xref: /openbmc/linux/fs/proc/array.c (revision c4ee0af3)
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 	"Z (zombie)",		/*  16 */
142 	"X (dead)",		/*  32 */
143 	"x (dead)",		/*  64 */
144 	"K (wakekill)",		/* 128 */
145 	"W (waking)",		/* 256 */
146 	"P (parked)",		/* 512 */
147 };
148 
149 static inline const char *get_task_state(struct task_struct *tsk)
150 {
151 	unsigned int state = (tsk->state & TASK_REPORT) | tsk->exit_state;
152 	const char * const *p = &task_state_array[0];
153 
154 	BUILD_BUG_ON(1 + ilog2(TASK_STATE_MAX) != ARRAY_SIZE(task_state_array));
155 
156 	while (state) {
157 		p++;
158 		state >>= 1;
159 	}
160 	return *p;
161 }
162 
163 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
164 				struct pid *pid, struct task_struct *p)
165 {
166 	struct user_namespace *user_ns = seq_user_ns(m);
167 	struct group_info *group_info;
168 	int g;
169 	struct fdtable *fdt = NULL;
170 	const struct cred *cred;
171 	pid_t ppid, tpid;
172 
173 	rcu_read_lock();
174 	ppid = pid_alive(p) ?
175 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
176 	tpid = 0;
177 	if (pid_alive(p)) {
178 		struct task_struct *tracer = ptrace_parent(p);
179 		if (tracer)
180 			tpid = task_pid_nr_ns(tracer, ns);
181 	}
182 	cred = get_task_cred(p);
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 		get_task_state(p),
193 		task_tgid_nr_ns(p, ns),
194 		task_numa_group_id(p),
195 		pid_nr_ns(pid, ns),
196 		ppid, tpid,
197 		from_kuid_munged(user_ns, cred->uid),
198 		from_kuid_munged(user_ns, cred->euid),
199 		from_kuid_munged(user_ns, cred->suid),
200 		from_kuid_munged(user_ns, cred->fsuid),
201 		from_kgid_munged(user_ns, cred->gid),
202 		from_kgid_munged(user_ns, cred->egid),
203 		from_kgid_munged(user_ns, cred->sgid),
204 		from_kgid_munged(user_ns, cred->fsgid));
205 
206 	task_lock(p);
207 	if (p->files)
208 		fdt = files_fdtable(p->files);
209 	seq_printf(m,
210 		"FDSize:\t%d\n"
211 		"Groups:\t",
212 		fdt ? fdt->max_fds : 0);
213 	rcu_read_unlock();
214 
215 	group_info = cred->group_info;
216 	task_unlock(p);
217 
218 	for (g = 0; g < group_info->ngroups; g++)
219 		seq_printf(m, "%d ",
220 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
221 	put_cred(cred);
222 
223 	seq_putc(m, '\n');
224 }
225 
226 void render_sigset_t(struct seq_file *m, const char *header,
227 				sigset_t *set)
228 {
229 	int i;
230 
231 	seq_puts(m, header);
232 
233 	i = _NSIG;
234 	do {
235 		int x = 0;
236 
237 		i -= 4;
238 		if (sigismember(set, i+1)) x |= 1;
239 		if (sigismember(set, i+2)) x |= 2;
240 		if (sigismember(set, i+3)) x |= 4;
241 		if (sigismember(set, i+4)) x |= 8;
242 		seq_printf(m, "%x", x);
243 	} while (i >= 4);
244 
245 	seq_putc(m, '\n');
246 }
247 
248 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
249 				    sigset_t *catch)
250 {
251 	struct k_sigaction *k;
252 	int i;
253 
254 	k = p->sighand->action;
255 	for (i = 1; i <= _NSIG; ++i, ++k) {
256 		if (k->sa.sa_handler == SIG_IGN)
257 			sigaddset(ign, i);
258 		else if (k->sa.sa_handler != SIG_DFL)
259 			sigaddset(catch, i);
260 	}
261 }
262 
263 static inline void task_sig(struct seq_file *m, struct task_struct *p)
264 {
265 	unsigned long flags;
266 	sigset_t pending, shpending, blocked, ignored, caught;
267 	int num_threads = 0;
268 	unsigned long qsize = 0;
269 	unsigned long qlim = 0;
270 
271 	sigemptyset(&pending);
272 	sigemptyset(&shpending);
273 	sigemptyset(&blocked);
274 	sigemptyset(&ignored);
275 	sigemptyset(&caught);
276 
277 	if (lock_task_sighand(p, &flags)) {
278 		pending = p->pending.signal;
279 		shpending = p->signal->shared_pending.signal;
280 		blocked = p->blocked;
281 		collect_sigign_sigcatch(p, &ignored, &caught);
282 		num_threads = get_nr_threads(p);
283 		rcu_read_lock();  /* FIXME: is this correct? */
284 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
285 		rcu_read_unlock();
286 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
287 		unlock_task_sighand(p, &flags);
288 	}
289 
290 	seq_printf(m, "Threads:\t%d\n", num_threads);
291 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
292 
293 	/* render them all */
294 	render_sigset_t(m, "SigPnd:\t", &pending);
295 	render_sigset_t(m, "ShdPnd:\t", &shpending);
296 	render_sigset_t(m, "SigBlk:\t", &blocked);
297 	render_sigset_t(m, "SigIgn:\t", &ignored);
298 	render_sigset_t(m, "SigCgt:\t", &caught);
299 }
300 
301 static void render_cap_t(struct seq_file *m, const char *header,
302 			kernel_cap_t *a)
303 {
304 	unsigned __capi;
305 
306 	seq_puts(m, header);
307 	CAP_FOR_EACH_U32(__capi) {
308 		seq_printf(m, "%08x",
309 			   a->cap[(_KERNEL_CAPABILITY_U32S-1) - __capi]);
310 	}
311 	seq_putc(m, '\n');
312 }
313 
314 /* Remove non-existent capabilities */
315 #define NORM_CAPS(v) (v.cap[CAP_TO_INDEX(CAP_LAST_CAP)] &= \
316 				CAP_TO_MASK(CAP_LAST_CAP + 1) - 1)
317 
318 static inline void task_cap(struct seq_file *m, struct task_struct *p)
319 {
320 	const struct cred *cred;
321 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
322 
323 	rcu_read_lock();
324 	cred = __task_cred(p);
325 	cap_inheritable	= cred->cap_inheritable;
326 	cap_permitted	= cred->cap_permitted;
327 	cap_effective	= cred->cap_effective;
328 	cap_bset	= cred->cap_bset;
329 	rcu_read_unlock();
330 
331 	NORM_CAPS(cap_inheritable);
332 	NORM_CAPS(cap_permitted);
333 	NORM_CAPS(cap_effective);
334 	NORM_CAPS(cap_bset);
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 }
341 
342 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
343 {
344 #ifdef CONFIG_SECCOMP
345 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
346 #endif
347 }
348 
349 static inline void task_context_switch_counts(struct seq_file *m,
350 						struct task_struct *p)
351 {
352 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
353 			"nonvoluntary_ctxt_switches:\t%lu\n",
354 			p->nvcsw,
355 			p->nivcsw);
356 }
357 
358 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
359 {
360 	seq_puts(m, "Cpus_allowed:\t");
361 	seq_cpumask(m, &task->cpus_allowed);
362 	seq_putc(m, '\n');
363 	seq_puts(m, "Cpus_allowed_list:\t");
364 	seq_cpumask_list(m, &task->cpus_allowed);
365 	seq_putc(m, '\n');
366 }
367 
368 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
369 			struct pid *pid, struct task_struct *task)
370 {
371 	struct mm_struct *mm = get_task_mm(task);
372 
373 	task_name(m, task);
374 	task_state(m, ns, pid, task);
375 
376 	if (mm) {
377 		task_mem(m, mm);
378 		mmput(mm);
379 	}
380 	task_sig(m, task);
381 	task_cap(m, task);
382 	task_seccomp(m, task);
383 	task_cpus_allowed(m, task);
384 	cpuset_task_status_allowed(m, task);
385 	task_context_switch_counts(m, task);
386 	return 0;
387 }
388 
389 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
390 			struct pid *pid, struct task_struct *task, int whole)
391 {
392 	unsigned long vsize, eip, esp, wchan = ~0UL;
393 	int priority, nice;
394 	int tty_pgrp = -1, tty_nr = 0;
395 	sigset_t sigign, sigcatch;
396 	char state;
397 	pid_t ppid = 0, pgid = -1, sid = -1;
398 	int num_threads = 0;
399 	int permitted;
400 	struct mm_struct *mm;
401 	unsigned long long start_time;
402 	unsigned long cmin_flt = 0, cmaj_flt = 0;
403 	unsigned long  min_flt = 0,  maj_flt = 0;
404 	cputime_t cutime, cstime, utime, stime;
405 	cputime_t cgtime, gtime;
406 	unsigned long rsslim = 0;
407 	char tcomm[sizeof(task->comm)];
408 	unsigned long flags;
409 
410 	state = *get_task_state(task);
411 	vsize = eip = esp = 0;
412 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
413 	mm = get_task_mm(task);
414 	if (mm) {
415 		vsize = task_vsize(mm);
416 		if (permitted) {
417 			eip = KSTK_EIP(task);
418 			esp = KSTK_ESP(task);
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 				t = next_thread(t);
457 			} while (t != task);
458 
459 			min_flt += sig->min_flt;
460 			maj_flt += sig->maj_flt;
461 			thread_group_cputime_adjusted(task, &utime, &stime);
462 			gtime += sig->gtime;
463 		}
464 
465 		sid = task_session_nr_ns(task, ns);
466 		ppid = task_tgid_nr_ns(task->real_parent, ns);
467 		pgid = task_pgrp_nr_ns(task, ns);
468 
469 		unlock_task_sighand(task, &flags);
470 	}
471 
472 	if (permitted && (!whole || num_threads < 2))
473 		wchan = get_wchan(task);
474 	if (!whole) {
475 		min_flt = task->min_flt;
476 		maj_flt = task->maj_flt;
477 		task_cputime_adjusted(task, &utime, &stime);
478 		gtime = task_gtime(task);
479 	}
480 
481 	/* scale priority and nice values from timeslices to -20..20 */
482 	/* to make it look like a "normal" Unix priority/nice value  */
483 	priority = task_prio(task);
484 	nice = task_nice(task);
485 
486 	/* Temporary variable needed for gcc-2.96 */
487 	/* convert timespec -> nsec*/
488 	start_time =
489 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
490 				+ task->real_start_time.tv_nsec;
491 	/* convert nsec -> ticks */
492 	start_time = nsec_to_clock_t(start_time);
493 
494 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
495 	seq_put_decimal_ll(m, ' ', ppid);
496 	seq_put_decimal_ll(m, ' ', pgid);
497 	seq_put_decimal_ll(m, ' ', sid);
498 	seq_put_decimal_ll(m, ' ', tty_nr);
499 	seq_put_decimal_ll(m, ' ', tty_pgrp);
500 	seq_put_decimal_ull(m, ' ', task->flags);
501 	seq_put_decimal_ull(m, ' ', min_flt);
502 	seq_put_decimal_ull(m, ' ', cmin_flt);
503 	seq_put_decimal_ull(m, ' ', maj_flt);
504 	seq_put_decimal_ull(m, ' ', cmaj_flt);
505 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
506 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
507 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
508 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
509 	seq_put_decimal_ll(m, ' ', priority);
510 	seq_put_decimal_ll(m, ' ', nice);
511 	seq_put_decimal_ll(m, ' ', num_threads);
512 	seq_put_decimal_ull(m, ' ', 0);
513 	seq_put_decimal_ull(m, ' ', start_time);
514 	seq_put_decimal_ull(m, ' ', vsize);
515 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
516 	seq_put_decimal_ull(m, ' ', rsslim);
517 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
518 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
519 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
520 	seq_put_decimal_ull(m, ' ', esp);
521 	seq_put_decimal_ull(m, ' ', eip);
522 	/* The signal information here is obsolete.
523 	 * It must be decimal for Linux 2.0 compatibility.
524 	 * Use /proc/#/status for real-time signals.
525 	 */
526 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
527 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
528 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
529 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
530 	seq_put_decimal_ull(m, ' ', wchan);
531 	seq_put_decimal_ull(m, ' ', 0);
532 	seq_put_decimal_ull(m, ' ', 0);
533 	seq_put_decimal_ll(m, ' ', task->exit_signal);
534 	seq_put_decimal_ll(m, ' ', task_cpu(task));
535 	seq_put_decimal_ull(m, ' ', task->rt_priority);
536 	seq_put_decimal_ull(m, ' ', task->policy);
537 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
538 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
539 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
540 
541 	if (mm && permitted) {
542 		seq_put_decimal_ull(m, ' ', mm->start_data);
543 		seq_put_decimal_ull(m, ' ', mm->end_data);
544 		seq_put_decimal_ull(m, ' ', mm->start_brk);
545 		seq_put_decimal_ull(m, ' ', mm->arg_start);
546 		seq_put_decimal_ull(m, ' ', mm->arg_end);
547 		seq_put_decimal_ull(m, ' ', mm->env_start);
548 		seq_put_decimal_ull(m, ' ', mm->env_end);
549 	} else
550 		seq_printf(m, " 0 0 0 0 0 0 0");
551 
552 	if (permitted)
553 		seq_put_decimal_ll(m, ' ', task->exit_code);
554 	else
555 		seq_put_decimal_ll(m, ' ', 0);
556 
557 	seq_putc(m, '\n');
558 	if (mm)
559 		mmput(mm);
560 	return 0;
561 }
562 
563 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
564 			struct pid *pid, struct task_struct *task)
565 {
566 	return do_task_stat(m, ns, pid, task, 0);
567 }
568 
569 int proc_tgid_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, 1);
573 }
574 
575 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
576 			struct pid *pid, struct task_struct *task)
577 {
578 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
579 	struct mm_struct *mm = get_task_mm(task);
580 
581 	if (mm) {
582 		size = task_statm(mm, &shared, &text, &data, &resident);
583 		mmput(mm);
584 	}
585 	/*
586 	 * For quick read, open code by putting numbers directly
587 	 * expected format is
588 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
589 	 *               size, resident, shared, text, data);
590 	 */
591 	seq_put_decimal_ull(m, 0, size);
592 	seq_put_decimal_ull(m, ' ', resident);
593 	seq_put_decimal_ull(m, ' ', shared);
594 	seq_put_decimal_ull(m, ' ', text);
595 	seq_put_decimal_ull(m, ' ', 0);
596 	seq_put_decimal_ull(m, ' ', data);
597 	seq_put_decimal_ull(m, ' ', 0);
598 	seq_putc(m, '\n');
599 
600 	return 0;
601 }
602 
603 #ifdef CONFIG_CHECKPOINT_RESTORE
604 static struct pid *
605 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
606 {
607 	struct task_struct *start, *task;
608 	struct pid *pid = NULL;
609 
610 	read_lock(&tasklist_lock);
611 
612 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
613 	if (!start)
614 		goto out;
615 
616 	/*
617 	 * Lets try to continue searching first, this gives
618 	 * us significant speedup on children-rich processes.
619 	 */
620 	if (pid_prev) {
621 		task = pid_task(pid_prev, PIDTYPE_PID);
622 		if (task && task->real_parent == start &&
623 		    !(list_empty(&task->sibling))) {
624 			if (list_is_last(&task->sibling, &start->children))
625 				goto out;
626 			task = list_first_entry(&task->sibling,
627 						struct task_struct, sibling);
628 			pid = get_pid(task_pid(task));
629 			goto out;
630 		}
631 	}
632 
633 	/*
634 	 * Slow search case.
635 	 *
636 	 * We might miss some children here if children
637 	 * are exited while we were not holding the lock,
638 	 * but it was never promised to be accurate that
639 	 * much.
640 	 *
641 	 * "Just suppose that the parent sleeps, but N children
642 	 *  exit after we printed their tids. Now the slow paths
643 	 *  skips N extra children, we miss N tasks." (c)
644 	 *
645 	 * So one need to stop or freeze the leader and all
646 	 * its children to get a precise result.
647 	 */
648 	list_for_each_entry(task, &start->children, sibling) {
649 		if (pos-- == 0) {
650 			pid = get_pid(task_pid(task));
651 			break;
652 		}
653 	}
654 
655 out:
656 	read_unlock(&tasklist_lock);
657 	return pid;
658 }
659 
660 static int children_seq_show(struct seq_file *seq, void *v)
661 {
662 	struct inode *inode = seq->private;
663 	pid_t pid;
664 
665 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
666 	return seq_printf(seq, "%d ", pid);
667 }
668 
669 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
670 {
671 	return get_children_pid(seq->private, NULL, *pos);
672 }
673 
674 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
675 {
676 	struct pid *pid;
677 
678 	pid = get_children_pid(seq->private, v, *pos + 1);
679 	put_pid(v);
680 
681 	++*pos;
682 	return pid;
683 }
684 
685 static void children_seq_stop(struct seq_file *seq, void *v)
686 {
687 	put_pid(v);
688 }
689 
690 static const struct seq_operations children_seq_ops = {
691 	.start	= children_seq_start,
692 	.next	= children_seq_next,
693 	.stop	= children_seq_stop,
694 	.show	= children_seq_show,
695 };
696 
697 static int children_seq_open(struct inode *inode, struct file *file)
698 {
699 	struct seq_file *m;
700 	int ret;
701 
702 	ret = seq_open(file, &children_seq_ops);
703 	if (ret)
704 		return ret;
705 
706 	m = file->private_data;
707 	m->private = inode;
708 
709 	return ret;
710 }
711 
712 int children_seq_release(struct inode *inode, struct file *file)
713 {
714 	seq_release(inode, file);
715 	return 0;
716 }
717 
718 const struct file_operations proc_tid_children_operations = {
719 	.open    = children_seq_open,
720 	.read    = seq_read,
721 	.llseek  = seq_lseek,
722 	.release = children_seq_release,
723 };
724 #endif /* CONFIG_CHECKPOINT_RESTORE */
725