| 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 | #include <linux/shm.h>
|
| 56 |
|
| 57 | #include <asm/uaccess.h>
|
| 58 | #include <asm/unistd.h>
|
| 59 | #include <asm/pgtable.h>
|
| 60 | #include <asm/mmu_context.h>
|
| 61 |
|
| 62 | static void exit_mm(struct task_struct * tsk);
|
| 63 |
|
| 64 | static void __unhash_process(struct task_struct *p, bool group_dead)
|
| 65 | {
|
| 66 | nr_threads--;
|
| 67 | detach_pid(p, PIDTYPE_PID);
|
| 68 | if (group_dead) {
|
| 69 | detach_pid(p, PIDTYPE_PGID);
|
| 70 | detach_pid(p, PIDTYPE_SID);
|
| 71 |
|
| 72 | list_del_rcu(&p->tasks);
|
| 73 | list_del_init(&p->sibling);
|
| 74 | __this_cpu_dec(process_counts);
|
| 75 | |
| 76 | * If we are the last child process in a pid namespace to be
|
| 77 | * reaped, notify the reaper sleeping zap_pid_ns_processes().
|
| 78 |
|
| 79 | if (IS_ENABLED(CONFIG_PID_NS)) {
|
| 80 | struct task_struct *parent = p->real_parent;
|
| 81 |
|
| 82 | if ((task_active_pid_ns(parent)->child_reaper == parent) &&
|
| 83 | list_empty(&parent->children) &&
|
| 84 | (parent->flags & PF_EXITING))
|
| 85 | wake_up_process(parent);
|
| 86 | }
|
| 87 | }
|
| 88 | list_del_rcu(&p->thread_group);
|
| 89 | }
|
| 90 |
|
| 91 | |
| 92 | * This function expects the tasklist_lock write-locked.
|
| 93 |
|
| 94 | static void __exit_signal(struct task_struct *tsk)
|
| 95 | {
|
| 96 | struct signal_struct *sig = tsk->signal;
|
| 97 | bool group_dead = thread_group_leader(tsk);
|
| 98 | struct sighand_struct *sighand;
|
| 99 | struct tty_struct *uninitialized_var(tty);
|
| 100 |
|
| 101 | sighand = rcu_dereference_check(tsk->sighand,
|
| 102 | lockdep_tasklist_lock_is_held());
|
| 103 | spin_lock(&sighand->siglock);
|
| 104 |
|
| 105 | posix_cpu_timers_exit(tsk);
|
| 106 | if (group_dead) {
|
| 107 | posix_cpu_timers_exit_group(tsk);
|
| 108 | tty = sig->tty;
|
| 109 | sig->tty = NULL;
|
| 110 | } else {
|
| 111 | |
| 112 | * This can only happen if the caller is de_thread().
|
| 113 | * FIXME: this is the temporary hack, we should teach
|
| 114 | * posix-cpu-timers to handle this case correctly.
|
| 115 |
|
| 116 | if (unlikely(has_group_leader_pid(tsk)))
|
| 117 | posix_cpu_timers_exit_group(tsk);
|
| 118 |
|
| 119 | |
| 120 | * If there is any task waiting for the group exit
|
| 121 | * then notify it:
|
| 122 |
|
| 123 | if (sig->notify_count > 0 && !--sig->notify_count)
|
| 124 | wake_up_process(sig->group_exit_task);
|
| 125 |
|
| 126 | if (tsk == sig->curr_target)
|
| 127 | sig->curr_target = next_thread(tsk);
|
| 128 | |
| 129 | * Accumulate here the counters for all threads but the
|
| 130 | * group leader as they die, so they can be added into
|
| 131 | * the process-wide totals when those are taken.
|
| 132 | * The group leader stays around as a zombie as long
|
| 133 | * as there are other threads. When it gets reaped,
|
| 134 | * the exit.c code will add its counts into these totals.
|
| 135 | * We won't ever get here for the group leader, since it
|
| 136 | * will have been the last reference on the signal_struct.
|
| 137 |
|
| 138 | sig->utime += tsk->utime;
|
| 139 | sig->stime += tsk->stime;
|
| 140 | sig->gtime += tsk->gtime;
|
| 141 | sig->min_flt += tsk->min_flt;
|
| 142 | sig->maj_flt += tsk->maj_flt;
|
| 143 | sig->nvcsw += tsk->nvcsw;
|
| 144 | sig->nivcsw += tsk->nivcsw;
|
| 145 | sig->inblock += task_io_get_inblock(tsk);
|
| 146 | sig->oublock += task_io_get_oublock(tsk);
|
| 147 | task_io_accounting_add(&sig->ioac, &tsk->ioac);
|
| 148 | sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
|
| 149 | }
|
| 150 |
|
| 151 | sig->nr_threads--;
|
| 152 | __unhash_process(tsk, group_dead);
|
| 153 |
|
| 154 | |
| 155 | * Do this under ->siglock, we can race with another thread
|
| 156 | * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
|
| 157 |
|
| 158 | flush_sigqueue(&tsk->pending);
|
| 159 | tsk->sighand = NULL;
|
| 160 | spin_unlock(&sighand->siglock);
|
| 161 |
|
| 162 | __cleanup_sighand(sighand);
|
| 163 | clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
|
| 164 | if (group_dead) {
|
| 165 | flush_sigqueue(&sig->shared_pending);
|
| 166 | tty_kref_put(tty);
|
| 167 | }
|
| 168 | }
|
| 169 |
|
| 170 | static void delayed_put_task_struct(struct rcu_head *rhp)
|
| 171 | {
|
| 172 | struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
|
| 173 |
|
| 174 | perf_event_delayed_put(tsk);
|
| 175 | trace_sched_process_free(tsk);
|
| 176 | put_task_struct(tsk);
|
| 177 | }
|
| 178 |
|
| 179 |
|
| 180 | void release_task(struct task_struct * p)
|
| 181 | {
|
| 182 | struct task_struct *leader;
|
| 183 | int zap_leader;
|
| 184 | repeat:
|
| 185 | |
| 186 | * can't be modifying its own credentials. But shut RCU-lockdep up
|
| 187 | rcu_read_lock();
|
| 188 | atomic_dec(&__task_cred(p)->user->processes);
|
| 189 | rcu_read_unlock();
|
| 190 |
|
| 191 | proc_flush_task(p);
|
| 192 |
|
| 193 | write_lock_irq(&tasklist_lock);
|
| 194 | ptrace_release_task(p);
|
| 195 | __exit_signal(p);
|
| 196 |
|
| 197 | |
| 198 | * If we are the last non-leader member of the thread
|
| 199 | * group, and the leader is zombie, then notify the
|
| 200 | * group leader's parent process. (if it wants notification.)
|
| 201 |
|
| 202 | zap_leader = 0;
|
| 203 | leader = p->group_leader;
|
| 204 | if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
|
| 205 | |
| 206 | * If we were the last child thread and the leader has
|
| 207 | * exited already, and the leader's parent ignores SIGCHLD,
|
| 208 | * then we are the one who should release the leader.
|
| 209 |
|
| 210 | zap_leader = do_notify_parent(leader, leader->exit_signal);
|
| 211 | if (zap_leader)
|
| 212 | leader->exit_state = EXIT_DEAD;
|
| 213 | }
|
| 214 |
|
| 215 | write_unlock_irq(&tasklist_lock);
|
| 216 | release_thread(p);
|
| 217 | call_rcu(&p->rcu, delayed_put_task_struct);
|
| 218 |
|
| 219 | p = leader;
|
| 220 | if (unlikely(zap_leader))
|
| 221 | goto repeat;
|
| 222 | }
|
| 223 |
|
| 224 | |
| 225 | * This checks not only the pgrp, but falls back on the pid if no
|
| 226 | * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
|
| 227 | * without this...
|
| 228 | *
|
| 229 | * The caller must hold rcu lock or the tasklist lock.
|
| 230 |
|
| 231 | struct pid *session_of_pgrp(struct pid *pgrp)
|
| 232 | {
|
| 233 | struct task_struct *p;
|
| 234 | struct pid *sid = NULL;
|
| 235 |
|
| 236 | p = pid_task(pgrp, PIDTYPE_PGID);
|
| 237 | if (p == NULL)
|
| 238 | p = pid_task(pgrp, PIDTYPE_PID);
|
| 239 | if (p != NULL)
|
| 240 | sid = task_session(p);
|
| 241 |
|
| 242 | return sid;
|
| 243 | }
|
| 244 |
|
| 245 | |
| 246 | * Determine if a process group is "orphaned", according to the POSIX
|
| 247 | * definition in 2.2.2.52. Orphaned process groups are not to be affected
|
| 248 | * by terminal-generated stop signals. Newly orphaned process groups are
|
| 249 | * to receive a SIGHUP and a SIGCONT.
|
| 250 | *
|
| 251 | * "I ask you, have you ever known what it is to be an orphan?"
|
| 252 |
|
| 253 | static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
|
| 254 | {
|
| 255 | struct task_struct *p;
|
| 256 |
|
| 257 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
|
| 258 | if ((p == ignored_task) ||
|
| 259 | (p->exit_state && thread_group_empty(p)) ||
|
| 260 | is_global_init(p->real_parent))
|
| 261 | continue;
|
| 262 |
|
| 263 | if (task_pgrp(p->real_parent) != pgrp &&
|
| 264 | task_session(p->real_parent) == task_session(p))
|
| 265 | return 0;
|
| 266 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
|
| 267 |
|
| 268 | return 1;
|
| 269 | }
|
| 270 |
|
| 271 | int is_current_pgrp_orphaned(void)
|
| 272 | {
|
| 273 | int retval;
|
| 274 |
|
| 275 | read_lock(&tasklist_lock);
|
| 276 | retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
|
| 277 | read_unlock(&tasklist_lock);
|
| 278 |
|
| 279 | return retval;
|
| 280 | }
|
| 281 |
|
| 282 | static bool has_stopped_jobs(struct pid *pgrp)
|
| 283 | {
|
| 284 | struct task_struct *p;
|
| 285 |
|
| 286 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
|
| 287 | if (p->signal->flags & SIGNAL_STOP_STOPPED)
|
| 288 | return true;
|
| 289 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
|
| 290 |
|
| 291 | return false;
|
| 292 | }
|
| 293 |
|
| 294 | |
| 295 | * Check to see if any process groups have become orphaned as
|
| 296 | * a result of our exiting, and if they have any stopped jobs,
|
| 297 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
|
| 298 |
|
| 299 | static void
|
| 300 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
|
| 301 | {
|
| 302 | struct pid *pgrp = task_pgrp(tsk);
|
| 303 | struct task_struct *ignored_task = tsk;
|
| 304 |
|
| 305 | if (!parent)
|
| 306 | |
| 307 | * we are and we were the only connection outside.
|
| 308 |
|
| 309 | parent = tsk->real_parent;
|
| 310 | else
|
| 311 | |
| 312 | * we are, and it was the only connection outside.
|
| 313 |
|
| 314 | ignored_task = NULL;
|
| 315 |
|
| 316 | if (task_pgrp(parent) != pgrp &&
|
| 317 | task_session(parent) == task_session(tsk) &&
|
| 318 | will_become_orphaned_pgrp(pgrp, ignored_task) &&
|
| 319 | has_stopped_jobs(pgrp)) {
|
| 320 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
|
| 321 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
|
| 322 | }
|
| 323 | }
|
| 324 |
|
| 325 | |
| 326 | * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
|
| 327 | *
|
| 328 | * If a kernel thread is launched as a result of a system call, or if
|
| 329 | * it ever exits, it should generally reparent itself to kthreadd so it
|
| 330 | * isn't in the way of other processes and is correctly cleaned up on exit.
|
| 331 | *
|
| 332 | * The various task state such as scheduling policy and priority may have
|
| 333 | * been inherited from a user process, so we reset them to sane values here.
|
| 334 | *
|
| 335 | * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
|
| 336 |
|
| 337 | static void reparent_to_kthreadd(void)
|
| 338 | {
|
| 339 | write_lock_irq(&tasklist_lock);
|
| 340 |
|
| 341 | ptrace_unlink(current);
|
| 342 |
|
| 343 | current->real_parent = current->parent = kthreadd_task;
|
| 344 | list_move_tail(¤t->sibling, ¤t->real_parent->children);
|
| 345 |
|
| 346 |
|
| 347 | current->exit_signal = SIGCHLD;
|
| 348 |
|
| 349 | if (task_nice(current) < 0)
|
| 350 | set_user_nice(current, 0);
|
| 351 |
|
| 352 |
|
| 353 |
|
| 354 | memcpy(current->signal->rlim, init_task.signal->rlim,
|
| 355 | sizeof(current->signal->rlim));
|
| 356 |
|
| 357 | atomic_inc(&init_cred.usage);
|
| 358 | commit_creds(&init_cred);
|
| 359 | write_unlock_irq(&tasklist_lock);
|
| 360 | }
|
| 361 |
|
| 362 | void __set_special_pids(struct pid *pid)
|
| 363 | {
|
| 364 | struct task_struct *curr = current->group_leader;
|
| 365 |
|
| 366 | if (task_session(curr) != pid)
|
| 367 | change_pid(curr, PIDTYPE_SID, pid);
|
| 368 |
|
| 369 | if (task_pgrp(curr) != pid)
|
| 370 | change_pid(curr, PIDTYPE_PGID, pid);
|
| 371 | }
|
| 372 |
|
| 373 | static void set_special_pids(struct pid *pid)
|
| 374 | {
|
| 375 | write_lock_irq(&tasklist_lock);
|
| 376 | __set_special_pids(pid);
|
| 377 | write_unlock_irq(&tasklist_lock);
|
| 378 | }
|
| 379 |
|
| 380 | |
| 381 | * Let kernel threads use this to say that they allow a certain signal.
|
| 382 | * Must not be used if kthread was cloned with CLONE_SIGHAND.
|
| 383 |
|
| 384 | int allow_signal(int sig)
|
| 385 | {
|
| 386 | if (!valid_signal(sig) || sig < 1)
|
| 387 | return -EINVAL;
|
| 388 |
|
| 389 | spin_lock_irq(¤t->sighand->siglock);
|
| 390 |
|
| 391 | sigdelset(¤t->blocked, sig);
|
| 392 | |
| 393 | * Kernel threads handle their own signals. Let the signal code
|
| 394 | * know it'll be handled, so that they don't get converted to
|
| 395 | * SIGKILL or just silently dropped.
|
| 396 |
|
| 397 | current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
|
| 398 | recalc_sigpending();
|
| 399 | spin_unlock_irq(¤t->sighand->siglock);
|
| 400 | return 0;
|
| 401 | }
|
| 402 |
|
| 403 | EXPORT_SYMBOL(allow_signal);
|
| 404 |
|
| 405 | int disallow_signal(int sig)
|
| 406 | {
|
| 407 | if (!valid_signal(sig) || sig < 1)
|
| 408 | return -EINVAL;
|
| 409 |
|
| 410 | spin_lock_irq(¤t->sighand->siglock);
|
| 411 | current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
|
| 412 | recalc_sigpending();
|
| 413 | spin_unlock_irq(¤t->sighand->siglock);
|
| 414 | return 0;
|
| 415 | }
|
| 416 |
|
| 417 | EXPORT_SYMBOL(disallow_signal);
|
| 418 |
|
| 419 | |
| 420 | * Put all the gunge required to become a kernel thread without
|
| 421 | * attached user resources in one place where it belongs.
|
| 422 |
|
| 423 |
|
| 424 | void daemonize(const char *name, ...)
|
| 425 | {
|
| 426 | va_list args;
|
| 427 | sigset_t blocked;
|
| 428 |
|
| 429 | va_start(args, name);
|
| 430 | vsnprintf(current->comm, sizeof(current->comm), name, args);
|
| 431 | va_end(args);
|
| 432 |
|
| 433 | |
| 434 | * If we were started as result of loading a module, close all of the
|
| 435 | * user space pages. We don't need them, and if we didn't close them
|
| 436 | * they would be locked into memory.
|
| 437 |
|
| 438 | exit_mm(current);
|
| 439 | |
| 440 | * We don't want to get frozen, in case system-wide hibernation
|
| 441 | * or suspend transition begins right now.
|
| 442 |
|
| 443 | current->flags |= (PF_NOFREEZE | PF_KTHREAD);
|
| 444 |
|
| 445 | if (current->nsproxy != &init_nsproxy) {
|
| 446 | get_nsproxy(&init_nsproxy);
|
| 447 | switch_task_namespaces(current, &init_nsproxy);
|
| 448 | }
|
| 449 | set_special_pids(&init_struct_pid);
|
| 450 | proc_clear_tty(current);
|
| 451 |
|
| 452 |
|
| 453 | sigfillset(&blocked);
|
| 454 | sigprocmask(SIG_BLOCK, &blocked, NULL);
|
| 455 | flush_signals(current);
|
| 456 |
|
| 457 |
|
| 458 |
|
| 459 | daemonize_fs_struct();
|
| 460 | exit_files(current);
|
| 461 | current->files = init_task.files;
|
| 462 | atomic_inc(¤t->files->count);
|
| 463 |
|
| 464 | reparent_to_kthreadd();
|
| 465 | }
|
| 466 |
|
| 467 | EXPORT_SYMBOL(daemonize);
|
| 468 |
|
| 469 | static void close_files(struct files_struct * files)
|
| 470 | {
|
| 471 | int i, j;
|
| 472 | struct fdtable *fdt;
|
| 473 |
|
| 474 | j = 0;
|
| 475 |
|
| 476 | |
| 477 | * It is safe to dereference the fd table without RCU or
|
| 478 | * ->file_lock because this is the last reference to the
|
| 479 | * files structure. But use RCU to shut RCU-lockdep up.
|
| 480 |
|
| 481 | rcu_read_lock();
|
| 482 | fdt = files_fdtable(files);
|
| 483 | rcu_read_unlock();
|
| 484 | for (;;) {
|
| 485 | unsigned long set;
|
| 486 | i = j * BITS_PER_LONG;
|
| 487 | if (i >= fdt->max_fds)
|
| 488 | break;
|
| 489 | set = fdt->open_fds[j++];
|
| 490 | while (set) {
|
| 491 | if (set & 1) {
|
| 492 | struct file * file = xchg(&fdt->fd[i], NULL);
|
| 493 | if (file) {
|
| 494 | filp_close(file, files);
|
| 495 | cond_resched();
|
| 496 | }
|
| 497 | }
|
| 498 | i++;
|
| 499 | set >>= 1;
|
| 500 | }
|
| 501 | }
|
| 502 | }
|
| 503 |
|
| 504 | struct files_struct *get_files_struct(struct task_struct *task)
|
| 505 | {
|
| 506 | struct files_struct *files;
|
| 507 |
|
| 508 | task_lock(task);
|
| 509 | files = task->files;
|
| 510 | if (files)
|
| 511 | atomic_inc(&files->count);
|
| 512 | task_unlock(task);
|
| 513 |
|
| 514 | return files;
|
| 515 | }
|
| 516 |
|
| 517 | void put_files_struct(struct files_struct *files)
|
| 518 | {
|
| 519 | struct fdtable *fdt;
|
| 520 |
|
| 521 | if (atomic_dec_and_test(&files->count)) {
|
| 522 | close_files(files);
|
| 523 | |
| 524 | * Free the fd and fdset arrays if we expanded them.
|
| 525 | * If the fdtable was embedded, pass files for freeing
|
| 526 | * at the end of the RCU grace period. Otherwise,
|
| 527 | * you can free files immediately.
|
| 528 |
|
| 529 | rcu_read_lock();
|
| 530 | fdt = files_fdtable(files);
|
| 531 | if (fdt != &files->fdtab)
|
| 532 | kmem_cache_free(files_cachep, files);
|
| 533 | free_fdtable(fdt);
|
| 534 | rcu_read_unlock();
|
| 535 | }
|
| 536 | }
|
| 537 |
|
| 538 | void reset_files_struct(struct files_struct *files)
|
| 539 | {
|
| 540 | struct task_struct *tsk = current;
|
| 541 | struct files_struct *old;
|
| 542 |
|
| 543 | old = tsk->files;
|
| 544 | task_lock(tsk);
|
| 545 | tsk->files = files;
|
| 546 | task_unlock(tsk);
|
| 547 | put_files_struct(old);
|
| 548 | }
|
| 549 |
|
| 550 | void exit_files(struct task_struct *tsk)
|
| 551 | {
|
| 552 | struct files_struct * files = tsk->files;
|
| 553 |
|
| 554 | if (files) {
|
| 555 | task_lock(tsk);
|
| 556 | tsk->files = NULL;
|
| 557 | task_unlock(tsk);
|
| 558 | put_files_struct(files);
|
| 559 | }
|
| 560 | }
|
| 561 |
|
| 562 | #ifdef CONFIG_MM_OWNER
|
| 563 | |
| 564 | * A task is exiting. If it owned this mm, find a new owner for the mm.
|
| 565 |
|
| 566 | void mm_update_next_owner(struct mm_struct *mm)
|
| 567 | {
|
| 568 | struct task_struct *c, *g, *p = current;
|
| 569 |
|
| 570 | retry:
|
| 571 | |
| 572 | * If the exiting or execing task is not the owner, it's
|
| 573 | * someone else's problem.
|
| 574 |
|
| 575 | if (mm->owner != p)
|
| 576 | return;
|
| 577 | |
| 578 | * The current owner is exiting/execing and there are no other
|
| 579 | * candidates. Do not leave the mm pointing to a possibly
|
| 580 | * freed task structure.
|
| 581 |
|
| 582 | if (atomic_read(&mm->mm_users) <= 1) {
|
| 583 | mm->owner = NULL;
|
| 584 | return;
|
| 585 | }
|
| 586 |
|
| 587 | read_lock(&tasklist_lock);
|
| 588 | |
| 589 | * Search in the children
|
| 590 |
|
| 591 | list_for_each_entry(c, &p->children, sibling) {
|
| 592 | if (c->mm == mm)
|
| 593 | goto assign_new_owner;
|
| 594 | }
|
| 595 |
|
| 596 | |
| 597 | * Search in the siblings
|
| 598 |
|
| 599 | list_for_each_entry(c, &p->real_parent->children, sibling) {
|
| 600 | if (c->mm == mm)
|
| 601 | goto assign_new_owner;
|
| 602 | }
|
| 603 |
|
| 604 | |
| 605 | * Search through everything else. We should not get
|
| 606 | * here often
|
| 607 |
|
| 608 | do_each_thread(g, c) {
|
| 609 | if (c->mm == mm)
|
| 610 | goto assign_new_owner;
|
| 611 | } while_each_thread(g, c);
|
| 612 |
|
| 613 | read_unlock(&tasklist_lock);
|
| 614 | |
| 615 | * We found no owner yet mm_users > 1: this implies that we are
|
| 616 | * most likely racing with swapoff (try_to_unuse()) or /proc or
|
| 617 | * ptrace or page migration (get_task_mm()). Mark owner as NULL.
|
| 618 |
|
| 619 | mm->owner = NULL;
|
| 620 | return;
|
| 621 |
|
| 622 | assign_new_owner:
|
| 623 | BUG_ON(c == p);
|
| 624 | get_task_struct(c);
|
| 625 | |
| 626 | * The task_lock protects c->mm from changing.
|
| 627 | * We always want mm->owner->mm == mm
|
| 628 |
|
| 629 | task_lock(c);
|
| 630 | |
| 631 | * Delay read_unlock() till we have the task_lock()
|
| 632 | * to ensure that c does not slip away underneath us
|
| 633 |
|
| 634 | read_unlock(&tasklist_lock);
|
| 635 | if (c->mm != mm) {
|
| 636 | task_unlock(c);
|
| 637 | put_task_struct(c);
|
| 638 | goto retry;
|
| 639 | }
|
| 640 | mm->owner = c;
|
| 641 | task_unlock(c);
|
| 642 | put_task_struct(c);
|
| 643 | }
|
| 644 | #endif
|
| 645 |
|
| 646 | |
| 647 | * Turn us into a lazy TLB process if we
|
| 648 | * aren't already..
|
| 649 |
|
| 650 | static void exit_mm(struct task_struct * tsk)
|
| 651 | {
|
| 652 | struct mm_struct *mm = tsk->mm;
|
| 653 | struct core_state *core_state;
|
| 654 |
|
| 655 | mm_release(tsk, mm);
|
| 656 | if (!mm)
|
| 657 | return;
|
| 658 | sync_mm_rss(mm);
|
| 659 | |
| 660 | * Serialize with any possible pending coredump.
|
| 661 | * We must hold mmap_sem around checking core_state
|
| 662 | * and clearing tsk->mm. The core-inducing thread
|
| 663 | * will increment ->nr_threads for each thread in the
|
| 664 | * group with ->mm != NULL.
|
| 665 |
|
| 666 | down_read(&mm->mmap_sem);
|
| 667 | core_state = mm->core_state;
|
| 668 | if (core_state) {
|
| 669 | struct core_thread self;
|
| 670 | up_read(&mm->mmap_sem);
|
| 671 |
|
| 672 | self.task = tsk;
|
| 673 | self.next = xchg(&core_state->dumper.next, &self);
|
| 674 | |
| 675 | * Implies mb(), the result of xchg() must be visible
|
| 676 | * to core_state->dumper.
|
| 677 |
|
| 678 | if (atomic_dec_and_test(&core_state->nr_threads))
|
| 679 | complete(&core_state->startup);
|
| 680 |
|
| 681 | for (;;) {
|
| 682 | set_task_state(tsk, TASK_UNINTERRUPTIBLE);
|
| 683 | if (!self.task)
|
| 684 | break;
|
| 685 | schedule();
|
| 686 | }
|
| 687 | __set_task_state(tsk, TASK_RUNNING);
|
| 688 | down_read(&mm->mmap_sem);
|
| 689 | }
|
| 690 | atomic_inc(&mm->mm_count);
|
| 691 | BUG_ON(mm != tsk->active_mm);
|
| 692 |
|
| 693 | task_lock(tsk);
|
| 694 | tsk->mm = NULL;
|
| 695 | up_read(&mm->mmap_sem);
|
| 696 | enter_lazy_tlb(mm, current);
|
| 697 | task_unlock(tsk);
|
| 698 | mm_update_next_owner(mm);
|
| 699 | mmput(mm);
|
| 700 | }
|
| 701 |
|
| 702 | |
| 703 | * When we die, we re-parent all our children, and try to:
|
| 704 | * 1. give them to another thread in our thread group, if such a member exists
|
| 705 | * 2. give it to the first ancestor process which prctl'd itself as a
|
| 706 | * child_subreaper for its children (like a service manager)
|
| 707 | * 3. give it to the init process (PID 1) in our pid namespace
|
| 708 |
|
| 709 | static struct task_struct *find_new_reaper(struct task_struct *father)
|
| 710 | __releases(&tasklist_lock)
|
| 711 | __acquires(&tasklist_lock)
|
| 712 | {
|
| 713 | struct pid_namespace *pid_ns = task_active_pid_ns(father);
|
| 714 | struct task_struct *thread;
|
| 715 |
|
| 716 | thread = father;
|
| 717 | while_each_thread(father, thread) {
|
| 718 | if (thread->flags & PF_EXITING)
|
| 719 | continue;
|
| 720 | if (unlikely(pid_ns->child_reaper == father))
|
| 721 | pid_ns->child_reaper = thread;
|
| 722 | return thread;
|
| 723 | }
|
| 724 |
|
| 725 | if (unlikely(pid_ns->child_reaper == father)) {
|
| 726 | write_unlock_irq(&tasklist_lock);
|
| 727 | if (unlikely(pid_ns == &init_pid_ns)) {
|
| 728 | panic("Attempted to kill init! exitcode=0x%08x\n",
|
| 729 | father->signal->group_exit_code ?:
|
| 730 | father->exit_code);
|
| 731 | }
|
| 732 |
|
| 733 | zap_pid_ns_processes(pid_ns);
|
| 734 | write_lock_irq(&tasklist_lock);
|
| 735 | } else if (father->signal->has_child_subreaper) {
|
| 736 | struct task_struct *reaper;
|
| 737 |
|
| 738 | |
| 739 | * Find the first ancestor marked as child_subreaper.
|
| 740 | * Note that the code below checks same_thread_group(reaper,
|
| 741 | * pid_ns->child_reaper). This is what we need to DTRT in a
|
| 742 | * PID namespace. However we still need the check above, see
|
| 743 | * http://marc.info/?l=linux-kernel&m=131385460420380
|
| 744 |
|
| 745 | for (reaper = father->real_parent;
|
| 746 | reaper != &init_task;
|
| 747 | reaper = reaper->real_parent) {
|
| 748 | if (same_thread_group(reaper, pid_ns->child_reaper))
|
| 749 | break;
|
| 750 | if (!reaper->signal->is_child_subreaper)
|
| 751 | continue;
|
| 752 | thread = reaper;
|
| 753 | do {
|
| 754 | if (!(thread->flags & PF_EXITING))
|
| 755 | return reaper;
|
| 756 | } while_each_thread(reaper, thread);
|
| 757 | }
|
| 758 | }
|
| 759 |
|
| 760 | return pid_ns->child_reaper;
|
| 761 | }
|
| 762 |
|
| 763 | |
| 764 | * Any that need to be release_task'd are put on the @dead list.
|
| 765 |
|
| 766 | static void reparent_leader(struct task_struct *father, struct task_struct *p,
|
| 767 | struct list_head *dead)
|
| 768 | {
|
| 769 | list_move_tail(&p->sibling, &p->real_parent->children);
|
| 770 |
|
| 771 | if (p->exit_state == EXIT_DEAD)
|
| 772 | return;
|
| 773 | |
| 774 | * If this is a threaded reparent there is no need to
|
| 775 | * notify anyone anything has happened.
|
| 776 |
|
| 777 | if (same_thread_group(p->real_parent, father))
|
| 778 | return;
|
| 779 |
|
| 780 |
|
| 781 | p->exit_signal = SIGCHLD;
|
| 782 |
|
| 783 |
|
| 784 | if (!p->ptrace &&
|
| 785 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
|
| 786 | if (do_notify_parent(p, p->exit_signal)) {
|
| 787 | p->exit_state = EXIT_DEAD;
|
| 788 | list_move_tail(&p->sibling, dead);
|
| 789 | }
|
| 790 | }
|
| 791 |
|
| 792 | kill_orphaned_pgrp(p, father);
|
| 793 | }
|
| 794 |
|
| 795 | static void forget_original_parent(struct task_struct *father)
|
| 796 | {
|
| 797 | struct task_struct *p, *n, *reaper;
|
| 798 | LIST_HEAD(dead_children);
|
| 799 |
|
| 800 | write_lock_irq(&tasklist_lock);
|
| 801 | |
| 802 | * Note that exit_ptrace() and find_new_reaper() might
|
| 803 | * drop tasklist_lock and reacquire it.
|
| 804 |
|
| 805 | exit_ptrace(father);
|
| 806 | reaper = find_new_reaper(father);
|
| 807 |
|
| 808 | list_for_each_entry_safe(p, n, &father->children, sibling) {
|
| 809 | struct task_struct *t = p;
|
| 810 | do {
|
| 811 | t->real_parent = reaper;
|
| 812 | if (t->parent == father) {
|
| 813 | BUG_ON(t->ptrace);
|
| 814 | t->parent = t->real_parent;
|
| 815 | }
|
| 816 | if (t->pdeath_signal)
|
| 817 | group_send_sig_info(t->pdeath_signal,
|
| 818 | SEND_SIG_NOINFO, t);
|
| 819 | } while_each_thread(p, t);
|
| 820 | reparent_leader(father, p, &dead_children);
|
| 821 | }
|
| 822 | write_unlock_irq(&tasklist_lock);
|
| 823 |
|
| 824 | BUG_ON(!list_empty(&father->children));
|
| 825 |
|
| 826 | list_for_each_entry_safe(p, n, &dead_children, sibling) {
|
| 827 | list_del_init(&p->sibling);
|
| 828 | release_task(p);
|
| 829 | }
|
| 830 | }
|
| 831 |
|
| 832 | |
| 833 | * Send signals to all our closest relatives so that they know
|
| 834 | * to properly mourn us..
|
| 835 |
|
| 836 | static void exit_notify(struct task_struct *tsk, int group_dead)
|
| 837 | {
|
| 838 | bool autoreap;
|
| 839 |
|
| 840 | |
| 841 | * This does two things:
|
| 842 | *
|
| 843 | * A. Make init inherit all the child processes
|
| 844 | * B. Check to see if any process groups have become orphaned
|
| 845 | * as a result of our exiting, and if they have any stopped
|
| 846 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
|
| 847 |
|
| 848 | forget_original_parent(tsk);
|
| 849 | exit_task_namespaces(tsk);
|
| 850 |
|
| 851 | write_lock_irq(&tasklist_lock);
|
| 852 | if (group_dead)
|
| 853 | kill_orphaned_pgrp(tsk->group_leader, NULL);
|
| 854 |
|
| 855 | if (unlikely(tsk->ptrace)) {
|
| 856 | int sig = thread_group_leader(tsk) &&
|
| 857 | thread_group_empty(tsk) &&
|
| 858 | !ptrace_reparented(tsk) ?
|
| 859 | tsk->exit_signal : SIGCHLD;
|
| 860 | autoreap = do_notify_parent(tsk, sig);
|
| 861 | } else if (thread_group_leader(tsk)) {
|
| 862 | autoreap = thread_group_empty(tsk) &&
|
| 863 | do_notify_parent(tsk, tsk->exit_signal);
|
| 864 | } else {
|
| 865 | autoreap = true;
|
| 866 | }
|
| 867 |
|
| 868 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
|
| 869 |
|
| 870 |
|
| 871 | if (unlikely(tsk->signal->notify_count < 0))
|
| 872 | wake_up_process(tsk->signal->group_exit_task);
|
| 873 | write_unlock_irq(&tasklist_lock);
|
| 874 |
|
| 875 |
|
| 876 | if (autoreap)
|
| 877 | release_task(tsk);
|
| 878 | }
|
| 879 |
|
| 880 | #ifdef CONFIG_DEBUG_STACK_USAGE
|
| 881 | static void check_stack_usage(void)
|
| 882 | {
|
| 883 | static DEFINE_SPINLOCK(low_water_lock);
|
| 884 | static int lowest_to_date = THREAD_SIZE;
|
| 885 | unsigned long free;
|
| 886 |
|
| 887 | free = stack_not_used(current);
|
| 888 |
|
| 889 | if (free >= lowest_to_date)
|
| 890 | return;
|
| 891 |
|
| 892 | spin_lock(&low_water_lock);
|
| 893 | if (free < lowest_to_date) {
|
| 894 | printk(KERN_WARNING "%s (%d) used greatest stack depth: "
|
| 895 | "%lu bytes left\n",
|
| 896 | current->comm, task_pid_nr(current), free);
|
| 897 | lowest_to_date = free;
|
| 898 | }
|
| 899 | spin_unlock(&low_water_lock);
|
| 900 | }
|
| 901 | #else
|
| 902 | static inline void check_stack_usage(void) {}
|
| 903 | #endif
|
| 904 |
|
| 905 | void do_exit(long code)
|
| 906 | {
|
| 907 | struct task_struct *tsk = current;
|
| 908 | int group_dead;
|
| 909 |
|
| 910 | profile_task_exit(tsk);
|
| 911 |
|
| 912 | WARN_ON(blk_needs_flush_plug(tsk));
|
| 913 |
|
| 914 | if (unlikely(in_interrupt()))
|
| 915 | panic("Aiee, killing interrupt handler!");
|
| 916 | if (unlikely(!tsk->pid))
|
| 917 | panic("Attempted to kill the idle task!");
|
| 918 |
|
| 919 | |
| 920 | * If do_exit is called because this processes oopsed, it's possible
|
| 921 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
|
| 922 | * continuing. Amongst other possible reasons, this is to prevent
|
| 923 | * mm_release()->clear_child_tid() from writing to a user-controlled
|
| 924 | * kernel address.
|
| 925 |
|
| 926 | set_fs(USER_DS);
|
| 927 |
|
| 928 | ptrace_event(PTRACE_EVENT_EXIT, code);
|
| 929 |
|
| 930 | validate_creds_for_do_exit(tsk);
|
| 931 |
|
| 932 | |
| 933 | * We're taking recursive faults here in do_exit. Safest is to just
|
| 934 | * leave this task alone and wait for reboot.
|
| 935 |
|
| 936 | if (unlikely(tsk->flags & PF_EXITING)) {
|
| 937 | printk(KERN_ALERT
|
| 938 | "Fixing recursive fault but reboot is needed!\n");
|
| 939 | |
| 940 | * We can do this unlocked here. The futex code uses
|
| 941 | * this flag just to verify whether the pi state
|
| 942 | * cleanup has been done or not. In the worst case it
|
| 943 | * loops once more. We pretend that the cleanup was
|
| 944 | * done as there is no way to return. Either the
|
| 945 | * OWNER_DIED bit is set by now or we push the blocked
|
| 946 | * task into the wait for ever nirwana as well.
|
| 947 |
|
| 948 | tsk->flags |= PF_EXITPIDONE;
|
| 949 | set_current_state(TASK_UNINTERRUPTIBLE);
|
| 950 | schedule();
|
| 951 | }
|
| 952 |
|
| 953 | exit_signals(tsk);
|
| 954 | |
| 955 | * tsk->flags are checked in the futex code to protect against
|
| 956 | * an exiting task cleaning up the robust pi futexes.
|
| 957 |
|
| 958 | smp_mb();
|
| 959 | raw_spin_unlock_wait(&tsk->pi_lock);
|
| 960 |
|
| 961 | if (unlikely(in_atomic()))
|
| 962 | printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
|
| 963 | current->comm, task_pid_nr(current),
|
| 964 | preempt_count());
|
| 965 |
|
| 966 | acct_update_integrals(tsk);
|
| 967 |
|
| 968 | if (tsk->mm)
|
| 969 | sync_mm_rss(tsk->mm);
|
| 970 | group_dead = atomic_dec_and_test(&tsk->signal->live);
|
| 971 | if (group_dead) {
|
| 972 | hrtimer_cancel(&tsk->signal->real_timer);
|
| 973 | exit_itimers(tsk->signal);
|
| 974 | if (tsk->mm)
|
| 975 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
|
| 976 | }
|
| 977 | acct_collect(code, group_dead);
|
| 978 | if (group_dead)
|
| 979 | tty_audit_exit();
|
| 980 | audit_free(tsk);
|
| 981 |
|
| 982 | tsk->exit_code = code;
|
| 983 | taskstats_exit(tsk, group_dead);
|
| 984 |
|
| 985 | exit_mm(tsk);
|
| 986 |
|
| 987 | if (group_dead)
|
| 988 | acct_process();
|
| 989 | trace_sched_process_exit(tsk);
|
| 990 |
|
| 991 | exit_sem(tsk);
|
| 992 | exit_shm(tsk);
|
| 993 | exit_files(tsk);
|
| 994 | exit_fs(tsk);
|
| 995 | exit_task_work(tsk);
|
| 996 | check_stack_usage();
|
| 997 | exit_thread();
|
| 998 |
|
| 999 | |
| 1000 | * Flush inherited counters to the parent - before the parent
|
| 1001 | * gets woken up by child-exit notifications.
|
| 1002 | *
|
| 1003 | * because of cgroup mode, must be called before cgroup_exit()
|
| 1004 |
|
| 1005 | perf_event_exit_task(tsk);
|
| 1006 |
|
| 1007 | cgroup_exit(tsk, 1);
|
| 1008 |
|
| 1009 | if (group_dead)
|
| 1010 | disassociate_ctty(1);
|
| 1011 |
|
| 1012 | module_put(task_thread_info(tsk)->exec_domain->module);
|
| 1013 |
|
| 1014 | proc_exit_connector(tsk);
|
| 1015 |
|
| 1016 | |
| 1017 | * FIXME: do that only when needed, using sched_exit tracepoint
|
| 1018 |
|
| 1019 | ptrace_put_breakpoints(tsk);
|
| 1020 |
|
| 1021 | exit_notify(tsk, group_dead);
|
| 1022 | #ifdef CONFIG_NUMA
|
| 1023 | task_lock(tsk);
|
| 1024 | mpol_put(tsk->mempolicy);
|
| 1025 | tsk->mempolicy = NULL;
|
| 1026 | task_unlock(tsk);
|
| 1027 | #endif
|
| 1028 | #ifdef CONFIG_FUTEX
|
| 1029 | if (unlikely(current->pi_state_cache))
|
| 1030 | kfree(current->pi_state_cache);
|
| 1031 | #endif
|
| 1032 | |
| 1033 | * Make sure we are holding no locks:
|
| 1034 |
|
| 1035 | debug_check_no_locks_held(tsk);
|
| 1036 | |
| 1037 | * We can do this unlocked here. The futex code uses this flag
|
| 1038 | * just to verify whether the pi state cleanup has been done
|
| 1039 | * or not. In the worst case it loops once more.
|
| 1040 |
|
| 1041 | tsk->flags |= PF_EXITPIDONE;
|
| 1042 |
|
| 1043 | if (tsk->io_context)
|
| 1044 | exit_io_context(tsk);
|
| 1045 |
|
| 1046 | if (tsk->splice_pipe)
|
| 1047 | __free_pipe_info(tsk->splice_pipe);
|
| 1048 |
|
| 1049 | validate_creds_for_do_exit(tsk);
|
| 1050 |
|
| 1051 | preempt_disable();
|
| 1052 | if (tsk->nr_dirtied)
|
| 1053 | __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
|
| 1054 | exit_rcu();
|
| 1055 |
|
| 1056 | |
| 1057 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
|
| 1058 | * when the following two conditions become true.
|
| 1059 | * - There is race condition of mmap_sem (It is acquired by
|
| 1060 | * exit_mm()), and
|
| 1061 | * - SMI occurs before setting TASK_RUNINNG.
|
| 1062 | * (or hypervisor of virtual machine switches to other guest)
|
| 1063 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
|
| 1064 | *
|
| 1065 | * To avoid it, we have to wait for releasing tsk->pi_lock which
|
| 1066 | * is held by try_to_wake_up()
|
| 1067 |
|
| 1068 | smp_mb();
|
| 1069 | raw_spin_unlock_wait(&tsk->pi_lock);
|
| 1070 |
|
| 1071 |
|
| 1072 | tsk->state = TASK_DEAD;
|
| 1073 | tsk->flags |= PF_NOFREEZE;
|
| 1074 | schedule();
|
| 1075 | BUG();
|
| 1076 |
|
| 1077 | for (;;)
|
| 1078 | cpu_relax();
|
| 1079 | }
|
| 1080 |
|
| 1081 | EXPORT_SYMBOL_GPL(do_exit);
|
| 1082 |
|
| 1083 | void complete_and_exit(struct completion *comp, long code)
|
| 1084 | {
|
| 1085 | if (comp)
|
| 1086 | complete(comp);
|
| 1087 |
|
| 1088 | do_exit(code);
|
| 1089 | }
|
| 1090 |
|
| 1091 | EXPORT_SYMBOL(complete_and_exit);
|
| 1092 |
|
| 1093 | SYSCALL_DEFINE1(exit, int, error_code)
|
| 1094 | {
|
| 1095 | do_exit((error_code&0xff)<<8);
|
| 1096 | }
|
| 1097 |
|
| 1098 | |
| 1099 | * Take down every thread in the group. This is called by fatal signals
|
| 1100 | * as well as by sys_exit_group (below).
|
| 1101 |
|
| 1102 | void
|
| 1103 | do_group_exit(int exit_code)
|
| 1104 | {
|
| 1105 | struct signal_struct *sig = current->signal;
|
| 1106 |
|
| 1107 | BUG_ON(exit_code & 0x80);
|
| 1108 |
|
| 1109 | if (signal_group_exit(sig))
|
| 1110 | exit_code = sig->group_exit_code;
|
| 1111 | else if (!thread_group_empty(current)) {
|
| 1112 | struct sighand_struct *const sighand = current->sighand;
|
| 1113 | spin_lock_irq(&sighand->siglock);
|
| 1114 | if (signal_group_exit(sig))
|
| 1115 |
|
| 1116 | exit_code = sig->group_exit_code;
|
| 1117 | else {
|
| 1118 | sig->group_exit_code = exit_code;
|
| 1119 | sig->flags = SIGNAL_GROUP_EXIT;
|
| 1120 | zap_other_threads(current);
|
| 1121 | }
|
| 1122 | spin_unlock_irq(&sighand->siglock);
|
| 1123 | }
|
| 1124 |
|
| 1125 | do_exit(exit_code);
|
| 1126 |
|
| 1127 | }
|
| 1128 |
|
| 1129 | |
| 1130 | * this kills every thread in the thread group. Note that any externally
|
| 1131 | * wait4()-ing process will get the correct exit code - even if this
|
| 1132 | * thread is not the thread group leader.
|
| 1133 |
|
| 1134 | SYSCALL_DEFINE1(exit_group, int, error_code)
|
| 1135 | {
|
| 1136 | do_group_exit((error_code & 0xff) << 8);
|
| 1137 |
|
| 1138 | return 0;
|
| 1139 | }
|
| 1140 |
|
| 1141 | struct wait_opts {
|
| 1142 | enum pid_type wo_type;
|
| 1143 | int wo_flags;
|
| 1144 | struct pid *wo_pid;
|
| 1145 |
|
| 1146 | struct siginfo __user *wo_info;
|
| 1147 | int __user *wo_stat;
|
| 1148 | struct rusage __user *wo_rusage;
|
| 1149 |
|
| 1150 | wait_queue_t child_wait;
|
| 1151 | int notask_error;
|
| 1152 | };
|
| 1153 |
|
| 1154 | static inline
|
| 1155 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
|
| 1156 | {
|
| 1157 | if (type != PIDTYPE_PID)
|
| 1158 | task = task->group_leader;
|
| 1159 | return task->pids[type].pid;
|
| 1160 | }
|
| 1161 |
|
| 1162 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
|
| 1163 | {
|
| 1164 | return wo->wo_type == PIDTYPE_MAX ||
|
| 1165 | task_pid_type(p, wo->wo_type) == wo->wo_pid;
|
| 1166 | }
|
| 1167 |
|
| 1168 | static int eligible_child(struct wait_opts *wo, struct task_struct *p)
|
| 1169 | {
|
| 1170 | if (!eligible_pid(wo, p))
|
| 1171 | return 0;
|
| 1172 | |
| 1173 | * otherwise, wait for clone children *only* if __WCLONE is
|
| 1174 | * set; otherwise, wait for non-clone children *only*. (Note:
|
| 1175 | * A "clone" child here is one that reports to its parent
|
| 1176 | * using a signal other than SIGCHLD.)
|
| 1177 | if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
|
| 1178 | && !(wo->wo_flags & __WALL))
|
| 1179 | return 0;
|
| 1180 |
|
| 1181 | return 1;
|
| 1182 | }
|
| 1183 |
|
| 1184 | static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
|
| 1185 | pid_t pid, uid_t uid, int why, int status)
|
| 1186 | {
|
| 1187 | struct siginfo __user *infop;
|
| 1188 | int retval = wo->wo_rusage
|
| 1189 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1190 |
|
| 1191 | put_task_struct(p);
|
| 1192 | infop = wo->wo_info;
|
| 1193 | if (infop) {
|
| 1194 | if (!retval)
|
| 1195 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1196 | if (!retval)
|
| 1197 | retval = put_user(0, &infop->si_errno);
|
| 1198 | if (!retval)
|
| 1199 | retval = put_user((short)why, &infop->si_code);
|
| 1200 | if (!retval)
|
| 1201 | retval = put_user(pid, &infop->si_pid);
|
| 1202 | if (!retval)
|
| 1203 | retval = put_user(uid, &infop->si_uid);
|
| 1204 | if (!retval)
|
| 1205 | retval = put_user(status, &infop->si_status);
|
| 1206 | }
|
| 1207 | if (!retval)
|
| 1208 | retval = pid;
|
| 1209 | return retval;
|
| 1210 | }
|
| 1211 |
|
| 1212 | |
| 1213 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
|
| 1214 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
|
| 1215 | * the lock and this task is uninteresting. If we return nonzero, we have
|
| 1216 | * released the lock and the system call should return.
|
| 1217 |
|
| 1218 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
|
| 1219 | {
|
| 1220 | unsigned long state;
|
| 1221 | int retval, status, traced;
|
| 1222 | pid_t pid = task_pid_vnr(p);
|
| 1223 | uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
|
| 1224 | struct siginfo __user *infop;
|
| 1225 |
|
| 1226 | if (!likely(wo->wo_flags & WEXITED))
|
| 1227 | return 0;
|
| 1228 |
|
| 1229 | if (unlikely(wo->wo_flags & WNOWAIT)) {
|
| 1230 | int exit_code = p->exit_code;
|
| 1231 | int why;
|
| 1232 |
|
| 1233 | get_task_struct(p);
|
| 1234 | read_unlock(&tasklist_lock);
|
| 1235 | if ((exit_code & 0x7f) == 0) {
|
| 1236 | why = CLD_EXITED;
|
| 1237 | status = exit_code >> 8;
|
| 1238 | } else {
|
| 1239 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
|
| 1240 | status = exit_code & 0x7f;
|
| 1241 | }
|
| 1242 | return wait_noreap_copyout(wo, p, pid, uid, why, status);
|
| 1243 | }
|
| 1244 |
|
| 1245 | |
| 1246 | * Try to move the task's state to DEAD
|
| 1247 | * only one thread is allowed to do this:
|
| 1248 |
|
| 1249 | state = xchg(&p->exit_state, EXIT_DEAD);
|
| 1250 | if (state != EXIT_ZOMBIE) {
|
| 1251 | BUG_ON(state != EXIT_DEAD);
|
| 1252 | return 0;
|
| 1253 | }
|
| 1254 |
|
| 1255 | traced = ptrace_reparented(p);
|
| 1256 | |
| 1257 | * It can be ptraced but not reparented, check
|
| 1258 | * thread_group_leader() to filter out sub-threads.
|
| 1259 |
|
| 1260 | if (likely(!traced) && thread_group_leader(p)) {
|
| 1261 | struct signal_struct *psig;
|
| 1262 | struct signal_struct *sig;
|
| 1263 | unsigned long maxrss;
|
| 1264 | cputime_t tgutime, tgstime;
|
| 1265 |
|
| 1266 | |
| 1267 | * The resource counters for the group leader are in its
|
| 1268 | * own task_struct. Those for dead threads in the group
|
| 1269 | * are in its signal_struct, as are those for the child
|
| 1270 | * processes it has previously reaped. All these
|
| 1271 | * accumulate in the parent's signal_struct c* fields.
|
| 1272 | *
|
| 1273 | * We don't bother to take a lock here to protect these
|
| 1274 | * p->signal fields, because they are only touched by
|
| 1275 | * __exit_signal, which runs with tasklist_lock
|
| 1276 | * write-locked anyway, and so is excluded here. We do
|
| 1277 | * need to protect the access to parent->signal fields,
|
| 1278 | * as other threads in the parent group can be right
|
| 1279 | * here reaping other children at the same time.
|
| 1280 | *
|
| 1281 | * We use thread_group_times() to get times for the thread
|
| 1282 | * group, which consolidates times for all threads in the
|
| 1283 | * group including the group leader.
|
| 1284 |
|
| 1285 | thread_group_times(p, &tgutime, &tgstime);
|
| 1286 | spin_lock_irq(&p->real_parent->sighand->siglock);
|
| 1287 | psig = p->real_parent->signal;
|
| 1288 | sig = p->signal;
|
| 1289 | psig->cutime += tgutime + sig->cutime;
|
| 1290 | psig->cstime += tgstime + sig->cstime;
|
| 1291 | psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
|
| 1292 | psig->cmin_flt +=
|
| 1293 | p->min_flt + sig->min_flt + sig->cmin_flt;
|
| 1294 | psig->cmaj_flt +=
|
| 1295 | p->maj_flt + sig->maj_flt + sig->cmaj_flt;
|
| 1296 | psig->cnvcsw +=
|
| 1297 | p->nvcsw + sig->nvcsw + sig->cnvcsw;
|
| 1298 | psig->cnivcsw +=
|
| 1299 | p->nivcsw + sig->nivcsw + sig->cnivcsw;
|
| 1300 | psig->cinblock +=
|
| 1301 | task_io_get_inblock(p) +
|
| 1302 | sig->inblock + sig->cinblock;
|
| 1303 | psig->coublock +=
|
| 1304 | task_io_get_oublock(p) +
|
| 1305 | sig->oublock + sig->coublock;
|
| 1306 | maxrss = max(sig->maxrss, sig->cmaxrss);
|
| 1307 | if (psig->cmaxrss < maxrss)
|
| 1308 | psig->cmaxrss = maxrss;
|
| 1309 | task_io_accounting_add(&psig->ioac, &p->ioac);
|
| 1310 | task_io_accounting_add(&psig->ioac, &sig->ioac);
|
| 1311 | spin_unlock_irq(&p->real_parent->sighand->siglock);
|
| 1312 | }
|
| 1313 |
|
| 1314 | |
| 1315 | * Now we are sure this task is interesting, and no other
|
| 1316 | * thread can reap it because we set its state to EXIT_DEAD.
|
| 1317 |
|
| 1318 | read_unlock(&tasklist_lock);
|
| 1319 |
|
| 1320 | retval = wo->wo_rusage
|
| 1321 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1322 | status = (p->signal->flags & SIGNAL_GROUP_EXIT)
|
| 1323 | ? p->signal->group_exit_code : p->exit_code;
|
| 1324 | if (!retval && wo->wo_stat)
|
| 1325 | retval = put_user(status, wo->wo_stat);
|
| 1326 |
|
| 1327 | infop = wo->wo_info;
|
| 1328 | if (!retval && infop)
|
| 1329 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1330 | if (!retval && infop)
|
| 1331 | retval = put_user(0, &infop->si_errno);
|
| 1332 | if (!retval && infop) {
|
| 1333 | int why;
|
| 1334 |
|
| 1335 | if ((status & 0x7f) == 0) {
|
| 1336 | why = CLD_EXITED;
|
| 1337 | status >>= 8;
|
| 1338 | } else {
|
| 1339 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
|
| 1340 | status &= 0x7f;
|
| 1341 | }
|
| 1342 | retval = put_user((short)why, &infop->si_code);
|
| 1343 | if (!retval)
|
| 1344 | retval = put_user(status, &infop->si_status);
|
| 1345 | }
|
| 1346 | if (!retval && infop)
|
| 1347 | retval = put_user(pid, &infop->si_pid);
|
| 1348 | if (!retval && infop)
|
| 1349 | retval = put_user(uid, &infop->si_uid);
|
| 1350 | if (!retval)
|
| 1351 | retval = pid;
|
| 1352 |
|
| 1353 | if (traced) {
|
| 1354 | write_lock_irq(&tasklist_lock);
|
| 1355 |
|
| 1356 | ptrace_unlink(p);
|
| 1357 | |
| 1358 | * If this is not a sub-thread, notify the parent.
|
| 1359 | * If parent wants a zombie, don't release it now.
|
| 1360 |
|
| 1361 | if (thread_group_leader(p) &&
|
| 1362 | !do_notify_parent(p, p->exit_signal)) {
|
| 1363 | p->exit_state = EXIT_ZOMBIE;
|
| 1364 | p = NULL;
|
| 1365 | }
|
| 1366 | write_unlock_irq(&tasklist_lock);
|
| 1367 | }
|
| 1368 | if (p != NULL)
|
| 1369 | release_task(p);
|
| 1370 |
|
| 1371 | return retval;
|
| 1372 | }
|
| 1373 |
|
| 1374 | static int *task_stopped_code(struct task_struct *p, bool ptrace)
|
| 1375 | {
|
| 1376 | if (ptrace) {
|
| 1377 | if (task_is_stopped_or_traced(p) &&
|
| 1378 | !(p->jobctl & JOBCTL_LISTENING))
|
| 1379 | return &p->exit_code;
|
| 1380 | } else {
|
| 1381 | if (p->signal->flags & SIGNAL_STOP_STOPPED)
|
| 1382 | return &p->signal->group_exit_code;
|
| 1383 | }
|
| 1384 | return NULL;
|
| 1385 | }
|
| 1386 |
|
| 1387 | |
| 1388 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
|
| 1389 | * @wo: wait options
|
| 1390 | * @ptrace: is the wait for ptrace
|
| 1391 | * @p: task to wait for
|
| 1392 | *
|
| 1393 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
|
| 1394 | *
|
| 1395 | * CONTEXT:
|
| 1396 | * read_lock(&tasklist_lock), which is released if return value is
|
| 1397 | * non-zero. Also, grabs and releases @p->sighand->siglock.
|
| 1398 | *
|
| 1399 | * RETURNS:
|
| 1400 | * 0 if wait condition didn't exist and search for other wait conditions
|
| 1401 | * should continue. Non-zero return, -errno on failure and @p's pid on
|
| 1402 | * success, implies that tasklist_lock is released and wait condition
|
| 1403 | * search should terminate.
|
| 1404 |
|
| 1405 | static int wait_task_stopped(struct wait_opts *wo,
|
| 1406 | int ptrace, struct task_struct *p)
|
| 1407 | {
|
| 1408 | struct siginfo __user *infop;
|
| 1409 | int retval, exit_code, *p_code, why;
|
| 1410 | uid_t uid = 0;
|
| 1411 | pid_t pid;
|
| 1412 |
|
| 1413 | |
| 1414 | * Traditionally we see ptrace'd stopped tasks regardless of options.
|
| 1415 |
|
| 1416 | if (!ptrace && !(wo->wo_flags & WUNTRACED))
|
| 1417 | return 0;
|
| 1418 |
|
| 1419 | if (!task_stopped_code(p, ptrace))
|
| 1420 | return 0;
|
| 1421 |
|
| 1422 | exit_code = 0;
|
| 1423 | spin_lock_irq(&p->sighand->siglock);
|
| 1424 |
|
| 1425 | p_code = task_stopped_code(p, ptrace);
|
| 1426 | if (unlikely(!p_code))
|
| 1427 | goto unlock_sig;
|
| 1428 |
|
| 1429 | exit_code = *p_code;
|
| 1430 | if (!exit_code)
|
| 1431 | goto unlock_sig;
|
| 1432 |
|
| 1433 | if (!unlikely(wo->wo_flags & WNOWAIT))
|
| 1434 | *p_code = 0;
|
| 1435 |
|
| 1436 | uid = from_kuid_munged(current_user_ns(), task_uid(p));
|
| 1437 | unlock_sig:
|
| 1438 | spin_unlock_irq(&p->sighand->siglock);
|
| 1439 | if (!exit_code)
|
| 1440 | return 0;
|
| 1441 |
|
| 1442 | |
| 1443 | * Now we are pretty sure this task is interesting.
|
| 1444 | * Make sure it doesn't get reaped out from under us while we
|
| 1445 | * give up the lock and then examine it below. We don't want to
|
| 1446 | * keep holding onto the tasklist_lock while we call getrusage and
|
| 1447 | * possibly take page faults for user memory.
|
| 1448 |
|
| 1449 | get_task_struct(p);
|
| 1450 | pid = task_pid_vnr(p);
|
| 1451 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
|
| 1452 | read_unlock(&tasklist_lock);
|
| 1453 |
|
| 1454 | if (unlikely(wo->wo_flags & WNOWAIT))
|
| 1455 | return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
|
| 1456 |
|
| 1457 | retval = wo->wo_rusage
|
| 1458 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1459 | if (!retval && wo->wo_stat)
|
| 1460 | retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
|
| 1461 |
|
| 1462 | infop = wo->wo_info;
|
| 1463 | if (!retval && infop)
|
| 1464 | retval = put_user(SIGCHLD, &infop->si_signo);
|
| 1465 | if (!retval && infop)
|
| 1466 | retval = put_user(0, &infop->si_errno);
|
| 1467 | if (!retval && infop)
|
| 1468 | retval = put_user((short)why, &infop->si_code);
|
| 1469 | if (!retval && infop)
|
| 1470 | retval = put_user(exit_code, &infop->si_status);
|
| 1471 | if (!retval && infop)
|
| 1472 | retval = put_user(pid, &infop->si_pid);
|
| 1473 | if (!retval && infop)
|
| 1474 | retval = put_user(uid, &infop->si_uid);
|
| 1475 | if (!retval)
|
| 1476 | retval = pid;
|
| 1477 | put_task_struct(p);
|
| 1478 |
|
| 1479 | BUG_ON(!retval);
|
| 1480 | return retval;
|
| 1481 | }
|
| 1482 |
|
| 1483 | |
| 1484 | * Handle do_wait work for one task in a live, non-stopped state.
|
| 1485 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
|
| 1486 | * the lock and this task is uninteresting. If we return nonzero, we have
|
| 1487 | * released the lock and the system call should return.
|
| 1488 |
|
| 1489 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
|
| 1490 | {
|
| 1491 | int retval;
|
| 1492 | pid_t pid;
|
| 1493 | uid_t uid;
|
| 1494 |
|
| 1495 | if (!unlikely(wo->wo_flags & WCONTINUED))
|
| 1496 | return 0;
|
| 1497 |
|
| 1498 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
|
| 1499 | return 0;
|
| 1500 |
|
| 1501 | spin_lock_irq(&p->sighand->siglock);
|
| 1502 |
|
| 1503 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
|
| 1504 | spin_unlock_irq(&p->sighand->siglock);
|
| 1505 | return 0;
|
| 1506 | }
|
| 1507 | if (!unlikely(wo->wo_flags & WNOWAIT))
|
| 1508 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
|
| 1509 | uid = from_kuid_munged(current_user_ns(), task_uid(p));
|
| 1510 | spin_unlock_irq(&p->sighand->siglock);
|
| 1511 |
|
| 1512 | pid = task_pid_vnr(p);
|
| 1513 | get_task_struct(p);
|
| 1514 | read_unlock(&tasklist_lock);
|
| 1515 |
|
| 1516 | if (!wo->wo_info) {
|
| 1517 | retval = wo->wo_rusage
|
| 1518 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
|
| 1519 | put_task_struct(p);
|
| 1520 | if (!retval && wo->wo_stat)
|
| 1521 | retval = put_user(0xffff, wo->wo_stat);
|
| 1522 | if (!retval)
|
| 1523 | retval = pid;
|
| 1524 | } else {
|
| 1525 | retval = wait_noreap_copyout(wo, p, pid, uid,
|
| 1526 | CLD_CONTINUED, SIGCONT);
|
| 1527 | BUG_ON(retval == 0);
|
| 1528 | }
|
| 1529 |
|
| 1530 | return retval;
|
| 1531 | }
|
| 1532 |
|
| 1533 | |
| 1534 | * Consider @p for a wait by @parent.
|
| 1535 | *
|
| 1536 | * -ECHILD should be in ->notask_error before the first call.
|
| 1537 | * Returns nonzero for a final return, when we have unlocked tasklist_lock.
|
| 1538 | * Returns zero if the search for a child should continue;
|
| 1539 | * then ->notask_error is 0 if @p is an eligible child,
|
| 1540 | * or another error from security_task_wait(), or still -ECHILD.
|
| 1541 |
|
| 1542 | static int wait_consider_task(struct wait_opts *wo, int ptrace,
|
| 1543 | struct task_struct *p)
|
| 1544 | {
|
| 1545 | int ret = eligible_child(wo, p);
|
| 1546 | if (!ret)
|
| 1547 | return ret;
|
| 1548 |
|
| 1549 | ret = security_task_wait(p);
|
| 1550 | if (unlikely(ret < 0)) {
|
| 1551 | |
| 1552 | * If we have not yet seen any eligible child,
|
| 1553 | * then let this error code replace -ECHILD.
|
| 1554 | * A permission error will give the user a clue
|
| 1555 | * to look for security policy problems, rather
|
| 1556 | * than for mysterious wait bugs.
|
| 1557 |
|
| 1558 | if (wo->notask_error)
|
| 1559 | wo->notask_error = ret;
|
| 1560 | return 0;
|
| 1561 | }
|
| 1562 |
|
| 1563 |
|
| 1564 | if (unlikely(p->exit_state == EXIT_DEAD)) {
|
| 1565 | |
| 1566 | * But do not ignore this task until the tracer does
|
| 1567 | * wait_task_zombie()->do_notify_parent().
|
| 1568 |
|
| 1569 | if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
|
| 1570 | wo->notask_error = 0;
|
| 1571 | return 0;
|
| 1572 | }
|
| 1573 |
|
| 1574 |
|
| 1575 | if (p->exit_state == EXIT_ZOMBIE) {
|
| 1576 | |
| 1577 | * A zombie ptracee is only visible to its ptracer.
|
| 1578 | * Notification and reaping will be cascaded to the real
|
| 1579 | * parent when the ptracer detaches.
|
| 1580 |
|
| 1581 | if (likely(!ptrace) && unlikely(p->ptrace)) {
|
| 1582 |
|
| 1583 | wo->notask_error = 0;
|
| 1584 | return 0;
|
| 1585 | }
|
| 1586 |
|
| 1587 |
|
| 1588 | if (!delay_group_leader(p))
|
| 1589 | return wait_task_zombie(wo, p);
|
| 1590 |
|
| 1591 | |
| 1592 | * Allow access to stopped/continued state via zombie by
|
| 1593 | * falling through. Clearing of notask_error is complex.
|
| 1594 | *
|
| 1595 | * When !@ptrace:
|
| 1596 | *
|
| 1597 | * If WEXITED is set, notask_error should naturally be
|
| 1598 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
|
| 1599 | * so, if there are live subthreads, there are events to
|
| 1600 | * wait for. If all subthreads are dead, it's still safe
|
| 1601 | * to clear - this function will be called again in finite
|
| 1602 | * amount time once all the subthreads are released and
|
| 1603 | * will then return without clearing.
|
| 1604 | *
|
| 1605 | * When @ptrace:
|
| 1606 | *
|
| 1607 | * Stopped state is per-task and thus can't change once the
|
| 1608 | * target task dies. Only continued and exited can happen.
|
| 1609 | * Clear notask_error if WCONTINUED | WEXITED.
|
| 1610 |
|
| 1611 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
|
| 1612 | wo->notask_error = 0;
|
| 1613 | } else {
|
| 1614 | |
| 1615 | * If @p is ptraced by a task in its real parent's group,
|
| 1616 | * hide group stop/continued state when looking at @p as
|
| 1617 | * the real parent; otherwise, a single stop can be
|
| 1618 | * reported twice as group and ptrace stops.
|
| 1619 | *
|
| 1620 | * If a ptracer wants to distinguish the two events for its
|
| 1621 | * own children, it should create a separate process which
|
| 1622 | * takes the role of real parent.
|
| 1623 |
|
| 1624 | if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
|
| 1625 | return 0;
|
| 1626 |
|
| 1627 | |
| 1628 | * @p is alive and it's gonna stop, continue or exit, so
|
| 1629 | * there always is something to wait for.
|
| 1630 |
|
| 1631 | wo->notask_error = 0;
|
| 1632 | }
|
| 1633 |
|
| 1634 | |
| 1635 | * Wait for stopped. Depending on @ptrace, different stopped state
|
| 1636 | * is used and the two don't interact with each other.
|
| 1637 |
|
| 1638 | ret = wait_task_stopped(wo, ptrace, p);
|
| 1639 | if (ret)
|
| 1640 | return ret;
|
| 1641 |
|
| 1642 | |
| 1643 | * Wait for continued. There's only one continued state and the
|
| 1644 | * ptracer can consume it which can confuse the real parent. Don't
|
| 1645 | * use WCONTINUED from ptracer. You don't need or want it.
|
| 1646 |
|
| 1647 | return wait_task_continued(wo, p);
|
| 1648 | }
|
| 1649 |
|
| 1650 | |
| 1651 | * Do the work of do_wait() for one thread in the group, @tsk.
|
| 1652 | *
|
| 1653 | * -ECHILD should be in ->notask_error before the first call.
|
| 1654 | * Returns nonzero for a final return, when we have unlocked tasklist_lock.
|
| 1655 | * Returns zero if the search for a child should continue; then
|
| 1656 | * ->notask_error is 0 if there were any eligible children,
|
| 1657 | * or another error from security_task_wait(), or still -ECHILD.
|
| 1658 |
|
| 1659 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
|
| 1660 | {
|
| 1661 | struct task_struct *p;
|
| 1662 |
|
| 1663 | list_for_each_entry(p, &tsk->children, sibling) {
|
| 1664 | int ret = wait_consider_task(wo, 0, p);
|
| 1665 | if (ret)
|
| 1666 | return ret;
|
| 1667 | }
|
| 1668 |
|
| 1669 | return 0;
|
| 1670 | }
|
| 1671 |
|
| 1672 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
|
| 1673 | {
|
| 1674 | struct task_struct *p;
|
| 1675 |
|
| 1676 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
|
| 1677 | int ret = wait_consider_task(wo, 1, p);
|
| 1678 | if (ret)
|
| 1679 | return ret;
|
| 1680 | }
|
| 1681 |
|
| 1682 | return 0;
|
| 1683 | }
|
| 1684 |
|
| 1685 | static int child_wait_callback(wait_queue_t *wait, unsigned mode,
|
| 1686 | int sync, void *key)
|
| 1687 | {
|
| 1688 | struct wait_opts *wo = container_of(wait, struct wait_opts,
|
| 1689 | child_wait);
|
| 1690 | struct task_struct *p = key;
|
| 1691 |
|
| 1692 | if (!eligible_pid(wo, p))
|
| 1693 | return 0;
|
| 1694 |
|
| 1695 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
|
| 1696 | return 0;
|
| 1697 |
|
| 1698 | return default_wake_function(wait, mode, sync, key);
|
| 1699 | }
|
| 1700 |
|
| 1701 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
|
| 1702 | {
|
| 1703 | __wake_up_sync_key(&parent->signal->wait_chldexit,
|
| 1704 | TASK_INTERRUPTIBLE, 1, p);
|
| 1705 | }
|
| 1706 |
|
| 1707 | static long do_wait(struct wait_opts *wo)
|
| 1708 | {
|
| 1709 | struct task_struct *tsk;
|
| 1710 | int retval;
|
| 1711 |
|
| 1712 | trace_sched_process_wait(wo->wo_pid);
|
| 1713 |
|
| 1714 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
|
| 1715 | wo->child_wait.private = current;
|
| 1716 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
|
| 1717 | repeat:
|
| 1718 | |
| 1719 | * If there is nothing that can match our critiera just get out.
|
| 1720 | * We will clear ->notask_error to zero if we see any child that
|
| 1721 | * might later match our criteria, even if we are not able to reap
|
| 1722 | * it yet.
|
| 1723 |
|
| 1724 | wo->notask_error = -ECHILD;
|
| 1725 | if ((wo->wo_type < PIDTYPE_MAX) &&
|
| 1726 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
|
| 1727 | goto notask;
|
| 1728 |
|
| 1729 | set_current_state(TASK_INTERRUPTIBLE);
|
| 1730 | read_lock(&tasklist_lock);
|
| 1731 | tsk = current;
|
| 1732 | do {
|
| 1733 | retval = do_wait_thread(wo, tsk);
|
| 1734 | if (retval)
|
| 1735 | goto end;
|
| 1736 |
|
| 1737 | retval = ptrace_do_wait(wo, tsk);
|
| 1738 | if (retval)
|
| 1739 | goto end;
|
| 1740 |
|
| 1741 | if (wo->wo_flags & __WNOTHREAD)
|
| 1742 | break;
|
| 1743 | } while_each_thread(current, tsk);
|
| 1744 | read_unlock(&tasklist_lock);
|
| 1745 |
|
| 1746 | notask:
|
| 1747 | retval = wo->notask_error;
|
| 1748 | if (!retval && !(wo->wo_flags & WNOHANG)) {
|
| 1749 | retval = -ERESTARTSYS;
|
| 1750 | if (!signal_pending(current)) {
|
| 1751 | schedule();
|
| 1752 | goto repeat;
|
| 1753 | }
|
| 1754 | }
|
| 1755 | end:
|
| 1756 | __set_current_state(TASK_RUNNING);
|
| 1757 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
|
| 1758 | return retval;
|
| 1759 | }
|
| 1760 |
|
| 1761 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
|
| 1762 | infop, int, options, struct rusage __user *, ru)
|
| 1763 | {
|
| 1764 | struct wait_opts wo;
|
| 1765 | struct pid *pid = NULL;
|
| 1766 | enum pid_type type;
|
| 1767 | long ret;
|
| 1768 |
|
| 1769 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
|
| 1770 | return -EINVAL;
|
| 1771 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
|
| 1772 | return -EINVAL;
|
| 1773 |
|
| 1774 | switch (which) {
|
| 1775 | case P_ALL:
|
| 1776 | type = PIDTYPE_MAX;
|
| 1777 | break;
|
| 1778 | case P_PID:
|
| 1779 | type = PIDTYPE_PID;
|
| 1780 | if (upid <= 0)
|
| 1781 | return -EINVAL;
|
| 1782 | break;
|
| 1783 | case P_PGID:
|
| 1784 | type = PIDTYPE_PGID;
|
| 1785 | if (upid <= 0)
|
| 1786 | return -EINVAL;
|
| 1787 | break;
|
| 1788 | default:
|
| 1789 | return -EINVAL;
|
| 1790 | }
|
| 1791 |
|
| 1792 | if (type < PIDTYPE_MAX)
|
| 1793 | pid = find_get_pid(upid);
|
| 1794 |
|
| 1795 | wo.wo_type = type;
|
| 1796 | wo.wo_pid = pid;
|
| 1797 | wo.wo_flags = options;
|
| 1798 | wo.wo_info = infop;
|
| 1799 | wo.wo_stat = NULL;
|
| 1800 | wo.wo_rusage = ru;
|
| 1801 | ret = do_wait(&wo);
|
| 1802 |
|
| 1803 | if (ret > 0) {
|
| 1804 | ret = 0;
|
| 1805 | } else if (infop) {
|
| 1806 | |
| 1807 | * For a WNOHANG return, clear out all the fields
|
| 1808 | * we would set so the user can easily tell the
|
| 1809 | * difference.
|
| 1810 |
|
| 1811 | if (!ret)
|
| 1812 | ret = put_user(0, &infop->si_signo);
|
| 1813 | if (!ret)
|
| 1814 | ret = put_user(0, &infop->si_errno);
|
| 1815 | if (!ret)
|
| 1816 | ret = put_user(0, &infop->si_code);
|
| 1817 | if (!ret)
|
| 1818 | ret = put_user(0, &infop->si_pid);
|
| 1819 | if (!ret)
|
| 1820 | ret = put_user(0, &infop->si_uid);
|
| 1821 | if (!ret)
|
| 1822 | ret = put_user(0, &infop->si_status);
|
| 1823 | }
|
| 1824 |
|
| 1825 | put_pid(pid);
|
| 1826 |
|
| 1827 |
|
| 1828 | asmlinkage_protect(5, ret, which, upid, infop, options, ru);
|
| 1829 | return ret;
|
| 1830 | }
|
| 1831 |
|
| 1832 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
|
| 1833 | int, options, struct rusage __user *, ru)
|
| 1834 | {
|
| 1835 | struct wait_opts wo;
|
| 1836 | struct pid *pid = NULL;
|
| 1837 | enum pid_type type;
|
| 1838 | long ret;
|
| 1839 |
|
| 1840 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
|
| 1841 | __WNOTHREAD|__WCLONE|__WALL))
|
| 1842 | return -EINVAL;
|
| 1843 |
|
| 1844 | if (upid == -1)
|
| 1845 | type = PIDTYPE_MAX;
|
| 1846 | else if (upid < 0) {
|
| 1847 | type = PIDTYPE_PGID;
|
| 1848 | pid = find_get_pid(-upid);
|
| 1849 | } else if (upid == 0) {
|
| 1850 | type = PIDTYPE_PGID;
|
| 1851 | pid = get_task_pid(current, PIDTYPE_PGID);
|
| 1852 | } else {
|
| 1853 | type = PIDTYPE_PID;
|
| 1854 | pid = find_get_pid(upid);
|
| 1855 | }
|
| 1856 |
|
| 1857 | wo.wo_type = type;
|
| 1858 | wo.wo_pid = pid;
|
| 1859 | wo.wo_flags = options | WEXITED;
|
| 1860 | wo.wo_info = NULL;
|
| 1861 | wo.wo_stat = stat_addr;
|
| 1862 | wo.wo_rusage = ru;
|
| 1863 | ret = do_wait(&wo);
|
| 1864 | put_pid(pid);
|
| 1865 |
|
| 1866 |
|
| 1867 | asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
|
| 1868 | return ret;
|
| 1869 | }
|
| 1870 |
|
| 1871 | #ifdef __ARCH_WANT_SYS_WAITPID
|
| 1872 |
|
| 1873 | |
| 1874 | * sys_waitpid() remains for compatibility. waitpid() should be
|
| 1875 | * implemented by calling sys_wait4() from libc.a.
|
| 1876 |
|
| 1877 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
|
| 1878 | {
|
| 1879 | return sys_wait4(pid, stat_addr, options, NULL);
|
| 1880 | }
|
| 1881 |
|
| 1882 | #endif
|