xref: /openbmc/linux/fs/proc/array.c (revision 0d456bad)
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 };
147 
148 static inline const char *get_task_state(struct task_struct *tsk)
149 {
150 	unsigned int state = (tsk->state & TASK_REPORT) | tsk->exit_state;
151 	const char * const *p = &task_state_array[0];
152 
153 	BUILD_BUG_ON(1 + ilog2(TASK_STATE_MAX) != ARRAY_SIZE(task_state_array));
154 
155 	while (state) {
156 		p++;
157 		state >>= 1;
158 	}
159 	return *p;
160 }
161 
162 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
163 				struct pid *pid, struct task_struct *p)
164 {
165 	struct user_namespace *user_ns = seq_user_ns(m);
166 	struct group_info *group_info;
167 	int g;
168 	struct fdtable *fdt = NULL;
169 	const struct cred *cred;
170 	pid_t ppid, tpid;
171 
172 	rcu_read_lock();
173 	ppid = pid_alive(p) ?
174 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
175 	tpid = 0;
176 	if (pid_alive(p)) {
177 		struct task_struct *tracer = ptrace_parent(p);
178 		if (tracer)
179 			tpid = task_pid_nr_ns(tracer, ns);
180 	}
181 	cred = get_task_cred(p);
182 	seq_printf(m,
183 		"State:\t%s\n"
184 		"Tgid:\t%d\n"
185 		"Pid:\t%d\n"
186 		"PPid:\t%d\n"
187 		"TracerPid:\t%d\n"
188 		"Uid:\t%d\t%d\t%d\t%d\n"
189 		"Gid:\t%d\t%d\t%d\t%d\n",
190 		get_task_state(p),
191 		task_tgid_nr_ns(p, ns),
192 		pid_nr_ns(pid, ns),
193 		ppid, tpid,
194 		from_kuid_munged(user_ns, cred->uid),
195 		from_kuid_munged(user_ns, cred->euid),
196 		from_kuid_munged(user_ns, cred->suid),
197 		from_kuid_munged(user_ns, cred->fsuid),
198 		from_kgid_munged(user_ns, cred->gid),
199 		from_kgid_munged(user_ns, cred->egid),
200 		from_kgid_munged(user_ns, cred->sgid),
201 		from_kgid_munged(user_ns, cred->fsgid));
202 
203 	task_lock(p);
204 	if (p->files)
205 		fdt = files_fdtable(p->files);
206 	seq_printf(m,
207 		"FDSize:\t%d\n"
208 		"Groups:\t",
209 		fdt ? fdt->max_fds : 0);
210 	rcu_read_unlock();
211 
212 	group_info = cred->group_info;
213 	task_unlock(p);
214 
215 	for (g = 0; g < group_info->ngroups; g++)
216 		seq_printf(m, "%d ",
217 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
218 	put_cred(cred);
219 
220 	seq_putc(m, '\n');
221 }
222 
223 void render_sigset_t(struct seq_file *m, const char *header,
224 				sigset_t *set)
225 {
226 	int i;
227 
228 	seq_puts(m, header);
229 
230 	i = _NSIG;
231 	do {
232 		int x = 0;
233 
234 		i -= 4;
235 		if (sigismember(set, i+1)) x |= 1;
236 		if (sigismember(set, i+2)) x |= 2;
237 		if (sigismember(set, i+3)) x |= 4;
238 		if (sigismember(set, i+4)) x |= 8;
239 		seq_printf(m, "%x", x);
240 	} while (i >= 4);
241 
242 	seq_putc(m, '\n');
243 }
244 
245 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
246 				    sigset_t *catch)
247 {
248 	struct k_sigaction *k;
249 	int i;
250 
251 	k = p->sighand->action;
252 	for (i = 1; i <= _NSIG; ++i, ++k) {
253 		if (k->sa.sa_handler == SIG_IGN)
254 			sigaddset(ign, i);
255 		else if (k->sa.sa_handler != SIG_DFL)
256 			sigaddset(catch, i);
257 	}
258 }
259 
260 static inline void task_sig(struct seq_file *m, struct task_struct *p)
261 {
262 	unsigned long flags;
263 	sigset_t pending, shpending, blocked, ignored, caught;
264 	int num_threads = 0;
265 	unsigned long qsize = 0;
266 	unsigned long qlim = 0;
267 
268 	sigemptyset(&pending);
269 	sigemptyset(&shpending);
270 	sigemptyset(&blocked);
271 	sigemptyset(&ignored);
272 	sigemptyset(&caught);
273 
274 	if (lock_task_sighand(p, &flags)) {
275 		pending = p->pending.signal;
276 		shpending = p->signal->shared_pending.signal;
277 		blocked = p->blocked;
278 		collect_sigign_sigcatch(p, &ignored, &caught);
279 		num_threads = get_nr_threads(p);
280 		rcu_read_lock();  /* FIXME: is this correct? */
281 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
282 		rcu_read_unlock();
283 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
284 		unlock_task_sighand(p, &flags);
285 	}
286 
287 	seq_printf(m, "Threads:\t%d\n", num_threads);
288 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
289 
290 	/* render them all */
291 	render_sigset_t(m, "SigPnd:\t", &pending);
292 	render_sigset_t(m, "ShdPnd:\t", &shpending);
293 	render_sigset_t(m, "SigBlk:\t", &blocked);
294 	render_sigset_t(m, "SigIgn:\t", &ignored);
295 	render_sigset_t(m, "SigCgt:\t", &caught);
296 }
297 
298 static void render_cap_t(struct seq_file *m, const char *header,
299 			kernel_cap_t *a)
300 {
301 	unsigned __capi;
302 
303 	seq_puts(m, header);
304 	CAP_FOR_EACH_U32(__capi) {
305 		seq_printf(m, "%08x",
306 			   a->cap[(_KERNEL_CAPABILITY_U32S-1) - __capi]);
307 	}
308 	seq_putc(m, '\n');
309 }
310 
311 /* Remove non-existent capabilities */
312 #define NORM_CAPS(v) (v.cap[CAP_TO_INDEX(CAP_LAST_CAP)] &= \
313 				CAP_TO_MASK(CAP_LAST_CAP + 1) - 1)
314 
315 static inline void task_cap(struct seq_file *m, struct task_struct *p)
316 {
317 	const struct cred *cred;
318 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
319 
320 	rcu_read_lock();
321 	cred = __task_cred(p);
322 	cap_inheritable	= cred->cap_inheritable;
323 	cap_permitted	= cred->cap_permitted;
324 	cap_effective	= cred->cap_effective;
325 	cap_bset	= cred->cap_bset;
326 	rcu_read_unlock();
327 
328 	NORM_CAPS(cap_inheritable);
329 	NORM_CAPS(cap_permitted);
330 	NORM_CAPS(cap_effective);
331 	NORM_CAPS(cap_bset);
332 
333 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
334 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
335 	render_cap_t(m, "CapEff:\t", &cap_effective);
336 	render_cap_t(m, "CapBnd:\t", &cap_bset);
337 }
338 
339 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
340 {
341 #ifdef CONFIG_SECCOMP
342 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
343 #endif
344 }
345 
346 static inline void task_context_switch_counts(struct seq_file *m,
347 						struct task_struct *p)
348 {
349 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
350 			"nonvoluntary_ctxt_switches:\t%lu\n",
351 			p->nvcsw,
352 			p->nivcsw);
353 }
354 
355 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
356 {
357 	seq_puts(m, "Cpus_allowed:\t");
358 	seq_cpumask(m, &task->cpus_allowed);
359 	seq_putc(m, '\n');
360 	seq_puts(m, "Cpus_allowed_list:\t");
361 	seq_cpumask_list(m, &task->cpus_allowed);
362 	seq_putc(m, '\n');
363 }
364 
365 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
366 			struct pid *pid, struct task_struct *task)
367 {
368 	struct mm_struct *mm = get_task_mm(task);
369 
370 	task_name(m, task);
371 	task_state(m, ns, pid, task);
372 
373 	if (mm) {
374 		task_mem(m, mm);
375 		mmput(mm);
376 	}
377 	task_sig(m, task);
378 	task_cap(m, task);
379 	task_seccomp(m, task);
380 	task_cpus_allowed(m, task);
381 	cpuset_task_status_allowed(m, task);
382 	task_context_switch_counts(m, task);
383 	return 0;
384 }
385 
386 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
387 			struct pid *pid, struct task_struct *task, int whole)
388 {
389 	unsigned long vsize, eip, esp, wchan = ~0UL;
390 	int priority, nice;
391 	int tty_pgrp = -1, tty_nr = 0;
392 	sigset_t sigign, sigcatch;
393 	char state;
394 	pid_t ppid = 0, pgid = -1, sid = -1;
395 	int num_threads = 0;
396 	int permitted;
397 	struct mm_struct *mm;
398 	unsigned long long start_time;
399 	unsigned long cmin_flt = 0, cmaj_flt = 0;
400 	unsigned long  min_flt = 0,  maj_flt = 0;
401 	cputime_t cutime, cstime, utime, stime;
402 	cputime_t cgtime, gtime;
403 	unsigned long rsslim = 0;
404 	char tcomm[sizeof(task->comm)];
405 	unsigned long flags;
406 
407 	state = *get_task_state(task);
408 	vsize = eip = esp = 0;
409 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
410 	mm = get_task_mm(task);
411 	if (mm) {
412 		vsize = task_vsize(mm);
413 		if (permitted) {
414 			eip = KSTK_EIP(task);
415 			esp = KSTK_ESP(task);
416 		}
417 	}
418 
419 	get_task_comm(tcomm, task);
420 
421 	sigemptyset(&sigign);
422 	sigemptyset(&sigcatch);
423 	cutime = cstime = utime = stime = 0;
424 	cgtime = gtime = 0;
425 
426 	if (lock_task_sighand(task, &flags)) {
427 		struct signal_struct *sig = task->signal;
428 
429 		if (sig->tty) {
430 			struct pid *pgrp = tty_get_pgrp(sig->tty);
431 			tty_pgrp = pid_nr_ns(pgrp, ns);
432 			put_pid(pgrp);
433 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
434 		}
435 
436 		num_threads = get_nr_threads(task);
437 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
438 
439 		cmin_flt = sig->cmin_flt;
440 		cmaj_flt = sig->cmaj_flt;
441 		cutime = sig->cutime;
442 		cstime = sig->cstime;
443 		cgtime = sig->cgtime;
444 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
445 
446 		/* add up live thread stats at the group level */
447 		if (whole) {
448 			struct task_struct *t = task;
449 			do {
450 				min_flt += t->min_flt;
451 				maj_flt += t->maj_flt;
452 				gtime += t->gtime;
453 				t = next_thread(t);
454 			} while (t != task);
455 
456 			min_flt += sig->min_flt;
457 			maj_flt += sig->maj_flt;
458 			thread_group_cputime_adjusted(task, &utime, &stime);
459 			gtime += sig->gtime;
460 		}
461 
462 		sid = task_session_nr_ns(task, ns);
463 		ppid = task_tgid_nr_ns(task->real_parent, ns);
464 		pgid = task_pgrp_nr_ns(task, ns);
465 
466 		unlock_task_sighand(task, &flags);
467 	}
468 
469 	if (permitted && (!whole || num_threads < 2))
470 		wchan = get_wchan(task);
471 	if (!whole) {
472 		min_flt = task->min_flt;
473 		maj_flt = task->maj_flt;
474 		task_cputime_adjusted(task, &utime, &stime);
475 		gtime = task->gtime;
476 	}
477 
478 	/* scale priority and nice values from timeslices to -20..20 */
479 	/* to make it look like a "normal" Unix priority/nice value  */
480 	priority = task_prio(task);
481 	nice = task_nice(task);
482 
483 	/* Temporary variable needed for gcc-2.96 */
484 	/* convert timespec -> nsec*/
485 	start_time =
486 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
487 				+ task->real_start_time.tv_nsec;
488 	/* convert nsec -> ticks */
489 	start_time = nsec_to_clock_t(start_time);
490 
491 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
492 	seq_put_decimal_ll(m, ' ', ppid);
493 	seq_put_decimal_ll(m, ' ', pgid);
494 	seq_put_decimal_ll(m, ' ', sid);
495 	seq_put_decimal_ll(m, ' ', tty_nr);
496 	seq_put_decimal_ll(m, ' ', tty_pgrp);
497 	seq_put_decimal_ull(m, ' ', task->flags);
498 	seq_put_decimal_ull(m, ' ', min_flt);
499 	seq_put_decimal_ull(m, ' ', cmin_flt);
500 	seq_put_decimal_ull(m, ' ', maj_flt);
501 	seq_put_decimal_ull(m, ' ', cmaj_flt);
502 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
503 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
504 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
505 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
506 	seq_put_decimal_ll(m, ' ', priority);
507 	seq_put_decimal_ll(m, ' ', nice);
508 	seq_put_decimal_ll(m, ' ', num_threads);
509 	seq_put_decimal_ull(m, ' ', 0);
510 	seq_put_decimal_ull(m, ' ', start_time);
511 	seq_put_decimal_ull(m, ' ', vsize);
512 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
513 	seq_put_decimal_ull(m, ' ', rsslim);
514 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
515 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
516 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
517 	seq_put_decimal_ull(m, ' ', esp);
518 	seq_put_decimal_ull(m, ' ', eip);
519 	/* The signal information here is obsolete.
520 	 * It must be decimal for Linux 2.0 compatibility.
521 	 * Use /proc/#/status for real-time signals.
522 	 */
523 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
524 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
525 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
526 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
527 	seq_put_decimal_ull(m, ' ', wchan);
528 	seq_put_decimal_ull(m, ' ', 0);
529 	seq_put_decimal_ull(m, ' ', 0);
530 	seq_put_decimal_ll(m, ' ', task->exit_signal);
531 	seq_put_decimal_ll(m, ' ', task_cpu(task));
532 	seq_put_decimal_ull(m, ' ', task->rt_priority);
533 	seq_put_decimal_ull(m, ' ', task->policy);
534 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
535 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
536 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
537 
538 	if (mm && permitted) {
539 		seq_put_decimal_ull(m, ' ', mm->start_data);
540 		seq_put_decimal_ull(m, ' ', mm->end_data);
541 		seq_put_decimal_ull(m, ' ', mm->start_brk);
542 		seq_put_decimal_ull(m, ' ', mm->arg_start);
543 		seq_put_decimal_ull(m, ' ', mm->arg_end);
544 		seq_put_decimal_ull(m, ' ', mm->env_start);
545 		seq_put_decimal_ull(m, ' ', mm->env_end);
546 	} else
547 		seq_printf(m, " 0 0 0 0 0 0 0");
548 
549 	if (permitted)
550 		seq_put_decimal_ll(m, ' ', task->exit_code);
551 	else
552 		seq_put_decimal_ll(m, ' ', 0);
553 
554 	seq_putc(m, '\n');
555 	if (mm)
556 		mmput(mm);
557 	return 0;
558 }
559 
560 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
561 			struct pid *pid, struct task_struct *task)
562 {
563 	return do_task_stat(m, ns, pid, task, 0);
564 }
565 
566 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
567 			struct pid *pid, struct task_struct *task)
568 {
569 	return do_task_stat(m, ns, pid, task, 1);
570 }
571 
572 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
573 			struct pid *pid, struct task_struct *task)
574 {
575 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
576 	struct mm_struct *mm = get_task_mm(task);
577 
578 	if (mm) {
579 		size = task_statm(mm, &shared, &text, &data, &resident);
580 		mmput(mm);
581 	}
582 	/*
583 	 * For quick read, open code by putting numbers directly
584 	 * expected format is
585 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
586 	 *               size, resident, shared, text, data);
587 	 */
588 	seq_put_decimal_ull(m, 0, size);
589 	seq_put_decimal_ull(m, ' ', resident);
590 	seq_put_decimal_ull(m, ' ', shared);
591 	seq_put_decimal_ull(m, ' ', text);
592 	seq_put_decimal_ull(m, ' ', 0);
593 	seq_put_decimal_ull(m, ' ', data);
594 	seq_put_decimal_ull(m, ' ', 0);
595 	seq_putc(m, '\n');
596 
597 	return 0;
598 }
599 
600 #ifdef CONFIG_CHECKPOINT_RESTORE
601 static struct pid *
602 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
603 {
604 	struct task_struct *start, *task;
605 	struct pid *pid = NULL;
606 
607 	read_lock(&tasklist_lock);
608 
609 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
610 	if (!start)
611 		goto out;
612 
613 	/*
614 	 * Lets try to continue searching first, this gives
615 	 * us significant speedup on children-rich processes.
616 	 */
617 	if (pid_prev) {
618 		task = pid_task(pid_prev, PIDTYPE_PID);
619 		if (task && task->real_parent == start &&
620 		    !(list_empty(&task->sibling))) {
621 			if (list_is_last(&task->sibling, &start->children))
622 				goto out;
623 			task = list_first_entry(&task->sibling,
624 						struct task_struct, sibling);
625 			pid = get_pid(task_pid(task));
626 			goto out;
627 		}
628 	}
629 
630 	/*
631 	 * Slow search case.
632 	 *
633 	 * We might miss some children here if children
634 	 * are exited while we were not holding the lock,
635 	 * but it was never promised to be accurate that
636 	 * much.
637 	 *
638 	 * "Just suppose that the parent sleeps, but N children
639 	 *  exit after we printed their tids. Now the slow paths
640 	 *  skips N extra children, we miss N tasks." (c)
641 	 *
642 	 * So one need to stop or freeze the leader and all
643 	 * its children to get a precise result.
644 	 */
645 	list_for_each_entry(task, &start->children, sibling) {
646 		if (pos-- == 0) {
647 			pid = get_pid(task_pid(task));
648 			break;
649 		}
650 	}
651 
652 out:
653 	read_unlock(&tasklist_lock);
654 	return pid;
655 }
656 
657 static int children_seq_show(struct seq_file *seq, void *v)
658 {
659 	struct inode *inode = seq->private;
660 	pid_t pid;
661 
662 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
663 	return seq_printf(seq, "%d ", pid);
664 }
665 
666 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
667 {
668 	return get_children_pid(seq->private, NULL, *pos);
669 }
670 
671 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
672 {
673 	struct pid *pid;
674 
675 	pid = get_children_pid(seq->private, v, *pos + 1);
676 	put_pid(v);
677 
678 	++*pos;
679 	return pid;
680 }
681 
682 static void children_seq_stop(struct seq_file *seq, void *v)
683 {
684 	put_pid(v);
685 }
686 
687 static const struct seq_operations children_seq_ops = {
688 	.start	= children_seq_start,
689 	.next	= children_seq_next,
690 	.stop	= children_seq_stop,
691 	.show	= children_seq_show,
692 };
693 
694 static int children_seq_open(struct inode *inode, struct file *file)
695 {
696 	struct seq_file *m;
697 	int ret;
698 
699 	ret = seq_open(file, &children_seq_ops);
700 	if (ret)
701 		return ret;
702 
703 	m = file->private_data;
704 	m->private = inode;
705 
706 	return ret;
707 }
708 
709 int children_seq_release(struct inode *inode, struct file *file)
710 {
711 	seq_release(inode, file);
712 	return 0;
713 }
714 
715 const struct file_operations proc_tid_children_operations = {
716 	.open    = children_seq_open,
717 	.read    = seq_read,
718 	.llseek  = seq_lseek,
719 	.release = children_seq_release,
720 };
721 #endif /* CONFIG_CHECKPOINT_RESTORE */
722