xref: /openbmc/linux/fs/proc/array.c (revision 4a44a19b) 
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/user_namespace.h>
85 
86 #include <asm/pgtable.h>
87 #include <asm/processor.h>
88 #include "internal.h"
89 
90 static inline void task_name(struct seq_file *m, struct task_struct *p)
91 {
92 	int i;
93 	char *buf, *end;
94 	char *name;
95 	char tcomm[sizeof(p->comm)];
96 
97 	get_task_comm(tcomm, p);
98 
99 	seq_puts(m, "Name:\t");
100 	end = m->buf + m->size;
101 	buf = m->buf + m->count;
102 	name = tcomm;
103 	i = sizeof(tcomm);
104 	while (i && (buf < end)) {
105 		unsigned char c = *name;
106 		name++;
107 		i--;
108 		*buf = c;
109 		if (!c)
110 			break;
111 		if (c == '\\') {
112 			buf++;
113 			if (buf < end)
114 				*buf++ = c;
115 			continue;
116 		}
117 		if (c == '\n') {
118 			*buf++ = '\\';
119 			if (buf < end)
120 				*buf++ = 'n';
121 			continue;
122 		}
123 		buf++;
124 	}
125 	m->count = buf - m->buf;
126 	seq_putc(m, '\n');
127 }
128 
129 /*
130  * The task state array is a strange "bitmap" of
131  * reasons to sleep. Thus "running" is zero, and
132  * you can test for combinations of others with
133  * simple bit tests.
134  */
135 static const char * const task_state_array[] = {
136 	"R (running)",		/*   0 */
137 	"S (sleeping)",		/*   1 */
138 	"D (disk sleep)",	/*   2 */
139 	"T (stopped)",		/*   4 */
140 	"t (tracing stop)",	/*   8 */
141 	"X (dead)",		/*  16 */
142 	"Z (zombie)",		/*  32 */
143 };
144 
145 static inline const char *get_task_state(struct task_struct *tsk)
146 {
147 	unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
148 
149 	BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
150 
151 	return task_state_array[fls(state)];
152 }
153 
154 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
155 				struct pid *pid, struct task_struct *p)
156 {
157 	struct user_namespace *user_ns = seq_user_ns(m);
158 	struct group_info *group_info;
159 	int g;
160 	struct fdtable *fdt = NULL;
161 	const struct cred *cred;
162 	pid_t ppid, tpid;
163 
164 	rcu_read_lock();
165 	ppid = pid_alive(p) ?
166 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
167 	tpid = 0;
168 	if (pid_alive(p)) {
169 		struct task_struct *tracer = ptrace_parent(p);
170 		if (tracer)
171 			tpid = task_pid_nr_ns(tracer, ns);
172 	}
173 	cred = get_task_cred(p);
174 	seq_printf(m,
175 		"State:\t%s\n"
176 		"Tgid:\t%d\n"
177 		"Ngid:\t%d\n"
178 		"Pid:\t%d\n"
179 		"PPid:\t%d\n"
180 		"TracerPid:\t%d\n"
181 		"Uid:\t%d\t%d\t%d\t%d\n"
182 		"Gid:\t%d\t%d\t%d\t%d\n",
183 		get_task_state(p),
184 		task_tgid_nr_ns(p, ns),
185 		task_numa_group_id(p),
186 		pid_nr_ns(pid, ns),
187 		ppid, tpid,
188 		from_kuid_munged(user_ns, cred->uid),
189 		from_kuid_munged(user_ns, cred->euid),
190 		from_kuid_munged(user_ns, cred->suid),
191 		from_kuid_munged(user_ns, cred->fsuid),
192 		from_kgid_munged(user_ns, cred->gid),
193 		from_kgid_munged(user_ns, cred->egid),
194 		from_kgid_munged(user_ns, cred->sgid),
195 		from_kgid_munged(user_ns, cred->fsgid));
196 
197 	task_lock(p);
198 	if (p->files)
199 		fdt = files_fdtable(p->files);
200 	seq_printf(m,
201 		"FDSize:\t%d\n"
202 		"Groups:\t",
203 		fdt ? fdt->max_fds : 0);
204 	rcu_read_unlock();
205 
206 	group_info = cred->group_info;
207 	task_unlock(p);
208 
209 	for (g = 0; g < group_info->ngroups; g++)
210 		seq_printf(m, "%d ",
211 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
212 	put_cred(cred);
213 
214 	seq_putc(m, '\n');
215 }
216 
217 void render_sigset_t(struct seq_file *m, const char *header,
218 				sigset_t *set)
219 {
220 	int i;
221 
222 	seq_puts(m, header);
223 
224 	i = _NSIG;
225 	do {
226 		int x = 0;
227 
228 		i -= 4;
229 		if (sigismember(set, i+1)) x |= 1;
230 		if (sigismember(set, i+2)) x |= 2;
231 		if (sigismember(set, i+3)) x |= 4;
232 		if (sigismember(set, i+4)) x |= 8;
233 		seq_printf(m, "%x", x);
234 	} while (i >= 4);
235 
236 	seq_putc(m, '\n');
237 }
238 
239 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
240 				    sigset_t *catch)
241 {
242 	struct k_sigaction *k;
243 	int i;
244 
245 	k = p->sighand->action;
246 	for (i = 1; i <= _NSIG; ++i, ++k) {
247 		if (k->sa.sa_handler == SIG_IGN)
248 			sigaddset(ign, i);
249 		else if (k->sa.sa_handler != SIG_DFL)
250 			sigaddset(catch, i);
251 	}
252 }
253 
254 static inline void task_sig(struct seq_file *m, struct task_struct *p)
255 {
256 	unsigned long flags;
257 	sigset_t pending, shpending, blocked, ignored, caught;
258 	int num_threads = 0;
259 	unsigned long qsize = 0;
260 	unsigned long qlim = 0;
261 
262 	sigemptyset(&pending);
263 	sigemptyset(&shpending);
264 	sigemptyset(&blocked);
265 	sigemptyset(&ignored);
266 	sigemptyset(&caught);
267 
268 	if (lock_task_sighand(p, &flags)) {
269 		pending = p->pending.signal;
270 		shpending = p->signal->shared_pending.signal;
271 		blocked = p->blocked;
272 		collect_sigign_sigcatch(p, &ignored, &caught);
273 		num_threads = get_nr_threads(p);
274 		rcu_read_lock();  /* FIXME: is this correct? */
275 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
276 		rcu_read_unlock();
277 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
278 		unlock_task_sighand(p, &flags);
279 	}
280 
281 	seq_printf(m, "Threads:\t%d\n", num_threads);
282 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
283 
284 	/* render them all */
285 	render_sigset_t(m, "SigPnd:\t", &pending);
286 	render_sigset_t(m, "ShdPnd:\t", &shpending);
287 	render_sigset_t(m, "SigBlk:\t", &blocked);
288 	render_sigset_t(m, "SigIgn:\t", &ignored);
289 	render_sigset_t(m, "SigCgt:\t", &caught);
290 }
291 
292 static void render_cap_t(struct seq_file *m, const char *header,
293 			kernel_cap_t *a)
294 {
295 	unsigned __capi;
296 
297 	seq_puts(m, header);
298 	CAP_FOR_EACH_U32(__capi) {
299 		seq_printf(m, "%08x",
300 			   a->cap[CAP_LAST_U32 - __capi]);
301 	}
302 	seq_putc(m, '\n');
303 }
304 
305 static inline void task_cap(struct seq_file *m, struct task_struct *p)
306 {
307 	const struct cred *cred;
308 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
309 
310 	rcu_read_lock();
311 	cred = __task_cred(p);
312 	cap_inheritable	= cred->cap_inheritable;
313 	cap_permitted	= cred->cap_permitted;
314 	cap_effective	= cred->cap_effective;
315 	cap_bset	= cred->cap_bset;
316 	rcu_read_unlock();
317 
318 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
319 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
320 	render_cap_t(m, "CapEff:\t", &cap_effective);
321 	render_cap_t(m, "CapBnd:\t", &cap_bset);
322 }
323 
324 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
325 {
326 #ifdef CONFIG_SECCOMP
327 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
328 #endif
329 }
330 
331 static inline void task_context_switch_counts(struct seq_file *m,
332 						struct task_struct *p)
333 {
334 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
335 			"nonvoluntary_ctxt_switches:\t%lu\n",
336 			p->nvcsw,
337 			p->nivcsw);
338 }
339 
340 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
341 {
342 	seq_puts(m, "Cpus_allowed:\t");
343 	seq_cpumask(m, &task->cpus_allowed);
344 	seq_putc(m, '\n');
345 	seq_puts(m, "Cpus_allowed_list:\t");
346 	seq_cpumask_list(m, &task->cpus_allowed);
347 	seq_putc(m, '\n');
348 }
349 
350 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
351 			struct pid *pid, struct task_struct *task)
352 {
353 	struct mm_struct *mm = get_task_mm(task);
354 
355 	task_name(m, task);
356 	task_state(m, ns, pid, task);
357 
358 	if (mm) {
359 		task_mem(m, mm);
360 		mmput(mm);
361 	}
362 	task_sig(m, task);
363 	task_cap(m, task);
364 	task_seccomp(m, task);
365 	task_cpus_allowed(m, task);
366 	cpuset_task_status_allowed(m, task);
367 	task_context_switch_counts(m, task);
368 	return 0;
369 }
370 
371 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
372 			struct pid *pid, struct task_struct *task, int whole)
373 {
374 	unsigned long vsize, eip, esp, wchan = ~0UL;
375 	int priority, nice;
376 	int tty_pgrp = -1, tty_nr = 0;
377 	sigset_t sigign, sigcatch;
378 	char state;
379 	pid_t ppid = 0, pgid = -1, sid = -1;
380 	int num_threads = 0;
381 	int permitted;
382 	struct mm_struct *mm;
383 	unsigned long long start_time;
384 	unsigned long cmin_flt = 0, cmaj_flt = 0;
385 	unsigned long  min_flt = 0,  maj_flt = 0;
386 	cputime_t cutime, cstime, utime, stime;
387 	cputime_t cgtime, gtime;
388 	unsigned long rsslim = 0;
389 	char tcomm[sizeof(task->comm)];
390 	unsigned long flags;
391 
392 	state = *get_task_state(task);
393 	vsize = eip = esp = 0;
394 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
395 	mm = get_task_mm(task);
396 	if (mm) {
397 		vsize = task_vsize(mm);
398 		if (permitted) {
399 			eip = KSTK_EIP(task);
400 			esp = KSTK_ESP(task);
401 		}
402 	}
403 
404 	get_task_comm(tcomm, task);
405 
406 	sigemptyset(&sigign);
407 	sigemptyset(&sigcatch);
408 	cutime = cstime = utime = stime = 0;
409 	cgtime = gtime = 0;
410 
411 	if (lock_task_sighand(task, &flags)) {
412 		struct signal_struct *sig = task->signal;
413 
414 		if (sig->tty) {
415 			struct pid *pgrp = tty_get_pgrp(sig->tty);
416 			tty_pgrp = pid_nr_ns(pgrp, ns);
417 			put_pid(pgrp);
418 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
419 		}
420 
421 		num_threads = get_nr_threads(task);
422 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
423 
424 		cmin_flt = sig->cmin_flt;
425 		cmaj_flt = sig->cmaj_flt;
426 		cutime = sig->cutime;
427 		cstime = sig->cstime;
428 		cgtime = sig->cgtime;
429 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
430 
431 		/* add up live thread stats at the group level */
432 		if (whole) {
433 			struct task_struct *t = task;
434 			do {
435 				min_flt += t->min_flt;
436 				maj_flt += t->maj_flt;
437 				gtime += task_gtime(t);
438 			} while_each_thread(task, t);
439 
440 			min_flt += sig->min_flt;
441 			maj_flt += sig->maj_flt;
442 			thread_group_cputime_adjusted(task, &utime, &stime);
443 			gtime += sig->gtime;
444 		}
445 
446 		sid = task_session_nr_ns(task, ns);
447 		ppid = task_tgid_nr_ns(task->real_parent, ns);
448 		pgid = task_pgrp_nr_ns(task, ns);
449 
450 		unlock_task_sighand(task, &flags);
451 	}
452 
453 	if (permitted && (!whole || num_threads < 2))
454 		wchan = get_wchan(task);
455 	if (!whole) {
456 		min_flt = task->min_flt;
457 		maj_flt = task->maj_flt;
458 		task_cputime_adjusted(task, &utime, &stime);
459 		gtime = task_gtime(task);
460 	}
461 
462 	/* scale priority and nice values from timeslices to -20..20 */
463 	/* to make it look like a "normal" Unix priority/nice value  */
464 	priority = task_prio(task);
465 	nice = task_nice(task);
466 
467 	/* convert nsec -> ticks */
468 	start_time = nsec_to_clock_t(task->real_start_time);
469 
470 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
471 	seq_put_decimal_ll(m, ' ', ppid);
472 	seq_put_decimal_ll(m, ' ', pgid);
473 	seq_put_decimal_ll(m, ' ', sid);
474 	seq_put_decimal_ll(m, ' ', tty_nr);
475 	seq_put_decimal_ll(m, ' ', tty_pgrp);
476 	seq_put_decimal_ull(m, ' ', task->flags);
477 	seq_put_decimal_ull(m, ' ', min_flt);
478 	seq_put_decimal_ull(m, ' ', cmin_flt);
479 	seq_put_decimal_ull(m, ' ', maj_flt);
480 	seq_put_decimal_ull(m, ' ', cmaj_flt);
481 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
482 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
483 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
484 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
485 	seq_put_decimal_ll(m, ' ', priority);
486 	seq_put_decimal_ll(m, ' ', nice);
487 	seq_put_decimal_ll(m, ' ', num_threads);
488 	seq_put_decimal_ull(m, ' ', 0);
489 	seq_put_decimal_ull(m, ' ', start_time);
490 	seq_put_decimal_ull(m, ' ', vsize);
491 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
492 	seq_put_decimal_ull(m, ' ', rsslim);
493 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
494 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
495 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
496 	seq_put_decimal_ull(m, ' ', esp);
497 	seq_put_decimal_ull(m, ' ', eip);
498 	/* The signal information here is obsolete.
499 	 * It must be decimal for Linux 2.0 compatibility.
500 	 * Use /proc/#/status for real-time signals.
501 	 */
502 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
503 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
504 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
505 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
506 	seq_put_decimal_ull(m, ' ', wchan);
507 	seq_put_decimal_ull(m, ' ', 0);
508 	seq_put_decimal_ull(m, ' ', 0);
509 	seq_put_decimal_ll(m, ' ', task->exit_signal);
510 	seq_put_decimal_ll(m, ' ', task_cpu(task));
511 	seq_put_decimal_ull(m, ' ', task->rt_priority);
512 	seq_put_decimal_ull(m, ' ', task->policy);
513 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
514 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
515 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
516 
517 	if (mm && permitted) {
518 		seq_put_decimal_ull(m, ' ', mm->start_data);
519 		seq_put_decimal_ull(m, ' ', mm->end_data);
520 		seq_put_decimal_ull(m, ' ', mm->start_brk);
521 		seq_put_decimal_ull(m, ' ', mm->arg_start);
522 		seq_put_decimal_ull(m, ' ', mm->arg_end);
523 		seq_put_decimal_ull(m, ' ', mm->env_start);
524 		seq_put_decimal_ull(m, ' ', mm->env_end);
525 	} else
526 		seq_printf(m, " 0 0 0 0 0 0 0");
527 
528 	if (permitted)
529 		seq_put_decimal_ll(m, ' ', task->exit_code);
530 	else
531 		seq_put_decimal_ll(m, ' ', 0);
532 
533 	seq_putc(m, '\n');
534 	if (mm)
535 		mmput(mm);
536 	return 0;
537 }
538 
539 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
540 			struct pid *pid, struct task_struct *task)
541 {
542 	return do_task_stat(m, ns, pid, task, 0);
543 }
544 
545 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
546 			struct pid *pid, struct task_struct *task)
547 {
548 	return do_task_stat(m, ns, pid, task, 1);
549 }
550 
551 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
552 			struct pid *pid, struct task_struct *task)
553 {
554 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
555 	struct mm_struct *mm = get_task_mm(task);
556 
557 	if (mm) {
558 		size = task_statm(mm, &shared, &text, &data, &resident);
559 		mmput(mm);
560 	}
561 	/*
562 	 * For quick read, open code by putting numbers directly
563 	 * expected format is
564 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
565 	 *               size, resident, shared, text, data);
566 	 */
567 	seq_put_decimal_ull(m, 0, size);
568 	seq_put_decimal_ull(m, ' ', resident);
569 	seq_put_decimal_ull(m, ' ', shared);
570 	seq_put_decimal_ull(m, ' ', text);
571 	seq_put_decimal_ull(m, ' ', 0);
572 	seq_put_decimal_ull(m, ' ', data);
573 	seq_put_decimal_ull(m, ' ', 0);
574 	seq_putc(m, '\n');
575 
576 	return 0;
577 }
578 
579 #ifdef CONFIG_CHECKPOINT_RESTORE
580 static struct pid *
581 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
582 {
583 	struct task_struct *start, *task;
584 	struct pid *pid = NULL;
585 
586 	read_lock(&tasklist_lock);
587 
588 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
589 	if (!start)
590 		goto out;
591 
592 	/*
593 	 * Lets try to continue searching first, this gives
594 	 * us significant speedup on children-rich processes.
595 	 */
596 	if (pid_prev) {
597 		task = pid_task(pid_prev, PIDTYPE_PID);
598 		if (task && task->real_parent == start &&
599 		    !(list_empty(&task->sibling))) {
600 			if (list_is_last(&task->sibling, &start->children))
601 				goto out;
602 			task = list_first_entry(&task->sibling,
603 						struct task_struct, sibling);
604 			pid = get_pid(task_pid(task));
605 			goto out;
606 		}
607 	}
608 
609 	/*
610 	 * Slow search case.
611 	 *
612 	 * We might miss some children here if children
613 	 * are exited while we were not holding the lock,
614 	 * but it was never promised to be accurate that
615 	 * much.
616 	 *
617 	 * "Just suppose that the parent sleeps, but N children
618 	 *  exit after we printed their tids. Now the slow paths
619 	 *  skips N extra children, we miss N tasks." (c)
620 	 *
621 	 * So one need to stop or freeze the leader and all
622 	 * its children to get a precise result.
623 	 */
624 	list_for_each_entry(task, &start->children, sibling) {
625 		if (pos-- == 0) {
626 			pid = get_pid(task_pid(task));
627 			break;
628 		}
629 	}
630 
631 out:
632 	read_unlock(&tasklist_lock);
633 	return pid;
634 }
635 
636 static int children_seq_show(struct seq_file *seq, void *v)
637 {
638 	struct inode *inode = seq->private;
639 	pid_t pid;
640 
641 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
642 	return seq_printf(seq, "%d ", pid);
643 }
644 
645 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
646 {
647 	return get_children_pid(seq->private, NULL, *pos);
648 }
649 
650 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
651 {
652 	struct pid *pid;
653 
654 	pid = get_children_pid(seq->private, v, *pos + 1);
655 	put_pid(v);
656 
657 	++*pos;
658 	return pid;
659 }
660 
661 static void children_seq_stop(struct seq_file *seq, void *v)
662 {
663 	put_pid(v);
664 }
665 
666 static const struct seq_operations children_seq_ops = {
667 	.start	= children_seq_start,
668 	.next	= children_seq_next,
669 	.stop	= children_seq_stop,
670 	.show	= children_seq_show,
671 };
672 
673 static int children_seq_open(struct inode *inode, struct file *file)
674 {
675 	struct seq_file *m;
676 	int ret;
677 
678 	ret = seq_open(file, &children_seq_ops);
679 	if (ret)
680 		return ret;
681 
682 	m = file->private_data;
683 	m->private = inode;
684 
685 	return ret;
686 }
687 
688 int children_seq_release(struct inode *inode, struct file *file)
689 {
690 	seq_release(inode, file);
691 	return 0;
692 }
693 
694 const struct file_operations proc_tid_children_operations = {
695 	.open    = children_seq_open,
696 	.read    = seq_read,
697 	.llseek  = seq_lseek,
698 	.release = children_seq_release,
699 };
700 #endif /* CONFIG_CHECKPOINT_RESTORE */
701