xref: /linux/fs/proc/array.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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/string_helpers.h>
85 #include <linux/user_namespace.h>
86 
87 #include <asm/pgtable.h>
88 #include <asm/processor.h>
89 #include "internal.h"
90 
91 static inline void task_name(struct seq_file *m, struct task_struct *p)
92 {
93 	char *buf;
94 	char tcomm[sizeof(p->comm)];
95 
96 	get_task_comm(tcomm, p);
97 
98 	seq_puts(m, "Name:\t");
99 	buf = m->buf + m->count;
100 
101 	/* Ignore error for now */
102 	buf += string_escape_str(tcomm, buf, m->size - m->count,
103 				 ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
104 
105 	m->count = buf - m->buf;
106 	seq_putc(m, '\n');
107 }
108 
109 /*
110  * The task state array is a strange "bitmap" of
111  * reasons to sleep. Thus "running" is zero, and
112  * you can test for combinations of others with
113  * simple bit tests.
114  */
115 static const char * const task_state_array[] = {
116 	"R (running)",		/*   0 */
117 	"S (sleeping)",		/*   1 */
118 	"D (disk sleep)",	/*   2 */
119 	"T (stopped)",		/*   4 */
120 	"t (tracing stop)",	/*   8 */
121 	"X (dead)",		/*  16 */
122 	"Z (zombie)",		/*  32 */
123 };
124 
125 static inline const char *get_task_state(struct task_struct *tsk)
126 {
127 	unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
128 
129 	/*
130 	 * Parked tasks do not run; they sit in __kthread_parkme().
131 	 * Without this check, we would report them as running, which is
132 	 * clearly wrong, so we report them as sleeping instead.
133 	 */
134 	if (tsk->state == TASK_PARKED)
135 		state = TASK_INTERRUPTIBLE;
136 
137 	BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
138 
139 	return task_state_array[fls(state)];
140 }
141 
142 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
143 				struct pid *pid, struct task_struct *p)
144 {
145 	struct user_namespace *user_ns = seq_user_ns(m);
146 	struct group_info *group_info;
147 	int g;
148 	struct task_struct *tracer;
149 	const struct cred *cred;
150 	pid_t ppid, tpid = 0, tgid, ngid;
151 	unsigned int max_fds = 0;
152 
153 	rcu_read_lock();
154 	ppid = pid_alive(p) ?
155 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
156 
157 	tracer = ptrace_parent(p);
158 	if (tracer)
159 		tpid = task_pid_nr_ns(tracer, ns);
160 
161 	tgid = task_tgid_nr_ns(p, ns);
162 	ngid = task_numa_group_id(p);
163 	cred = get_task_cred(p);
164 
165 	task_lock(p);
166 	if (p->files)
167 		max_fds = files_fdtable(p->files)->max_fds;
168 	task_unlock(p);
169 	rcu_read_unlock();
170 
171 	seq_printf(m,
172 		"State:\t%s\n"
173 		"Tgid:\t%d\n"
174 		"Ngid:\t%d\n"
175 		"Pid:\t%d\n"
176 		"PPid:\t%d\n"
177 		"TracerPid:\t%d\n"
178 		"Uid:\t%d\t%d\t%d\t%d\n"
179 		"Gid:\t%d\t%d\t%d\t%d\n"
180 		"FDSize:\t%d\nGroups:\t",
181 		get_task_state(p),
182 		tgid, ngid, pid_nr_ns(pid, ns), ppid, tpid,
183 		from_kuid_munged(user_ns, cred->uid),
184 		from_kuid_munged(user_ns, cred->euid),
185 		from_kuid_munged(user_ns, cred->suid),
186 		from_kuid_munged(user_ns, cred->fsuid),
187 		from_kgid_munged(user_ns, cred->gid),
188 		from_kgid_munged(user_ns, cred->egid),
189 		from_kgid_munged(user_ns, cred->sgid),
190 		from_kgid_munged(user_ns, cred->fsgid),
191 		max_fds);
192 
193 	group_info = cred->group_info;
194 	for (g = 0; g < group_info->ngroups; g++)
195 		seq_printf(m, "%d ",
196 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
197 	put_cred(cred);
198 
199 #ifdef CONFIG_PID_NS
200 	seq_puts(m, "\nNStgid:");
201 	for (g = ns->level; g <= pid->level; g++)
202 		seq_printf(m, "\t%d",
203 			task_tgid_nr_ns(p, pid->numbers[g].ns));
204 	seq_puts(m, "\nNSpid:");
205 	for (g = ns->level; g <= pid->level; g++)
206 		seq_printf(m, "\t%d",
207 			task_pid_nr_ns(p, pid->numbers[g].ns));
208 	seq_puts(m, "\nNSpgid:");
209 	for (g = ns->level; g <= pid->level; g++)
210 		seq_printf(m, "\t%d",
211 			task_pgrp_nr_ns(p, pid->numbers[g].ns));
212 	seq_puts(m, "\nNSsid:");
213 	for (g = ns->level; g <= pid->level; g++)
214 		seq_printf(m, "\t%d",
215 			task_session_nr_ns(p, pid->numbers[g].ns));
216 #endif
217 	seq_putc(m, '\n');
218 }
219 
220 void render_sigset_t(struct seq_file *m, const char *header,
221 				sigset_t *set)
222 {
223 	int i;
224 
225 	seq_puts(m, header);
226 
227 	i = _NSIG;
228 	do {
229 		int x = 0;
230 
231 		i -= 4;
232 		if (sigismember(set, i+1)) x |= 1;
233 		if (sigismember(set, i+2)) x |= 2;
234 		if (sigismember(set, i+3)) x |= 4;
235 		if (sigismember(set, i+4)) x |= 8;
236 		seq_printf(m, "%x", x);
237 	} while (i >= 4);
238 
239 	seq_putc(m, '\n');
240 }
241 
242 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
243 				    sigset_t *catch)
244 {
245 	struct k_sigaction *k;
246 	int i;
247 
248 	k = p->sighand->action;
249 	for (i = 1; i <= _NSIG; ++i, ++k) {
250 		if (k->sa.sa_handler == SIG_IGN)
251 			sigaddset(ign, i);
252 		else if (k->sa.sa_handler != SIG_DFL)
253 			sigaddset(catch, i);
254 	}
255 }
256 
257 static inline void task_sig(struct seq_file *m, struct task_struct *p)
258 {
259 	unsigned long flags;
260 	sigset_t pending, shpending, blocked, ignored, caught;
261 	int num_threads = 0;
262 	unsigned long qsize = 0;
263 	unsigned long qlim = 0;
264 
265 	sigemptyset(&pending);
266 	sigemptyset(&shpending);
267 	sigemptyset(&blocked);
268 	sigemptyset(&ignored);
269 	sigemptyset(&caught);
270 
271 	if (lock_task_sighand(p, &flags)) {
272 		pending = p->pending.signal;
273 		shpending = p->signal->shared_pending.signal;
274 		blocked = p->blocked;
275 		collect_sigign_sigcatch(p, &ignored, &caught);
276 		num_threads = get_nr_threads(p);
277 		rcu_read_lock();  /* FIXME: is this correct? */
278 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
279 		rcu_read_unlock();
280 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
281 		unlock_task_sighand(p, &flags);
282 	}
283 
284 	seq_printf(m, "Threads:\t%d\n", num_threads);
285 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
286 
287 	/* render them all */
288 	render_sigset_t(m, "SigPnd:\t", &pending);
289 	render_sigset_t(m, "ShdPnd:\t", &shpending);
290 	render_sigset_t(m, "SigBlk:\t", &blocked);
291 	render_sigset_t(m, "SigIgn:\t", &ignored);
292 	render_sigset_t(m, "SigCgt:\t", &caught);
293 }
294 
295 static void render_cap_t(struct seq_file *m, const char *header,
296 			kernel_cap_t *a)
297 {
298 	unsigned __capi;
299 
300 	seq_puts(m, header);
301 	CAP_FOR_EACH_U32(__capi) {
302 		seq_printf(m, "%08x",
303 			   a->cap[CAP_LAST_U32 - __capi]);
304 	}
305 	seq_putc(m, '\n');
306 }
307 
308 static inline void task_cap(struct seq_file *m, struct task_struct *p)
309 {
310 	const struct cred *cred;
311 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
312 			cap_bset, cap_ambient;
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 	cap_ambient	= cred->cap_ambient;
321 	rcu_read_unlock();
322 
323 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
324 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
325 	render_cap_t(m, "CapEff:\t", &cap_effective);
326 	render_cap_t(m, "CapBnd:\t", &cap_bset);
327 	render_cap_t(m, "CapAmb:\t", &cap_ambient);
328 }
329 
330 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
331 {
332 #ifdef CONFIG_SECCOMP
333 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
334 #endif
335 }
336 
337 static inline void task_context_switch_counts(struct seq_file *m,
338 						struct task_struct *p)
339 {
340 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
341 			"nonvoluntary_ctxt_switches:\t%lu\n",
342 			p->nvcsw,
343 			p->nivcsw);
344 }
345 
346 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
347 {
348 	seq_printf(m, "Cpus_allowed:\t%*pb\n",
349 		   cpumask_pr_args(&task->cpus_allowed));
350 	seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
351 		   cpumask_pr_args(&task->cpus_allowed));
352 }
353 
354 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
355 			struct pid *pid, struct task_struct *task)
356 {
357 	struct mm_struct *mm = get_task_mm(task);
358 
359 	task_name(m, task);
360 	task_state(m, ns, pid, task);
361 
362 	if (mm) {
363 		task_mem(m, mm);
364 		mmput(mm);
365 	}
366 	task_sig(m, task);
367 	task_cap(m, task);
368 	task_seccomp(m, task);
369 	task_cpus_allowed(m, task);
370 	cpuset_task_status_allowed(m, task);
371 	task_context_switch_counts(m, task);
372 	return 0;
373 }
374 
375 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
376 			struct pid *pid, struct task_struct *task, int whole)
377 {
378 	unsigned long vsize, eip, esp, wchan = ~0UL;
379 	int priority, nice;
380 	int tty_pgrp = -1, tty_nr = 0;
381 	sigset_t sigign, sigcatch;
382 	char state;
383 	pid_t ppid = 0, pgid = -1, sid = -1;
384 	int num_threads = 0;
385 	int permitted;
386 	struct mm_struct *mm;
387 	unsigned long long start_time;
388 	unsigned long cmin_flt = 0, cmaj_flt = 0;
389 	unsigned long  min_flt = 0,  maj_flt = 0;
390 	cputime_t cutime, cstime, utime, stime;
391 	cputime_t cgtime, gtime;
392 	unsigned long rsslim = 0;
393 	char tcomm[sizeof(task->comm)];
394 	unsigned long flags;
395 
396 	state = *get_task_state(task);
397 	vsize = eip = esp = 0;
398 	permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
399 	mm = get_task_mm(task);
400 	if (mm) {
401 		vsize = task_vsize(mm);
402 		if (permitted) {
403 			eip = KSTK_EIP(task);
404 			esp = KSTK_ESP(task);
405 		}
406 	}
407 
408 	get_task_comm(tcomm, task);
409 
410 	sigemptyset(&sigign);
411 	sigemptyset(&sigcatch);
412 	cutime = cstime = utime = stime = 0;
413 	cgtime = gtime = 0;
414 
415 	if (lock_task_sighand(task, &flags)) {
416 		struct signal_struct *sig = task->signal;
417 
418 		if (sig->tty) {
419 			struct pid *pgrp = tty_get_pgrp(sig->tty);
420 			tty_pgrp = pid_nr_ns(pgrp, ns);
421 			put_pid(pgrp);
422 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
423 		}
424 
425 		num_threads = get_nr_threads(task);
426 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
427 
428 		cmin_flt = sig->cmin_flt;
429 		cmaj_flt = sig->cmaj_flt;
430 		cutime = sig->cutime;
431 		cstime = sig->cstime;
432 		cgtime = sig->cgtime;
433 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
434 
435 		/* add up live thread stats at the group level */
436 		if (whole) {
437 			struct task_struct *t = task;
438 			do {
439 				min_flt += t->min_flt;
440 				maj_flt += t->maj_flt;
441 				gtime += task_gtime(t);
442 			} while_each_thread(task, t);
443 
444 			min_flt += sig->min_flt;
445 			maj_flt += sig->maj_flt;
446 			thread_group_cputime_adjusted(task, &utime, &stime);
447 			gtime += sig->gtime;
448 		}
449 
450 		sid = task_session_nr_ns(task, ns);
451 		ppid = task_tgid_nr_ns(task->real_parent, ns);
452 		pgid = task_pgrp_nr_ns(task, ns);
453 
454 		unlock_task_sighand(task, &flags);
455 	}
456 
457 	if (permitted && (!whole || num_threads < 2))
458 		wchan = get_wchan(task);
459 	if (!whole) {
460 		min_flt = task->min_flt;
461 		maj_flt = task->maj_flt;
462 		task_cputime_adjusted(task, &utime, &stime);
463 		gtime = task_gtime(task);
464 	}
465 
466 	/* scale priority and nice values from timeslices to -20..20 */
467 	/* to make it look like a "normal" Unix priority/nice value  */
468 	priority = task_prio(task);
469 	nice = task_nice(task);
470 
471 	/* convert nsec -> ticks */
472 	start_time = nsec_to_clock_t(task->real_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_PROC_CHILDREN
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 	seq_printf(seq, "%d ", pid);
647 
648 	return 0;
649 }
650 
651 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
652 {
653 	return get_children_pid(seq->private, NULL, *pos);
654 }
655 
656 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
657 {
658 	struct pid *pid;
659 
660 	pid = get_children_pid(seq->private, v, *pos + 1);
661 	put_pid(v);
662 
663 	++*pos;
664 	return pid;
665 }
666 
667 static void children_seq_stop(struct seq_file *seq, void *v)
668 {
669 	put_pid(v);
670 }
671 
672 static const struct seq_operations children_seq_ops = {
673 	.start	= children_seq_start,
674 	.next	= children_seq_next,
675 	.stop	= children_seq_stop,
676 	.show	= children_seq_show,
677 };
678 
679 static int children_seq_open(struct inode *inode, struct file *file)
680 {
681 	struct seq_file *m;
682 	int ret;
683 
684 	ret = seq_open(file, &children_seq_ops);
685 	if (ret)
686 		return ret;
687 
688 	m = file->private_data;
689 	m->private = inode;
690 
691 	return ret;
692 }
693 
694 int children_seq_release(struct inode *inode, struct file *file)
695 {
696 	seq_release(inode, file);
697 	return 0;
698 }
699 
700 const struct file_operations proc_tid_children_operations = {
701 	.open    = children_seq_open,
702 	.read    = seq_read,
703 	.llseek  = seq_lseek,
704 	.release = children_seq_release,
705 };
706 #endif /* CONFIG_PROC_CHILDREN */
707