xref: /linux/fs/proc/array.c (revision 5e8c0fb6a95728b852d56c0a9244425d474670c0)
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 task_struct *tracer;
161 	const struct cred *cred;
162 	pid_t ppid, tpid = 0, tgid, ngid;
163 	unsigned int max_fds = 0;
164 
165 	rcu_read_lock();
166 	ppid = pid_alive(p) ?
167 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
168 
169 	tracer = ptrace_parent(p);
170 	if (tracer)
171 		tpid = task_pid_nr_ns(tracer, ns);
172 
173 	tgid = task_tgid_nr_ns(p, ns);
174 	ngid = task_numa_group_id(p);
175 	cred = get_task_cred(p);
176 
177 	task_lock(p);
178 	if (p->files)
179 		max_fds = files_fdtable(p->files)->max_fds;
180 	task_unlock(p);
181 	rcu_read_unlock();
182 
183 	seq_printf(m,
184 		"State:\t%s\n"
185 		"Tgid:\t%d\n"
186 		"Ngid:\t%d\n"
187 		"Pid:\t%d\n"
188 		"PPid:\t%d\n"
189 		"TracerPid:\t%d\n"
190 		"Uid:\t%d\t%d\t%d\t%d\n"
191 		"Gid:\t%d\t%d\t%d\t%d\n"
192 		"FDSize:\t%d\nGroups:\t",
193 		get_task_state(p),
194 		tgid, ngid, pid_nr_ns(pid, ns), ppid, tpid,
195 		from_kuid_munged(user_ns, cred->uid),
196 		from_kuid_munged(user_ns, cred->euid),
197 		from_kuid_munged(user_ns, cred->suid),
198 		from_kuid_munged(user_ns, cred->fsuid),
199 		from_kgid_munged(user_ns, cred->gid),
200 		from_kgid_munged(user_ns, cred->egid),
201 		from_kgid_munged(user_ns, cred->sgid),
202 		from_kgid_munged(user_ns, cred->fsgid),
203 		max_fds);
204 
205 	group_info = cred->group_info;
206 	for (g = 0; g < group_info->ngroups; g++)
207 		seq_printf(m, "%d ",
208 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
209 	put_cred(cred);
210 
211 	seq_putc(m, '\n');
212 }
213 
214 void render_sigset_t(struct seq_file *m, const char *header,
215 				sigset_t *set)
216 {
217 	int i;
218 
219 	seq_puts(m, header);
220 
221 	i = _NSIG;
222 	do {
223 		int x = 0;
224 
225 		i -= 4;
226 		if (sigismember(set, i+1)) x |= 1;
227 		if (sigismember(set, i+2)) x |= 2;
228 		if (sigismember(set, i+3)) x |= 4;
229 		if (sigismember(set, i+4)) x |= 8;
230 		seq_printf(m, "%x", x);
231 	} while (i >= 4);
232 
233 	seq_putc(m, '\n');
234 }
235 
236 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
237 				    sigset_t *catch)
238 {
239 	struct k_sigaction *k;
240 	int i;
241 
242 	k = p->sighand->action;
243 	for (i = 1; i <= _NSIG; ++i, ++k) {
244 		if (k->sa.sa_handler == SIG_IGN)
245 			sigaddset(ign, i);
246 		else if (k->sa.sa_handler != SIG_DFL)
247 			sigaddset(catch, i);
248 	}
249 }
250 
251 static inline void task_sig(struct seq_file *m, struct task_struct *p)
252 {
253 	unsigned long flags;
254 	sigset_t pending, shpending, blocked, ignored, caught;
255 	int num_threads = 0;
256 	unsigned long qsize = 0;
257 	unsigned long qlim = 0;
258 
259 	sigemptyset(&pending);
260 	sigemptyset(&shpending);
261 	sigemptyset(&blocked);
262 	sigemptyset(&ignored);
263 	sigemptyset(&caught);
264 
265 	if (lock_task_sighand(p, &flags)) {
266 		pending = p->pending.signal;
267 		shpending = p->signal->shared_pending.signal;
268 		blocked = p->blocked;
269 		collect_sigign_sigcatch(p, &ignored, &caught);
270 		num_threads = get_nr_threads(p);
271 		rcu_read_lock();  /* FIXME: is this correct? */
272 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
273 		rcu_read_unlock();
274 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
275 		unlock_task_sighand(p, &flags);
276 	}
277 
278 	seq_printf(m, "Threads:\t%d\n", num_threads);
279 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
280 
281 	/* render them all */
282 	render_sigset_t(m, "SigPnd:\t", &pending);
283 	render_sigset_t(m, "ShdPnd:\t", &shpending);
284 	render_sigset_t(m, "SigBlk:\t", &blocked);
285 	render_sigset_t(m, "SigIgn:\t", &ignored);
286 	render_sigset_t(m, "SigCgt:\t", &caught);
287 }
288 
289 static void render_cap_t(struct seq_file *m, const char *header,
290 			kernel_cap_t *a)
291 {
292 	unsigned __capi;
293 
294 	seq_puts(m, header);
295 	CAP_FOR_EACH_U32(__capi) {
296 		seq_printf(m, "%08x",
297 			   a->cap[CAP_LAST_U32 - __capi]);
298 	}
299 	seq_putc(m, '\n');
300 }
301 
302 static inline void task_cap(struct seq_file *m, struct task_struct *p)
303 {
304 	const struct cred *cred;
305 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
306 
307 	rcu_read_lock();
308 	cred = __task_cred(p);
309 	cap_inheritable	= cred->cap_inheritable;
310 	cap_permitted	= cred->cap_permitted;
311 	cap_effective	= cred->cap_effective;
312 	cap_bset	= cred->cap_bset;
313 	rcu_read_unlock();
314 
315 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
316 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
317 	render_cap_t(m, "CapEff:\t", &cap_effective);
318 	render_cap_t(m, "CapBnd:\t", &cap_bset);
319 }
320 
321 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
322 {
323 #ifdef CONFIG_SECCOMP
324 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
325 #endif
326 }
327 
328 static inline void task_context_switch_counts(struct seq_file *m,
329 						struct task_struct *p)
330 {
331 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
332 			"nonvoluntary_ctxt_switches:\t%lu\n",
333 			p->nvcsw,
334 			p->nivcsw);
335 }
336 
337 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
338 {
339 	seq_puts(m, "Cpus_allowed:\t");
340 	seq_cpumask(m, &task->cpus_allowed);
341 	seq_putc(m, '\n');
342 	seq_puts(m, "Cpus_allowed_list:\t");
343 	seq_cpumask_list(m, &task->cpus_allowed);
344 	seq_putc(m, '\n');
345 }
346 
347 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
348 			struct pid *pid, struct task_struct *task)
349 {
350 	struct mm_struct *mm = get_task_mm(task);
351 
352 	task_name(m, task);
353 	task_state(m, ns, pid, task);
354 
355 	if (mm) {
356 		task_mem(m, mm);
357 		mmput(mm);
358 	}
359 	task_sig(m, task);
360 	task_cap(m, task);
361 	task_seccomp(m, task);
362 	task_cpus_allowed(m, task);
363 	cpuset_task_status_allowed(m, task);
364 	task_context_switch_counts(m, task);
365 	return 0;
366 }
367 
368 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
369 			struct pid *pid, struct task_struct *task, int whole)
370 {
371 	unsigned long vsize, eip, esp, wchan = ~0UL;
372 	int priority, nice;
373 	int tty_pgrp = -1, tty_nr = 0;
374 	sigset_t sigign, sigcatch;
375 	char state;
376 	pid_t ppid = 0, pgid = -1, sid = -1;
377 	int num_threads = 0;
378 	int permitted;
379 	struct mm_struct *mm;
380 	unsigned long long start_time;
381 	unsigned long cmin_flt = 0, cmaj_flt = 0;
382 	unsigned long  min_flt = 0,  maj_flt = 0;
383 	cputime_t cutime, cstime, utime, stime;
384 	cputime_t cgtime, gtime;
385 	unsigned long rsslim = 0;
386 	char tcomm[sizeof(task->comm)];
387 	unsigned long flags;
388 
389 	state = *get_task_state(task);
390 	vsize = eip = esp = 0;
391 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
392 	mm = get_task_mm(task);
393 	if (mm) {
394 		vsize = task_vsize(mm);
395 		if (permitted) {
396 			eip = KSTK_EIP(task);
397 			esp = KSTK_ESP(task);
398 		}
399 	}
400 
401 	get_task_comm(tcomm, task);
402 
403 	sigemptyset(&sigign);
404 	sigemptyset(&sigcatch);
405 	cutime = cstime = utime = stime = 0;
406 	cgtime = gtime = 0;
407 
408 	if (lock_task_sighand(task, &flags)) {
409 		struct signal_struct *sig = task->signal;
410 
411 		if (sig->tty) {
412 			struct pid *pgrp = tty_get_pgrp(sig->tty);
413 			tty_pgrp = pid_nr_ns(pgrp, ns);
414 			put_pid(pgrp);
415 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
416 		}
417 
418 		num_threads = get_nr_threads(task);
419 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
420 
421 		cmin_flt = sig->cmin_flt;
422 		cmaj_flt = sig->cmaj_flt;
423 		cutime = sig->cutime;
424 		cstime = sig->cstime;
425 		cgtime = sig->cgtime;
426 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
427 
428 		/* add up live thread stats at the group level */
429 		if (whole) {
430 			struct task_struct *t = task;
431 			do {
432 				min_flt += t->min_flt;
433 				maj_flt += t->maj_flt;
434 				gtime += task_gtime(t);
435 			} while_each_thread(task, t);
436 
437 			min_flt += sig->min_flt;
438 			maj_flt += sig->maj_flt;
439 			thread_group_cputime_adjusted(task, &utime, &stime);
440 			gtime += sig->gtime;
441 		}
442 
443 		sid = task_session_nr_ns(task, ns);
444 		ppid = task_tgid_nr_ns(task->real_parent, ns);
445 		pgid = task_pgrp_nr_ns(task, ns);
446 
447 		unlock_task_sighand(task, &flags);
448 	}
449 
450 	if (permitted && (!whole || num_threads < 2))
451 		wchan = get_wchan(task);
452 	if (!whole) {
453 		min_flt = task->min_flt;
454 		maj_flt = task->maj_flt;
455 		task_cputime_adjusted(task, &utime, &stime);
456 		gtime = task_gtime(task);
457 	}
458 
459 	/* scale priority and nice values from timeslices to -20..20 */
460 	/* to make it look like a "normal" Unix priority/nice value  */
461 	priority = task_prio(task);
462 	nice = task_nice(task);
463 
464 	/* convert nsec -> ticks */
465 	start_time = nsec_to_clock_t(task->real_start_time);
466 
467 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
468 	seq_put_decimal_ll(m, ' ', ppid);
469 	seq_put_decimal_ll(m, ' ', pgid);
470 	seq_put_decimal_ll(m, ' ', sid);
471 	seq_put_decimal_ll(m, ' ', tty_nr);
472 	seq_put_decimal_ll(m, ' ', tty_pgrp);
473 	seq_put_decimal_ull(m, ' ', task->flags);
474 	seq_put_decimal_ull(m, ' ', min_flt);
475 	seq_put_decimal_ull(m, ' ', cmin_flt);
476 	seq_put_decimal_ull(m, ' ', maj_flt);
477 	seq_put_decimal_ull(m, ' ', cmaj_flt);
478 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
479 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
480 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
481 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
482 	seq_put_decimal_ll(m, ' ', priority);
483 	seq_put_decimal_ll(m, ' ', nice);
484 	seq_put_decimal_ll(m, ' ', num_threads);
485 	seq_put_decimal_ull(m, ' ', 0);
486 	seq_put_decimal_ull(m, ' ', start_time);
487 	seq_put_decimal_ull(m, ' ', vsize);
488 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
489 	seq_put_decimal_ull(m, ' ', rsslim);
490 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
491 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
492 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
493 	seq_put_decimal_ull(m, ' ', esp);
494 	seq_put_decimal_ull(m, ' ', eip);
495 	/* The signal information here is obsolete.
496 	 * It must be decimal for Linux 2.0 compatibility.
497 	 * Use /proc/#/status for real-time signals.
498 	 */
499 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
500 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
501 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
502 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
503 	seq_put_decimal_ull(m, ' ', wchan);
504 	seq_put_decimal_ull(m, ' ', 0);
505 	seq_put_decimal_ull(m, ' ', 0);
506 	seq_put_decimal_ll(m, ' ', task->exit_signal);
507 	seq_put_decimal_ll(m, ' ', task_cpu(task));
508 	seq_put_decimal_ull(m, ' ', task->rt_priority);
509 	seq_put_decimal_ull(m, ' ', task->policy);
510 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
511 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
512 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
513 
514 	if (mm && permitted) {
515 		seq_put_decimal_ull(m, ' ', mm->start_data);
516 		seq_put_decimal_ull(m, ' ', mm->end_data);
517 		seq_put_decimal_ull(m, ' ', mm->start_brk);
518 		seq_put_decimal_ull(m, ' ', mm->arg_start);
519 		seq_put_decimal_ull(m, ' ', mm->arg_end);
520 		seq_put_decimal_ull(m, ' ', mm->env_start);
521 		seq_put_decimal_ull(m, ' ', mm->env_end);
522 	} else
523 		seq_printf(m, " 0 0 0 0 0 0 0");
524 
525 	if (permitted)
526 		seq_put_decimal_ll(m, ' ', task->exit_code);
527 	else
528 		seq_put_decimal_ll(m, ' ', 0);
529 
530 	seq_putc(m, '\n');
531 	if (mm)
532 		mmput(mm);
533 	return 0;
534 }
535 
536 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
537 			struct pid *pid, struct task_struct *task)
538 {
539 	return do_task_stat(m, ns, pid, task, 0);
540 }
541 
542 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
543 			struct pid *pid, struct task_struct *task)
544 {
545 	return do_task_stat(m, ns, pid, task, 1);
546 }
547 
548 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
549 			struct pid *pid, struct task_struct *task)
550 {
551 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
552 	struct mm_struct *mm = get_task_mm(task);
553 
554 	if (mm) {
555 		size = task_statm(mm, &shared, &text, &data, &resident);
556 		mmput(mm);
557 	}
558 	/*
559 	 * For quick read, open code by putting numbers directly
560 	 * expected format is
561 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
562 	 *               size, resident, shared, text, data);
563 	 */
564 	seq_put_decimal_ull(m, 0, size);
565 	seq_put_decimal_ull(m, ' ', resident);
566 	seq_put_decimal_ull(m, ' ', shared);
567 	seq_put_decimal_ull(m, ' ', text);
568 	seq_put_decimal_ull(m, ' ', 0);
569 	seq_put_decimal_ull(m, ' ', data);
570 	seq_put_decimal_ull(m, ' ', 0);
571 	seq_putc(m, '\n');
572 
573 	return 0;
574 }
575 
576 #ifdef CONFIG_CHECKPOINT_RESTORE
577 static struct pid *
578 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
579 {
580 	struct task_struct *start, *task;
581 	struct pid *pid = NULL;
582 
583 	read_lock(&tasklist_lock);
584 
585 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
586 	if (!start)
587 		goto out;
588 
589 	/*
590 	 * Lets try to continue searching first, this gives
591 	 * us significant speedup on children-rich processes.
592 	 */
593 	if (pid_prev) {
594 		task = pid_task(pid_prev, PIDTYPE_PID);
595 		if (task && task->real_parent == start &&
596 		    !(list_empty(&task->sibling))) {
597 			if (list_is_last(&task->sibling, &start->children))
598 				goto out;
599 			task = list_first_entry(&task->sibling,
600 						struct task_struct, sibling);
601 			pid = get_pid(task_pid(task));
602 			goto out;
603 		}
604 	}
605 
606 	/*
607 	 * Slow search case.
608 	 *
609 	 * We might miss some children here if children
610 	 * are exited while we were not holding the lock,
611 	 * but it was never promised to be accurate that
612 	 * much.
613 	 *
614 	 * "Just suppose that the parent sleeps, but N children
615 	 *  exit after we printed their tids. Now the slow paths
616 	 *  skips N extra children, we miss N tasks." (c)
617 	 *
618 	 * So one need to stop or freeze the leader and all
619 	 * its children to get a precise result.
620 	 */
621 	list_for_each_entry(task, &start->children, sibling) {
622 		if (pos-- == 0) {
623 			pid = get_pid(task_pid(task));
624 			break;
625 		}
626 	}
627 
628 out:
629 	read_unlock(&tasklist_lock);
630 	return pid;
631 }
632 
633 static int children_seq_show(struct seq_file *seq, void *v)
634 {
635 	struct inode *inode = seq->private;
636 	pid_t pid;
637 
638 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
639 	return seq_printf(seq, "%d ", pid);
640 }
641 
642 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
643 {
644 	return get_children_pid(seq->private, NULL, *pos);
645 }
646 
647 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
648 {
649 	struct pid *pid;
650 
651 	pid = get_children_pid(seq->private, v, *pos + 1);
652 	put_pid(v);
653 
654 	++*pos;
655 	return pid;
656 }
657 
658 static void children_seq_stop(struct seq_file *seq, void *v)
659 {
660 	put_pid(v);
661 }
662 
663 static const struct seq_operations children_seq_ops = {
664 	.start	= children_seq_start,
665 	.next	= children_seq_next,
666 	.stop	= children_seq_stop,
667 	.show	= children_seq_show,
668 };
669 
670 static int children_seq_open(struct inode *inode, struct file *file)
671 {
672 	struct seq_file *m;
673 	int ret;
674 
675 	ret = seq_open(file, &children_seq_ops);
676 	if (ret)
677 		return ret;
678 
679 	m = file->private_data;
680 	m->private = inode;
681 
682 	return ret;
683 }
684 
685 int children_seq_release(struct inode *inode, struct file *file)
686 {
687 	seq_release(inode, file);
688 	return 0;
689 }
690 
691 const struct file_operations proc_tid_children_operations = {
692 	.open    = children_seq_open,
693 	.read    = seq_read,
694 	.llseek  = seq_lseek,
695 	.release = children_seq_release,
696 };
697 #endif /* CONFIG_CHECKPOINT_RESTORE */
698