xref: /openbmc/linux/fs/proc/array.c (revision 95e9fd10) 
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 = current_user_ns();
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 < min(group_info->ngroups, NGROUPS_SMALL); 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 static 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 static inline void task_cap(struct seq_file *m, struct task_struct *p)
312 {
313 	const struct cred *cred;
314 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective, cap_bset;
315 
316 	rcu_read_lock();
317 	cred = __task_cred(p);
318 	cap_inheritable	= cred->cap_inheritable;
319 	cap_permitted	= cred->cap_permitted;
320 	cap_effective	= cred->cap_effective;
321 	cap_bset	= cred->cap_bset;
322 	rcu_read_unlock();
323 
324 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
325 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
326 	render_cap_t(m, "CapEff:\t", &cap_effective);
327 	render_cap_t(m, "CapBnd:\t", &cap_bset);
328 }
329 
330 static inline void task_context_switch_counts(struct seq_file *m,
331 						struct task_struct *p)
332 {
333 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
334 			"nonvoluntary_ctxt_switches:\t%lu\n",
335 			p->nvcsw,
336 			p->nivcsw);
337 }
338 
339 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
340 {
341 	seq_puts(m, "Cpus_allowed:\t");
342 	seq_cpumask(m, &task->cpus_allowed);
343 	seq_putc(m, '\n');
344 	seq_puts(m, "Cpus_allowed_list:\t");
345 	seq_cpumask_list(m, &task->cpus_allowed);
346 	seq_putc(m, '\n');
347 }
348 
349 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
350 			struct pid *pid, struct task_struct *task)
351 {
352 	struct mm_struct *mm = get_task_mm(task);
353 
354 	task_name(m, task);
355 	task_state(m, ns, pid, task);
356 
357 	if (mm) {
358 		task_mem(m, mm);
359 		mmput(mm);
360 	}
361 	task_sig(m, task);
362 	task_cap(m, task);
363 	task_cpus_allowed(m, task);
364 	cpuset_task_status_allowed(m, task);
365 	task_context_switch_counts(m, task);
366 	return 0;
367 }
368 
369 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
370 			struct pid *pid, struct task_struct *task, int whole)
371 {
372 	unsigned long vsize, eip, esp, wchan = ~0UL;
373 	int priority, nice;
374 	int tty_pgrp = -1, tty_nr = 0;
375 	sigset_t sigign, sigcatch;
376 	char state;
377 	pid_t ppid = 0, pgid = -1, sid = -1;
378 	int num_threads = 0;
379 	int permitted;
380 	struct mm_struct *mm;
381 	unsigned long long start_time;
382 	unsigned long cmin_flt = 0, cmaj_flt = 0;
383 	unsigned long  min_flt = 0,  maj_flt = 0;
384 	cputime_t cutime, cstime, utime, stime;
385 	cputime_t cgtime, gtime;
386 	unsigned long rsslim = 0;
387 	char tcomm[sizeof(task->comm)];
388 	unsigned long flags;
389 
390 	state = *get_task_state(task);
391 	vsize = eip = esp = 0;
392 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
393 	mm = get_task_mm(task);
394 	if (mm) {
395 		vsize = task_vsize(mm);
396 		if (permitted) {
397 			eip = KSTK_EIP(task);
398 			esp = KSTK_ESP(task);
399 		}
400 	}
401 
402 	get_task_comm(tcomm, task);
403 
404 	sigemptyset(&sigign);
405 	sigemptyset(&sigcatch);
406 	cutime = cstime = utime = stime = 0;
407 	cgtime = gtime = 0;
408 
409 	if (lock_task_sighand(task, &flags)) {
410 		struct signal_struct *sig = task->signal;
411 
412 		if (sig->tty) {
413 			struct pid *pgrp = tty_get_pgrp(sig->tty);
414 			tty_pgrp = pid_nr_ns(pgrp, ns);
415 			put_pid(pgrp);
416 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
417 		}
418 
419 		num_threads = get_nr_threads(task);
420 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
421 
422 		cmin_flt = sig->cmin_flt;
423 		cmaj_flt = sig->cmaj_flt;
424 		cutime = sig->cutime;
425 		cstime = sig->cstime;
426 		cgtime = sig->cgtime;
427 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
428 
429 		/* add up live thread stats at the group level */
430 		if (whole) {
431 			struct task_struct *t = task;
432 			do {
433 				min_flt += t->min_flt;
434 				maj_flt += t->maj_flt;
435 				gtime += t->gtime;
436 				t = next_thread(t);
437 			} while (t != task);
438 
439 			min_flt += sig->min_flt;
440 			maj_flt += sig->maj_flt;
441 			thread_group_times(task, &utime, &stime);
442 			gtime += sig->gtime;
443 		}
444 
445 		sid = task_session_nr_ns(task, ns);
446 		ppid = task_tgid_nr_ns(task->real_parent, ns);
447 		pgid = task_pgrp_nr_ns(task, ns);
448 
449 		unlock_task_sighand(task, &flags);
450 	}
451 
452 	if (permitted && (!whole || num_threads < 2))
453 		wchan = get_wchan(task);
454 	if (!whole) {
455 		min_flt = task->min_flt;
456 		maj_flt = task->maj_flt;
457 		task_times(task, &utime, &stime);
458 		gtime = task->gtime;
459 	}
460 
461 	/* scale priority and nice values from timeslices to -20..20 */
462 	/* to make it look like a "normal" Unix priority/nice value  */
463 	priority = task_prio(task);
464 	nice = task_nice(task);
465 
466 	/* Temporary variable needed for gcc-2.96 */
467 	/* convert timespec -> nsec*/
468 	start_time =
469 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
470 				+ task->real_start_time.tv_nsec;
471 	/* convert nsec -> ticks */
472 	start_time = nsec_to_clock_t(start_time);
473 
474 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
475 	seq_put_decimal_ll(m, ' ', ppid);
476 	seq_put_decimal_ll(m, ' ', pgid);
477 	seq_put_decimal_ll(m, ' ', sid);
478 	seq_put_decimal_ll(m, ' ', tty_nr);
479 	seq_put_decimal_ll(m, ' ', tty_pgrp);
480 	seq_put_decimal_ull(m, ' ', task->flags);
481 	seq_put_decimal_ull(m, ' ', min_flt);
482 	seq_put_decimal_ull(m, ' ', cmin_flt);
483 	seq_put_decimal_ull(m, ' ', maj_flt);
484 	seq_put_decimal_ull(m, ' ', cmaj_flt);
485 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
486 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
487 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
488 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
489 	seq_put_decimal_ll(m, ' ', priority);
490 	seq_put_decimal_ll(m, ' ', nice);
491 	seq_put_decimal_ll(m, ' ', num_threads);
492 	seq_put_decimal_ull(m, ' ', 0);
493 	seq_put_decimal_ull(m, ' ', start_time);
494 	seq_put_decimal_ull(m, ' ', vsize);
495 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
496 	seq_put_decimal_ull(m, ' ', rsslim);
497 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
498 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
499 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
500 	seq_put_decimal_ull(m, ' ', esp);
501 	seq_put_decimal_ull(m, ' ', eip);
502 	/* The signal information here is obsolete.
503 	 * It must be decimal for Linux 2.0 compatibility.
504 	 * Use /proc/#/status for real-time signals.
505 	 */
506 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
507 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
508 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
509 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
510 	seq_put_decimal_ull(m, ' ', wchan);
511 	seq_put_decimal_ull(m, ' ', 0);
512 	seq_put_decimal_ull(m, ' ', 0);
513 	seq_put_decimal_ll(m, ' ', task->exit_signal);
514 	seq_put_decimal_ll(m, ' ', task_cpu(task));
515 	seq_put_decimal_ull(m, ' ', task->rt_priority);
516 	seq_put_decimal_ull(m, ' ', task->policy);
517 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
518 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
519 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
520 
521 	if (mm && permitted) {
522 		seq_put_decimal_ull(m, ' ', mm->start_data);
523 		seq_put_decimal_ull(m, ' ', mm->end_data);
524 		seq_put_decimal_ull(m, ' ', mm->start_brk);
525 		seq_put_decimal_ull(m, ' ', mm->arg_start);
526 		seq_put_decimal_ull(m, ' ', mm->arg_end);
527 		seq_put_decimal_ull(m, ' ', mm->env_start);
528 		seq_put_decimal_ull(m, ' ', mm->env_end);
529 	} else
530 		seq_printf(m, " 0 0 0 0 0 0 0");
531 
532 	if (permitted)
533 		seq_put_decimal_ll(m, ' ', task->exit_code);
534 	else
535 		seq_put_decimal_ll(m, ' ', 0);
536 
537 	seq_putc(m, '\n');
538 	if (mm)
539 		mmput(mm);
540 	return 0;
541 }
542 
543 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
544 			struct pid *pid, struct task_struct *task)
545 {
546 	return do_task_stat(m, ns, pid, task, 0);
547 }
548 
549 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
550 			struct pid *pid, struct task_struct *task)
551 {
552 	return do_task_stat(m, ns, pid, task, 1);
553 }
554 
555 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
556 			struct pid *pid, struct task_struct *task)
557 {
558 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
559 	struct mm_struct *mm = get_task_mm(task);
560 
561 	if (mm) {
562 		size = task_statm(mm, &shared, &text, &data, &resident);
563 		mmput(mm);
564 	}
565 	/*
566 	 * For quick read, open code by putting numbers directly
567 	 * expected format is
568 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
569 	 *               size, resident, shared, text, data);
570 	 */
571 	seq_put_decimal_ull(m, 0, size);
572 	seq_put_decimal_ull(m, ' ', resident);
573 	seq_put_decimal_ull(m, ' ', shared);
574 	seq_put_decimal_ull(m, ' ', text);
575 	seq_put_decimal_ull(m, ' ', 0);
576 	seq_put_decimal_ull(m, ' ', data);
577 	seq_put_decimal_ull(m, ' ', 0);
578 	seq_putc(m, '\n');
579 
580 	return 0;
581 }
582 
583 #ifdef CONFIG_CHECKPOINT_RESTORE
584 static struct pid *
585 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
586 {
587 	struct task_struct *start, *task;
588 	struct pid *pid = NULL;
589 
590 	read_lock(&tasklist_lock);
591 
592 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
593 	if (!start)
594 		goto out;
595 
596 	/*
597 	 * Lets try to continue searching first, this gives
598 	 * us significant speedup on children-rich processes.
599 	 */
600 	if (pid_prev) {
601 		task = pid_task(pid_prev, PIDTYPE_PID);
602 		if (task && task->real_parent == start &&
603 		    !(list_empty(&task->sibling))) {
604 			if (list_is_last(&task->sibling, &start->children))
605 				goto out;
606 			task = list_first_entry(&task->sibling,
607 						struct task_struct, sibling);
608 			pid = get_pid(task_pid(task));
609 			goto out;
610 		}
611 	}
612 
613 	/*
614 	 * Slow search case.
615 	 *
616 	 * We might miss some children here if children
617 	 * are exited while we were not holding the lock,
618 	 * but it was never promised to be accurate that
619 	 * much.
620 	 *
621 	 * "Just suppose that the parent sleeps, but N children
622 	 *  exit after we printed their tids. Now the slow paths
623 	 *  skips N extra children, we miss N tasks." (c)
624 	 *
625 	 * So one need to stop or freeze the leader and all
626 	 * its children to get a precise result.
627 	 */
628 	list_for_each_entry(task, &start->children, sibling) {
629 		if (pos-- == 0) {
630 			pid = get_pid(task_pid(task));
631 			break;
632 		}
633 	}
634 
635 out:
636 	read_unlock(&tasklist_lock);
637 	return pid;
638 }
639 
640 static int children_seq_show(struct seq_file *seq, void *v)
641 {
642 	struct inode *inode = seq->private;
643 	pid_t pid;
644 
645 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
646 	return seq_printf(seq, "%d ", pid);
647 }
648 
649 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
650 {
651 	return get_children_pid(seq->private, NULL, *pos);
652 }
653 
654 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
655 {
656 	struct pid *pid;
657 
658 	pid = get_children_pid(seq->private, v, *pos + 1);
659 	put_pid(v);
660 
661 	++*pos;
662 	return pid;
663 }
664 
665 static void children_seq_stop(struct seq_file *seq, void *v)
666 {
667 	put_pid(v);
668 }
669 
670 static const struct seq_operations children_seq_ops = {
671 	.start	= children_seq_start,
672 	.next	= children_seq_next,
673 	.stop	= children_seq_stop,
674 	.show	= children_seq_show,
675 };
676 
677 static int children_seq_open(struct inode *inode, struct file *file)
678 {
679 	struct seq_file *m;
680 	int ret;
681 
682 	ret = seq_open(file, &children_seq_ops);
683 	if (ret)
684 		return ret;
685 
686 	m = file->private_data;
687 	m->private = inode;
688 
689 	return ret;
690 }
691 
692 int children_seq_release(struct inode *inode, struct file *file)
693 {
694 	seq_release(inode, file);
695 	return 0;
696 }
697 
698 const struct file_operations proc_tid_children_operations = {
699 	.open    = children_seq_open,
700 	.read    = seq_read,
701 	.llseek  = seq_lseek,
702 	.release = children_seq_release,
703 };
704 #endif /* CONFIG_CHECKPOINT_RESTORE */
705