xref: /openbmc/linux/fs/proc/array.c (revision 2572f00d)
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 	size_t size;
95 	char tcomm[sizeof(p->comm)];
96 	int ret;
97 
98 	get_task_comm(tcomm, p);
99 
100 	seq_puts(m, "Name:\t");
101 
102 	size = seq_get_buf(m, &buf);
103 	ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
104 	seq_commit(m, ret < size ? ret : -1);
105 
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 = 0;
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 
511 	/*
512 	 * We used to output the absolute kernel address, but that's an
513 	 * information leak - so instead we show a 0/1 flag here, to signal
514 	 * to user-space whether there's a wchan field in /proc/PID/wchan.
515 	 *
516 	 * This works with older implementations of procps as well.
517 	 */
518 	if (wchan)
519 		seq_puts(m, " 1");
520 	else
521 		seq_puts(m, " 0");
522 
523 	seq_put_decimal_ull(m, ' ', 0);
524 	seq_put_decimal_ull(m, ' ', 0);
525 	seq_put_decimal_ll(m, ' ', task->exit_signal);
526 	seq_put_decimal_ll(m, ' ', task_cpu(task));
527 	seq_put_decimal_ull(m, ' ', task->rt_priority);
528 	seq_put_decimal_ull(m, ' ', task->policy);
529 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
530 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
531 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
532 
533 	if (mm && permitted) {
534 		seq_put_decimal_ull(m, ' ', mm->start_data);
535 		seq_put_decimal_ull(m, ' ', mm->end_data);
536 		seq_put_decimal_ull(m, ' ', mm->start_brk);
537 		seq_put_decimal_ull(m, ' ', mm->arg_start);
538 		seq_put_decimal_ull(m, ' ', mm->arg_end);
539 		seq_put_decimal_ull(m, ' ', mm->env_start);
540 		seq_put_decimal_ull(m, ' ', mm->env_end);
541 	} else
542 		seq_printf(m, " 0 0 0 0 0 0 0");
543 
544 	if (permitted)
545 		seq_put_decimal_ll(m, ' ', task->exit_code);
546 	else
547 		seq_put_decimal_ll(m, ' ', 0);
548 
549 	seq_putc(m, '\n');
550 	if (mm)
551 		mmput(mm);
552 	return 0;
553 }
554 
555 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
556 			struct pid *pid, struct task_struct *task)
557 {
558 	return do_task_stat(m, ns, pid, task, 0);
559 }
560 
561 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
562 			struct pid *pid, struct task_struct *task)
563 {
564 	return do_task_stat(m, ns, pid, task, 1);
565 }
566 
567 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
568 			struct pid *pid, struct task_struct *task)
569 {
570 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
571 	struct mm_struct *mm = get_task_mm(task);
572 
573 	if (mm) {
574 		size = task_statm(mm, &shared, &text, &data, &resident);
575 		mmput(mm);
576 	}
577 	/*
578 	 * For quick read, open code by putting numbers directly
579 	 * expected format is
580 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
581 	 *               size, resident, shared, text, data);
582 	 */
583 	seq_put_decimal_ull(m, 0, size);
584 	seq_put_decimal_ull(m, ' ', resident);
585 	seq_put_decimal_ull(m, ' ', shared);
586 	seq_put_decimal_ull(m, ' ', text);
587 	seq_put_decimal_ull(m, ' ', 0);
588 	seq_put_decimal_ull(m, ' ', data);
589 	seq_put_decimal_ull(m, ' ', 0);
590 	seq_putc(m, '\n');
591 
592 	return 0;
593 }
594 
595 #ifdef CONFIG_PROC_CHILDREN
596 static struct pid *
597 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
598 {
599 	struct task_struct *start, *task;
600 	struct pid *pid = NULL;
601 
602 	read_lock(&tasklist_lock);
603 
604 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
605 	if (!start)
606 		goto out;
607 
608 	/*
609 	 * Lets try to continue searching first, this gives
610 	 * us significant speedup on children-rich processes.
611 	 */
612 	if (pid_prev) {
613 		task = pid_task(pid_prev, PIDTYPE_PID);
614 		if (task && task->real_parent == start &&
615 		    !(list_empty(&task->sibling))) {
616 			if (list_is_last(&task->sibling, &start->children))
617 				goto out;
618 			task = list_first_entry(&task->sibling,
619 						struct task_struct, sibling);
620 			pid = get_pid(task_pid(task));
621 			goto out;
622 		}
623 	}
624 
625 	/*
626 	 * Slow search case.
627 	 *
628 	 * We might miss some children here if children
629 	 * are exited while we were not holding the lock,
630 	 * but it was never promised to be accurate that
631 	 * much.
632 	 *
633 	 * "Just suppose that the parent sleeps, but N children
634 	 *  exit after we printed their tids. Now the slow paths
635 	 *  skips N extra children, we miss N tasks." (c)
636 	 *
637 	 * So one need to stop or freeze the leader and all
638 	 * its children to get a precise result.
639 	 */
640 	list_for_each_entry(task, &start->children, sibling) {
641 		if (pos-- == 0) {
642 			pid = get_pid(task_pid(task));
643 			break;
644 		}
645 	}
646 
647 out:
648 	read_unlock(&tasklist_lock);
649 	return pid;
650 }
651 
652 static int children_seq_show(struct seq_file *seq, void *v)
653 {
654 	struct inode *inode = seq->private;
655 	pid_t pid;
656 
657 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
658 	seq_printf(seq, "%d ", pid);
659 
660 	return 0;
661 }
662 
663 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
664 {
665 	return get_children_pid(seq->private, NULL, *pos);
666 }
667 
668 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
669 {
670 	struct pid *pid;
671 
672 	pid = get_children_pid(seq->private, v, *pos + 1);
673 	put_pid(v);
674 
675 	++*pos;
676 	return pid;
677 }
678 
679 static void children_seq_stop(struct seq_file *seq, void *v)
680 {
681 	put_pid(v);
682 }
683 
684 static const struct seq_operations children_seq_ops = {
685 	.start	= children_seq_start,
686 	.next	= children_seq_next,
687 	.stop	= children_seq_stop,
688 	.show	= children_seq_show,
689 };
690 
691 static int children_seq_open(struct inode *inode, struct file *file)
692 {
693 	struct seq_file *m;
694 	int ret;
695 
696 	ret = seq_open(file, &children_seq_ops);
697 	if (ret)
698 		return ret;
699 
700 	m = file->private_data;
701 	m->private = inode;
702 
703 	return ret;
704 }
705 
706 int children_seq_release(struct inode *inode, struct file *file)
707 {
708 	seq_release(inode, file);
709 	return 0;
710 }
711 
712 const struct file_operations proc_tid_children_operations = {
713 	.open    = children_seq_open,
714 	.read    = seq_read,
715 	.llseek  = seq_lseek,
716 	.release = children_seq_release,
717 };
718 #endif /* CONFIG_PROC_CHILDREN */
719