xref: /openbmc/linux/fs/proc/array.c (revision 94c7b6fc)
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[(_KERNEL_CAPABILITY_U32S-1) - __capi]);
301 	}
302 	seq_putc(m, '\n');
303 }
304 
305 /* Remove non-existent capabilities */
306 #define NORM_CAPS(v) (v.cap[CAP_TO_INDEX(CAP_LAST_CAP)] &= \
307 				CAP_TO_MASK(CAP_LAST_CAP + 1) - 1)
308 
309 static inline void task_cap(struct seq_file *m, struct task_struct *p)
310 {
311 	const struct cred *cred;
312 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
313 
314 	rcu_read_lock();
315 	cred = __task_cred(p);
316 	cap_inheritable	= cred->cap_inheritable;
317 	cap_permitted	= cred->cap_permitted;
318 	cap_effective	= cred->cap_effective;
319 	cap_bset	= cred->cap_bset;
320 	rcu_read_unlock();
321 
322 	NORM_CAPS(cap_inheritable);
323 	NORM_CAPS(cap_permitted);
324 	NORM_CAPS(cap_effective);
325 	NORM_CAPS(cap_bset);
326 
327 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
328 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
329 	render_cap_t(m, "CapEff:\t", &cap_effective);
330 	render_cap_t(m, "CapBnd:\t", &cap_bset);
331 }
332 
333 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
334 {
335 #ifdef CONFIG_SECCOMP
336 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
337 #endif
338 }
339 
340 static inline void task_context_switch_counts(struct seq_file *m,
341 						struct task_struct *p)
342 {
343 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
344 			"nonvoluntary_ctxt_switches:\t%lu\n",
345 			p->nvcsw,
346 			p->nivcsw);
347 }
348 
349 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
350 {
351 	seq_puts(m, "Cpus_allowed:\t");
352 	seq_cpumask(m, &task->cpus_allowed);
353 	seq_putc(m, '\n');
354 	seq_puts(m, "Cpus_allowed_list:\t");
355 	seq_cpumask_list(m, &task->cpus_allowed);
356 	seq_putc(m, '\n');
357 }
358 
359 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
360 			struct pid *pid, struct task_struct *task)
361 {
362 	struct mm_struct *mm = get_task_mm(task);
363 
364 	task_name(m, task);
365 	task_state(m, ns, pid, task);
366 
367 	if (mm) {
368 		task_mem(m, mm);
369 		mmput(mm);
370 	}
371 	task_sig(m, task);
372 	task_cap(m, task);
373 	task_seccomp(m, task);
374 	task_cpus_allowed(m, task);
375 	cpuset_task_status_allowed(m, task);
376 	task_context_switch_counts(m, task);
377 	return 0;
378 }
379 
380 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
381 			struct pid *pid, struct task_struct *task, int whole)
382 {
383 	unsigned long vsize, eip, esp, wchan = ~0UL;
384 	int priority, nice;
385 	int tty_pgrp = -1, tty_nr = 0;
386 	sigset_t sigign, sigcatch;
387 	char state;
388 	pid_t ppid = 0, pgid = -1, sid = -1;
389 	int num_threads = 0;
390 	int permitted;
391 	struct mm_struct *mm;
392 	unsigned long long start_time;
393 	unsigned long cmin_flt = 0, cmaj_flt = 0;
394 	unsigned long  min_flt = 0,  maj_flt = 0;
395 	cputime_t cutime, cstime, utime, stime;
396 	cputime_t cgtime, gtime;
397 	unsigned long rsslim = 0;
398 	char tcomm[sizeof(task->comm)];
399 	unsigned long flags;
400 
401 	state = *get_task_state(task);
402 	vsize = eip = esp = 0;
403 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
404 	mm = get_task_mm(task);
405 	if (mm) {
406 		vsize = task_vsize(mm);
407 		if (permitted) {
408 			eip = KSTK_EIP(task);
409 			esp = KSTK_ESP(task);
410 		}
411 	}
412 
413 	get_task_comm(tcomm, task);
414 
415 	sigemptyset(&sigign);
416 	sigemptyset(&sigcatch);
417 	cutime = cstime = utime = stime = 0;
418 	cgtime = gtime = 0;
419 
420 	if (lock_task_sighand(task, &flags)) {
421 		struct signal_struct *sig = task->signal;
422 
423 		if (sig->tty) {
424 			struct pid *pgrp = tty_get_pgrp(sig->tty);
425 			tty_pgrp = pid_nr_ns(pgrp, ns);
426 			put_pid(pgrp);
427 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
428 		}
429 
430 		num_threads = get_nr_threads(task);
431 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
432 
433 		cmin_flt = sig->cmin_flt;
434 		cmaj_flt = sig->cmaj_flt;
435 		cutime = sig->cutime;
436 		cstime = sig->cstime;
437 		cgtime = sig->cgtime;
438 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
439 
440 		/* add up live thread stats at the group level */
441 		if (whole) {
442 			struct task_struct *t = task;
443 			do {
444 				min_flt += t->min_flt;
445 				maj_flt += t->maj_flt;
446 				gtime += task_gtime(t);
447 			} while_each_thread(task, t);
448 
449 			min_flt += sig->min_flt;
450 			maj_flt += sig->maj_flt;
451 			thread_group_cputime_adjusted(task, &utime, &stime);
452 			gtime += sig->gtime;
453 		}
454 
455 		sid = task_session_nr_ns(task, ns);
456 		ppid = task_tgid_nr_ns(task->real_parent, ns);
457 		pgid = task_pgrp_nr_ns(task, ns);
458 
459 		unlock_task_sighand(task, &flags);
460 	}
461 
462 	if (permitted && (!whole || num_threads < 2))
463 		wchan = get_wchan(task);
464 	if (!whole) {
465 		min_flt = task->min_flt;
466 		maj_flt = task->maj_flt;
467 		task_cputime_adjusted(task, &utime, &stime);
468 		gtime = task_gtime(task);
469 	}
470 
471 	/* scale priority and nice values from timeslices to -20..20 */
472 	/* to make it look like a "normal" Unix priority/nice value  */
473 	priority = task_prio(task);
474 	nice = task_nice(task);
475 
476 	/* Temporary variable needed for gcc-2.96 */
477 	/* convert timespec -> nsec*/
478 	start_time =
479 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
480 				+ task->real_start_time.tv_nsec;
481 	/* convert nsec -> ticks */
482 	start_time = nsec_to_clock_t(start_time);
483 
484 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
485 	seq_put_decimal_ll(m, ' ', ppid);
486 	seq_put_decimal_ll(m, ' ', pgid);
487 	seq_put_decimal_ll(m, ' ', sid);
488 	seq_put_decimal_ll(m, ' ', tty_nr);
489 	seq_put_decimal_ll(m, ' ', tty_pgrp);
490 	seq_put_decimal_ull(m, ' ', task->flags);
491 	seq_put_decimal_ull(m, ' ', min_flt);
492 	seq_put_decimal_ull(m, ' ', cmin_flt);
493 	seq_put_decimal_ull(m, ' ', maj_flt);
494 	seq_put_decimal_ull(m, ' ', cmaj_flt);
495 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
496 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
497 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
498 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
499 	seq_put_decimal_ll(m, ' ', priority);
500 	seq_put_decimal_ll(m, ' ', nice);
501 	seq_put_decimal_ll(m, ' ', num_threads);
502 	seq_put_decimal_ull(m, ' ', 0);
503 	seq_put_decimal_ull(m, ' ', start_time);
504 	seq_put_decimal_ull(m, ' ', vsize);
505 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
506 	seq_put_decimal_ull(m, ' ', rsslim);
507 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
508 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
509 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
510 	seq_put_decimal_ull(m, ' ', esp);
511 	seq_put_decimal_ull(m, ' ', eip);
512 	/* The signal information here is obsolete.
513 	 * It must be decimal for Linux 2.0 compatibility.
514 	 * Use /proc/#/status for real-time signals.
515 	 */
516 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
517 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
518 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
519 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
520 	seq_put_decimal_ull(m, ' ', wchan);
521 	seq_put_decimal_ull(m, ' ', 0);
522 	seq_put_decimal_ull(m, ' ', 0);
523 	seq_put_decimal_ll(m, ' ', task->exit_signal);
524 	seq_put_decimal_ll(m, ' ', task_cpu(task));
525 	seq_put_decimal_ull(m, ' ', task->rt_priority);
526 	seq_put_decimal_ull(m, ' ', task->policy);
527 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
528 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
529 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
530 
531 	if (mm && permitted) {
532 		seq_put_decimal_ull(m, ' ', mm->start_data);
533 		seq_put_decimal_ull(m, ' ', mm->end_data);
534 		seq_put_decimal_ull(m, ' ', mm->start_brk);
535 		seq_put_decimal_ull(m, ' ', mm->arg_start);
536 		seq_put_decimal_ull(m, ' ', mm->arg_end);
537 		seq_put_decimal_ull(m, ' ', mm->env_start);
538 		seq_put_decimal_ull(m, ' ', mm->env_end);
539 	} else
540 		seq_printf(m, " 0 0 0 0 0 0 0");
541 
542 	if (permitted)
543 		seq_put_decimal_ll(m, ' ', task->exit_code);
544 	else
545 		seq_put_decimal_ll(m, ' ', 0);
546 
547 	seq_putc(m, '\n');
548 	if (mm)
549 		mmput(mm);
550 	return 0;
551 }
552 
553 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
554 			struct pid *pid, struct task_struct *task)
555 {
556 	return do_task_stat(m, ns, pid, task, 0);
557 }
558 
559 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
560 			struct pid *pid, struct task_struct *task)
561 {
562 	return do_task_stat(m, ns, pid, task, 1);
563 }
564 
565 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
566 			struct pid *pid, struct task_struct *task)
567 {
568 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
569 	struct mm_struct *mm = get_task_mm(task);
570 
571 	if (mm) {
572 		size = task_statm(mm, &shared, &text, &data, &resident);
573 		mmput(mm);
574 	}
575 	/*
576 	 * For quick read, open code by putting numbers directly
577 	 * expected format is
578 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
579 	 *               size, resident, shared, text, data);
580 	 */
581 	seq_put_decimal_ull(m, 0, size);
582 	seq_put_decimal_ull(m, ' ', resident);
583 	seq_put_decimal_ull(m, ' ', shared);
584 	seq_put_decimal_ull(m, ' ', text);
585 	seq_put_decimal_ull(m, ' ', 0);
586 	seq_put_decimal_ull(m, ' ', data);
587 	seq_put_decimal_ull(m, ' ', 0);
588 	seq_putc(m, '\n');
589 
590 	return 0;
591 }
592 
593 #ifdef CONFIG_CHECKPOINT_RESTORE
594 static struct pid *
595 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
596 {
597 	struct task_struct *start, *task;
598 	struct pid *pid = NULL;
599 
600 	read_lock(&tasklist_lock);
601 
602 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
603 	if (!start)
604 		goto out;
605 
606 	/*
607 	 * Lets try to continue searching first, this gives
608 	 * us significant speedup on children-rich processes.
609 	 */
610 	if (pid_prev) {
611 		task = pid_task(pid_prev, PIDTYPE_PID);
612 		if (task && task->real_parent == start &&
613 		    !(list_empty(&task->sibling))) {
614 			if (list_is_last(&task->sibling, &start->children))
615 				goto out;
616 			task = list_first_entry(&task->sibling,
617 						struct task_struct, sibling);
618 			pid = get_pid(task_pid(task));
619 			goto out;
620 		}
621 	}
622 
623 	/*
624 	 * Slow search case.
625 	 *
626 	 * We might miss some children here if children
627 	 * are exited while we were not holding the lock,
628 	 * but it was never promised to be accurate that
629 	 * much.
630 	 *
631 	 * "Just suppose that the parent sleeps, but N children
632 	 *  exit after we printed their tids. Now the slow paths
633 	 *  skips N extra children, we miss N tasks." (c)
634 	 *
635 	 * So one need to stop or freeze the leader and all
636 	 * its children to get a precise result.
637 	 */
638 	list_for_each_entry(task, &start->children, sibling) {
639 		if (pos-- == 0) {
640 			pid = get_pid(task_pid(task));
641 			break;
642 		}
643 	}
644 
645 out:
646 	read_unlock(&tasklist_lock);
647 	return pid;
648 }
649 
650 static int children_seq_show(struct seq_file *seq, void *v)
651 {
652 	struct inode *inode = seq->private;
653 	pid_t pid;
654 
655 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
656 	return seq_printf(seq, "%d ", pid);
657 }
658 
659 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
660 {
661 	return get_children_pid(seq->private, NULL, *pos);
662 }
663 
664 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
665 {
666 	struct pid *pid;
667 
668 	pid = get_children_pid(seq->private, v, *pos + 1);
669 	put_pid(v);
670 
671 	++*pos;
672 	return pid;
673 }
674 
675 static void children_seq_stop(struct seq_file *seq, void *v)
676 {
677 	put_pid(v);
678 }
679 
680 static const struct seq_operations children_seq_ops = {
681 	.start	= children_seq_start,
682 	.next	= children_seq_next,
683 	.stop	= children_seq_stop,
684 	.show	= children_seq_show,
685 };
686 
687 static int children_seq_open(struct inode *inode, struct file *file)
688 {
689 	struct seq_file *m;
690 	int ret;
691 
692 	ret = seq_open(file, &children_seq_ops);
693 	if (ret)
694 		return ret;
695 
696 	m = file->private_data;
697 	m->private = inode;
698 
699 	return ret;
700 }
701 
702 int children_seq_release(struct inode *inode, struct file *file)
703 {
704 	seq_release(inode, file);
705 	return 0;
706 }
707 
708 const struct file_operations proc_tid_children_operations = {
709 	.open    = children_seq_open,
710 	.read    = seq_read,
711 	.llseek  = seq_lseek,
712 	.release = children_seq_release,
713 };
714 #endif /* CONFIG_CHECKPOINT_RESTORE */
715