| 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> /* for audit_free() */ |
| 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 | |
| 55 | #include <asm/uaccess.h> |
| 56 | #include <asm/unistd.h> |
| 57 | #include <asm/pgtable.h> |
| 58 | #include <asm/mmu_context.h> |
| 59 | |
| 60 | static void exit_mm(struct task_struct * tsk); |
| 61 | |
| 62 | static void __unhash_process(struct task_struct *p, bool group_dead) |
| 63 | { |
| 64 | nr_threads--; |
| 65 | detach_pid(p, PIDTYPE_PID); |
| 66 | if (group_dead) { |
| 67 | detach_pid(p, PIDTYPE_PGID); |
| 68 | detach_pid(p, PIDTYPE_SID); |
| 69 | |
| 70 | list_del_rcu(&p->tasks); |
| 71 | list_del_init(&p->sibling); |
| 72 | __this_cpu_dec(process_counts); |
| 73 | } |
| 74 | list_del_rcu(&p->thread_group); |
| 75 | } |
| 76 | |
| 77 | /* |
| 78 | * This function expects the tasklist_lock write-locked. |
| 79 | */ |
| 80 | static void __exit_signal(struct task_struct *tsk) |
| 81 | { |
| 82 | struct signal_struct *sig = tsk->signal; |
| 83 | bool group_dead = thread_group_leader(tsk); |
| 84 | struct sighand_struct *sighand; |
| 85 | struct tty_struct *uninitialized_var(tty); |
| 86 | |
| 87 | sighand = rcu_dereference_check(tsk->sighand, |
| 88 | rcu_read_lock_held() || |
| 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 = cputime_add(sig->utime, tsk->utime); |
| 126 | sig->stime = cputime_add(sig->stime, tsk->stime); |
| 127 | sig->gtime = cputime_add(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 | /* don't need to get the RCU readlock here - the process is dead and |
| 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 int has_stopped_jobs(struct pid *pgrp) |
| 270 | { |
| 271 | int retval = 0; |
| 272 | struct task_struct *p; |
| 273 | |
| 274 | do_each_pid_task(pgrp, PIDTYPE_PGID, p) { |
| 275 | if (!task_is_stopped(p)) |
| 276 | continue; |
| 277 | retval = 1; |
| 278 | break; |
| 279 | } while_each_pid_task(pgrp, PIDTYPE_PGID, p); |
| 280 | return retval; |
| 281 | } |
| 282 | |
| 283 | /* |
| 284 | * Check to see if any process groups have become orphaned as |
| 285 | * a result of our exiting, and if they have any stopped jobs, |
| 286 | * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) |
| 287 | */ |
| 288 | static void |
| 289 | kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) |
| 290 | { |
| 291 | struct pid *pgrp = task_pgrp(tsk); |
| 292 | struct task_struct *ignored_task = tsk; |
| 293 | |
| 294 | if (!parent) |
| 295 | /* exit: our father is in a different pgrp than |
| 296 | * we are and we were the only connection outside. |
| 297 | */ |
| 298 | parent = tsk->real_parent; |
| 299 | else |
| 300 | /* reparent: our child is in a different pgrp than |
| 301 | * we are, and it was the only connection outside. |
| 302 | */ |
| 303 | ignored_task = NULL; |
| 304 | |
| 305 | if (task_pgrp(parent) != pgrp && |
| 306 | task_session(parent) == task_session(tsk) && |
| 307 | will_become_orphaned_pgrp(pgrp, ignored_task) && |
| 308 | has_stopped_jobs(pgrp)) { |
| 309 | __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); |
| 310 | __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); |
| 311 | } |
| 312 | } |
| 313 | |
| 314 | /** |
| 315 | * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd |
| 316 | * |
| 317 | * If a kernel thread is launched as a result of a system call, or if |
| 318 | * it ever exits, it should generally reparent itself to kthreadd so it |
| 319 | * isn't in the way of other processes and is correctly cleaned up on exit. |
| 320 | * |
| 321 | * The various task state such as scheduling policy and priority may have |
| 322 | * been inherited from a user process, so we reset them to sane values here. |
| 323 | * |
| 324 | * NOTE that reparent_to_kthreadd() gives the caller full capabilities. |
| 325 | */ |
| 326 | static void reparent_to_kthreadd(void) |
| 327 | { |
| 328 | write_lock_irq(&tasklist_lock); |
| 329 | |
| 330 | ptrace_unlink(current); |
| 331 | /* Reparent to init */ |
| 332 | current->real_parent = current->parent = kthreadd_task; |
| 333 | list_move_tail(¤t->sibling, ¤t->real_parent->children); |
| 334 | |
| 335 | /* Set the exit signal to SIGCHLD so we signal init on exit */ |
| 336 | current->exit_signal = SIGCHLD; |
| 337 | |
| 338 | if (task_nice(current) < 0) |
| 339 | set_user_nice(current, 0); |
| 340 | /* cpus_allowed? */ |
| 341 | /* rt_priority? */ |
| 342 | /* signals? */ |
| 343 | memcpy(current->signal->rlim, init_task.signal->rlim, |
| 344 | sizeof(current->signal->rlim)); |
| 345 | |
| 346 | atomic_inc(&init_cred.usage); |
| 347 | commit_creds(&init_cred); |
| 348 | write_unlock_irq(&tasklist_lock); |
| 349 | } |
| 350 | |
| 351 | void __set_special_pids(struct pid *pid) |
| 352 | { |
| 353 | struct task_struct *curr = current->group_leader; |
| 354 | |
| 355 | if (task_session(curr) != pid) |
| 356 | change_pid(curr, PIDTYPE_SID, pid); |
| 357 | |
| 358 | if (task_pgrp(curr) != pid) |
| 359 | change_pid(curr, PIDTYPE_PGID, pid); |
| 360 | } |
| 361 | |
| 362 | static void set_special_pids(struct pid *pid) |
| 363 | { |
| 364 | write_lock_irq(&tasklist_lock); |
| 365 | __set_special_pids(pid); |
| 366 | write_unlock_irq(&tasklist_lock); |
| 367 | } |
| 368 | |
| 369 | /* |
| 370 | * Let kernel threads use this to say that they allow a certain signal. |
| 371 | * Must not be used if kthread was cloned with CLONE_SIGHAND. |
| 372 | */ |
| 373 | int allow_signal(int sig) |
| 374 | { |
| 375 | if (!valid_signal(sig) || sig < 1) |
| 376 | return -EINVAL; |
| 377 | |
| 378 | spin_lock_irq(¤t->sighand->siglock); |
| 379 | /* This is only needed for daemonize()'ed kthreads */ |
| 380 | sigdelset(¤t->blocked, sig); |
| 381 | /* |
| 382 | * Kernel threads handle their own signals. Let the signal code |
| 383 | * know it'll be handled, so that they don't get converted to |
| 384 | * SIGKILL or just silently dropped. |
| 385 | */ |
| 386 | current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2; |
| 387 | recalc_sigpending(); |
| 388 | spin_unlock_irq(¤t->sighand->siglock); |
| 389 | return 0; |
| 390 | } |
| 391 | |
| 392 | EXPORT_SYMBOL(allow_signal); |
| 393 | |
| 394 | int disallow_signal(int sig) |
| 395 | { |
| 396 | if (!valid_signal(sig) || sig < 1) |
| 397 | return -EINVAL; |
| 398 | |
| 399 | spin_lock_irq(¤t->sighand->siglock); |
| 400 | current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN; |
| 401 | recalc_sigpending(); |
| 402 | spin_unlock_irq(¤t->sighand->siglock); |
| 403 | return 0; |
| 404 | } |
| 405 | |
| 406 | EXPORT_SYMBOL(disallow_signal); |
| 407 | |
| 408 | /* |
| 409 | * Put all the gunge required to become a kernel thread without |
| 410 | * attached user resources in one place where it belongs. |
| 411 | */ |
| 412 | |
| 413 | void daemonize(const char *name, ...) |
| 414 | { |
| 415 | va_list args; |
| 416 | sigset_t blocked; |
| 417 | |
| 418 | va_start(args, name); |
| 419 | vsnprintf(current->comm, sizeof(current->comm), name, args); |
| 420 | va_end(args); |
| 421 | |
| 422 | /* |
| 423 | * If we were started as result of loading a module, close all of the |
| 424 | * user space pages. We don't need them, and if we didn't close them |
| 425 | * they would be locked into memory. |
| 426 | */ |
| 427 | exit_mm(current); |
| 428 | /* |
| 429 | * We don't want to have TIF_FREEZE set if the system-wide hibernation |
| 430 | * or suspend transition begins right now. |
| 431 | */ |
| 432 | current->flags |= (PF_NOFREEZE | PF_KTHREAD); |
| 433 | |
| 434 | if (current->nsproxy != &init_nsproxy) { |
| 435 | get_nsproxy(&init_nsproxy); |
| 436 | switch_task_namespaces(current, &init_nsproxy); |
| 437 | } |
| 438 | set_special_pids(&init_struct_pid); |
| 439 | proc_clear_tty(current); |
| 440 | |
| 441 | /* Block and flush all signals */ |
| 442 | sigfillset(&blocked); |
| 443 | sigprocmask(SIG_BLOCK, &blocked, NULL); |
| 444 | flush_signals(current); |
| 445 | |
| 446 | /* Become as one with the init task */ |
| 447 | |
| 448 | daemonize_fs_struct(); |
| 449 | exit_files(current); |
| 450 | current->files = init_task.files; |
| 451 | atomic_inc(¤t->files->count); |
| 452 | |
| 453 | reparent_to_kthreadd(); |
| 454 | } |
| 455 | |
| 456 | EXPORT_SYMBOL(daemonize); |
| 457 | |
| 458 | static void close_files(struct files_struct * files) |
| 459 | { |
| 460 | int i, j; |
| 461 | struct fdtable *fdt; |
| 462 | |
| 463 | j = 0; |
| 464 | |
| 465 | /* |
| 466 | * It is safe to dereference the fd table without RCU or |
| 467 | * ->file_lock because this is the last reference to the |
| 468 | * files structure. But use RCU to shut RCU-lockdep up. |
| 469 | */ |
| 470 | rcu_read_lock(); |
| 471 | fdt = files_fdtable(files); |
| 472 | rcu_read_unlock(); |
| 473 | for (;;) { |
| 474 | unsigned long set; |
| 475 | i = j * __NFDBITS; |
| 476 | if (i >= fdt->max_fds) |
| 477 | break; |
| 478 | set = fdt->open_fds->fds_bits[j++]; |
| 479 | while (set) { |
| 480 | if (set & 1) { |
| 481 | struct file * file = xchg(&fdt->fd[i], NULL); |
| 482 | if (file) { |
| 483 | filp_close(file, files); |
| 484 | cond_resched(); |
| 485 | } |
| 486 | } |
| 487 | i++; |
| 488 | set >>= 1; |
| 489 | } |
| 490 | } |
| 491 | } |
| 492 | |
| 493 | struct files_struct *get_files_struct(struct task_struct *task) |
| 494 | { |
| 495 | struct files_struct *files; |
| 496 | |
| 497 | task_lock(task); |
| 498 | files = task->files; |
| 499 | if (files) |
| 500 | atomic_inc(&files->count); |
| 501 | task_unlock(task); |
| 502 | |
| 503 | return files; |
| 504 | } |
| 505 | |
| 506 | void put_files_struct(struct files_struct *files) |
| 507 | { |
| 508 | struct fdtable *fdt; |
| 509 | |
| 510 | if (atomic_dec_and_test(&files->count)) { |
| 511 | close_files(files); |
| 512 | /* |
| 513 | * Free the fd and fdset arrays if we expanded them. |
| 514 | * If the fdtable was embedded, pass files for freeing |
| 515 | * at the end of the RCU grace period. Otherwise, |
| 516 | * you can free files immediately. |
| 517 | */ |
| 518 | rcu_read_lock(); |
| 519 | fdt = files_fdtable(files); |
| 520 | if (fdt != &files->fdtab) |
| 521 | kmem_cache_free(files_cachep, files); |
| 522 | free_fdtable(fdt); |
| 523 | rcu_read_unlock(); |
| 524 | } |
| 525 | } |
| 526 | |
| 527 | void reset_files_struct(struct files_struct *files) |
| 528 | { |
| 529 | struct task_struct *tsk = current; |
| 530 | struct files_struct *old; |
| 531 | |
| 532 | old = tsk->files; |
| 533 | task_lock(tsk); |
| 534 | tsk->files = files; |
| 535 | task_unlock(tsk); |
| 536 | put_files_struct(old); |
| 537 | } |
| 538 | |
| 539 | void exit_files(struct task_struct *tsk) |
| 540 | { |
| 541 | struct files_struct * files = tsk->files; |
| 542 | |
| 543 | if (files) { |
| 544 | task_lock(tsk); |
| 545 | tsk->files = NULL; |
| 546 | task_unlock(tsk); |
| 547 | put_files_struct(files); |
| 548 | } |
| 549 | } |
| 550 | |
| 551 | #ifdef CONFIG_MM_OWNER |
| 552 | /* |
| 553 | * Task p is exiting and it owned mm, lets find a new owner for it |
| 554 | */ |
| 555 | static inline int |
| 556 | mm_need_new_owner(struct mm_struct *mm, struct task_struct *p) |
| 557 | { |
| 558 | /* |
| 559 | * If there are other users of the mm and the owner (us) is exiting |
| 560 | * we need to find a new owner to take on the responsibility. |
| 561 | */ |
| 562 | if (atomic_read(&mm->mm_users) <= 1) |
| 563 | return 0; |
| 564 | if (mm->owner != p) |
| 565 | return 0; |
| 566 | return 1; |
| 567 | } |
| 568 | |
| 569 | void mm_update_next_owner(struct mm_struct *mm) |
| 570 | { |
| 571 | struct task_struct *c, *g, *p = current; |
| 572 | |
| 573 | retry: |
| 574 | if (!mm_need_new_owner(mm, p)) |
| 575 | return; |
| 576 | |
| 577 | read_lock(&tasklist_lock); |
| 578 | /* |
| 579 | * Search in the children |
| 580 | */ |
| 581 | list_for_each_entry(c, &p->children, sibling) { |
| 582 | if (c->mm == mm) |
| 583 | goto assign_new_owner; |
| 584 | } |
| 585 | |
| 586 | /* |
| 587 | * Search in the siblings |
| 588 | */ |
| 589 | list_for_each_entry(c, &p->real_parent->children, sibling) { |
| 590 | if (c->mm == mm) |
| 591 | goto assign_new_owner; |
| 592 | } |
| 593 | |
| 594 | /* |
| 595 | * Search through everything else. We should not get |
| 596 | * here often |
| 597 | */ |
| 598 | do_each_thread(g, c) { |
| 599 | if (c->mm == mm) |
| 600 | goto assign_new_owner; |
| 601 | } while_each_thread(g, c); |
| 602 | |
| 603 | read_unlock(&tasklist_lock); |
| 604 | /* |
| 605 | * We found no owner yet mm_users > 1: this implies that we are |
| 606 | * most likely racing with swapoff (try_to_unuse()) or /proc or |
| 607 | * ptrace or page migration (get_task_mm()). Mark owner as NULL. |
| 608 | */ |
| 609 | mm->owner = NULL; |
| 610 | return; |
| 611 | |
| 612 | assign_new_owner: |
| 613 | BUG_ON(c == p); |
| 614 | get_task_struct(c); |
| 615 | /* |
| 616 | * The task_lock protects c->mm from changing. |
| 617 | * We always want mm->owner->mm == mm |
| 618 | */ |
| 619 | task_lock(c); |
| 620 | /* |
| 621 | * Delay read_unlock() till we have the task_lock() |
| 622 | * to ensure that c does not slip away underneath us |
| 623 | */ |
| 624 | read_unlock(&tasklist_lock); |
| 625 | if (c->mm != mm) { |
| 626 | task_unlock(c); |
| 627 | put_task_struct(c); |
| 628 | goto retry; |
| 629 | } |
| 630 | mm->owner = c; |
| 631 | task_unlock(c); |
| 632 | put_task_struct(c); |
| 633 | } |
| 634 | #endif /* CONFIG_MM_OWNER */ |
| 635 | |
| 636 | /* |
| 637 | * Turn us into a lazy TLB process if we |
| 638 | * aren't already.. |
| 639 | */ |
| 640 | static void exit_mm(struct task_struct * tsk) |
| 641 | { |
| 642 | struct mm_struct *mm = tsk->mm; |
| 643 | struct core_state *core_state; |
| 644 | |
| 645 | mm_release(tsk, mm); |
| 646 | if (!mm) |
| 647 | return; |
| 648 | /* |
| 649 | * Serialize with any possible pending coredump. |
| 650 | * We must hold mmap_sem around checking core_state |
| 651 | * and clearing tsk->mm. The core-inducing thread |
| 652 | * will increment ->nr_threads for each thread in the |
| 653 | * group with ->mm != NULL. |
| 654 | */ |
| 655 | down_read(&mm->mmap_sem); |
| 656 | core_state = mm->core_state; |
| 657 | if (core_state) { |
| 658 | struct core_thread self; |
| 659 | up_read(&mm->mmap_sem); |
| 660 | |
| 661 | self.task = tsk; |
| 662 | self.next = xchg(&core_state->dumper.next, &self); |
| 663 | /* |
| 664 | * Implies mb(), the result of xchg() must be visible |
| 665 | * to core_state->dumper. |
| 666 | */ |
| 667 | if (atomic_dec_and_test(&core_state->nr_threads)) |
| 668 | complete(&core_state->startup); |
| 669 | |
| 670 | for (;;) { |
| 671 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
| 672 | if (!self.task) /* see coredump_finish() */ |
| 673 | break; |
| 674 | schedule(); |
| 675 | } |
| 676 | __set_task_state(tsk, TASK_RUNNING); |
| 677 | down_read(&mm->mmap_sem); |
| 678 | } |
| 679 | atomic_inc(&mm->mm_count); |
| 680 | BUG_ON(mm != tsk->active_mm); |
| 681 | /* more a memory barrier than a real lock */ |
| 682 | task_lock(tsk); |
| 683 | tsk->mm = NULL; |
| 684 | up_read(&mm->mmap_sem); |
| 685 | enter_lazy_tlb(mm, current); |
| 686 | /* We don't want this task to be frozen prematurely */ |
| 687 | clear_freeze_flag(tsk); |
| 688 | if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) |
| 689 | atomic_dec(&mm->oom_disable_count); |
| 690 | task_unlock(tsk); |
| 691 | mm_update_next_owner(mm); |
| 692 | mmput(mm); |
| 693 | } |
| 694 | |
| 695 | /* |
| 696 | * When we die, we re-parent all our children. |
| 697 | * Try to give them to another thread in our thread |
| 698 | * group, and if no such member exists, give it to |
| 699 | * the child reaper process (ie "init") in our pid |
| 700 | * space. |
| 701 | */ |
| 702 | static struct task_struct *find_new_reaper(struct task_struct *father) |
| 703 | __releases(&tasklist_lock) |
| 704 | __acquires(&tasklist_lock) |
| 705 | { |
| 706 | struct pid_namespace *pid_ns = task_active_pid_ns(father); |
| 707 | struct task_struct *thread; |
| 708 | |
| 709 | thread = father; |
| 710 | while_each_thread(father, thread) { |
| 711 | if (thread->flags & PF_EXITING) |
| 712 | continue; |
| 713 | if (unlikely(pid_ns->child_reaper == father)) |
| 714 | pid_ns->child_reaper = thread; |
| 715 | return thread; |
| 716 | } |
| 717 | |
| 718 | if (unlikely(pid_ns->child_reaper == father)) { |
| 719 | write_unlock_irq(&tasklist_lock); |
| 720 | if (unlikely(pid_ns == &init_pid_ns)) |
| 721 | panic("Attempted to kill init!"); |
| 722 | |
| 723 | zap_pid_ns_processes(pid_ns); |
| 724 | write_lock_irq(&tasklist_lock); |
| 725 | /* |
| 726 | * We can not clear ->child_reaper or leave it alone. |
| 727 | * There may by stealth EXIT_DEAD tasks on ->children, |
| 728 | * forget_original_parent() must move them somewhere. |
| 729 | */ |
| 730 | pid_ns->child_reaper = init_pid_ns.child_reaper; |
| 731 | } |
| 732 | |
| 733 | return pid_ns->child_reaper; |
| 734 | } |
| 735 | |
| 736 | /* |
| 737 | * Any that need to be release_task'd are put on the @dead list. |
| 738 | */ |
| 739 | static void reparent_leader(struct task_struct *father, struct task_struct *p, |
| 740 | struct list_head *dead) |
| 741 | { |
| 742 | list_move_tail(&p->sibling, &p->real_parent->children); |
| 743 | |
| 744 | if (p->exit_state == EXIT_DEAD) |
| 745 | return; |
| 746 | /* |
| 747 | * If this is a threaded reparent there is no need to |
| 748 | * notify anyone anything has happened. |
| 749 | */ |
| 750 | if (same_thread_group(p->real_parent, father)) |
| 751 | return; |
| 752 | |
| 753 | /* We don't want people slaying init. */ |
| 754 | p->exit_signal = SIGCHLD; |
| 755 | |
| 756 | /* If it has exited notify the new parent about this child's death. */ |
| 757 | if (!p->ptrace && |
| 758 | p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { |
| 759 | if (do_notify_parent(p, p->exit_signal)) { |
| 760 | p->exit_state = EXIT_DEAD; |
| 761 | list_move_tail(&p->sibling, dead); |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | kill_orphaned_pgrp(p, father); |
| 766 | } |
| 767 | |
| 768 | static void forget_original_parent(struct task_struct *father) |
| 769 | { |
| 770 | struct task_struct *p, *n, *reaper; |
| 771 | LIST_HEAD(dead_children); |
| 772 | |
| 773 | write_lock_irq(&tasklist_lock); |
| 774 | /* |
| 775 | * Note that exit_ptrace() and find_new_reaper() might |
| 776 | * drop tasklist_lock and reacquire it. |
| 777 | */ |
| 778 | exit_ptrace(father); |
| 779 | reaper = find_new_reaper(father); |
| 780 | |
| 781 | list_for_each_entry_safe(p, n, &father->children, sibling) { |
| 782 | struct task_struct *t = p; |
| 783 | do { |
| 784 | t->real_parent = reaper; |
| 785 | if (t->parent == father) { |
| 786 | BUG_ON(t->ptrace); |
| 787 | t->parent = t->real_parent; |
| 788 | } |
| 789 | if (t->pdeath_signal) |
| 790 | group_send_sig_info(t->pdeath_signal, |
| 791 | SEND_SIG_NOINFO, t); |
| 792 | } while_each_thread(p, t); |
| 793 | reparent_leader(father, p, &dead_children); |
| 794 | } |
| 795 | write_unlock_irq(&tasklist_lock); |
| 796 | |
| 797 | BUG_ON(!list_empty(&father->children)); |
| 798 | |
| 799 | list_for_each_entry_safe(p, n, &dead_children, sibling) { |
| 800 | list_del_init(&p->sibling); |
| 801 | release_task(p); |
| 802 | } |
| 803 | } |
| 804 | |
| 805 | /* |
| 806 | * Send signals to all our closest relatives so that they know |
| 807 | * to properly mourn us.. |
| 808 | */ |
| 809 | static void exit_notify(struct task_struct *tsk, int group_dead) |
| 810 | { |
| 811 | bool autoreap; |
| 812 | |
| 813 | /* |
| 814 | * This does two things: |
| 815 | * |
| 816 | * A. Make init inherit all the child processes |
| 817 | * B. Check to see if any process groups have become orphaned |
| 818 | * as a result of our exiting, and if they have any stopped |
| 819 | * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) |
| 820 | */ |
| 821 | forget_original_parent(tsk); |
| 822 | exit_task_namespaces(tsk); |
| 823 | |
| 824 | write_lock_irq(&tasklist_lock); |
| 825 | if (group_dead) |
| 826 | kill_orphaned_pgrp(tsk->group_leader, NULL); |
| 827 | |
| 828 | /* Let father know we died |
| 829 | * |
| 830 | * Thread signals are configurable, but you aren't going to use |
| 831 | * that to send signals to arbitrary processes. |
| 832 | * That stops right now. |
| 833 | * |
| 834 | * If the parent exec id doesn't match the exec id we saved |
| 835 | * when we started then we know the parent has changed security |
| 836 | * domain. |
| 837 | * |
| 838 | * If our self_exec id doesn't match our parent_exec_id then |
| 839 | * we have changed execution domain as these two values started |
| 840 | * the same after a fork. |
| 841 | */ |
| 842 | if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD && |
| 843 | (tsk->parent_exec_id != tsk->real_parent->self_exec_id || |
| 844 | tsk->self_exec_id != tsk->parent_exec_id)) |
| 845 | tsk->exit_signal = SIGCHLD; |
| 846 | |
| 847 | if (unlikely(tsk->ptrace)) { |
| 848 | int sig = thread_group_leader(tsk) && |
| 849 | thread_group_empty(tsk) && |
| 850 | !ptrace_reparented(tsk) ? |
| 851 | tsk->exit_signal : SIGCHLD; |
| 852 | autoreap = do_notify_parent(tsk, sig); |
| 853 | } else if (thread_group_leader(tsk)) { |
| 854 | autoreap = thread_group_empty(tsk) && |
| 855 | do_notify_parent(tsk, tsk->exit_signal); |
| 856 | } else { |
| 857 | autoreap = true; |
| 858 | } |
| 859 | |
| 860 | tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; |
| 861 | |
| 862 | /* mt-exec, de_thread() is waiting for group leader */ |
| 863 | if (unlikely(tsk->signal->notify_count < 0)) |
| 864 | wake_up_process(tsk->signal->group_exit_task); |
| 865 | write_unlock_irq(&tasklist_lock); |
| 866 | |
| 867 | /* If the process is dead, release it - nobody will wait for it */ |
| 868 | if (autoreap) |
| 869 | release_task(tsk); |
| 870 | } |
| 871 | |
| 872 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 873 | static void check_stack_usage(void) |
| 874 | { |
| 875 | static DEFINE_SPINLOCK(low_water_lock); |
| 876 | static int lowest_to_date = THREAD_SIZE; |
| 877 | unsigned long free; |
| 878 | |
| 879 | free = stack_not_used(current); |
| 880 | |
| 881 | if (free >= lowest_to_date) |
| 882 | return; |
| 883 | |
| 884 | spin_lock(&low_water_lock); |
| 885 | if (free < lowest_to_date) { |
| 886 | printk(KERN_WARNING "%s used greatest stack depth: %lu bytes " |
| 887 | "left\n", |
| 888 | current->comm, free); |
| 889 | lowest_to_date = free; |
| 890 | } |
| 891 | spin_unlock(&low_water_lock); |
| 892 | } |
| 893 | #else |
| 894 | static inline void check_stack_usage(void) {} |
| 895 | #endif |
| 896 | |
| 897 | NORET_TYPE void do_exit(long code) |
| 898 | { |
| 899 | struct task_struct *tsk = current; |
| 900 | int group_dead; |
| 901 | |
| 902 | profile_task_exit(tsk); |
| 903 | |
| 904 | WARN_ON(atomic_read(&tsk->fs_excl)); |
| 905 | WARN_ON(blk_needs_flush_plug(tsk)); |
| 906 | |
| 907 | if (unlikely(in_interrupt())) |
| 908 | panic("Aiee, killing interrupt handler!"); |
| 909 | if (unlikely(!tsk->pid)) |
| 910 | panic("Attempted to kill the idle task!"); |
| 911 | |
| 912 | /* |
| 913 | * If do_exit is called because this processes oopsed, it's possible |
| 914 | * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before |
| 915 | * continuing. Amongst other possible reasons, this is to prevent |
| 916 | * mm_release()->clear_child_tid() from writing to a user-controlled |
| 917 | * kernel address. |
| 918 | */ |
| 919 | set_fs(USER_DS); |
| 920 | |
| 921 | ptrace_event(PTRACE_EVENT_EXIT, code); |
| 922 | |
| 923 | validate_creds_for_do_exit(tsk); |
| 924 | |
| 925 | /* |
| 926 | * We're taking recursive faults here in do_exit. Safest is to just |
| 927 | * leave this task alone and wait for reboot. |
| 928 | */ |
| 929 | if (unlikely(tsk->flags & PF_EXITING)) { |
| 930 | printk(KERN_ALERT |
| 931 | "Fixing recursive fault but reboot is needed!\n"); |
| 932 | /* |
| 933 | * We can do this unlocked here. The futex code uses |
| 934 | * this flag just to verify whether the pi state |
| 935 | * cleanup has been done or not. In the worst case it |
| 936 | * loops once more. We pretend that the cleanup was |
| 937 | * done as there is no way to return. Either the |
| 938 | * OWNER_DIED bit is set by now or we push the blocked |
| 939 | * task into the wait for ever nirwana as well. |
| 940 | */ |
| 941 | tsk->flags |= PF_EXITPIDONE; |
| 942 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 943 | schedule(); |
| 944 | } |
| 945 | |
| 946 | exit_irq_thread(); |
| 947 | |
| 948 | exit_signals(tsk); /* sets PF_EXITING */ |
| 949 | /* |
| 950 | * tsk->flags are checked in the futex code to protect against |
| 951 | * an exiting task cleaning up the robust pi futexes. |
| 952 | */ |
| 953 | smp_mb(); |
| 954 | raw_spin_unlock_wait(&tsk->pi_lock); |
| 955 | |
| 956 | if (unlikely(in_atomic())) |
| 957 | printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n", |
| 958 | current->comm, task_pid_nr(current), |
| 959 | preempt_count()); |
| 960 | |
| 961 | acct_update_integrals(tsk); |
| 962 | /* sync mm's RSS info before statistics gathering */ |
| 963 | if (tsk->mm) |
| 964 | sync_mm_rss(tsk, tsk->mm); |
| 965 | group_dead = atomic_dec_and_test(&tsk->signal->live); |
| 966 | if (group_dead) { |
| 967 | hrtimer_cancel(&tsk->signal->real_timer); |
| 968 | exit_itimers(tsk->signal); |
| 969 | if (tsk->mm) |
| 970 | setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); |
| 971 | } |
| 972 | acct_collect(code, group_dead); |
| 973 | if (group_dead) |
| 974 | tty_audit_exit(); |
| 975 | if (unlikely(tsk->audit_context)) |
| 976 | audit_free(tsk); |
| 977 | |
| 978 | tsk->exit_code = code; |
| 979 | taskstats_exit(tsk, group_dead); |
| 980 | |
| 981 | exit_mm(tsk); |
| 982 | |
| 983 | if (group_dead) |
| 984 | acct_process(); |
| 985 | trace_sched_process_exit(tsk); |
| 986 | |
| 987 | exit_sem(tsk); |
| 988 | exit_files(tsk); |
| 989 | exit_fs(tsk); |
| 990 | check_stack_usage(); |
| 991 | exit_thread(); |
| 992 | |
| 993 | /* |
| 994 | * Flush inherited counters to the parent - before the parent |
| 995 | * gets woken up by child-exit notifications. |
| 996 | * |
| 997 | * because of cgroup mode, must be called before cgroup_exit() |
| 998 | */ |
| 999 | perf_event_exit_task(tsk); |
| 1000 | |
| 1001 | cgroup_exit(tsk, 1); |
| 1002 | |
| 1003 | if (group_dead) |
| 1004 | disassociate_ctty(1); |
| 1005 | |
| 1006 | module_put(task_thread_info(tsk)->exec_domain->module); |
| 1007 | |
| 1008 | proc_exit_connector(tsk); |
| 1009 | |
| 1010 | /* |
| 1011 | * FIXME: do that only when needed, using sched_exit tracepoint |
| 1012 | */ |
| 1013 | ptrace_put_breakpoints(tsk); |
| 1014 | |
| 1015 | exit_notify(tsk, group_dead); |
| 1016 | #ifdef CONFIG_NUMA |
| 1017 | task_lock(tsk); |
| 1018 | mpol_put(tsk->mempolicy); |
| 1019 | tsk->mempolicy = NULL; |
| 1020 | task_unlock(tsk); |
| 1021 | #endif |
| 1022 | #ifdef CONFIG_FUTEX |
| 1023 | if (unlikely(current->pi_state_cache)) |
| 1024 | kfree(current->pi_state_cache); |
| 1025 | #endif |
| 1026 | /* |
| 1027 | * Make sure we are holding no locks: |
| 1028 | */ |
| 1029 | debug_check_no_locks_held(tsk); |
| 1030 | /* |
| 1031 | * We can do this unlocked here. The futex code uses this flag |
| 1032 | * just to verify whether the pi state cleanup has been done |
| 1033 | * or not. In the worst case it loops once more. |
| 1034 | */ |
| 1035 | tsk->flags |= PF_EXITPIDONE; |
| 1036 | |
| 1037 | if (tsk->io_context) |
| 1038 | exit_io_context(tsk); |
| 1039 | |
| 1040 | if (tsk->splice_pipe) |
| 1041 | __free_pipe_info(tsk->splice_pipe); |
| 1042 | |
| 1043 | validate_creds_for_do_exit(tsk); |
| 1044 | |
| 1045 | preempt_disable(); |
| 1046 | exit_rcu(); |
| 1047 | /* causes final put_task_struct in finish_task_switch(). */ |
| 1048 | tsk->state = TASK_DEAD; |
| 1049 | schedule(); |
| 1050 | BUG(); |
| 1051 | /* Avoid "noreturn function does return". */ |
| 1052 | for (;;) |
| 1053 | cpu_relax(); /* For when BUG is null */ |
| 1054 | } |
| 1055 | |
| 1056 | EXPORT_SYMBOL_GPL(do_exit); |
| 1057 | |
| 1058 | NORET_TYPE void complete_and_exit(struct completion *comp, long code) |
| 1059 | { |
| 1060 | if (comp) |
| 1061 | complete(comp); |
| 1062 | |
| 1063 | do_exit(code); |
| 1064 | } |
| 1065 | |
| 1066 | EXPORT_SYMBOL(complete_and_exit); |
| 1067 | |
| 1068 | SYSCALL_DEFINE1(exit, int, error_code) |
| 1069 | { |
| 1070 | do_exit((error_code&0xff)<<8); |
| 1071 | } |
| 1072 | |
| 1073 | /* |
| 1074 | * Take down every thread in the group. This is called by fatal signals |
| 1075 | * as well as by sys_exit_group (below). |
| 1076 | */ |
| 1077 | NORET_TYPE void |
| 1078 | do_group_exit(int exit_code) |
| 1079 | { |
| 1080 | struct signal_struct *sig = current->signal; |
| 1081 | |
| 1082 | BUG_ON(exit_code & 0x80); /* core dumps don't get here */ |
| 1083 | |
| 1084 | if (signal_group_exit(sig)) |
| 1085 | exit_code = sig->group_exit_code; |
| 1086 | else if (!thread_group_empty(current)) { |
| 1087 | struct sighand_struct *const sighand = current->sighand; |
| 1088 | spin_lock_irq(&sighand->siglock); |
| 1089 | if (signal_group_exit(sig)) |
| 1090 | /* Another thread got here before we took the lock. */ |
| 1091 | exit_code = sig->group_exit_code; |
| 1092 | else { |
| 1093 | sig->group_exit_code = exit_code; |
| 1094 | sig->flags = SIGNAL_GROUP_EXIT; |
| 1095 | zap_other_threads(current); |
| 1096 | } |
| 1097 | spin_unlock_irq(&sighand->siglock); |
| 1098 | } |
| 1099 | |
| 1100 | do_exit(exit_code); |
| 1101 | /* NOTREACHED */ |
| 1102 | } |
| 1103 | |
| 1104 | /* |
| 1105 | * this kills every thread in the thread group. Note that any externally |
| 1106 | * wait4()-ing process will get the correct exit code - even if this |
| 1107 | * thread is not the thread group leader. |
| 1108 | */ |
| 1109 | SYSCALL_DEFINE1(exit_group, int, error_code) |
| 1110 | { |
| 1111 | do_group_exit((error_code & 0xff) << 8); |
| 1112 | /* NOTREACHED */ |
| 1113 | return 0; |
| 1114 | } |
| 1115 | |
| 1116 | struct wait_opts { |
| 1117 | enum pid_type wo_type; |
| 1118 | int wo_flags; |
| 1119 | struct pid *wo_pid; |
| 1120 | |
| 1121 | struct siginfo __user *wo_info; |
| 1122 | int __user *wo_stat; |
| 1123 | struct rusage __user *wo_rusage; |
| 1124 | |
| 1125 | wait_queue_t child_wait; |
| 1126 | int notask_error; |
| 1127 | }; |
| 1128 | |
| 1129 | static inline |
| 1130 | struct pid *task_pid_type(struct task_struct *task, enum pid_type type) |
| 1131 | { |
| 1132 | if (type != PIDTYPE_PID) |
| 1133 | task = task->group_leader; |
| 1134 | return task->pids[type].pid; |
| 1135 | } |
| 1136 | |
| 1137 | static int eligible_pid(struct wait_opts *wo, struct task_struct *p) |
| 1138 | { |
| 1139 | return wo->wo_type == PIDTYPE_MAX || |
| 1140 | task_pid_type(p, wo->wo_type) == wo->wo_pid; |
| 1141 | } |
| 1142 | |
| 1143 | static int eligible_child(struct wait_opts *wo, struct task_struct *p) |
| 1144 | { |
| 1145 | if (!eligible_pid(wo, p)) |
| 1146 | return 0; |
| 1147 | /* Wait for all children (clone and not) if __WALL is set; |
| 1148 | * otherwise, wait for clone children *only* if __WCLONE is |
| 1149 | * set; otherwise, wait for non-clone children *only*. (Note: |
| 1150 | * A "clone" child here is one that reports to its parent |
| 1151 | * using a signal other than SIGCHLD.) */ |
| 1152 | if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) |
| 1153 | && !(wo->wo_flags & __WALL)) |
| 1154 | return 0; |
| 1155 | |
| 1156 | return 1; |
| 1157 | } |
| 1158 | |
| 1159 | static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p, |
| 1160 | pid_t pid, uid_t uid, int why, int status) |
| 1161 | { |
| 1162 | struct siginfo __user *infop; |
| 1163 | int retval = wo->wo_rusage |
| 1164 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; |
| 1165 | |
| 1166 | put_task_struct(p); |
| 1167 | infop = wo->wo_info; |
| 1168 | if (infop) { |
| 1169 | if (!retval) |
| 1170 | retval = put_user(SIGCHLD, &infop->si_signo); |
| 1171 | if (!retval) |
| 1172 | retval = put_user(0, &infop->si_errno); |
| 1173 | if (!retval) |
| 1174 | retval = put_user((short)why, &infop->si_code); |
| 1175 | if (!retval) |
| 1176 | retval = put_user(pid, &infop->si_pid); |
| 1177 | if (!retval) |
| 1178 | retval = put_user(uid, &infop->si_uid); |
| 1179 | if (!retval) |
| 1180 | retval = put_user(status, &infop->si_status); |
| 1181 | } |
| 1182 | if (!retval) |
| 1183 | retval = pid; |
| 1184 | return retval; |
| 1185 | } |
| 1186 | |
| 1187 | /* |
| 1188 | * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold |
| 1189 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold |
| 1190 | * the lock and this task is uninteresting. If we return nonzero, we have |
| 1191 | * released the lock and the system call should return. |
| 1192 | */ |
| 1193 | static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) |
| 1194 | { |
| 1195 | unsigned long state; |
| 1196 | int retval, status, traced; |
| 1197 | pid_t pid = task_pid_vnr(p); |
| 1198 | uid_t uid = __task_cred(p)->uid; |
| 1199 | struct siginfo __user *infop; |
| 1200 | |
| 1201 | if (!likely(wo->wo_flags & WEXITED)) |
| 1202 | return 0; |
| 1203 | |
| 1204 | if (unlikely(wo->wo_flags & WNOWAIT)) { |
| 1205 | int exit_code = p->exit_code; |
| 1206 | int why; |
| 1207 | |
| 1208 | get_task_struct(p); |
| 1209 | read_unlock(&tasklist_lock); |
| 1210 | if ((exit_code & 0x7f) == 0) { |
| 1211 | why = CLD_EXITED; |
| 1212 | status = exit_code >> 8; |
| 1213 | } else { |
| 1214 | why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; |
| 1215 | status = exit_code & 0x7f; |
| 1216 | } |
| 1217 | return wait_noreap_copyout(wo, p, pid, uid, why, status); |
| 1218 | } |
| 1219 | |
| 1220 | /* |
| 1221 | * Try to move the task's state to DEAD |
| 1222 | * only one thread is allowed to do this: |
| 1223 | */ |
| 1224 | state = xchg(&p->exit_state, EXIT_DEAD); |
| 1225 | if (state != EXIT_ZOMBIE) { |
| 1226 | BUG_ON(state != EXIT_DEAD); |
| 1227 | return 0; |
| 1228 | } |
| 1229 | |
| 1230 | traced = ptrace_reparented(p); |
| 1231 | /* |
| 1232 | * It can be ptraced but not reparented, check |
| 1233 | * thread_group_leader() to filter out sub-threads. |
| 1234 | */ |
| 1235 | if (likely(!traced) && thread_group_leader(p)) { |
| 1236 | struct signal_struct *psig; |
| 1237 | struct signal_struct *sig; |
| 1238 | unsigned long maxrss; |
| 1239 | cputime_t tgutime, tgstime; |
| 1240 | |
| 1241 | /* |
| 1242 | * The resource counters for the group leader are in its |
| 1243 | * own task_struct. Those for dead threads in the group |
| 1244 | * are in its signal_struct, as are those for the child |
| 1245 | * processes it has previously reaped. All these |
| 1246 | * accumulate in the parent's signal_struct c* fields. |
| 1247 | * |
| 1248 | * We don't bother to take a lock here to protect these |
| 1249 | * p->signal fields, because they are only touched by |
| 1250 | * __exit_signal, which runs with tasklist_lock |
| 1251 | * write-locked anyway, and so is excluded here. We do |
| 1252 | * need to protect the access to parent->signal fields, |
| 1253 | * as other threads in the parent group can be right |
| 1254 | * here reaping other children at the same time. |
| 1255 | * |
| 1256 | * We use thread_group_times() to get times for the thread |
| 1257 | * group, which consolidates times for all threads in the |
| 1258 | * group including the group leader. |
| 1259 | */ |
| 1260 | thread_group_times(p, &tgutime, &tgstime); |
| 1261 | spin_lock_irq(&p->real_parent->sighand->siglock); |
| 1262 | psig = p->real_parent->signal; |
| 1263 | sig = p->signal; |
| 1264 | psig->cutime = |
| 1265 | cputime_add(psig->cutime, |
| 1266 | cputime_add(tgutime, |
| 1267 | sig->cutime)); |
| 1268 | psig->cstime = |
| 1269 | cputime_add(psig->cstime, |
| 1270 | cputime_add(tgstime, |
| 1271 | sig->cstime)); |
| 1272 | psig->cgtime = |
| 1273 | cputime_add(psig->cgtime, |
| 1274 | cputime_add(p->gtime, |
| 1275 | cputime_add(sig->gtime, |
| 1276 | sig->cgtime))); |
| 1277 | psig->cmin_flt += |
| 1278 | p->min_flt + sig->min_flt + sig->cmin_flt; |
| 1279 | psig->cmaj_flt += |
| 1280 | p->maj_flt + sig->maj_flt + sig->cmaj_flt; |
| 1281 | psig->cnvcsw += |
| 1282 | p->nvcsw + sig->nvcsw + sig->cnvcsw; |
| 1283 | psig->cnivcsw += |
| 1284 | p->nivcsw + sig->nivcsw + sig->cnivcsw; |
| 1285 | psig->cinblock += |
| 1286 | task_io_get_inblock(p) + |
| 1287 | sig->inblock + sig->cinblock; |
| 1288 | psig->coublock += |
| 1289 | task_io_get_oublock(p) + |
| 1290 | sig->oublock + sig->coublock; |
| 1291 | maxrss = max(sig->maxrss, sig->cmaxrss); |
| 1292 | if (psig->cmaxrss < maxrss) |
| 1293 | psig->cmaxrss = maxrss; |
| 1294 | task_io_accounting_add(&psig->ioac, &p->ioac); |
| 1295 | task_io_accounting_add(&psig->ioac, &sig->ioac); |
| 1296 | spin_unlock_irq(&p->real_parent->sighand->siglock); |
| 1297 | } |
| 1298 | |
| 1299 | /* |
| 1300 | * Now we are sure this task is interesting, and no other |
| 1301 | * thread can reap it because we set its state to EXIT_DEAD. |
| 1302 | */ |
| 1303 | read_unlock(&tasklist_lock); |
| 1304 | |
| 1305 | retval = wo->wo_rusage |
| 1306 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; |
| 1307 | status = (p->signal->flags & SIGNAL_GROUP_EXIT) |
| 1308 | ? p->signal->group_exit_code : p->exit_code; |
| 1309 | if (!retval && wo->wo_stat) |
| 1310 | retval = put_user(status, wo->wo_stat); |
| 1311 | |
| 1312 | infop = wo->wo_info; |
| 1313 | if (!retval && infop) |
| 1314 | retval = put_user(SIGCHLD, &infop->si_signo); |
| 1315 | if (!retval && infop) |
| 1316 | retval = put_user(0, &infop->si_errno); |
| 1317 | if (!retval && infop) { |
| 1318 | int why; |
| 1319 | |
| 1320 | if ((status & 0x7f) == 0) { |
| 1321 | why = CLD_EXITED; |
| 1322 | status >>= 8; |
| 1323 | } else { |
| 1324 | why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; |
| 1325 | status &= 0x7f; |
| 1326 | } |
| 1327 | retval = put_user((short)why, &infop->si_code); |
| 1328 | if (!retval) |
| 1329 | retval = put_user(status, &infop->si_status); |
| 1330 | } |
| 1331 | if (!retval && infop) |
| 1332 | retval = put_user(pid, &infop->si_pid); |
| 1333 | if (!retval && infop) |
| 1334 | retval = put_user(uid, &infop->si_uid); |
| 1335 | if (!retval) |
| 1336 | retval = pid; |
| 1337 | |
| 1338 | if (traced) { |
| 1339 | write_lock_irq(&tasklist_lock); |
| 1340 | /* We dropped tasklist, ptracer could die and untrace */ |
| 1341 | ptrace_unlink(p); |
| 1342 | /* |
| 1343 | * If this is not a sub-thread, notify the parent. |
| 1344 | * If parent wants a zombie, don't release it now. |
| 1345 | */ |
| 1346 | if (thread_group_leader(p) && |
| 1347 | !do_notify_parent(p, p->exit_signal)) { |
| 1348 | p->exit_state = EXIT_ZOMBIE; |
| 1349 | p = NULL; |
| 1350 | } |
| 1351 | write_unlock_irq(&tasklist_lock); |
| 1352 | } |
| 1353 | if (p != NULL) |
| 1354 | release_task(p); |
| 1355 | |
| 1356 | return retval; |
| 1357 | } |
| 1358 | |
| 1359 | static int *task_stopped_code(struct task_struct *p, bool ptrace) |
| 1360 | { |
| 1361 | if (ptrace) { |
| 1362 | if (task_is_stopped_or_traced(p) && |
| 1363 | !(p->jobctl & JOBCTL_LISTENING)) |
| 1364 | return &p->exit_code; |
| 1365 | } else { |
| 1366 | if (p->signal->flags & SIGNAL_STOP_STOPPED) |
| 1367 | return &p->signal->group_exit_code; |
| 1368 | } |
| 1369 | return NULL; |
| 1370 | } |
| 1371 | |
| 1372 | /** |
| 1373 | * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED |
| 1374 | * @wo: wait options |
| 1375 | * @ptrace: is the wait for ptrace |
| 1376 | * @p: task to wait for |
| 1377 | * |
| 1378 | * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. |
| 1379 | * |
| 1380 | * CONTEXT: |
| 1381 | * read_lock(&tasklist_lock), which is released if return value is |
| 1382 | * non-zero. Also, grabs and releases @p->sighand->siglock. |
| 1383 | * |
| 1384 | * RETURNS: |
| 1385 | * 0 if wait condition didn't exist and search for other wait conditions |
| 1386 | * should continue. Non-zero return, -errno on failure and @p's pid on |
| 1387 | * success, implies that tasklist_lock is released and wait condition |
| 1388 | * search should terminate. |
| 1389 | */ |
| 1390 | static int wait_task_stopped(struct wait_opts *wo, |
| 1391 | int ptrace, struct task_struct *p) |
| 1392 | { |
| 1393 | struct siginfo __user *infop; |
| 1394 | int retval, exit_code, *p_code, why; |
| 1395 | uid_t uid = 0; /* unneeded, required by compiler */ |
| 1396 | pid_t pid; |
| 1397 | |
| 1398 | /* |
| 1399 | * Traditionally we see ptrace'd stopped tasks regardless of options. |
| 1400 | */ |
| 1401 | if (!ptrace && !(wo->wo_flags & WUNTRACED)) |
| 1402 | return 0; |
| 1403 | |
| 1404 | if (!task_stopped_code(p, ptrace)) |
| 1405 | return 0; |
| 1406 | |
| 1407 | exit_code = 0; |
| 1408 | spin_lock_irq(&p->sighand->siglock); |
| 1409 | |
| 1410 | p_code = task_stopped_code(p, ptrace); |
| 1411 | if (unlikely(!p_code)) |
| 1412 | goto unlock_sig; |
| 1413 | |
| 1414 | exit_code = *p_code; |
| 1415 | if (!exit_code) |
| 1416 | goto unlock_sig; |
| 1417 | |
| 1418 | if (!unlikely(wo->wo_flags & WNOWAIT)) |
| 1419 | *p_code = 0; |
| 1420 | |
| 1421 | uid = task_uid(p); |
| 1422 | unlock_sig: |
| 1423 | spin_unlock_irq(&p->sighand->siglock); |
| 1424 | if (!exit_code) |
| 1425 | return 0; |
| 1426 | |
| 1427 | /* |
| 1428 | * Now we are pretty sure this task is interesting. |
| 1429 | * Make sure it doesn't get reaped out from under us while we |
| 1430 | * give up the lock and then examine it below. We don't want to |
| 1431 | * keep holding onto the tasklist_lock while we call getrusage and |
| 1432 | * possibly take page faults for user memory. |
| 1433 | */ |
| 1434 | get_task_struct(p); |
| 1435 | pid = task_pid_vnr(p); |
| 1436 | why = ptrace ? CLD_TRAPPED : CLD_STOPPED; |
| 1437 | read_unlock(&tasklist_lock); |
| 1438 | |
| 1439 | if (unlikely(wo->wo_flags & WNOWAIT)) |
| 1440 | return wait_noreap_copyout(wo, p, pid, uid, why, exit_code); |
| 1441 | |
| 1442 | retval = wo->wo_rusage |
| 1443 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; |
| 1444 | if (!retval && wo->wo_stat) |
| 1445 | retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat); |
| 1446 | |
| 1447 | infop = wo->wo_info; |
| 1448 | if (!retval && infop) |
| 1449 | retval = put_user(SIGCHLD, &infop->si_signo); |
| 1450 | if (!retval && infop) |
| 1451 | retval = put_user(0, &infop->si_errno); |
| 1452 | if (!retval && infop) |
| 1453 | retval = put_user((short)why, &infop->si_code); |
| 1454 | if (!retval && infop) |
| 1455 | retval = put_user(exit_code, &infop->si_status); |
| 1456 | if (!retval && infop) |
| 1457 | retval = put_user(pid, &infop->si_pid); |
| 1458 | if (!retval && infop) |
| 1459 | retval = put_user(uid, &infop->si_uid); |
| 1460 | if (!retval) |
| 1461 | retval = pid; |
| 1462 | put_task_struct(p); |
| 1463 | |
| 1464 | BUG_ON(!retval); |
| 1465 | return retval; |
| 1466 | } |
| 1467 | |
| 1468 | /* |
| 1469 | * Handle do_wait work for one task in a live, non-stopped state. |
| 1470 | * read_lock(&tasklist_lock) on entry. If we return zero, we still hold |
| 1471 | * the lock and this task is uninteresting. If we return nonzero, we have |
| 1472 | * released the lock and the system call should return. |
| 1473 | */ |
| 1474 | static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) |
| 1475 | { |
| 1476 | int retval; |
| 1477 | pid_t pid; |
| 1478 | uid_t uid; |
| 1479 | |
| 1480 | if (!unlikely(wo->wo_flags & WCONTINUED)) |
| 1481 | return 0; |
| 1482 | |
| 1483 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) |
| 1484 | return 0; |
| 1485 | |
| 1486 | spin_lock_irq(&p->sighand->siglock); |
| 1487 | /* Re-check with the lock held. */ |
| 1488 | if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { |
| 1489 | spin_unlock_irq(&p->sighand->siglock); |
| 1490 | return 0; |
| 1491 | } |
| 1492 | if (!unlikely(wo->wo_flags & WNOWAIT)) |
| 1493 | p->signal->flags &= ~SIGNAL_STOP_CONTINUED; |
| 1494 | uid = task_uid(p); |
| 1495 | spin_unlock_irq(&p->sighand->siglock); |
| 1496 | |
| 1497 | pid = task_pid_vnr(p); |
| 1498 | get_task_struct(p); |
| 1499 | read_unlock(&tasklist_lock); |
| 1500 | |
| 1501 | if (!wo->wo_info) { |
| 1502 | retval = wo->wo_rusage |
| 1503 | ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; |
| 1504 | put_task_struct(p); |
| 1505 | if (!retval && wo->wo_stat) |
| 1506 | retval = put_user(0xffff, wo->wo_stat); |
| 1507 | if (!retval) |
| 1508 | retval = pid; |
| 1509 | } else { |
| 1510 | retval = wait_noreap_copyout(wo, p, pid, uid, |
| 1511 | CLD_CONTINUED, SIGCONT); |
| 1512 | BUG_ON(retval == 0); |
| 1513 | } |
| 1514 | |
| 1515 | return retval; |
| 1516 | } |
| 1517 | |
| 1518 | /* |
| 1519 | * Consider @p for a wait by @parent. |
| 1520 | * |
| 1521 | * -ECHILD should be in ->notask_error before the first call. |
| 1522 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. |
| 1523 | * Returns zero if the search for a child should continue; |
| 1524 | * then ->notask_error is 0 if @p is an eligible child, |
| 1525 | * or another error from security_task_wait(), or still -ECHILD. |
| 1526 | */ |
| 1527 | static int wait_consider_task(struct wait_opts *wo, int ptrace, |
| 1528 | struct task_struct *p) |
| 1529 | { |
| 1530 | int ret = eligible_child(wo, p); |
| 1531 | if (!ret) |
| 1532 | return ret; |
| 1533 | |
| 1534 | ret = security_task_wait(p); |
| 1535 | if (unlikely(ret < 0)) { |
| 1536 | /* |
| 1537 | * If we have not yet seen any eligible child, |
| 1538 | * then let this error code replace -ECHILD. |
| 1539 | * A permission error will give the user a clue |
| 1540 | * to look for security policy problems, rather |
| 1541 | * than for mysterious wait bugs. |
| 1542 | */ |
| 1543 | if (wo->notask_error) |
| 1544 | wo->notask_error = ret; |
| 1545 | return 0; |
| 1546 | } |
| 1547 | |
| 1548 | /* dead body doesn't have much to contribute */ |
| 1549 | if (p->exit_state == EXIT_DEAD) |
| 1550 | return 0; |
| 1551 | |
| 1552 | /* slay zombie? */ |
| 1553 | if (p->exit_state == EXIT_ZOMBIE) { |
| 1554 | /* |
| 1555 | * A zombie ptracee is only visible to its ptracer. |
| 1556 | * Notification and reaping will be cascaded to the real |
| 1557 | * parent when the ptracer detaches. |
| 1558 | */ |
| 1559 | if (likely(!ptrace) && unlikely(p->ptrace)) { |
| 1560 | /* it will become visible, clear notask_error */ |
| 1561 | wo->notask_error = 0; |
| 1562 | return 0; |
| 1563 | } |
| 1564 | |
| 1565 | /* we don't reap group leaders with subthreads */ |
| 1566 | if (!delay_group_leader(p)) |
| 1567 | return wait_task_zombie(wo, p); |
| 1568 | |
| 1569 | /* |
| 1570 | * Allow access to stopped/continued state via zombie by |
| 1571 | * falling through. Clearing of notask_error is complex. |
| 1572 | * |
| 1573 | * When !@ptrace: |
| 1574 | * |
| 1575 | * If WEXITED is set, notask_error should naturally be |
| 1576 | * cleared. If not, subset of WSTOPPED|WCONTINUED is set, |
| 1577 | * so, if there are live subthreads, there are events to |
| 1578 | * wait for. If all subthreads are dead, it's still safe |
| 1579 | * to clear - this function will be called again in finite |
| 1580 | * amount time once all the subthreads are released and |
| 1581 | * will then return without clearing. |
| 1582 | * |
| 1583 | * When @ptrace: |
| 1584 | * |
| 1585 | * Stopped state is per-task and thus can't change once the |
| 1586 | * target task dies. Only continued and exited can happen. |
| 1587 | * Clear notask_error if WCONTINUED | WEXITED. |
| 1588 | */ |
| 1589 | if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) |
| 1590 | wo->notask_error = 0; |
| 1591 | } else { |
| 1592 | /* |
| 1593 | * If @p is ptraced by a task in its real parent's group, |
| 1594 | * hide group stop/continued state when looking at @p as |
| 1595 | * the real parent; otherwise, a single stop can be |
| 1596 | * reported twice as group and ptrace stops. |
| 1597 | * |
| 1598 | * If a ptracer wants to distinguish the two events for its |
| 1599 | * own children, it should create a separate process which |
| 1600 | * takes the role of real parent. |
| 1601 | */ |
| 1602 | if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p)) |
| 1603 | return 0; |
| 1604 | |
| 1605 | /* |
| 1606 | * @p is alive and it's gonna stop, continue or exit, so |
| 1607 | * there always is something to wait for. |
| 1608 | */ |
| 1609 | wo->notask_error = 0; |
| 1610 | } |
| 1611 | |
| 1612 | /* |
| 1613 | * Wait for stopped. Depending on @ptrace, different stopped state |
| 1614 | * is used and the two don't interact with each other. |
| 1615 | */ |
| 1616 | ret = wait_task_stopped(wo, ptrace, p); |
| 1617 | if (ret) |
| 1618 | return ret; |
| 1619 | |
| 1620 | /* |
| 1621 | * Wait for continued. There's only one continued state and the |
| 1622 | * ptracer can consume it which can confuse the real parent. Don't |
| 1623 | * use WCONTINUED from ptracer. You don't need or want it. |
| 1624 | */ |
| 1625 | return wait_task_continued(wo, p); |
| 1626 | } |
| 1627 | |
| 1628 | /* |
| 1629 | * Do the work of do_wait() for one thread in the group, @tsk. |
| 1630 | * |
| 1631 | * -ECHILD should be in ->notask_error before the first call. |
| 1632 | * Returns nonzero for a final return, when we have unlocked tasklist_lock. |
| 1633 | * Returns zero if the search for a child should continue; then |
| 1634 | * ->notask_error is 0 if there were any eligible children, |
| 1635 | * or another error from security_task_wait(), or still -ECHILD. |
| 1636 | */ |
| 1637 | static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) |
| 1638 | { |
| 1639 | struct task_struct *p; |
| 1640 | |
| 1641 | list_for_each_entry(p, &tsk->children, sibling) { |
| 1642 | int ret = wait_consider_task(wo, 0, p); |
| 1643 | if (ret) |
| 1644 | return ret; |
| 1645 | } |
| 1646 | |
| 1647 | return 0; |
| 1648 | } |
| 1649 | |
| 1650 | static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) |
| 1651 | { |
| 1652 | struct task_struct *p; |
| 1653 | |
| 1654 | list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { |
| 1655 | int ret = wait_consider_task(wo, 1, p); |
| 1656 | if (ret) |
| 1657 | return ret; |
| 1658 | } |
| 1659 | |
| 1660 | return 0; |
| 1661 | } |
| 1662 | |
| 1663 | static int child_wait_callback(wait_queue_t *wait, unsigned mode, |
| 1664 | int sync, void *key) |
| 1665 | { |
| 1666 | struct wait_opts *wo = container_of(wait, struct wait_opts, |
| 1667 | child_wait); |
| 1668 | struct task_struct *p = key; |
| 1669 | |
| 1670 | if (!eligible_pid(wo, p)) |
| 1671 | return 0; |
| 1672 | |
| 1673 | if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) |
| 1674 | return 0; |
| 1675 | |
| 1676 | return default_wake_function(wait, mode, sync, key); |
| 1677 | } |
| 1678 | |
| 1679 | void __wake_up_parent(struct task_struct *p, struct task_struct *parent) |
| 1680 | { |
| 1681 | __wake_up_sync_key(&parent->signal->wait_chldexit, |
| 1682 | TASK_INTERRUPTIBLE, 1, p); |
| 1683 | } |
| 1684 | |
| 1685 | static long do_wait(struct wait_opts *wo) |
| 1686 | { |
| 1687 | struct task_struct *tsk; |
| 1688 | int retval; |
| 1689 | |
| 1690 | trace_sched_process_wait(wo->wo_pid); |
| 1691 | |
| 1692 | init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); |
| 1693 | wo->child_wait.private = current; |
| 1694 | add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); |
| 1695 | repeat: |
| 1696 | /* |
| 1697 | * If there is nothing that can match our critiera just get out. |
| 1698 | * We will clear ->notask_error to zero if we see any child that |
| 1699 | * might later match our criteria, even if we are not able to reap |
| 1700 | * it yet. |
| 1701 | */ |
| 1702 | wo->notask_error = -ECHILD; |
| 1703 | if ((wo->wo_type < PIDTYPE_MAX) && |
| 1704 | (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) |
| 1705 | goto notask; |
| 1706 | |
| 1707 | set_current_state(TASK_INTERRUPTIBLE); |
| 1708 | read_lock(&tasklist_lock); |
| 1709 | tsk = current; |
| 1710 | do { |
| 1711 | retval = do_wait_thread(wo, tsk); |
| 1712 | if (retval) |
| 1713 | goto end; |
| 1714 | |
| 1715 | retval = ptrace_do_wait(wo, tsk); |
| 1716 | if (retval) |
| 1717 | goto end; |
| 1718 | |
| 1719 | if (wo->wo_flags & __WNOTHREAD) |
| 1720 | break; |
| 1721 | } while_each_thread(current, tsk); |
| 1722 | read_unlock(&tasklist_lock); |
| 1723 | |
| 1724 | notask: |
| 1725 | retval = wo->notask_error; |
| 1726 | if (!retval && !(wo->wo_flags & WNOHANG)) { |
| 1727 | retval = -ERESTARTSYS; |
| 1728 | if (!signal_pending(current)) { |
| 1729 | schedule(); |
| 1730 | goto repeat; |
| 1731 | } |
| 1732 | } |
| 1733 | end: |
| 1734 | __set_current_state(TASK_RUNNING); |
| 1735 | remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); |
| 1736 | return retval; |
| 1737 | } |
| 1738 | |
| 1739 | SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, |
| 1740 | infop, int, options, struct rusage __user *, ru) |
| 1741 | { |
| 1742 | struct wait_opts wo; |
| 1743 | struct pid *pid = NULL; |
| 1744 | enum pid_type type; |
| 1745 | long ret; |
| 1746 | |
| 1747 | if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED)) |
| 1748 | return -EINVAL; |
| 1749 | if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) |
| 1750 | return -EINVAL; |
| 1751 | |
| 1752 | switch (which) { |
| 1753 | case P_ALL: |
| 1754 | type = PIDTYPE_MAX; |
| 1755 | break; |
| 1756 | case P_PID: |
| 1757 | type = PIDTYPE_PID; |
| 1758 | if (upid <= 0) |
| 1759 | return -EINVAL; |
| 1760 | break; |
| 1761 | case P_PGID: |
| 1762 | type = PIDTYPE_PGID; |
| 1763 | if (upid <= 0) |
| 1764 | return -EINVAL; |
| 1765 | break; |
| 1766 | default: |
| 1767 | return -EINVAL; |
| 1768 | } |
| 1769 | |
| 1770 | if (type < PIDTYPE_MAX) |
| 1771 | pid = find_get_pid(upid); |
| 1772 | |
| 1773 | wo.wo_type = type; |
| 1774 | wo.wo_pid = pid; |
| 1775 | wo.wo_flags = options; |
| 1776 | wo.wo_info = infop; |
| 1777 | wo.wo_stat = NULL; |
| 1778 | wo.wo_rusage = ru; |
| 1779 | ret = do_wait(&wo); |
| 1780 | |
| 1781 | if (ret > 0) { |
| 1782 | ret = 0; |
| 1783 | } else if (infop) { |
| 1784 | /* |
| 1785 | * For a WNOHANG return, clear out all the fields |
| 1786 | * we would set so the user can easily tell the |
| 1787 | * difference. |
| 1788 | */ |
| 1789 | if (!ret) |
| 1790 | ret = put_user(0, &infop->si_signo); |
| 1791 | if (!ret) |
| 1792 | ret = put_user(0, &infop->si_errno); |
| 1793 | if (!ret) |
| 1794 | ret = put_user(0, &infop->si_code); |
| 1795 | if (!ret) |
| 1796 | ret = put_user(0, &infop->si_pid); |
| 1797 | if (!ret) |
| 1798 | ret = put_user(0, &infop->si_uid); |
| 1799 | if (!ret) |
| 1800 | ret = put_user(0, &infop->si_status); |
| 1801 | } |
| 1802 | |
| 1803 | put_pid(pid); |
| 1804 | |
| 1805 | /* avoid REGPARM breakage on x86: */ |
| 1806 | asmlinkage_protect(5, ret, which, upid, infop, options, ru); |
| 1807 | return ret; |
| 1808 | } |
| 1809 | |
| 1810 | SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, |
| 1811 | int, options, struct rusage __user *, ru) |
| 1812 | { |
| 1813 | struct wait_opts wo; |
| 1814 | struct pid *pid = NULL; |
| 1815 | enum pid_type type; |
| 1816 | long ret; |
| 1817 | |
| 1818 | if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| |
| 1819 | __WNOTHREAD|__WCLONE|__WALL)) |
| 1820 | return -EINVAL; |
| 1821 | |
| 1822 | if (upid == -1) |
| 1823 | type = PIDTYPE_MAX; |
| 1824 | else if (upid < 0) { |
| 1825 | type = PIDTYPE_PGID; |
| 1826 | pid = find_get_pid(-upid); |
| 1827 | } else if (upid == 0) { |
| 1828 | type = PIDTYPE_PGID; |
| 1829 | pid = get_task_pid(current, PIDTYPE_PGID); |
| 1830 | } else /* upid > 0 */ { |
| 1831 | type = PIDTYPE_PID; |
| 1832 | pid = find_get_pid(upid); |
| 1833 | } |
| 1834 | |
| 1835 | wo.wo_type = type; |
| 1836 | wo.wo_pid = pid; |
| 1837 | wo.wo_flags = options | WEXITED; |
| 1838 | wo.wo_info = NULL; |
| 1839 | wo.wo_stat = stat_addr; |
| 1840 | wo.wo_rusage = ru; |
| 1841 | ret = do_wait(&wo); |
| 1842 | put_pid(pid); |
| 1843 | |
| 1844 | /* avoid REGPARM breakage on x86: */ |
| 1845 | asmlinkage_protect(4, ret, upid, stat_addr, options, ru); |
| 1846 | return ret; |
| 1847 | } |
| 1848 | |
| 1849 | #ifdef __ARCH_WANT_SYS_WAITPID |
| 1850 | |
| 1851 | /* |
| 1852 | * sys_waitpid() remains for compatibility. waitpid() should be |
| 1853 | * implemented by calling sys_wait4() from libc.a. |
| 1854 | */ |
| 1855 | SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) |
| 1856 | { |
| 1857 | return sys_wait4(pid, stat_addr, options, NULL); |
| 1858 | } |
| 1859 | |
| 1860 | #endif |