| 1 | |
| 2 | * linux/kernel/exit.c
|
| 3 | *
|
| 4 | * Copyright (C) 1991, 1992 Linus Torvalds
|
| 5 |
|
| 6 |
|
| 7 | #include <linux/mm.h>
|
| 8 | #include <linux/slab.h>
|
| 9 | #include <linux/interrupt.h>
|
| 10 | #include <linux/module.h>
|
| 11 | #include <linux/capability.h>
|
| 12 | #include <linux/completion.h>
|
| 13 | #include <linux/personality.h>
|
| 14 | #include <linux/tty.h>
|
| 15 | #include <linux/iocontext.h>
|
| 16 | #include <linux/key.h>
|
| 17 | #include <linux/security.h>
|
| 18 | #include <linux/cpu.h>
|
| 19 | #include <linux/acct.h>
|
| 20 | #include <linux/tsacct_kern.h>
|
| 21 | #include <linux/file.h>
|
| 22 | #include <linux/fdtable.h>
|
| 23 | #include <linux/binfmts.h>
|
| 24 | #include <linux/nsproxy.h>
|
| 25 | #include <linux/pid_namespace.h>
|
| 26 | #include <linux/ptrace.h>
|
| 27 | #include <linux/profile.h>
|
| 28 | #include <linux/mount.h>
|
| 29 | #include <linux/proc_fs.h>
|
| 30 | #include <linux/kthread.h>
|
| 31 | #include <linux/mempolicy.h>
|
| 32 | #include <linux/taskstats_kern.h>
|
| 33 | #include <linux/delayacct.h>
|
| 34 | #include <linux/freezer.h>
|
| 35 | #include <linux/cgroup.h>
|
| 36 | #include <linux/syscalls.h>
|
| 37 | #include <linux/signal.h>
|
| 38 | #include <linux/posix-timers.h>
|
| 39 | #include <linux/cn_proc.h>
|
| 40 | #include <linux/mutex.h>
|
| 41 | #include <linux/futex.h>
|
| 42 | #include <linux/pipe_fs_i.h>
|
| 43 | #include <linux/audit.h>
|
| 44 | #include <linux/resource.h>
|
| 45 | #include <linux/blkdev.h>
|
| 46 | #include <linux/task_io_accounting_ops.h>
|
| 47 | #include <linux/tracehook.h>
|
| 48 | #include <linux/fs_struct.h>
|
| 49 | #include <linux/init_task.h>
|
| 50 | #include <linux/perf_event.h>
|
| 51 | #include <trace/events/sched.h>
|
| 52 | #include <linux/hw_breakpoint.h>
|
| 53 | #include <linux/oom.h>
|
| 54 | #include <linux/writeback.h>
|
| 55 |
|
| 56 | #include <asm/uaccess.h>
|
| 57 | #include <asm/unistd.h>
|
| 58 | #include <asm/pgtable.h>
|
| 59 | #include <asm/mmu_context.h>
|
| 60 |
|
| 61 | static void exit_mm(struct task_struct * tsk);
|
| 62 |
|
| 63 | static void __unhash_process(struct task_struct *p, bool group_dead)
|
| 64 | {
|
| 65 | nr_threads--;
|
| 66 | detach_pid(p, PIDTYPE_PID);
|
| 67 | if (group_dead) {
|
| 68 | detach_pid(p, PIDTYPE_PGID);
|
| 69 | detach_pid(p, PIDTYPE_SID);
|
| 70 |
|
| 71 | list_del_rcu(&p->tasks);
|
| 72 | list_del_init(&p->sibling);
|
| 73 | __this_cpu_dec(process_counts);
|
| 74 | }
|
| 75 | list_del_rcu(&p->thread_group);
|
| 76 | }
|
| 77 |
|
| 78 | |
| 79 | * This function expects the tasklist_lock write-locked.
|
| 80 |
|
| 81 | static void __exit_signal(struct task_struct *tsk)
|
| 82 | {
|
| 83 | struct signal_struct *sig = tsk->signal;
|
| 84 | bool group_dead = thread_group_leader(tsk);
|
| 85 | struct sighand_struct *sighand;
|
| 86 | struct tty_struct *uninitialized_var(tty);
|
| 87 |
|
| 88 | sighand = rcu_dereference_check(tsk->sighand,
|
| 89 | lockdep_tasklist_lock_is_held());
|
| 90 | spin_lock(&sighand->siglock);
|
| 91 |
|
| 92 | posix_cpu_timers_exit(tsk);
|
| 93 | if (group_dead) {
|
| 94 | posix_cpu_timers_exit_group(tsk);
|
| 95 | tty = sig->tty;
|
| 96 | sig->tty = NULL;
|
| 97 | } else {
|
| 98 | |
| 99 | * This can only happen if the caller is de_thread().
|
| 100 | * FIXME: this is the temporary hack, we should teach
|
| 101 | * posix-cpu-timers to handle this case correctly.
|
| 102 |
|
| 103 | if (unlikely(has_group_leader_pid(tsk)))
|
| 104 | posix_cpu_timers_exit_group(tsk);
|
| 105 |
|
| 106 | |
| 107 | * If there is any task waiting for the group exit
|
| 108 | * then notify it:
|
| 109 |
|
| 110 | if (sig->notify_count > 0 && !--sig->notify_count)
|
| 111 | wake_up_process(sig->group_exit_task);
|
| 112 |
|
| 113 | if (tsk == sig->curr_target)
|
| 114 | sig->curr_target = next_thread(tsk);
|
| 115 | |
| 116 | * Accumulate here the counters for all threads but the
|
| 117 | * group leader as they die, so they can be added into
|
| 118 | * the process-wide totals when those are taken.
|
| 119 | * The group leader stays around as a zombie as long
|
| 120 | * as there are other threads. When it gets reaped,
|
| 121 | * the exit.c code will add its counts into these totals.
|
| 122 | * We won't ever get here for the group leader, since it
|
| 123 | * will have been the last reference on the signal_struct.
|
| 124 |
|
| 125 | sig->utime += tsk->utime;
|
| 126 | sig->stime += tsk->stime;
|
| 127 | sig->gtime += tsk->gtime;
|
| 128 | sig->min_flt += tsk->min_flt;
|
| 129 | sig->maj_flt += tsk->maj_flt;
|
| 130 | sig->nvcsw += tsk->nvcsw;
|
| 131 | sig->nivcsw += tsk->nivcsw;
|
| 132 | sig->inblock += task_io_get_inblock(tsk);
|
| 133 | sig->oublock += task_io_get_oublock(tsk);
|
| 134 | task_io_accounting_add(&sig->ioac, &tsk->ioac);
|
| 135 | sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
|
| 136 | }
|
| 137 |
|
| 138 | sig->nr_threads--;
|
| 139 | __unhash_process(tsk, group_dead);
|
| 140 |
|
| 141 | |
| 142 | * Do this under ->siglock, we can race with another thread
|
| 143 | * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
|
| 144 |
|
| 145 | flush_sigqueue(&tsk->pending);
|
| 146 | tsk->sighand = NULL;
|
| 147 | spin_unlock(&sighand->siglock);
|
| 148 |
|
| 149 | __cleanup_sighand(sighand);
|
| 150 | clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
|
| 151 | if (group_dead) {
|
| 152 | flush_sigqueue(&sig->shared_pending);
|
| 153 | tty_kref_put(tty);
|
| 154 | }
|
| 155 | }
|
| 156 |
|
| 157 | static void delayed_put_task_struct(struct rcu_head *rhp)
|
| 158 | {
|
| 159 | struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
|
| 160 |
|
| 161 | perf_event_delayed_put(tsk);
|
| 162 | trace_sched_process_free(tsk);
|
| 163 | put_task_struct(tsk);
|
| 164 | }
|
| 165 |
|
| 166 |
|
| 167 | void release_task(struct task_struct * p)
|
| 168 | {
|
| 169 | struct task_struct *leader;
|
| 170 | int zap_leader;
|
| 171 | repeat:
|
| 172 | |
| 173 | * can't be modifying its own credentials. But shut RCU-lockdep up
|
| 174 | rcu_read_lock();
|
| 175 | atomic_dec(&__task_cred(p)->user->processes);
|
| 176 | rcu_read_unlock();
|
| 177 |
|
| 178 | proc_flush_task(p);
|
| 179 |
|
| 180 | write_lock_irq(&tasklist_lock);
|
| 181 | ptrace_release_task(p);
|
| 182 | __exit_signal(p);
|
| 183 |
|
| 184 | |
| 185 | * If we are the last non-leader member of the thread
|
| 186 | * group, and the leader is zombie, then notify the
|
| 187 | * group leader's parent process. (if it wants notification.)
|
| 188 |
|
| 189 | zap_leader = 0;
|
| 190 | leader = p->group_leader;
|
| 191 | if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
|
| 192 | |
| 193 | * If we were the last child thread and the leader has
|
| 194 | * exited already, and the leader's parent ignores SIGCHLD,
|
| 195 | * then we are the one who should release the leader.
|
| 196 |
|
| 197 | zap_leader = do_notify_parent(leader, leader->exit_signal);
|
| 198 | if (zap_leader)
|
| 199 | leader->exit_state = EXIT_DEAD;
|
| 200 | }
|
| 201 |
|
| 202 | write_unlock_irq(&tasklist_lock);
|
| 203 | release_thread(p);
|
| 204 | call_rcu(&p->rcu, delayed_put_task_struct);
|
| 205 |
|
| 206 | p = leader;
|
| 207 | if (unlikely(zap_leader))
|
| 208 | goto repeat;
|
| 209 | }
|
| 210 |
|
| 211 | |
| 212 | * This checks not only the pgrp, but falls back on the pid if no
|
| 213 | * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
|
| 214 | * without this...
|
| 215 | *
|
| 216 | * The caller must hold rcu lock or the tasklist lock.
|
| 217 |
|
| 218 | struct pid *session_of_pgrp(struct pid *pgrp)
|
| 219 | {
|
| 220 | struct task_struct *p;
|
| 221 | struct pid *sid = NULL;
|
| 222 |
|
| 223 | p = pid_task(pgrp, PIDTYPE_PGID);
|
| 224 | if (p == NULL)
|
| 225 | p = pid_task(pgrp, PIDTYPE_PID);
|
| 226 | if (p != NULL)
|
| 227 | sid = task_session(p);
|
| 228 |
|
| 229 | return sid;
|
| 230 | }
|
| 231 |
|
| 232 | |
| 233 | * Determine if a process group is "orphaned", according to the POSIX
|
| 234 | * definition in 2.2.2.52. Orphaned process groups are not to be affected
|
| 235 | * by terminal-generated stop signals. Newly orphaned process groups are
|
| 236 | * to receive a SIGHUP and a SIGCONT.
|
| 237 | *
|
| 238 | * "I ask you, have you ever known what it is to be an orphan?"
|
| 239 |
|
| 240 | static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
|
| 241 | {
|
| 242 | struct task_struct *p;
|
| 243 |
|
| 244 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
|
| 245 | if ((p == ignored_task) ||
|
| 246 | (p->exit_state && thread_group_empty(p)) ||
|
| 247 | is_global_init(p->real_parent))
|
| 248 | continue;
|
| 249 |
|
| 250 | if (task_pgrp(p->real_parent) != pgrp &&
|
| 251 | task_session(p->real_parent) == task_session(p))
|
| 252 | return 0;
|
| 253 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
|
| 254 |
|
| 255 | return 1;
|
| 256 | }
|
| 257 |
|
| 258 | int is_current_pgrp_orphaned(void)
|
| 259 | {
|
| 260 | int retval;
|
| 261 |
|
| 262 | read_lock(&tasklist_lock);
|
| 263 | retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
|
| 264 | read_unlock(&tasklist_lock);
|
| 265 |
|
| 266 | return retval;
|
| 267 | }
|
| 268 |
|
| 269 | static bool has_stopped_jobs(struct pid *pgrp)
|
| 270 | {
|
| 271 | struct task_struct *p;
|
| 272 |
|
| 273 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
|
| 274 | if (p->signal->flags & SIGNAL_STOP_STOPPED)
|
| 275 | return true;
|
| 276 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
|
| 277 |
|
| 278 | return false;
|
| 279 | }
|
| 280 |
|
| 281 | |
| 282 | * Check to see if any process groups have become orphaned as
|
| 283 | * a result of our exiting, and if they have any stopped jobs,
|
| 284 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
|
| 285 |
|
| 286 | static void
|
| 287 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
|
| 288 | {
|
| 289 | struct pid *pgrp = task_pgrp(tsk);
|
| 290 | struct task_struct *ignored_task = tsk;
|
| 291 |
|
| 292 | if (!parent)
|
| 293 | |
| 294 | * we are and we were the only connection outside.
|
| 295 |
|
| 296 | parent = tsk->real_parent;
|
| 297 | else
|
| 298 | |
| 299 | * we are, and it was the only connection outside.
|
| 300 |
|
| 301 | ignored_task = NULL;
|
| 302 |
|
| 303 | if (task_pgrp(parent) != pgrp &&
|
| 304 | task_session(parent) == task_session(tsk) &&
|
| 305 | will_become_orphaned_pgrp(pgrp, ignored_task) &&
|
| 306 | has_stopped_jobs(pgrp)) {
|
| 307 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
|
| 308 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
|
| 309 | }
|
| 310 | }
|
| 311 |
|
| 312 | |
| 313 | * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
|
| 314 | *
|
| 315 | * If a kernel thread is launched as a result of a system call, or if
|
| 316 | * it ever exits, it should generally reparent itself to kthreadd so it
|
| 317 | * isn't in the way of other processes and is correctly cleaned up on exit.
|
| 318 | *
|
| 319 | * The various task state such as scheduling policy and priority may have
|
| 320 | * been inherited from a user process, so we reset them to sane values here.
|
| 321 | *
|
| 322 | * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
|
| 323 |
|
| 324 | static void reparent_to_kthreadd(void)
|
| 325 | {
|
| 326 | write_lock_irq(&tasklist_lock);
|
| 327 |
|
| 328 | ptrace_unlink(current);
|
| 329 |
|
| 330 | current->real_parent = current->parent = kthreadd_task;
|
| 331 | list_move_tail(¤t->sibling, ¤t->real_parent->children);
|
| 332 |
|
| 333 |
|
| 334 | current->exit_signal = SIGCHLD;
|
| 335 |
|
| 336 | if (task_nice(current) < 0)
|
| 337 | set_user_nice(current, 0);
|
| 338 |
|
| 339 |
|
| 340 |
|
| 341 | memcpy(current->signal->rlim, init_task.signal->rlim,
|
| 342 | sizeof(current->signal->rlim));
|
| 343 |
|
| 344 | atomic_inc(&init_cred.usage);
|
| 345 | commit_creds(&init_cred);
|
| 346 | write_unlock_irq(&tasklist_lock);
|
| 347 | }
|
| 348 |
|
| 349 | void __set_special_pids(struct pid *pid)
|
| 350 | {
|
| 351 | struct task_struct *curr = current->group_leader;
|
| 352 |
|
| 353 | if (task_session(curr) != pid)
|
| 354 | change_pid(curr, PIDTYPE_SID, pid);
|
| 355 |
|
| 356 | if (task_pgrp(curr) != pid)
|
| 357 | change_pid(curr, PIDTYPE_PGID, pid);
|
| 358 | }
|
| 359 |
|
| 360 | static void set_special_pids(struct pid *pid)
|
| 361 | {
|
| 362 | write_lock_irq(&tasklist_lock);
|
| 363 | __set_special_pids(pid);
|
| 364 | write_unlock_irq(&tasklist_lock);
|
| 365 | }
|
| 366 |
|
| 367 | |
| 368 | * Let kernel threads use this to say that they allow a certain signal.
|
| 369 | * Must not be used if kthread was cloned with CLONE_SIGHAND.
|
| 370 |
|
| 371 | int allow_signal(int sig)
|
| 372 | {
|
| 373 | if (!valid_signal(sig) || sig < 1)
|
| 374 | return -EINVAL;
|
| 375 |
|
| 376 | spin_lock_irq(¤t->sighand->siglock);
|
| 377 |
|
| 378 | sigdelset(¤t->blocked, sig);
|
| 379 | |
| 380 | * Kernel threads handle their own signals. Let the signal code
|
| 381 | * know it'll be handled, so that they don't get converted to
|
| 382 | * SIGKILL or just silently dropped.
|
| 383 |
|
| 384 | current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
|
| 385 | recalc_sigpending();
|
| 386 | spin_unlock_irq(¤t->sighand->siglock);
|
| 387 | return 0;
|
| 388 | }
|
| 389 |
|
| 390 | EXPORT_SYMBOL(allow_signal);
|
| 391 |
|
| 392 | int disallow_signal(int sig)
|
| 393 | {
|
| 394 | if (!valid_signal(sig) || sig < 1)
|
| 395 | return -EINVAL;
|
| 396 |
|
| 397 | spin_lock_irq(¤t->sighand->siglock);
|
| 398 | current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
|
| 399 | recalc_sigpending();
|
| 400 | spin_unlock_irq(¤t->sighand->siglock);
|
| 401 | return 0;
|
| 402 | }
|
| 403 |
|
| 404 | EXPORT_SYMBOL(disallow_signal);
|
| 405 |
|
| 406 | |
| 407 | * Put all the gunge required to become a kernel thread without
|
| 408 | * attached user resources in one place where it belongs.
|
| 409 |
|
| 410 |
|
| 411 | void daemonize(const char *name, ...)
|
| 412 | {
|
| 413 | va_list args;
|
| 414 | sigset_t blocked;
|
| 415 |
|
| 416 | va_start(args, name);
|
| 417 | vsnprintf(current->comm, sizeof(current->comm), name, args);
|
| 418 | va_end(args);
|
| 419 |
|
| 420 | |
| 421 | * If we were started as result of loading a module, close all of the
|
| 422 | * user space pages. We don't need them, and if we didn't close them
|
| 423 | * they would be locked into memory.
|
| 424 |
|
| 425 | exit_mm(current);
|
| 426 | |
| 427 | * We don't want to have TIF_FREEZE set if the system-wide hibernation
|
| 428 | * or suspend transition begins right now.
|
| 429 |
|
| 430 | current->flags |= (PF_NOFREEZE | PF_KTHREAD);
|
| 431 |
|
| 432 | if (current->nsproxy != &init_nsproxy) {
|
| 433 | get_nsproxy(&init_nsproxy);
|
| 434 | switch_task_namespaces(current, &init_nsproxy);
|
| 435 | }
|
| 436 | set_special_pids(&init_struct_pid);
|
| 437 | proc_clear_tty(current);
|
| 438 |
|
| 439 |
|
| 440 | sigfillset(&blocked);
|
| 441 | sigprocmask(SIG_BLOCK, &blocked, NULL);
|
| 442 | flush_signals(current);
|
| 443 |
|
| 444 |
|
| 445 |
|
| 446 | daemonize_fs_struct();
|
| 447 | exit_files(current);
|
| 448 | current->files = init_task.files;
|
| 449 | atomic_inc(¤t->files->count);
|
| 450 |
|
| 451 | reparent_to_kthreadd();
|
| 452 | }
|
| 453 |
|
| 454 | EXPORT_SYMBOL(daemonize);
|
| 455 |
|
| 456 | static void close_files(struct files_struct * files)
|
| 457 | {
|
| 458 | int i, j;
|
| 459 | struct fdtable *fdt;
|
| 460 |
|
| 461 | j = 0;
|
| 462 |
|
| 463 | |
| 464 | * It is safe to dereference the fd table without RCU or
|
| 465 | * ->file_lock because this is the last reference to the
|
| 466 | * files structure. But use RCU to shut RCU-lockdep up.
|
| 467 |
|
| 468 | rcu_read_lock();
|
| 469 | fdt = files_fdtable(files);
|
| 470 | rcu_read_unlock();
|
| 471 | for (;;) {
|
| 472 | unsigned long set;
|
| 473 | i = j * __NFDBITS;
|
| 474 | if (i >= fdt->max_fds)
|
| 475 | break;
|
| 476 | set = fdt->open_fds->fds_bits[j++];
|
| 477 | while (set) {
|
| 478 | if (set & 1) {
|
| 479 | struct file * file = xchg(&fdt->fd[i], NULL);
|
| 480 | if (file) {
|
| 481 | filp_close(file, files);
|
| 482 | cond_resched();
|
| 483 | }
|
| 484 | }
|
| 485 | i++;
|
| 486 | set >>= 1;
|
| 487 | }
|
| 488 | }
|
| 489 | }
|
| 490 |
|
| 491 | struct files_struct *get_files_struct(struct task_struct *task)
|
| 492 | {
|
| 493 | struct files_struct *files;
|
| 494 |
|
| 495 | task_lock(task);
|
| 496 | files = task->files;
|
| 497 | if (files)
|
| 498 | atomic_inc(&files->count);
|
| 499 | task_unlock(task);
|
| 500 |
|
| 501 | return files;
|
| 502 | }
|
| 503 |
|
| 504 | void put_files_struct(struct files_struct *files)
|
| 505 | {
|
| 506 | struct fdtable *fdt;
|
| 507 |
|
| 508 | if (atomic_dec_and_test(&files->count)) {
|
| 509 | close_files(files);
|
| 510 | |
| 511 | * Free the fd and fdset arrays if we expanded them.
|
| 512 | * If the fdtable was embedded, pass files for freeing
|
| 513 | * at the end of the RCU grace period. Otherwise,
|
| 514 | * you can free files immediately.
|
| 515 |
|
| 516 | rcu_read_lock();
|
| 517 | fdt = files_fdtable(files);
|
| 518 | if (fdt != &files->fdtab)
|
| 519 | kmem_cache_free(files_cachep, files);
|
| 520 | free_fdtable(fdt);
|
| 521 | rcu_read_unlock();
|
| 522 | }
|
| 523 | }
|
| 524 |
|
| 525 | void reset_files_struct(struct files_struct *files)
|
| 526 | {
|
| 527 | struct task_struct *tsk = current;
|
| 528 | struct files_struct *old;
|
| 529 |
|
| 530 | old = tsk->files;
|
| 531 | task_lock(tsk);
|
| 532 | tsk->files = files;
|
| 533 | task_unlock(tsk);
|
| 534 | put_files_struct(old);
|
| 535 | }
|
| 536 |
|
| 537 | void exit_files(struct task_struct *tsk)
|
| 538 | {
|
| 539 | struct files_struct * files = tsk->files;
|
| 540 |
|
| 541 | if (files) {
|
| 542 | task_lock(tsk);
|
| 543 | tsk->files = NULL;
|
| 544 | task_unlock(tsk);
|
| 545 | put_files_struct(files);
|
| 546 | }
|
| 547 | }
|
| 548 |
|
| 549 | #ifdef CONFIG_MM_OWNER
|
| 550 | |
| 551 | * A task is exiting. If it owned this mm, find a new owner for the mm.
|
| 552 |
|
| 553 | void mm_update_next_owner(struct mm_struct *mm)
|
| 554 | {
|
| 555 | struct task_struct *c, *g, *p = current;
|
| 556 |
|
| 557 | retry:
|
| 558 | |
| 559 | * If the exiting or execing task is not the owner, it's
|
| 560 | * someone else's problem.
|
| 561 |
|
| 562 | if (mm->owner != p)
|
| 563 | return;
|
| 564 | |
| 565 | * The current owner is exiting/execing and there are no other
|
| 566 | * candidates. Do not leave the mm pointing to a possibly
|
| 567 | * freed task structure.
|
| 568 |
|
| 569 | if (atomic_read(&mm->mm_users) <= 1) {
|
| 570 | mm->owner = NULL;
|
| 571 | return;
|
| 572 | }
|
| 573 |
|
| 574 | read_lock(&tasklist_lock);
|
| 575 | |
| 576 | * Search in the children
|
| 577 |
|
| 578 | list_for_each_entry(c, &p->children, sibling) {
|
| 579 | if (c->mm == mm)
|
| 580 | goto assign_new_owner;
|
| 581 | }
|
| 582 |
|
| 583 | |
| 584 | * Search in the siblings
|
| 585 |
|
| 586 | list_for_each_entry(c, &p->real_parent->children, sibling) {
|
| 587 | if (c->mm == mm)
|
| 588 | goto assign_new_owner;
|
| 589 | }
|
| 590 |
|
| 591 | |
| 592 | * Search through everything else. We should not get
|
| 593 | * here often
|
| 594 |
|
| 595 | do_each_thread(g, c) {
|
| 596 | if (c->mm == mm)
|
| 597 | goto assign_new_owner;
|
| 598 | } while_each_thread(g, c);
|
| 599 |
|
| 600 | read_unlock(&tasklist_lock);
|
| 601 | |
| 602 | * We found no owner yet mm_users > 1: this implies that we are
|
| 603 | * most likely racing with swapoff (try_to_unuse()) or /proc or
|
| 604 | * ptrace or page migration (get_task_mm()). Mark owner as NULL.
|
| 605 |
|
| 606 | mm->owner = NULL;
|
| 607 | return;
|
| 608 |
|
| 609 | assign_new_owner:
|
| 610 | BUG_ON(c == p);
|
| 611 | get_task_struct(c);
|
| 612 | |
| 613 | * The task_lock protects c->mm from changing.
|
| 614 | * We always want mm->owner->mm == mm
|
| 615 |
|
| 616 | task_lock(c);
|
| 617 | |
| 618 | * Delay read_unlock() till we have the task_lock()
|
| 619 | * to ensure that c does not slip away underneath us
|
| 620 |
|
| 621 | read_unlock(&tasklist_lock);
|
| 622 | if (c->mm != mm) {
|
| 623 | task_unlock(c);
|
| 624 | put_task_struct(c);
|
| 625 | goto retry;
|
| 626 | }
|
| 627 | mm->owner = c;
|
| 628 | task_unlock(c);
|
| 629 | put_task_struct(c);
|
| 630 | }
|
| 631 | #endif
|
| 632 |
|
| 633 | |
| 634 | * Turn us into a lazy TLB process if we
|
| 635 | * aren't already..
|
| 636 |
|
| 637 | static void exit_mm(struct task_struct * tsk)
|
| 638 | {
|
| 639 | struct mm_struct *mm = tsk->mm;
|
| 640 | struct core_state *core_state;
|
| 641 |
|
| 642 | mm_release(tsk, mm);
|
| 643 | if (!mm)
|
| 644 | return;
|
| 645 | |
| 646 | * Serialize with any possible pending coredump.
|
| 647 | * We must hold mmap_sem around checking core_state
|
| 648 | * and clearing tsk->mm. The core-inducing thread
|
| 649 | * will increment ->nr_threads for each thread in the
|
| 650 | * group with ->mm != NULL.
|
| 651 |
|
| 652 | down_read(&mm->mmap_sem);
|
| 653 | core_state = mm->core_state;
|
| 654 | if (core_state) {
|
| 655 | struct core_thread self;
|
| 656 | up_read(&mm->mmap_sem);
|
| 657 |
|
| 658 | self.task = tsk;
|
| 659 | self.next = xchg(&core_state->dumper.next, &self);
|
| 660 | |
| 661 | * Implies mb(), the result of xchg() must be visible
|
| 662 | * to core_state->dumper.
|
| 663 |
|
| 664 | if (atomic_dec_and_test(&core_state->nr_threads))
|
| 665 | complete(&core_state->startup);
|
| 666 |
|
| 667 | for (;;) {
|
| 668 | set_task_state(tsk, TASK_UNINTERRUPTIBLE);
|
| 669 | if (!self.task)
|
| 670 | break;
|
| 671 | schedule();
|
| 672 | }
|
| 673 | __set_task_state(tsk, TASK_RUNNING);
|
| 674 | down_read(&mm->mmap_sem);
|
| 675 | }
|
| 676 | atomic_inc(&mm->mm_count);
|
| 677 | BUG_ON(mm != tsk->active_mm);
|
| 678 |
|
| 679 | task_lock(tsk);
|
| 680 | tsk->mm = NULL;
|
| 681 | up_read(&mm->mmap_sem);
|
| 682 | enter_lazy_tlb(mm, current);
|
| 683 | task_unlock(tsk);
|
| 684 | mm_update_next_owner(mm);
|
| 685 | mmput(mm);
|
| 686 | }
|
| 687 |
|
| 688 | |
| 689 | * When we die, we re-parent all our children.
|
| 690 | * Try to give them to another thread in our thread
|
| 691 | * group, and if no such member exists, give it to
|
| 692 | * the child reaper process (ie "init") in our pid
|
| 693 | * space.
|
| 694 |
|
| 695 | static struct task_struct *find_new_reaper(struct task_struct *father)
|
| 696 | __releases(&tasklist_lock)
|
| 697 | __acquires(&tasklist_lock)
|
| 698 | {
|
| 699 | struct pid_namespace *pid_ns = task_active_pid_ns(father);
|
| 700 | struct task_struct *thread;
|
| 701 |
|
| 702 | thread = father;
|
| 703 | while_each_thread(father, thread) {
|
| 704 | if (thread->flags & PF_EXITING)
|
| 705 | continue;
|
| 706 | if (unlikely(pid_ns->child_reaper == father))
|
| 707 | pid_ns->child_reaper = thread;
|
| 708 | return thread;
|
| 709 | }
|
| 710 |
|
| 711 | if (unlikely(pid_ns->child_reaper == father)) {
|
| 712 | write_unlock_irq(&tasklist_lock);
|
| 713 | if (unlikely(pid_ns == &init_pid_ns))
|
| 714 | panic("Attempted to kill init!");
|
| 715 |
|
| 716 | zap_pid_ns_processes(pid_ns);
|
| 717 | write_lock_irq(&tasklist_lock);
|
| 718 | |
| 719 | * We can not clear ->child_reaper or leave it alone.
|
| 720 | * There may by stealth EXIT_DEAD tasks on ->children,
|
| 721 | * forget_original_parent() must move them somewhere.
|
| 722 |
|
| 723 | pid_ns->child_reaper = init_pid_ns.child_reaper;
|
| 724 | }
|
| 725 |
|
| 726 | return pid_ns->child_reaper;
|
| 727 | }
|
| 728 |
|
| 729 | |
| 730 | * Any that need to be release_task'd are put on the @dead list.
|
| 731 |
|
| 732 | static void reparent_leader(struct task_struct *father, struct task_struct *p,
|
| 733 | struct list_head *dead)
|
| 734 | {
|
| 735 | list_move_tail(&p->sibling, &p->real_parent->children);
|
| 736 |
|
| 737 | if (p->exit_state == EXIT_DEAD)
|
| 738 | return;
|
| 739 | |
| 740 | * If this is a threaded reparent there is no need to
|
| 741 | * notify anyone anything has happened.
|
| 742 |
|
| 743 | if (same_thread_group(p->real_parent, father))
|
| 744 | return;
|
| 745 |
|
| 746 |
|
| 747 | p->exit_signal = SIGCHLD;
|
| 748 |
|
| 749 |
|
| 750 | if (!p->ptrace &&
|
| 751 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
|
| 752 | if (do_notify_parent(p, p->exit_signal)) {
|
| 753 | p->exit_state = EXIT_DEAD;
|
| 754 | list_move_tail(&p->sibling, dead);
|
| 755 | }
|
| 756 | }
|
| 757 |
|
| 758 | kill_orphaned_pgrp(p, father);
|
| 759 | }
|
| 760 |
|
| 761 | static void forget_original_parent(struct task_struct *father)
|
| 762 | {
|
| 763 | struct task_struct *p, *n, *reaper;
|
| 764 | LIST_HEAD(dead_children);
|
| 765 |
|
| 766 | write_lock_irq(&tasklist_lock);
|
| 767 | |
| 768 | * Note that exit_ptrace() and find_new_reaper() might
|
| 769 | * drop tasklist_lock and reacquire it.
|
| 770 |
|
| 771 | exit_ptrace(father);
|
| 772 | reaper = find_new_reaper(father);
|
| 773 |
|
| 774 | list_for_each_entry_safe(p, n, &father->children, sibling) {
|
| 775 | struct task_struct *t = p;
|
| 776 | do {
|
| 777 | t->real_parent = reaper;
|
| 778 | if (t->parent == father) {
|
| 779 | BUG_ON(t->ptrace);
|
| 780 | t->parent = t->real_parent;
|
| 781 | }
|
| 782 | if (t->pdeath_signal)
|
| 783 | group_send_sig_info(t->pdeath_signal,
|
| 784 | SEND_SIG_NOINFO, t);
|
| 785 | } while_each_thread(p, t);
|
| 786 | reparent_leader(father, p, &dead_children);
|
| 787 | }
|
| 788 | write_unlock_irq(&tasklist_lock);
|
| 789 |
|
| 790 | BUG_ON(!list_empty(&father->children));
|
| 791 |
|
| 792 | list_for_each_entry_safe(p, n, &dead_children, sibling) {
|
| 793 | list_del_init(&p->sibling);
|
| 794 | release_task(p);
|
| 795 | }
|
| 796 | }
|
| 797 |
|
| 798 | |
| 799 | * Send signals to all our closest relatives so that they know
|
| 800 | * to properly mourn us..
|
| 801 |
|
| 802 | static void exit_notify(struct task_struct *tsk, int group_dead)
|
| 803 | {
|
| 804 | bool autoreap;
|
| 805 |
|
| 806 | |
| 807 | * This does two things:
|
| 808 | *
|
| 809 | * A. Make init inherit all the child processes
|
| 810 | * B. Check to see if any process groups have become orphaned
|
| 811 | * as a result of our exiting, and if they have any stopped
|
| 812 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
|
| 813 |
|
| 814 | forget_original_parent(tsk);
|
| 815 | exit_task_namespaces(tsk);
|
| 816 |
|
| 817 | write_lock_irq(&tasklist_lock);
|
| 818 | if (group_dead)
|
| 819 | kill_orphaned_pgrp(tsk->group_leader, NULL);
|
| 820 |
|
| 821 | |
| 822 | *
|
| 823 | * Thread signals are configurable, but you aren't going to use
|
| 824 | * that to send signals to arbitrary processes.
|
| 825 | * That stops right now.
|
| 826 | *
|
| 827 | * If the parent exec id doesn't match the exec id we saved
|
| 828 | * when we started then we know the parent has changed security
|
| 829 | * domain.
|
| 830 | *
|
| 831 | * If our self_exec id doesn't match our parent_exec_id then
|
| 832 | * we have changed execution domain as these two values started
|
| 833 | * the same after a fork.
|
| 834 |
|
| 835 | if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
|
| 836 | (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
|
| 837 | tsk->self_exec_id != tsk->parent_exec_id))
|
| 838 | tsk->exit_signal = SIGCHLD;
|
| 839 |
|
| 840 | if (unlikely(tsk->ptrace)) {
|
| 841 | int sig = thread_group_leader(tsk) &&
|
| 842 | thread_group_empty(tsk) &&
|
| 843 | !ptrace_reparented(tsk) ?
|
| 844 | tsk->exit_signal : SIGCHLD;
|
| 845 | autoreap = do_notify_parent(tsk, sig);
|
| 846 | } else if (thread_group_leader(tsk)) {
|
| 847 | autoreap = thread_group_empty(tsk) &&
|
| 848 | do_notify_parent(tsk, tsk->exit_signal);
|
| 849 | } else {
|
| 850 | autoreap = true;
|
| 851 | }
|
| 852 |
|
| 853 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
|
| 854 |
|
| 855 |
|
| 856 | if (unlikely(tsk->signal->notify_count < 0))
|
| 857 | wake_up_process(tsk->signal->group_exit_task);
|
| 858 | write_unlock_irq(&tasklist_lock);
|
| 859 |
|
| 860 |
|
| 861 | if (autoreap)
|
| 862 | release_task(tsk);
|
| 863 | }
|
| 864 |
|
| 865 | #ifdef CONFIG_DEBUG_STACK_USAGE
|
| 866 | static void check_stack_usage(void)
|
| 867 | {
|
| 868 | static DEFINE_SPINLOCK(low_water_lock);
|
| 869 | static int lowest_to_date = THREAD_SIZE;
|
| 870 | unsigned long free;
|
| 871 |
|
| 872 | free = stack_not_used(current);
|
| 873 |
|
| 874 | if (free >= lowest_to_date)
|
| 875 | return;
|
| 876 |
|
| 877 | spin_lock(&low_water_lock);
|
| 878 | if (free < lowest_to_date) {
|
| 879 | printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
|
| 880 | "left\n",
|
| 881 | current->comm, free);
|
| 882 | lowest_to_date = free;
|
| 883 | }
|
| 884 | spin_unlock(&low_water_lock);
|
| 885 | }
|
| 886 | #else
|
| 887 | static inline void check_stack_usage(void) {}
|
| 888 | #endif
|
| 889 |
|
| 890 | void do_exit(long code)
|
| 891 | {
|
| 892 | struct task_struct *tsk = current;
|
| 893 | int group_dead;
|
| 894 |
|
| 895 | profile_task_exit(tsk);
|
| 896 |
|
| 897 | WARN_ON(blk_needs_flush_plug(tsk));
|
| 898 |
|
| 899 | if (unlikely(in_interrupt()))
|
| 900 | panic("Aiee, killing interrupt handler!");
|
| 901 | if (unlikely(!tsk->pid))
|
| 902 | panic("Attempted to kill the idle task!");
|
| 903 |
|
| 904 | |
| 905 | * If do_exit is called because this processes oopsed, it's possible
|
| 906 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
|
| 907 | * continuing. Amongst other possible reasons, this is to prevent
|
| 908 | * mm_release()->clear_child_tid() from writing to a user-controlled
|
| 909 | * kernel address.
|
| 910 |
|
| 911 | set_fs(USER_DS);
|
| 912 |
|
| 913 | ptrace_event(PTRACE_EVENT_EXIT, code);
|
| 914 |
|
| 915 | validate_creds_for_do_exit(tsk);
|
| 916 |
|
| 917 | |
| 918 | * We're taking recursive faults here in do_exit. Safest is to just
|
| 919 | * leave this task alone and wait for reboot.
|
| 920 |
|
| 921 | if (unlikely(tsk->flags & PF_EXITING)) {
|
| 922 | printk(KERN_ALERT
|
| 923 | "Fixing recursive fault but reboot is needed!\n");
|
| 924 | |
| 925 | * We can do this unlocked here. The futex code uses
|
| 926 | * this flag just to verify whether the pi state
|
| 927 | * cleanup has been done or not. In the worst case it
|
| 928 | * loops once more. We pretend that the cleanup was
|
| 929 | * done as there is no way to return. Either the
|
| 930 | * OWNER_DIED bit is set by now or we push the blocked
|
| 931 | * task into the wait for ever nirwana as well.
|
| 932 |
|
| 933 | tsk->flags |= PF_EXITPIDONE;
|
| 934 | set_current_state(TASK_UNINTERRUPTIBLE);
|
| 935 | schedule();
|
| 936 | }
|
| 937 |
|
| 938 | exit_irq_thread();
|
| 939 |
|
| 940 | exit_signals(tsk);
|
| 941 | |
| 942 | * tsk->flags are checked in the futex code to protect against
|
| 943 | * an exiting task cleaning up the robust pi futexes.
|
| 944 |
|
| 945 | smp_mb();
|
| 946 | raw_spin_unlock_wait(&tsk->pi_lock);
|
| 947 |
|
| 948 | if (unlikely(in_atomic()))
|
| 949 | printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
|
| 950 | current->comm, task_pid_nr(current),
|
| 951 | preempt_count());
|
| 952 |
|
| 953 | acct_update_integrals(tsk);
|
| 954 |
|
| 955 | if (tsk->mm)
|
| 956 | sync_mm_rss(tsk, tsk->mm);
|
| 957 | group_dead = atomic_dec_and_test(&tsk->signal->live);
|
| 958 | if (group_dead) {
|
| 959 | hrtimer_cancel(&tsk->signal->real_timer);
|
| 960 | exit_itimers(tsk->signal);
|
| 961 | if (tsk->mm)
|
| 962 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
|
| 963 | }
|
| 964 | acct_collect(code, group_dead);
|
| 965 | if (group_dead)
|
| 966 | tty_audit_exit();
|
| 967 | audit_free(tsk);
|
| 968 |
|
| 969 | tsk->exit_code = code;
|
| 970 | taskstats_exit(tsk, group_dead);
|
| 971 |
|
| 972 | exit_mm(tsk);
|
| 973 |
|
| 974 | if (group_dead)
|
| 975 | acct_process();
|
| 976 | trace_sched_process_exit(tsk);
|
| 977 |
|
| 978 | exit_sem(tsk);
|
| 979 | exit_shm(tsk);
|
| 980 | exit_files(tsk);
|
| 981 | exit_fs(tsk);
|
| 982 | check_stack_usage();
|
| 983 | exit_thread();
|
| 984 |
|
| 985 | |
| 986 | * Flush inherited counters to the parent - before the parent
|
| 987 | * gets woken up by child-exit notifications.
|
| 988 | *
|
| 989 | * because of cgroup mode, must be called before cgroup_exit()
|
| 990 |
|
| 991 | perf_event_exit_task(tsk);
|
| 992 |
|
| 993 | cgroup_exit(tsk, 1);
|
| 994 |
|
| 995 | if (group_dead)
|
| 996 | disassociate_ctty(1);
|
| 997 |
|
| 998 | module_put(task_thread_info(tsk)->exec_domain->module);
|
| 999 |
|
| 1000 | proc_exit_connector(tsk);
|
| 1001 |
|
| 1002 | |
| 1003 | * FIXME: do that only when needed, using sched_exit tracepoint
|
| 1004 |
|
| 1005 | ptrace_put_breakpoints(tsk);
|
| 1006 |
|
| 1007 | exit_notify(tsk, group_dead);
|
| 1008 | #ifdef CONFIG_NUMA
|
| 1009 | task_lock(tsk);
|
| 1010 | mpol_put(tsk->mempolicy);
|
| 1011 | tsk->mempolicy = NULL;
|
| 1012 | task_unlock(tsk);
|
| 1013 | #endif
|
| 1014 | #ifdef CONFIG_FUTEX
|
| 1015 | if (unlikely(current->pi_state_cache))
|
| 1016 | kfree(current->pi_state_cache);
|
| 1017 | #endif
|
| 1018 | |
| 1019 | * Make sure we are holding no locks:
|
| 1020 |
|
| 1021 | debug_check_no_locks_held(tsk);
|
| 1022 | |
| 1023 | * We can do this unlocked here. The futex code uses this flag
|
| 1024 | * just to verify whether the pi state cleanup has been done
|
| 1025 | * or not. In the worst case it loops once more.
|
| 1026 |
|
| 1027 | tsk->flags |= PF_EXITPIDONE;
|
| 1028 |
|
| 1029 | if (tsk->io_context)
|
| 1030 | exit_io_context(tsk);
|
| 1031 |
|
| 1032 | if (tsk->splice_pipe)
|
| 1033 | __free_pipe_info(tsk->splice_pipe);
|
| 1034 |
|
| 1035 | validate_creds_for_do_exit(tsk);
|
| 1036 |
|
| 1037 | preempt_disable();
|
| 1038 | if (tsk->nr_dirtied)
|
| 1039 | __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
|
| 1040 | exit_rcu();
|
| 1041 |
|
| 1042 | |
| 1043 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
|
| 1044 | * when the following two conditions become true.
|
| 1045 | * - There is race condition of mmap_sem (It is acquired by
|
| 1046 | * exit_mm()), and
|
| 1047 | * - SMI occurs before setting TASK_RUNINNG.
|
| 1048 | * (or hypervisor of virtual machine switches to other guest)
|
| 1049 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
|
| 1050 | *
|
| 1051 | * To avoid it, we have to wait for releasing tsk->pi_lock which
|
| 1052 | * is held by try_to_wake_up()
|
| 1053 |
|
| 1054 | smp_mb();
|
| 1055 | raw_spin_unlock_wait(&tsk->pi_lock);
|
| 1056 |
|
| 1057 |
|
| 1058 | tsk->state = TASK_DEAD;
|
| 1059 | tsk->flags |= PF_NOFREEZE;
|
| 1060 | schedule();
|
| 1061 | BUG();
|
| 1062 |
|
| 1063 | for (;;)
|
| 1064 | cpu_relax();
|
| 1065 | }
|
| 1066 |
|
| 1067 | EXPORT_SYMBOL_GPL(do_exit);
|
| 1068 |
|
| 1069 | void complete_and_exit(struct completion *comp, long code)
|
| 1070 | {
|
| 1071 | if (comp)
|
| 1072 | complete(comp);
|
| 1073 |
|
| 1074 | do_exit(code);
|
| 1075 | }
|
| 1076 |
|
| 1077 | EXPORT_SYMBOL(complete_and_exit);
|
| 1078 |
|
| 1079 | SYSCALL_DEFINE1(exit, int, error_code)
|
| 1080 | {
|
| 1081 | do_exit((error_code&0xff)<<8);
|
| 1082 | }
|
| 1083 |
|
| 1084 | |
| 1085 | * Take down every thread in the group. This is called by fatal signals
|
| 1086 | * as well as by sys_exit_group (below).
|
| 1087 |
|
| 1088 | void
|
| 1089 | do_group_exit(int exit_code)
|
| 1090 | {
|
| 1091 | struct signal_struct *sig = current->signal;
|
| 1092 |
|
| 1093 | BUG_ON(exit_code & 0x80);
|
| 1094 |
|
| 1095 | if (signal_group_exit(sig))
|
| 1096 | exit_code = sig->group_exit_code;
|
| 1097 | else if (!thread_group_empty(current)) {
|
| 1098 | struct sighand_struct *const sighand = current->sighand;
|
| 1099 | spin_lock_irq(&sighand->siglock);
|
| 1100 | if (signal_group_exit(sig))
|
| 1101 |
|
| 1102 | exit_code = sig->group_exit_code;
|
| 1103 | else {
|
| 1104 | sig->group_exit_code = exit_code;
|
| 1105 | sig->flags = SIGNAL_GROUP_EXIT;
|
| 1106 | zap_other_threads(current);
|
| 1107 | }
|
| 1108 | spin_unlock_irq(&sighand->siglock);
|
| 1109 | }
|
| 1110 |
|
| 1111 | do_exit(exit_code);
|
| 1112 |
|
| 1113 | }
|
| 1114 |
|
| 1115 | |
| 1116 | * this kills every thread in the thread group. Note that any externally
|
| 1117 | * wait4()-ing process will get the correct exit code - even if this
|
| 1118 | * thread is not the thread group leader.
|
| 1119 |
|
| 1120 | SYSCALL_DEFINE1(exit_group, int, error_code)
|
| 1121 | {
|
| 1122 | do_group_exit((error_code & 0xff) << 8);
|
| 1123 |
|
| 1124 | return 0;
|
| 1125 | }
|
| 1126 |
|
| 1127 | struct wait_opts {
|
| 1128 | enum pid_type wo_type;
|
| 1129 | int wo_flags;
|
| 1130 | struct pid *wo_pid;
|
| 1131 |
|
| 1132 | struct siginfo __user *wo_info;
|
| 1133 | int __user *wo_stat;
|
| 1134 | struct rusage __user *wo_rusage;
|
| 1135 |
|
| 1136 | wait_queue_t child_wait;
|
| 1137 | int notask_error;
|
| 1138 | };
|
| 1139 |
|
| 1140 | static inline
|
| 1141 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
|
| 1142 | {
|
| 1143 | if (type != PIDTYPE_PID)
|
| 1144 | task = task->group_leader;
|
| 1145 | return task->pids[type].pid;
|
| 1146 | }
|
| 1147 |
|
| 1148 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
|
| 1149 | {
|
| 1150 | return wo->wo_type == PIDTYPE_MAX ||
|
| 1151 | task_pid_type(p, wo->wo_type) == wo->wo_pid;
|
| 1152 | }
|
| 1153 |
|
| 1154 | static int eligible_child(struct wait_opts *wo, struct task_struct *p)
|
| 1155 | {
|
| 1156 | if (!eligible_pid(wo, p))
|
| 1157 | return 0;
|
| 1158 | |
| 1159 | * otherwise, wait for clone children *only* if __WCLONE is
|
| 1160 | * set; otherwise, wait for non-clone children *only*. (Note:
|
| 1161 | * A "clone" child here is one that reports to its parent
|
| 1162 | * using a signal other than SIGCHLD.)
|
| 1163 | if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
|
| 1164 | && !(wo->wo_flags & __WALL))
|
| 1165 | return 0;
|
| 1166 |
|
| 1167 | return 1;
|
| 1168 | }
|
| 1169 |
|
| 1170 | static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
|
| 1171 | pid_t pid, uid_t uid, int why, int status)
|
| 1172 | {
|
| 1173 | struct siginfo __user *infop;
|
| 1174 | int retval = wo->wo_rusage
|
| 1175 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1176 |
|
| 1177 | put_task_struct(p);
|
| 1178 | infop = wo->wo_info;
|
| 1179 | if (infop) {
|
| 1180 | if (!retval)
|
| 1181 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1182 | if (!retval)
|
| 1183 | retval = put_user(0, &infop->si_errno);
|
| 1184 | if (!retval)
|
| 1185 | retval = put_user((short)why, &infop->si_code);
|
| 1186 | if (!retval)
|
| 1187 | retval = put_user(pid, &infop->si_pid);
|
| 1188 | if (!retval)
|
| 1189 | retval = put_user(uid, &infop->si_uid);
|
| 1190 | if (!retval)
|
| 1191 | retval = put_user(status, &infop->si_status);
|
| 1192 | }
|
| 1193 | if (!retval)
|
| 1194 | retval = pid;
|
| 1195 | return retval;
|
| 1196 | }
|
| 1197 |
|
| 1198 | |
| 1199 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
|
| 1200 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
|
| 1201 | * the lock and this task is uninteresting. If we return nonzero, we have
|
| 1202 | * released the lock and the system call should return.
|
| 1203 |
|
| 1204 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
|
| 1205 | {
|
| 1206 | unsigned long state;
|
| 1207 | int retval, status, traced;
|
| 1208 | pid_t pid = task_pid_vnr(p);
|
| 1209 | uid_t uid = __task_cred(p)->uid;
|
| 1210 | struct siginfo __user *infop;
|
| 1211 |
|
| 1212 | if (!likely(wo->wo_flags & WEXITED))
|
| 1213 | return 0;
|
| 1214 |
|
| 1215 | if (unlikely(wo->wo_flags & WNOWAIT)) {
|
| 1216 | int exit_code = p->exit_code;
|
| 1217 | int why;
|
| 1218 |
|
| 1219 | get_task_struct(p);
|
| 1220 | read_unlock(&tasklist_lock);
|
| 1221 | if ((exit_code & 0x7f) == 0) {
|
| 1222 | why = CLD_EXITED;
|
| 1223 | status = exit_code >> 8;
|
| 1224 | } else {
|
| 1225 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
|
| 1226 | status = exit_code & 0x7f;
|
| 1227 | }
|
| 1228 | return wait_noreap_copyout(wo, p, pid, uid, why, status);
|
| 1229 | }
|
| 1230 |
|
| 1231 | |
| 1232 | * Try to move the task's state to DEAD
|
| 1233 | * only one thread is allowed to do this:
|
| 1234 |
|
| 1235 | state = xchg(&p->exit_state, EXIT_DEAD);
|
| 1236 | if (state != EXIT_ZOMBIE) {
|
| 1237 | BUG_ON(state != EXIT_DEAD);
|
| 1238 | return 0;
|
| 1239 | }
|
| 1240 |
|
| 1241 | traced = ptrace_reparented(p);
|
| 1242 | |
| 1243 | * It can be ptraced but not reparented, check
|
| 1244 | * thread_group_leader() to filter out sub-threads.
|
| 1245 |
|
| 1246 | if (likely(!traced) && thread_group_leader(p)) {
|
| 1247 | struct signal_struct *psig;
|
| 1248 | struct signal_struct *sig;
|
| 1249 | unsigned long maxrss;
|
| 1250 | cputime_t tgutime, tgstime;
|
| 1251 |
|
| 1252 | |
| 1253 | * The resource counters for the group leader are in its
|
| 1254 | * own task_struct. Those for dead threads in the group
|
| 1255 | * are in its signal_struct, as are those for the child
|
| 1256 | * processes it has previously reaped. All these
|
| 1257 | * accumulate in the parent's signal_struct c* fields.
|
| 1258 | *
|
| 1259 | * We don't bother to take a lock here to protect these
|
| 1260 | * p->signal fields, because they are only touched by
|
| 1261 | * __exit_signal, which runs with tasklist_lock
|
| 1262 | * write-locked anyway, and so is excluded here. We do
|
| 1263 | * need to protect the access to parent->signal fields,
|
| 1264 | * as other threads in the parent group can be right
|
| 1265 | * here reaping other children at the same time.
|
| 1266 | *
|
| 1267 | * We use thread_group_times() to get times for the thread
|
| 1268 | * group, which consolidates times for all threads in the
|
| 1269 | * group including the group leader.
|
| 1270 |
|
| 1271 | thread_group_times(p, &tgutime, &tgstime);
|
| 1272 | spin_lock_irq(&p->real_parent->sighand->siglock);
|
| 1273 | psig = p->real_parent->signal;
|
| 1274 | sig = p->signal;
|
| 1275 | psig->cutime += tgutime + sig->cutime;
|
| 1276 | psig->cstime += tgstime + sig->cstime;
|
| 1277 | psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
|
| 1278 | psig->cmin_flt +=
|
| 1279 | p->min_flt + sig->min_flt + sig->cmin_flt;
|
| 1280 | psig->cmaj_flt +=
|
| 1281 | p->maj_flt + sig->maj_flt + sig->cmaj_flt;
|
| 1282 | psig->cnvcsw +=
|
| 1283 | p->nvcsw + sig->nvcsw + sig->cnvcsw;
|
| 1284 | psig->cnivcsw +=
|
| 1285 | p->nivcsw + sig->nivcsw + sig->cnivcsw;
|
| 1286 | psig->cinblock +=
|
| 1287 | task_io_get_inblock(p) +
|
| 1288 | sig->inblock + sig->cinblock;
|
| 1289 | psig->coublock +=
|
| 1290 | task_io_get_oublock(p) +
|
| 1291 | sig->oublock + sig->coublock;
|
| 1292 | maxrss = max(sig->maxrss, sig->cmaxrss);
|
| 1293 | if (psig->cmaxrss < maxrss)
|
| 1294 | psig->cmaxrss = maxrss;
|
| 1295 | task_io_accounting_add(&psig->ioac, &p->ioac);
|
| 1296 | task_io_accounting_add(&psig->ioac, &sig->ioac);
|
| 1297 | spin_unlock_irq(&p->real_parent->sighand->siglock);
|
| 1298 | }
|
| 1299 |
|
| 1300 | |
| 1301 | * Now we are sure this task is interesting, and no other
|
| 1302 | * thread can reap it because we set its state to EXIT_DEAD.
|
| 1303 |
|
| 1304 | read_unlock(&tasklist_lock);
|
| 1305 |
|
| 1306 | retval = wo->wo_rusage
|
| 1307 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1308 | status = (p->signal->flags & SIGNAL_GROUP_EXIT)
|
| 1309 | ? p->signal->group_exit_code : p->exit_code;
|
| 1310 | if (!retval && wo->wo_stat)
|
| 1311 | retval = put_user(status, wo->wo_stat);
|
| 1312 |
|
| 1313 | infop = wo->wo_info;
|
| 1314 | if (!retval && infop)
|
| 1315 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1316 | if (!retval && infop)
|
| 1317 | retval = put_user(0, &infop->si_errno);
|
| 1318 | if (!retval && infop) {
|
| 1319 | int why;
|
| 1320 |
|
| 1321 | if ((status & 0x7f) == 0) {
|
| 1322 | why = CLD_EXITED;
|
| 1323 | status >>= 8;
|
| 1324 | } else {
|
| 1325 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
|
| 1326 | status &= 0x7f;
|
| 1327 | }
|
| 1328 | retval = put_user((short)why, &infop->si_code);
|
| 1329 | if (!retval)
|
| 1330 | retval = put_user(status, &infop->si_status);
|
| 1331 | }
|
| 1332 | if (!retval && infop)
|
| 1333 | retval = put_user(pid, &infop->si_pid);
|
| 1334 | if (!retval && infop)
|
| 1335 | retval = put_user(uid, &infop->si_uid);
|
| 1336 | if (!retval)
|
| 1337 | retval = pid;
|
| 1338 |
|
| 1339 | if (traced) {
|
| 1340 | write_lock_irq(&tasklist_lock);
|
| 1341 |
|
| 1342 | ptrace_unlink(p);
|
| 1343 | |
| 1344 | * If this is not a sub-thread, notify the parent.
|
| 1345 | * If parent wants a zombie, don't release it now.
|
| 1346 |
|
| 1347 | if (thread_group_leader(p) &&
|
| 1348 | !do_notify_parent(p, p->exit_signal)) {
|
| 1349 | p->exit_state = EXIT_ZOMBIE;
|
| 1350 | p = NULL;
|
| 1351 | }
|
| 1352 | write_unlock_irq(&tasklist_lock);
|
| 1353 | }
|
| 1354 | if (p != NULL)
|
| 1355 | release_task(p);
|
| 1356 |
|
| 1357 | return retval;
|
| 1358 | }
|
| 1359 |
|
| 1360 | static int *task_stopped_code(struct task_struct *p, bool ptrace)
|
| 1361 | {
|
| 1362 | if (ptrace) {
|
| 1363 | if (task_is_stopped_or_traced(p) &&
|
| 1364 | !(p->jobctl & JOBCTL_LISTENING))
|
| 1365 | return &p->exit_code;
|
| 1366 | } else {
|
| 1367 | if (p->signal->flags & SIGNAL_STOP_STOPPED)
|
| 1368 | return &p->signal->group_exit_code;
|
| 1369 | }
|
| 1370 | return NULL;
|
| 1371 | }
|
| 1372 |
|
| 1373 | |
| 1374 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
|
| 1375 | * @wo: wait options
|
| 1376 | * @ptrace: is the wait for ptrace
|
| 1377 | * @p: task to wait for
|
| 1378 | *
|
| 1379 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
|
| 1380 | *
|
| 1381 | * CONTEXT:
|
| 1382 | * read_lock(&tasklist_lock), which is released if return value is
|
| 1383 | * non-zero. Also, grabs and releases @p->sighand->siglock.
|
| 1384 | *
|
| 1385 | * RETURNS:
|
| 1386 | * 0 if wait condition didn't exist and search for other wait conditions
|
| 1387 | * should continue. Non-zero return, -errno on failure and @p's pid on
|
| 1388 | * success, implies that tasklist_lock is released and wait condition
|
| 1389 | * search should terminate.
|
| 1390 |
|
| 1391 | static int wait_task_stopped(struct wait_opts *wo,
|
| 1392 | int ptrace, struct task_struct *p)
|
| 1393 | {
|
| 1394 | struct siginfo __user *infop;
|
| 1395 | int retval, exit_code, *p_code, why;
|
| 1396 | uid_t uid = 0;
|
| 1397 | pid_t pid;
|
| 1398 |
|
| 1399 | |
| 1400 | * Traditionally we see ptrace'd stopped tasks regardless of options.
|
| 1401 |
|
| 1402 | if (!ptrace && !(wo->wo_flags & WUNTRACED))
|
| 1403 | return 0;
|
| 1404 |
|
| 1405 | if (!task_stopped_code(p, ptrace))
|
| 1406 | return 0;
|
| 1407 |
|
| 1408 | exit_code = 0;
|
| 1409 | spin_lock_irq(&p->sighand->siglock);
|
| 1410 |
|
| 1411 | p_code = task_stopped_code(p, ptrace);
|
| 1412 | if (unlikely(!p_code))
|
| 1413 | goto unlock_sig;
|
| 1414 |
|
| 1415 | exit_code = *p_code;
|
| 1416 | if (!exit_code)
|
| 1417 | goto unlock_sig;
|
| 1418 |
|
| 1419 | if (!unlikely(wo->wo_flags & WNOWAIT))
|
| 1420 | *p_code = 0;
|
| 1421 |
|
| 1422 | uid = task_uid(p);
|
| 1423 | unlock_sig:
|
| 1424 | spin_unlock_irq(&p->sighand->siglock);
|
| 1425 | if (!exit_code)
|
| 1426 | return 0;
|
| 1427 |
|
| 1428 | |
| 1429 | * Now we are pretty sure this task is interesting.
|
| 1430 | * Make sure it doesn't get reaped out from under us while we
|
| 1431 | * give up the lock and then examine it below. We don't want to
|
| 1432 | * keep holding onto the tasklist_lock while we call getrusage and
|
| 1433 | * possibly take page faults for user memory.
|
| 1434 |
|
| 1435 | get_task_struct(p);
|
| 1436 | pid = task_pid_vnr(p);
|
| 1437 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
|
| 1438 | read_unlock(&tasklist_lock);
|
| 1439 |
|
| 1440 | if (unlikely(wo->wo_flags & WNOWAIT))
|
| 1441 | return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
|
| 1442 |
|
| 1443 | retval = wo->wo_rusage
|
| 1444 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1445 | if (!retval && wo->wo_stat)
|
| 1446 | retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
|
| 1447 |
|
| 1448 | infop = wo->wo_info;
|
| 1449 | if (!retval && infop)
|
| 1450 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1451 | if (!retval && infop)
|
| 1452 | retval = put_user(0, &infop->si_errno);
|
| 1453 | if (!retval && infop)
|
| 1454 | retval = put_user((short)why, &infop->si_code);
|
| 1455 | if (!retval && infop)
|
| 1456 | retval = put_user(exit_code, &infop->si_status);
|
| 1457 | if (!retval && infop)
|
| 1458 | retval = put_user(pid, &infop->si_pid);
|
| 1459 | if (!retval && infop)
|
| 1460 | retval = put_user(uid, &infop->si_uid);
|
| 1461 | if (!retval)
|
| 1462 | retval = pid;
|
| 1463 | put_task_struct(p);
|
| 1464 |
|
| 1465 | BUG_ON(!retval);
|
| 1466 | return retval;
|
| 1467 | }
|
| 1468 |
|
| 1469 | |
| 1470 | * Handle do_wait work for one task in a live, non-stopped state.
|
| 1471 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
|
| 1472 | * the lock and this task is uninteresting. If we return nonzero, we have
|
| 1473 | * released the lock and the system call should return.
|
| 1474 |
|
| 1475 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
|
| 1476 | {
|
| 1477 | int retval;
|
| 1478 | pid_t pid;
|
| 1479 | uid_t uid;
|
| 1480 |
|
| 1481 | if (!unlikely(wo->wo_flags & WCONTINUED))
|
| 1482 | return 0;
|
| 1483 |
|
| 1484 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
|
| 1485 | return 0;
|
| 1486 |
|
| 1487 | spin_lock_irq(&p->sighand->siglock);
|
| 1488 |
|
| 1489 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
|
| 1490 | spin_unlock_irq(&p->sighand->siglock);
|
| 1491 | return 0;
|
| 1492 | }
|
| 1493 | if (!unlikely(wo->wo_flags & WNOWAIT))
|
| 1494 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
|
| 1495 | uid = task_uid(p);
|
| 1496 | spin_unlock_irq(&p->sighand->siglock);
|
| 1497 |
|
| 1498 | pid = task_pid_vnr(p);
|
| 1499 | get_task_struct(p);
|
| 1500 | read_unlock(&tasklist_lock);
|
| 1501 |
|
| 1502 | if (!wo->wo_info) {
|
| 1503 | retval = wo->wo_rusage
|
| 1504 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1505 | put_task_struct(p);
|
| 1506 | if (!retval && wo->wo_stat)
|
| 1507 | retval = put_user(0xffff, wo->wo_stat);
|
| 1508 | if (!retval)
|
| 1509 | retval = pid;
|
| 1510 | } else {
|
| 1511 | retval = wait_noreap_copyout(wo, p, pid, uid,
|
| 1512 | CLD_CONTINUED, SIGCONT);
|
| 1513 | BUG_ON(retval == 0);
|
| 1514 | }
|
| 1515 |
|
| 1516 | return retval;
|
| 1517 | }
|
| 1518 |
|
| 1519 | |
| 1520 | * Consider @p for a wait by @parent.
|
| 1521 | *
|
| 1522 | * -ECHILD should be in ->notask_error before the first call.
|
| 1523 | * Returns nonzero for a final return, when we have unlocked tasklist_lock.
|
| 1524 | * Returns zero if the search for a child should continue;
|
| 1525 | * then ->notask_error is 0 if @p is an eligible child,
|
| 1526 | * or another error from security_task_wait(), or still -ECHILD.
|
| 1527 |
|
| 1528 | static int wait_consider_task(struct wait_opts *wo, int ptrace,
|
| 1529 | struct task_struct *p)
|
| 1530 | {
|
| 1531 | int ret = eligible_child(wo, p);
|
| 1532 | if (!ret)
|
| 1533 | return ret;
|
| 1534 |
|
| 1535 | ret = security_task_wait(p);
|
| 1536 | if (unlikely(ret < 0)) {
|
| 1537 | |
| 1538 | * If we have not yet seen any eligible child,
|
| 1539 | * then let this error code replace -ECHILD.
|
| 1540 | * A permission error will give the user a clue
|
| 1541 | * to look for security policy problems, rather
|
| 1542 | * than for mysterious wait bugs.
|
| 1543 |
|
| 1544 | if (wo->notask_error)
|
| 1545 | wo->notask_error = ret;
|
| 1546 | return 0;
|
| 1547 | }
|
| 1548 |
|
| 1549 |
|
| 1550 | if (unlikely(p->exit_state == EXIT_DEAD)) {
|
| 1551 | |
| 1552 | * But do not ignore this task until the tracer does
|
| 1553 | * wait_task_zombie()->do_notify_parent().
|
| 1554 |
|
| 1555 | if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
|
| 1556 | wo->notask_error = 0;
|
| 1557 | return 0;
|
| 1558 | }
|
| 1559 |
|
| 1560 |
|
| 1561 | if (p->exit_state == EXIT_ZOMBIE) {
|
| 1562 | |
| 1563 | * A zombie ptracee is only visible to its ptracer.
|
| 1564 | * Notification and reaping will be cascaded to the real
|
| 1565 | * parent when the ptracer detaches.
|
| 1566 |
|
| 1567 | if (likely(!ptrace) && unlikely(p->ptrace)) {
|
| 1568 |
|
| 1569 | wo->notask_error = 0;
|
| 1570 | return 0;
|
| 1571 | }
|
| 1572 |
|
| 1573 |
|
| 1574 | if (!delay_group_leader(p))
|
| 1575 | return wait_task_zombie(wo, p);
|
| 1576 |
|
| 1577 | |
| 1578 | * Allow access to stopped/continued state via zombie by
|
| 1579 | * falling through. Clearing of notask_error is complex.
|
| 1580 | *
|
| 1581 | * When !@ptrace:
|
| 1582 | *
|
| 1583 | * If WEXITED is set, notask_error should naturally be
|
| 1584 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
|
| 1585 | * so, if there are live subthreads, there are events to
|
| 1586 | * wait for. If all subthreads are dead, it's still safe
|
| 1587 | * to clear - this function will be called again in finite
|
| 1588 | * amount time once all the subthreads are released and
|
| 1589 | * will then return without clearing.
|
| 1590 | *
|
| 1591 | * When @ptrace:
|
| 1592 | *
|
| 1593 | * Stopped state is per-task and thus can't change once the
|
| 1594 | * target task dies. Only continued and exited can happen.
|
| 1595 | * Clear notask_error if WCONTINUED | WEXITED.
|
| 1596 |
|
| 1597 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
|
| 1598 | wo->notask_error = 0;
|
| 1599 | } else {
|
| 1600 | |
| 1601 | * If @p is ptraced by a task in its real parent's group,
|
| 1602 | * hide group stop/continued state when looking at @p as
|
| 1603 | * the real parent; otherwise, a single stop can be
|
| 1604 | * reported twice as group and ptrace stops.
|
| 1605 | *
|
| 1606 | * If a ptracer wants to distinguish the two events for its
|
| 1607 | * own children, it should create a separate process which
|
| 1608 | * takes the role of real parent.
|
| 1609 |
|
| 1610 | if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
|
| 1611 | return 0;
|
| 1612 |
|
| 1613 | |
| 1614 | * @p is alive and it's gonna stop, continue or exit, so
|
| 1615 | * there always is something to wait for.
|
| 1616 |
|
| 1617 | wo->notask_error = 0;
|
| 1618 | }
|
| 1619 |
|
| 1620 | |
| 1621 | * Wait for stopped. Depending on @ptrace, different stopped state
|
| 1622 | * is used and the two don't interact with each other.
|
| 1623 |
|
| 1624 | ret = wait_task_stopped(wo, ptrace, p);
|
| 1625 | if (ret)
|
| 1626 | return ret;
|
| 1627 |
|
| 1628 | |
| 1629 | * Wait for continued. There's only one continued state and the
|
| 1630 | * ptracer can consume it which can confuse the real parent. Don't
|
| 1631 | * use WCONTINUED from ptracer. You don't need or want it.
|
| 1632 |
|
| 1633 | return wait_task_continued(wo, p);
|
| 1634 | }
|
| 1635 |
|
| 1636 | |
| 1637 | * Do the work of do_wait() for one thread in the group, @tsk.
|
| 1638 | *
|
| 1639 | * -ECHILD should be in ->notask_error before the first call.
|
| 1640 | * Returns nonzero for a final return, when we have unlocked tasklist_lock.
|
| 1641 | * Returns zero if the search for a child should continue; then
|
| 1642 | * ->notask_error is 0 if there were any eligible children,
|
| 1643 | * or another error from security_task_wait(), or still -ECHILD.
|
| 1644 |
|
| 1645 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
|
| 1646 | {
|
| 1647 | struct task_struct *p;
|
| 1648 |
|
| 1649 | list_for_each_entry(p, &tsk->children, sibling) {
|
| 1650 | int ret = wait_consider_task(wo, 0, p);
|
| 1651 | if (ret)
|
| 1652 | return ret;
|
| 1653 | }
|
| 1654 |
|
| 1655 | return 0;
|
| 1656 | }
|
| 1657 |
|
| 1658 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
|
| 1659 | {
|
| 1660 | struct task_struct *p;
|
| 1661 |
|
| 1662 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
|
| 1663 | int ret = wait_consider_task(wo, 1, p);
|
| 1664 | if (ret)
|
| 1665 | return ret;
|
| 1666 | }
|
| 1667 |
|
| 1668 | return 0;
|
| 1669 | }
|
| 1670 |
|
| 1671 | static int child_wait_callback(wait_queue_t *wait, unsigned mode,
|
| 1672 | int sync, void *key)
|
| 1673 | {
|
| 1674 | struct wait_opts *wo = container_of(wait, struct wait_opts,
|
| 1675 | child_wait);
|
| 1676 | struct task_struct *p = key;
|
| 1677 |
|
| 1678 | if (!eligible_pid(wo, p))
|
| 1679 | return 0;
|
| 1680 |
|
| 1681 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
|
| 1682 | return 0;
|
| 1683 |
|
| 1684 | return default_wake_function(wait, mode, sync, key);
|
| 1685 | }
|
| 1686 |
|
| 1687 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
|
| 1688 | {
|
| 1689 | __wake_up_sync_key(&parent->signal->wait_chldexit,
|
| 1690 | TASK_INTERRUPTIBLE, 1, p);
|
| 1691 | }
|
| 1692 |
|
| 1693 | static long do_wait(struct wait_opts *wo)
|
| 1694 | {
|
| 1695 | struct task_struct *tsk;
|
| 1696 | int retval;
|
| 1697 |
|
| 1698 | trace_sched_process_wait(wo->wo_pid);
|
| 1699 |
|
| 1700 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
|
| 1701 | wo->child_wait.private = current;
|
| 1702 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
|
| 1703 | repeat:
|
| 1704 | |
| 1705 | * If there is nothing that can match our critiera just get out.
|
| 1706 | * We will clear ->notask_error to zero if we see any child that
|
| 1707 | * might later match our criteria, even if we are not able to reap
|
| 1708 | * it yet.
|
| 1709 |
|
| 1710 | wo->notask_error = -ECHILD;
|
| 1711 | if ((wo->wo_type < PIDTYPE_MAX) &&
|
| 1712 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
|
| 1713 | goto notask;
|
| 1714 |
|
| 1715 | set_current_state(TASK_INTERRUPTIBLE);
|
| 1716 | read_lock(&tasklist_lock);
|
| 1717 | tsk = current;
|
| 1718 | do {
|
| 1719 | retval = do_wait_thread(wo, tsk);
|
| 1720 | if (retval)
|
| 1721 | goto end;
|
| 1722 |
|
| 1723 | retval = ptrace_do_wait(wo, tsk);
|
| 1724 | if (retval)
|
| 1725 | goto end;
|
| 1726 |
|
| 1727 | if (wo->wo_flags & __WNOTHREAD)
|
| 1728 | break;
|
| 1729 | } while_each_thread(current, tsk);
|
| 1730 | read_unlock(&tasklist_lock);
|
| 1731 |
|
| 1732 | notask:
|
| 1733 | retval = wo->notask_error;
|
| 1734 | if (!retval && !(wo->wo_flags & WNOHANG)) {
|
| 1735 | retval = -ERESTARTSYS;
|
| 1736 | if (!signal_pending(current)) {
|
| 1737 | schedule();
|
| 1738 | goto repeat;
|
| 1739 | }
|
| 1740 | }
|
| 1741 | end:
|
| 1742 | __set_current_state(TASK_RUNNING);
|
| 1743 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
|
| 1744 | return retval;
|
| 1745 | }
|
| 1746 |
|
| 1747 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
|
| 1748 | infop, int, options, struct rusage __user *, ru)
|
| 1749 | {
|
| 1750 | struct wait_opts wo;
|
| 1751 | struct pid *pid = NULL;
|
| 1752 | enum pid_type type;
|
| 1753 | long ret;
|
| 1754 |
|
| 1755 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
|
| 1756 | return -EINVAL;
|
| 1757 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
|
| 1758 | return -EINVAL;
|
| 1759 |
|
| 1760 | switch (which) {
|
| 1761 | case P_ALL:
|
| 1762 | type = PIDTYPE_MAX;
|
| 1763 | break;
|
| 1764 | case P_PID:
|
| 1765 | type = PIDTYPE_PID;
|
| 1766 | if (upid <= 0)
|
| 1767 | return -EINVAL;
|
| 1768 | break;
|
| 1769 | case P_PGID:
|
| 1770 | type = PIDTYPE_PGID;
|
| 1771 | if (upid <= 0)
|
| 1772 | return -EINVAL;
|
| 1773 | break;
|
| 1774 | default:
|
| 1775 | return -EINVAL;
|
| 1776 | }
|
| 1777 |
|
| 1778 | if (type < PIDTYPE_MAX)
|
| 1779 | pid = find_get_pid(upid);
|
| 1780 |
|
| 1781 | wo.wo_type = type;
|
| 1782 | wo.wo_pid = pid;
|
| 1783 | wo.wo_flags = options;
|
| 1784 | wo.wo_info = infop;
|
| 1785 | wo.wo_stat = NULL;
|
| 1786 | wo.wo_rusage = ru;
|
| 1787 | ret = do_wait(&wo);
|
| 1788 |
|
| 1789 | if (ret > 0) {
|
| 1790 | ret = 0;
|
| 1791 | } else if (infop) {
|
| 1792 | |
| 1793 | * For a WNOHANG return, clear out all the fields
|
| 1794 | * we would set so the user can easily tell the
|
| 1795 | * difference.
|
| 1796 |
|
| 1797 | if (!ret)
|
| 1798 | ret = put_user(0, &infop->si_signo);
|
| 1799 | if (!ret)
|
| 1800 | ret = put_user(0, &infop->si_errno);
|
| 1801 | if (!ret)
|
| 1802 | ret = put_user(0, &infop->si_code);
|
| 1803 | if (!ret)
|
| 1804 | ret = put_user(0, &infop->si_pid);
|
| 1805 | if (!ret)
|
| 1806 | ret = put_user(0, &infop->si_uid);
|
| 1807 | if (!ret)
|
| 1808 | ret = put_user(0, &infop->si_status);
|
| 1809 | }
|
| 1810 |
|
| 1811 | put_pid(pid);
|
| 1812 |
|
| 1813 |
|
| 1814 | asmlinkage_protect(5, ret, which, upid, infop, options, ru);
|
| 1815 | return ret;
|
| 1816 | }
|
| 1817 |
|
| 1818 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
|
| 1819 | int, options, struct rusage __user *, ru)
|
| 1820 | {
|
| 1821 | struct wait_opts wo;
|
| 1822 | struct pid *pid = NULL;
|
| 1823 | enum pid_type type;
|
| 1824 | long ret;
|
| 1825 |
|
| 1826 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
|
| 1827 | __WNOTHREAD|__WCLONE|__WALL))
|
| 1828 | return -EINVAL;
|
| 1829 |
|
| 1830 | if (upid == -1)
|
| 1831 | type = PIDTYPE_MAX;
|
| 1832 | else if (upid < 0) {
|
| 1833 | type = PIDTYPE_PGID;
|
| 1834 | pid = find_get_pid(-upid);
|
| 1835 | } else if (upid == 0) {
|
| 1836 | type = PIDTYPE_PGID;
|
| 1837 | pid = get_task_pid(current, PIDTYPE_PGID);
|
| 1838 | } else {
|
| 1839 | type = PIDTYPE_PID;
|
| 1840 | pid = find_get_pid(upid);
|
| 1841 | }
|
| 1842 |
|
| 1843 | wo.wo_type = type;
|
| 1844 | wo.wo_pid = pid;
|
| 1845 | wo.wo_flags = options | WEXITED;
|
| 1846 | wo.wo_info = NULL;
|
| 1847 | wo.wo_stat = stat_addr;
|
| 1848 | wo.wo_rusage = ru;
|
| 1849 | ret = do_wait(&wo);
|
| 1850 | put_pid(pid);
|
| 1851 |
|
| 1852 |
|
| 1853 | asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
|
| 1854 | return ret;
|
| 1855 | }
|
| 1856 |
|
| 1857 | #ifdef __ARCH_WANT_SYS_WAITPID
|
| 1858 |
|
| 1859 | |
| 1860 | * sys_waitpid() remains for compatibility. waitpid() should be
|
| 1861 | * implemented by calling sys_wait4() from libc.a.
|
| 1862 |
|
| 1863 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
|
| 1864 | {
|
| 1865 | return sys_wait4(pid, stat_addr, options, NULL);
|
| 1866 | }
|
| 1867 |
|
| 1868 | #endif
|