(tsk); | |
1541 | if (tsk->signal != current->signal) | |
1542 | BUG(); | |
1543 | } while (tsk != current); | |
1544 | ||
1545 | read_unlock(&tasklist_lock); | |
1546 | if (flag) { | |
1547 | retval = 0; | |
1548 | if (options & WNOHANG) | |
1549 | goto end; | |
1550 | retval = -ERESTARTSYS; | |
1551 | if (signal_pending(current)) | |
1552 | goto end; | |
1553 | schedule(); | |
1554 | goto repeat; | |
1555 | } | |
1556 | retval = -ECHILD; | |
1557 | end: | |
1558 | current->state = TASK_RUNNING; | |
1559 | remove_wait_queue(¤t->signal->wait_chldexit,&wait); | |
1560 | if (infop) { | |
1561 | if (retval > 0) | |
1562 | retval = 0; | |
1563 | else { | |
1564 | /* | |
1565 | * For a WNOHANG return, clear out all the fields | |
1566 | * we would set so the user can easily tell the | |
1567 | * difference. | |
1568 | */ | |
1569 | if (!retval) | |
1570 | retval = put_user(0, &infop->si_signo); | |
1571 | if (!retval) | |
1572 | retval = put_user(0, &infop->si_errno); | |
1573 | if (!retval) | |
1574 | retval = put_user(0, &infop->si_code); | |
1575 | if (!retval) | |
1576 | retval = put_user(0, &infop->si_pid); | |
1577 | if (!retval) | |
1578 | retval = put_user(0, &infop->si_uid); | |
1579 | if (!retval) | |
1580 | retval = put_user(0, &infop->si_status); | |
1581 | } | |
1582 | } | |
1583 | return retval; | |
1584 | } | |
1585 | ||
1586 | asmlinkage long sys_waitid(int which, pid_t pid, | |
1587 | struct siginfo __user *infop, int options, | |
1588 | struct rusage __user *ru) | |
1589 | { | |
1590 | long ret; | |
1591 | ||
1592 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED)) | |
1593 | return -EINVAL; | |
1594 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) | |
1595 | return -EINVAL; | |
1596 | ||
1597 | switch (which) { | |
1598 | case P_ALL: | |
1599 | pid = -1; | |
1600 | break; | |
1601 | case P_PID: | |
1602 | if (pid <= 0) | |
1603 | return -EINVAL; | |
1604 | break; | |
1605 | case P_PGID: | |
1606 | if (pid <= 0) | |
1607 | return -EINVAL; | |
1608 | pid = -pid; | |
1609 | break; | |
1610 | default: | |
1611 | return -EINVAL; | |
1612 | } | |
1613 | ||
1614 | ret = do_wait(pid, options, infop, NULL, ru); | |
1615 | ||
1616 | /* avoid REGPARM breakage on x86: */ | |
1617 | prevent_tail_call(ret); | |
1618 | return ret; | |
1619 | } | |
1620 | ||
1621 | asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr, | |
1622 | int options, struct rusage __user *ru) | |
1623 | { | |
1624 | long ret; | |
1625 | ||
1626 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| | |
1627 | __WNOTHREAD|__WCLONE|__WALL)) | |
1628 | return -EINVAL; | |
1629 | ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru); | |
1630 | ||
1631 | /* avoid REGPARM breakage on x86: */ | |
1632 | prevent_tail_call(ret); | |
1633 | return ret; | |
1634 | } | |
1635 | ||
1636 | #ifdef __ARCH_WANT_SYS_WAITPID | |
1637 | ||
1638 | /* | |
1639 | * sys_waitpid() remains for compatibility. waitpid() should be | |
1640 | * implemented by calling sys_wait4() from libc.a. | |
1641 | */ | |
1642 | asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options) | |
1643 | { | |
1644 | return sys_wait4(pid, stat_addr, options, NULL); | |
1645 | } | |
1646 | ||
1647 | #endif |