Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * linux/fs/proc/array.c
3 *
4 * Copyright (C) 1992 by Linus Torvalds
5 * based on ideas by Darren Senn
6 *
7 * Fixes:
8 * Michael. K. Johnson: stat,statm extensions.
9 * <johnsonm@stolaf.edu>
10 *
11 * Pauline Middelink : Made cmdline,envline only break at '\0's, to
12 * make sure SET_PROCTITLE works. Also removed
13 * bad '!' which forced address recalculation for
14 * EVERY character on the current page.
15 * <middelin@polyware.iaf.nl>
16 *
17 * Danny ter Haar : added cpuinfo
18 * <dth@cistron.nl>
19 *
20 * Alessandro Rubini : profile extension.
21 * <rubini@ipvvis.unipv.it>
22 *
23 * Jeff Tranter : added BogoMips field to cpuinfo
24 * <Jeff_Tranter@Mitel.COM>
25 *
26 * Bruno Haible : remove 4K limit for the maps file
27 * <haible@ma2s2.mathematik.uni-karlsruhe.de>
28 *
29 * Yves Arrouye : remove removal of trailing spaces in get_array.
30 * <Yves.Arrouye@marin.fdn.fr>
31 *
32 * Jerome Forissier : added per-CPU time information to /proc/stat
33 * and /proc/<pid>/cpu extension
34 * <forissier@isia.cma.fr>
35 * - Incorporation and non-SMP safe operation
36 * of forissier patch in 2.1.78 by
37 * Hans Marcus <crowbar@concepts.nl>
38 *
39 * aeb@cwi.nl : /proc/partitions
40 *
41 *
42 * Alan Cox : security fixes.
43 * <alan@lxorguk.ukuu.org.uk>
44 *
45 * Al Viro : safe handling of mm_struct
46 *
47 * Gerhard Wichert : added BIGMEM support
48 * Siemens AG <Gerhard.Wichert@pdb.siemens.de>
49 *
50 * Al Viro & Jeff Garzik : moved most of the thing into base.c and
51 * : proc_misc.c. The rest may eventually go into
52 * : base.c too.
53 */
54
55#include <linux/types.h>
56#include <linux/errno.h>
57#include <linux/time.h>
58#include <linux/kernel.h>
59#include <linux/kernel_stat.h>
60#include <linux/tty.h>
61#include <linux/string.h>
62#include <linux/mman.h>
63#include <linux/proc_fs.h>
64#include <linux/ioport.h>
65#include <linux/uaccess.h>
66#include <linux/io.h>
67#include <linux/mm.h>
68#include <linux/hugetlb.h>
69#include <linux/pagemap.h>
70#include <linux/swap.h>
71#include <linux/smp.h>
72#include <linux/signal.h>
73#include <linux/highmem.h>
74#include <linux/file.h>
75#include <linux/fdtable.h>
76#include <linux/times.h>
77#include <linux/cpuset.h>
78#include <linux/rcupdate.h>
79#include <linux/delayacct.h>
80#include <linux/seq_file.h>
81#include <linux/pid_namespace.h>
82#include <linux/ptrace.h>
83#include <linux/tracehook.h>
84#include <linux/string_helpers.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87
88#include <asm/pgtable.h>
89#include <asm/processor.h>
90#include "internal.h"
91
92static inline void task_name(struct seq_file *m, struct task_struct *p)
93{
94 char *buf;
95 size_t size;
96 char tcomm[sizeof(p->comm)];
97 int ret;
98
99 get_task_comm(tcomm, p);
100
101 seq_puts(m, "Name:\t");
102
103 size = seq_get_buf(m, &buf);
104 ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
105 seq_commit(m, ret < size ? ret : -1);
106
107 seq_putc(m, '\n');
108}
109
110/*
111 * The task state array is a strange "bitmap" of
112 * reasons to sleep. Thus "running" is zero, and
113 * you can test for combinations of others with
114 * simple bit tests.
115 */
116static const char * const task_state_array[] = {
117 "R (running)", /* 0 */
118 "S (sleeping)", /* 1 */
119 "D (disk sleep)", /* 2 */
120 "T (stopped)", /* 4 */
121 "t (tracing stop)", /* 8 */
122 "X (dead)", /* 16 */
123 "Z (zombie)", /* 32 */
124};
125
126static inline const char *get_task_state(struct task_struct *tsk)
127{
128 unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
129
130 /*
131 * Parked tasks do not run; they sit in __kthread_parkme().
132 * Without this check, we would report them as running, which is
133 * clearly wrong, so we report them as sleeping instead.
134 */
135 if (tsk->state == TASK_PARKED)
136 state = TASK_INTERRUPTIBLE;
137
138 BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
139
140 return task_state_array[fls(state)];
141}
142
143static inline int get_task_umask(struct task_struct *tsk)
144{
145 struct fs_struct *fs;
146 int umask = -ENOENT;
147
148 task_lock(tsk);
149 fs = tsk->fs;
150 if (fs)
151 umask = fs->umask;
152 task_unlock(tsk);
153 return umask;
154}
155
156static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
157 struct pid *pid, struct task_struct *p)
158{
159 struct user_namespace *user_ns = seq_user_ns(m);
160 struct group_info *group_info;
161 int g, umask;
162 struct task_struct *tracer;
163 const struct cred *cred;
164 pid_t ppid, tpid = 0, tgid, ngid;
165 unsigned int max_fds = 0;
166
167 rcu_read_lock();
168 ppid = pid_alive(p) ?
169 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
170
171 tracer = ptrace_parent(p);
172 if (tracer)
173 tpid = task_pid_nr_ns(tracer, ns);
174
175 tgid = task_tgid_nr_ns(p, ns);
176 ngid = task_numa_group_id(p);
177 cred = get_task_cred(p);
178
179 umask = get_task_umask(p);
180 if (umask >= 0)
181 seq_printf(m, "Umask:\t%#04o\n", umask);
182
183 task_lock(p);
184 if (p->files)
185 max_fds = files_fdtable(p->files)->max_fds;
186 task_unlock(p);
187 rcu_read_unlock();
188
189 seq_printf(m, "State:\t%s", get_task_state(p));
190
191 seq_put_decimal_ull(m, "\nTgid:\t", tgid);
192 seq_put_decimal_ull(m, "\nNgid:\t", ngid);
193 seq_put_decimal_ull(m, "\nPid:\t", pid_nr_ns(pid, ns));
194 seq_put_decimal_ull(m, "\nPPid:\t", ppid);
195 seq_put_decimal_ull(m, "\nTracerPid:\t", tpid);
196 seq_put_decimal_ull(m, "\nUid:\t", from_kuid_munged(user_ns, cred->uid));
197 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->euid));
198 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->suid));
199 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->fsuid));
200 seq_put_decimal_ull(m, "\nGid:\t", from_kgid_munged(user_ns, cred->gid));
201 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->egid));
202 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->sgid));
203 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->fsgid));
204 seq_put_decimal_ull(m, "\nFDSize:\t", max_fds);
205
206 seq_puts(m, "\nGroups:\t");
207 group_info = cred->group_info;
208 for (g = 0; g < group_info->ngroups; g++)
209 seq_put_decimal_ull(m, g ? " " : "",
210 from_kgid_munged(user_ns, group_info->gid[g]));
211 put_cred(cred);
212 /* Trailing space shouldn't have been added in the first place. */
213 seq_putc(m, ' ');
214
215#ifdef CONFIG_PID_NS
216 seq_puts(m, "\nNStgid:");
217 for (g = ns->level; g <= pid->level; g++)
218 seq_put_decimal_ull(m, "\t", task_tgid_nr_ns(p, pid->numbers[g].ns));
219 seq_puts(m, "\nNSpid:");
220 for (g = ns->level; g <= pid->level; g++)
221 seq_put_decimal_ull(m, "\t", task_pid_nr_ns(p, pid->numbers[g].ns));
222 seq_puts(m, "\nNSpgid:");
223 for (g = ns->level; g <= pid->level; g++)
224 seq_put_decimal_ull(m, "\t", task_pgrp_nr_ns(p, pid->numbers[g].ns));
225 seq_puts(m, "\nNSsid:");
226 for (g = ns->level; g <= pid->level; g++)
227 seq_put_decimal_ull(m, "\t", task_session_nr_ns(p, pid->numbers[g].ns));
228#endif
229 seq_putc(m, '\n');
230}
231
232void render_sigset_t(struct seq_file *m, const char *header,
233 sigset_t *set)
234{
235 int i;
236
237 seq_puts(m, header);
238
239 i = _NSIG;
240 do {
241 int x = 0;
242
243 i -= 4;
244 if (sigismember(set, i+1)) x |= 1;
245 if (sigismember(set, i+2)) x |= 2;
246 if (sigismember(set, i+3)) x |= 4;
247 if (sigismember(set, i+4)) x |= 8;
248 seq_printf(m, "%x", x);
249 } while (i >= 4);
250
251 seq_putc(m, '\n');
252}
253
254static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
255 sigset_t *catch)
256{
257 struct k_sigaction *k;
258 int i;
259
260 k = p->sighand->action;
261 for (i = 1; i <= _NSIG; ++i, ++k) {
262 if (k->sa.sa_handler == SIG_IGN)
263 sigaddset(ign, i);
264 else if (k->sa.sa_handler != SIG_DFL)
265 sigaddset(catch, i);
266 }
267}
268
269static inline void task_sig(struct seq_file *m, struct task_struct *p)
270{
271 unsigned long flags;
272 sigset_t pending, shpending, blocked, ignored, caught;
273 int num_threads = 0;
274 unsigned long qsize = 0;
275 unsigned long qlim = 0;
276
277 sigemptyset(&pending);
278 sigemptyset(&shpending);
279 sigemptyset(&blocked);
280 sigemptyset(&ignored);
281 sigemptyset(&caught);
282
283 if (lock_task_sighand(p, &flags)) {
284 pending = p->pending.signal;
285 shpending = p->signal->shared_pending.signal;
286 blocked = p->blocked;
287 collect_sigign_sigcatch(p, &ignored, &caught);
288 num_threads = get_nr_threads(p);
289 rcu_read_lock(); /* FIXME: is this correct? */
290 qsize = atomic_read(&__task_cred(p)->user->sigpending);
291 rcu_read_unlock();
292 qlim = task_rlimit(p, RLIMIT_SIGPENDING);
293 unlock_task_sighand(p, &flags);
294 }
295
296 seq_put_decimal_ull(m, "Threads:\t", num_threads);
297 seq_put_decimal_ull(m, "\nSigQ:\t", qsize);
298 seq_put_decimal_ull(m, "/", qlim);
299
300 /* render them all */
301 render_sigset_t(m, "\nSigPnd:\t", &pending);
302 render_sigset_t(m, "ShdPnd:\t", &shpending);
303 render_sigset_t(m, "SigBlk:\t", &blocked);
304 render_sigset_t(m, "SigIgn:\t", &ignored);
305 render_sigset_t(m, "SigCgt:\t", &caught);
306}
307
308static void render_cap_t(struct seq_file *m, const char *header,
309 kernel_cap_t *a)
310{
311 unsigned __capi;
312
313 seq_puts(m, header);
314 CAP_FOR_EACH_U32(__capi) {
315 seq_printf(m, "%08x",
316 a->cap[CAP_LAST_U32 - __capi]);
317 }
318 seq_putc(m, '\n');
319}
320
321static inline void task_cap(struct seq_file *m, struct task_struct *p)
322{
323 const struct cred *cred;
324 kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
325 cap_bset, cap_ambient;
326
327 rcu_read_lock();
328 cred = __task_cred(p);
329 cap_inheritable = cred->cap_inheritable;
330 cap_permitted = cred->cap_permitted;
331 cap_effective = cred->cap_effective;
332 cap_bset = cred->cap_bset;
333 cap_ambient = cred->cap_ambient;
334 rcu_read_unlock();
335
336 render_cap_t(m, "CapInh:\t", &cap_inheritable);
337 render_cap_t(m, "CapPrm:\t", &cap_permitted);
338 render_cap_t(m, "CapEff:\t", &cap_effective);
339 render_cap_t(m, "CapBnd:\t", &cap_bset);
340 render_cap_t(m, "CapAmb:\t", &cap_ambient);
341}
342
343static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
344{
345#ifdef CONFIG_SECCOMP
346 seq_put_decimal_ull(m, "Seccomp:\t", p->seccomp.mode);
347 seq_putc(m, '\n');
348#endif
349}
350
351static inline void task_context_switch_counts(struct seq_file *m,
352 struct task_struct *p)
353{
354 seq_put_decimal_ull(m, "voluntary_ctxt_switches:\t", p->nvcsw);
355 seq_put_decimal_ull(m, "\nnonvoluntary_ctxt_switches:\t", p->nivcsw);
356 seq_putc(m, '\n');
357}
358
359static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
360{
361 seq_printf(m, "Cpus_allowed:\t%*pb\n",
362 cpumask_pr_args(&task->cpus_allowed));
363 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
364 cpumask_pr_args(&task->cpus_allowed));
365}
366
367int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
368 struct pid *pid, struct task_struct *task)
369{
370 struct mm_struct *mm = get_task_mm(task);
371
372 task_name(m, task);
373 task_state(m, ns, pid, task);
374
375 if (mm) {
376 task_mem(m, mm);
377 mmput(mm);
378 }
379 task_sig(m, task);
380 task_cap(m, task);
381 task_seccomp(m, task);
382 task_cpus_allowed(m, task);
383 cpuset_task_status_allowed(m, task);
384 task_context_switch_counts(m, task);
385 return 0;
386}
387
388static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
389 struct pid *pid, struct task_struct *task, int whole)
390{
391 unsigned long vsize, eip, esp, wchan = 0;
392 int priority, nice;
393 int tty_pgrp = -1, tty_nr = 0;
394 sigset_t sigign, sigcatch;
395 char state;
396 pid_t ppid = 0, pgid = -1, sid = -1;
397 int num_threads = 0;
398 int permitted;
399 struct mm_struct *mm;
400 unsigned long long start_time;
401 unsigned long cmin_flt = 0, cmaj_flt = 0;
402 unsigned long min_flt = 0, maj_flt = 0;
403 cputime_t cutime, cstime, utime, stime;
404 cputime_t cgtime, gtime;
405 unsigned long rsslim = 0;
406 char tcomm[sizeof(task->comm)];
407 unsigned long flags;
408
409 state = *get_task_state(task);
410 vsize = eip = esp = 0;
411 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
412 mm = get_task_mm(task);
413 if (mm) {
414 vsize = task_vsize(mm);
415 /*
416 * esp and eip are intentionally zeroed out. There is no
417 * non-racy way to read them without freezing the task.
418 * Programs that need reliable values can use ptrace(2).
419 */
420 }
421
422 get_task_comm(tcomm, task);
423
424 sigemptyset(&sigign);
425 sigemptyset(&sigcatch);
426 cutime = cstime = utime = stime = 0;
427 cgtime = gtime = 0;
428
429 if (lock_task_sighand(task, &flags)) {
430 struct signal_struct *sig = task->signal;
431
432 if (sig->tty) {
433 struct pid *pgrp = tty_get_pgrp(sig->tty);
434 tty_pgrp = pid_nr_ns(pgrp, ns);
435 put_pid(pgrp);
436 tty_nr = new_encode_dev(tty_devnum(sig->tty));
437 }
438
439 num_threads = get_nr_threads(task);
440 collect_sigign_sigcatch(task, &sigign, &sigcatch);
441
442 cmin_flt = sig->cmin_flt;
443 cmaj_flt = sig->cmaj_flt;
444 cutime = sig->cutime;
445 cstime = sig->cstime;
446 cgtime = sig->cgtime;
447 rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
448
449 /* add up live thread stats at the group level */
450 if (whole) {
451 struct task_struct *t = task;
452 do {
453 min_flt += t->min_flt;
454 maj_flt += t->maj_flt;
455 gtime += task_gtime(t);
456 } while_each_thread(task, t);
457
458 min_flt += sig->min_flt;
459 maj_flt += sig->maj_flt;
460 thread_group_cputime_adjusted(task, &utime, &stime);
461 gtime += sig->gtime;
462 }
463
464 sid = task_session_nr_ns(task, ns);
465 ppid = task_tgid_nr_ns(task->real_parent, ns);
466 pgid = task_pgrp_nr_ns(task, ns);
467
468 unlock_task_sighand(task, &flags);
469 }
470
471 if (permitted && (!whole || num_threads < 2))
472 wchan = get_wchan(task);
473 if (!whole) {
474 min_flt = task->min_flt;
475 maj_flt = task->maj_flt;
476 task_cputime_adjusted(task, &utime, &stime);
477 gtime = task_gtime(task);
478 }
479
480 /* scale priority and nice values from timeslices to -20..20 */
481 /* to make it look like a "normal" Unix priority/nice value */
482 priority = task_prio(task);
483 nice = task_nice(task);
484
485 /* convert nsec -> ticks */
486 start_time = nsec_to_clock_t(task->real_start_time);
487
488 seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
489 seq_put_decimal_ll(m, " ", ppid);
490 seq_put_decimal_ll(m, " ", pgid);
491 seq_put_decimal_ll(m, " ", sid);
492 seq_put_decimal_ll(m, " ", tty_nr);
493 seq_put_decimal_ll(m, " ", tty_pgrp);
494 seq_put_decimal_ull(m, " ", task->flags);
495 seq_put_decimal_ull(m, " ", min_flt);
496 seq_put_decimal_ull(m, " ", cmin_flt);
497 seq_put_decimal_ull(m, " ", maj_flt);
498 seq_put_decimal_ull(m, " ", cmaj_flt);
499 seq_put_decimal_ull(m, " ", cputime_to_clock_t(utime));
500 seq_put_decimal_ull(m, " ", cputime_to_clock_t(stime));
501 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cutime));
502 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cstime));
503 seq_put_decimal_ll(m, " ", priority);
504 seq_put_decimal_ll(m, " ", nice);
505 seq_put_decimal_ll(m, " ", num_threads);
506 seq_put_decimal_ull(m, " ", 0);
507 seq_put_decimal_ull(m, " ", start_time);
508 seq_put_decimal_ull(m, " ", vsize);
509 seq_put_decimal_ull(m, " ", mm ? get_mm_rss(mm) : 0);
510 seq_put_decimal_ull(m, " ", rsslim);
511 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->start_code : 1) : 0);
512 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->end_code : 1) : 0);
513 seq_put_decimal_ull(m, " ", (permitted && mm) ? mm->start_stack : 0);
514 seq_put_decimal_ull(m, " ", esp);
515 seq_put_decimal_ull(m, " ", eip);
516 /* The signal information here is obsolete.
517 * It must be decimal for Linux 2.0 compatibility.
518 * Use /proc/#/status for real-time signals.
519 */
520 seq_put_decimal_ull(m, " ", task->pending.signal.sig[0] & 0x7fffffffUL);
521 seq_put_decimal_ull(m, " ", task->blocked.sig[0] & 0x7fffffffUL);
522 seq_put_decimal_ull(m, " ", sigign.sig[0] & 0x7fffffffUL);
523 seq_put_decimal_ull(m, " ", sigcatch.sig[0] & 0x7fffffffUL);
524
525 /*
526 * We used to output the absolute kernel address, but that's an
527 * information leak - so instead we show a 0/1 flag here, to signal
528 * to user-space whether there's a wchan field in /proc/PID/wchan.
529 *
530 * This works with older implementations of procps as well.
531 */
532 if (wchan)
533 seq_puts(m, " 1");
534 else
535 seq_puts(m, " 0");
536
537 seq_put_decimal_ull(m, " ", 0);
538 seq_put_decimal_ull(m, " ", 0);
539 seq_put_decimal_ll(m, " ", task->exit_signal);
540 seq_put_decimal_ll(m, " ", task_cpu(task));
541 seq_put_decimal_ull(m, " ", task->rt_priority);
542 seq_put_decimal_ull(m, " ", task->policy);
543 seq_put_decimal_ull(m, " ", delayacct_blkio_ticks(task));
544 seq_put_decimal_ull(m, " ", cputime_to_clock_t(gtime));
545 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cgtime));
546
547 if (mm && permitted) {
548 seq_put_decimal_ull(m, " ", mm->start_data);
549 seq_put_decimal_ull(m, " ", mm->end_data);
550 seq_put_decimal_ull(m, " ", mm->start_brk);
551 seq_put_decimal_ull(m, " ", mm->arg_start);
552 seq_put_decimal_ull(m, " ", mm->arg_end);
553 seq_put_decimal_ull(m, " ", mm->env_start);
554 seq_put_decimal_ull(m, " ", mm->env_end);
555 } else
556 seq_puts(m, " 0 0 0 0 0 0 0");
557
558 if (permitted)
559 seq_put_decimal_ll(m, " ", task->exit_code);
560 else
561 seq_puts(m, " 0");
562
563 seq_putc(m, '\n');
564 if (mm)
565 mmput(mm);
566 return 0;
567}
568
569int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
570 struct pid *pid, struct task_struct *task)
571{
572 return do_task_stat(m, ns, pid, task, 0);
573}
574
575int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
576 struct pid *pid, struct task_struct *task)
577{
578 return do_task_stat(m, ns, pid, task, 1);
579}
580
581int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
582 struct pid *pid, struct task_struct *task)
583{
584 unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
585 struct mm_struct *mm = get_task_mm(task);
586
587 if (mm) {
588 size = task_statm(mm, &shared, &text, &data, &resident);
589 mmput(mm);
590 }
591 /*
592 * For quick read, open code by putting numbers directly
593 * expected format is
594 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
595 * size, resident, shared, text, data);
596 */
597 seq_put_decimal_ull(m, "", size);
598 seq_put_decimal_ull(m, " ", resident);
599 seq_put_decimal_ull(m, " ", shared);
600 seq_put_decimal_ull(m, " ", text);
601 seq_put_decimal_ull(m, " ", 0);
602 seq_put_decimal_ull(m, " ", data);
603 seq_put_decimal_ull(m, " ", 0);
604 seq_putc(m, '\n');
605
606 return 0;
607}
608
609#ifdef CONFIG_PROC_CHILDREN
610static struct pid *
611get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
612{
613 struct task_struct *start, *task;
614 struct pid *pid = NULL;
615
616 read_lock(&tasklist_lock);
617
618 start = pid_task(proc_pid(inode), PIDTYPE_PID);
619 if (!start)
620 goto out;
621
622 /*
623 * Lets try to continue searching first, this gives
624 * us significant speedup on children-rich processes.
625 */
626 if (pid_prev) {
627 task = pid_task(pid_prev, PIDTYPE_PID);
628 if (task && task->real_parent == start &&
629 !(list_empty(&task->sibling))) {
630 if (list_is_last(&task->sibling, &start->children))
631 goto out;
632 task = list_first_entry(&task->sibling,
633 struct task_struct, sibling);
634 pid = get_pid(task_pid(task));
635 goto out;
636 }
637 }
638
639 /*
640 * Slow search case.
641 *
642 * We might miss some children here if children
643 * are exited while we were not holding the lock,
644 * but it was never promised to be accurate that
645 * much.
646 *
647 * "Just suppose that the parent sleeps, but N children
648 * exit after we printed their tids. Now the slow paths
649 * skips N extra children, we miss N tasks." (c)
650 *
651 * So one need to stop or freeze the leader and all
652 * its children to get a precise result.
653 */
654 list_for_each_entry(task, &start->children, sibling) {
655 if (pos-- == 0) {
656 pid = get_pid(task_pid(task));
657 break;
658 }
659 }
660
661out:
662 read_unlock(&tasklist_lock);
663 return pid;
664}
665
666static int children_seq_show(struct seq_file *seq, void *v)
667{
668 struct inode *inode = seq->private;
669 pid_t pid;
670
671 pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
672 seq_printf(seq, "%d ", pid);
673
674 return 0;
675}
676
677static void *children_seq_start(struct seq_file *seq, loff_t *pos)
678{
679 return get_children_pid(seq->private, NULL, *pos);
680}
681
682static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
683{
684 struct pid *pid;
685
686 pid = get_children_pid(seq->private, v, *pos + 1);
687 put_pid(v);
688
689 ++*pos;
690 return pid;
691}
692
693static void children_seq_stop(struct seq_file *seq, void *v)
694{
695 put_pid(v);
696}
697
698static const struct seq_operations children_seq_ops = {
699 .start = children_seq_start,
700 .next = children_seq_next,
701 .stop = children_seq_stop,
702 .show = children_seq_show,
703};
704
705static int children_seq_open(struct inode *inode, struct file *file)
706{
707 struct seq_file *m;
708 int ret;
709
710 ret = seq_open(file, &children_seq_ops);
711 if (ret)
712 return ret;
713
714 m = file->private_data;
715 m->private = inode;
716
717 return ret;
718}
719
720int children_seq_release(struct inode *inode, struct file *file)
721{
722 seq_release(inode, file);
723 return 0;
724}
725
726const struct file_operations proc_tid_children_operations = {
727 .open = children_seq_open,
728 .read = seq_read,
729 .llseek = seq_lseek,
730 .release = children_seq_release,
731};
732#endif /* CONFIG_PROC_CHILDREN */