xref: /openbmc/linux/fs/proc/array.c (revision 32981ea5)
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 #include <linux/fs_struct.h>
87 
88 #include <asm/pgtable.h>
89 #include <asm/processor.h>
90 #include "internal.h"
91 
92 static inline void task_name(struct seq_file *m, struct task_struct *p)
93 {
94 	char *buf;
95 	size_t size;
96 	char tcomm[sizeof(p->comm)];
97 	int ret;
98 
99 	get_task_comm(tcomm, p);
100 
101 	seq_puts(m, "Name:\t");
102 
103 	size = seq_get_buf(m, &buf);
104 	ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
105 	seq_commit(m, ret < size ? ret : -1);
106 
107 	seq_putc(m, '\n');
108 }
109 
110 /*
111  * The task state array is a strange "bitmap" of
112  * reasons to sleep. Thus "running" is zero, and
113  * you can test for combinations of others with
114  * simple bit tests.
115  */
116 static const char * const task_state_array[] = {
117 	"R (running)",		/*   0 */
118 	"S (sleeping)",		/*   1 */
119 	"D (disk sleep)",	/*   2 */
120 	"T (stopped)",		/*   4 */
121 	"t (tracing stop)",	/*   8 */
122 	"X (dead)",		/*  16 */
123 	"Z (zombie)",		/*  32 */
124 };
125 
126 static inline const char *get_task_state(struct task_struct *tsk)
127 {
128 	unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
129 
130 	/*
131 	 * Parked tasks do not run; they sit in __kthread_parkme().
132 	 * Without this check, we would report them as running, which is
133 	 * clearly wrong, so we report them as sleeping instead.
134 	 */
135 	if (tsk->state == TASK_PARKED)
136 		state = TASK_INTERRUPTIBLE;
137 
138 	BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
139 
140 	return task_state_array[fls(state)];
141 }
142 
143 static inline int get_task_umask(struct task_struct *tsk)
144 {
145 	struct fs_struct *fs;
146 	int umask = -ENOENT;
147 
148 	task_lock(tsk);
149 	fs = tsk->fs;
150 	if (fs)
151 		umask = fs->umask;
152 	task_unlock(tsk);
153 	return umask;
154 }
155 
156 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
157 				struct pid *pid, struct task_struct *p)
158 {
159 	struct user_namespace *user_ns = seq_user_ns(m);
160 	struct group_info *group_info;
161 	int g, umask;
162 	struct task_struct *tracer;
163 	const struct cred *cred;
164 	pid_t ppid, tpid = 0, tgid, ngid;
165 	unsigned int max_fds = 0;
166 
167 	rcu_read_lock();
168 	ppid = pid_alive(p) ?
169 		task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
170 
171 	tracer = ptrace_parent(p);
172 	if (tracer)
173 		tpid = task_pid_nr_ns(tracer, ns);
174 
175 	tgid = task_tgid_nr_ns(p, ns);
176 	ngid = task_numa_group_id(p);
177 	cred = get_task_cred(p);
178 
179 	umask = get_task_umask(p);
180 	if (umask >= 0)
181 		seq_printf(m, "Umask:\t%#04o\n", umask);
182 
183 	task_lock(p);
184 	if (p->files)
185 		max_fds = files_fdtable(p->files)->max_fds;
186 	task_unlock(p);
187 	rcu_read_unlock();
188 
189 	seq_printf(m,
190 		"State:\t%s\n"
191 		"Tgid:\t%d\n"
192 		"Ngid:\t%d\n"
193 		"Pid:\t%d\n"
194 		"PPid:\t%d\n"
195 		"TracerPid:\t%d\n"
196 		"Uid:\t%d\t%d\t%d\t%d\n"
197 		"Gid:\t%d\t%d\t%d\t%d\n"
198 		"FDSize:\t%d\nGroups:\t",
199 		get_task_state(p),
200 		tgid, ngid, pid_nr_ns(pid, ns), ppid, tpid,
201 		from_kuid_munged(user_ns, cred->uid),
202 		from_kuid_munged(user_ns, cred->euid),
203 		from_kuid_munged(user_ns, cred->suid),
204 		from_kuid_munged(user_ns, cred->fsuid),
205 		from_kgid_munged(user_ns, cred->gid),
206 		from_kgid_munged(user_ns, cred->egid),
207 		from_kgid_munged(user_ns, cred->sgid),
208 		from_kgid_munged(user_ns, cred->fsgid),
209 		max_fds);
210 
211 	group_info = cred->group_info;
212 	for (g = 0; g < group_info->ngroups; g++)
213 		seq_printf(m, "%d ",
214 			   from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
215 	put_cred(cred);
216 
217 #ifdef CONFIG_PID_NS
218 	seq_puts(m, "\nNStgid:");
219 	for (g = ns->level; g <= pid->level; g++)
220 		seq_printf(m, "\t%d",
221 			task_tgid_nr_ns(p, pid->numbers[g].ns));
222 	seq_puts(m, "\nNSpid:");
223 	for (g = ns->level; g <= pid->level; g++)
224 		seq_printf(m, "\t%d",
225 			task_pid_nr_ns(p, pid->numbers[g].ns));
226 	seq_puts(m, "\nNSpgid:");
227 	for (g = ns->level; g <= pid->level; g++)
228 		seq_printf(m, "\t%d",
229 			task_pgrp_nr_ns(p, pid->numbers[g].ns));
230 	seq_puts(m, "\nNSsid:");
231 	for (g = ns->level; g <= pid->level; g++)
232 		seq_printf(m, "\t%d",
233 			task_session_nr_ns(p, pid->numbers[g].ns));
234 #endif
235 	seq_putc(m, '\n');
236 }
237 
238 void render_sigset_t(struct seq_file *m, const char *header,
239 				sigset_t *set)
240 {
241 	int i;
242 
243 	seq_puts(m, header);
244 
245 	i = _NSIG;
246 	do {
247 		int x = 0;
248 
249 		i -= 4;
250 		if (sigismember(set, i+1)) x |= 1;
251 		if (sigismember(set, i+2)) x |= 2;
252 		if (sigismember(set, i+3)) x |= 4;
253 		if (sigismember(set, i+4)) x |= 8;
254 		seq_printf(m, "%x", x);
255 	} while (i >= 4);
256 
257 	seq_putc(m, '\n');
258 }
259 
260 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
261 				    sigset_t *catch)
262 {
263 	struct k_sigaction *k;
264 	int i;
265 
266 	k = p->sighand->action;
267 	for (i = 1; i <= _NSIG; ++i, ++k) {
268 		if (k->sa.sa_handler == SIG_IGN)
269 			sigaddset(ign, i);
270 		else if (k->sa.sa_handler != SIG_DFL)
271 			sigaddset(catch, i);
272 	}
273 }
274 
275 static inline void task_sig(struct seq_file *m, struct task_struct *p)
276 {
277 	unsigned long flags;
278 	sigset_t pending, shpending, blocked, ignored, caught;
279 	int num_threads = 0;
280 	unsigned long qsize = 0;
281 	unsigned long qlim = 0;
282 
283 	sigemptyset(&pending);
284 	sigemptyset(&shpending);
285 	sigemptyset(&blocked);
286 	sigemptyset(&ignored);
287 	sigemptyset(&caught);
288 
289 	if (lock_task_sighand(p, &flags)) {
290 		pending = p->pending.signal;
291 		shpending = p->signal->shared_pending.signal;
292 		blocked = p->blocked;
293 		collect_sigign_sigcatch(p, &ignored, &caught);
294 		num_threads = get_nr_threads(p);
295 		rcu_read_lock();  /* FIXME: is this correct? */
296 		qsize = atomic_read(&__task_cred(p)->user->sigpending);
297 		rcu_read_unlock();
298 		qlim = task_rlimit(p, RLIMIT_SIGPENDING);
299 		unlock_task_sighand(p, &flags);
300 	}
301 
302 	seq_printf(m, "Threads:\t%d\n", num_threads);
303 	seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
304 
305 	/* render them all */
306 	render_sigset_t(m, "SigPnd:\t", &pending);
307 	render_sigset_t(m, "ShdPnd:\t", &shpending);
308 	render_sigset_t(m, "SigBlk:\t", &blocked);
309 	render_sigset_t(m, "SigIgn:\t", &ignored);
310 	render_sigset_t(m, "SigCgt:\t", &caught);
311 }
312 
313 static void render_cap_t(struct seq_file *m, const char *header,
314 			kernel_cap_t *a)
315 {
316 	unsigned __capi;
317 
318 	seq_puts(m, header);
319 	CAP_FOR_EACH_U32(__capi) {
320 		seq_printf(m, "%08x",
321 			   a->cap[CAP_LAST_U32 - __capi]);
322 	}
323 	seq_putc(m, '\n');
324 }
325 
326 static inline void task_cap(struct seq_file *m, struct task_struct *p)
327 {
328 	const struct cred *cred;
329 	kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
330 			cap_bset, cap_ambient;
331 
332 	rcu_read_lock();
333 	cred = __task_cred(p);
334 	cap_inheritable	= cred->cap_inheritable;
335 	cap_permitted	= cred->cap_permitted;
336 	cap_effective	= cred->cap_effective;
337 	cap_bset	= cred->cap_bset;
338 	cap_ambient	= cred->cap_ambient;
339 	rcu_read_unlock();
340 
341 	render_cap_t(m, "CapInh:\t", &cap_inheritable);
342 	render_cap_t(m, "CapPrm:\t", &cap_permitted);
343 	render_cap_t(m, "CapEff:\t", &cap_effective);
344 	render_cap_t(m, "CapBnd:\t", &cap_bset);
345 	render_cap_t(m, "CapAmb:\t", &cap_ambient);
346 }
347 
348 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
349 {
350 #ifdef CONFIG_SECCOMP
351 	seq_printf(m, "Seccomp:\t%d\n", p->seccomp.mode);
352 #endif
353 }
354 
355 static inline void task_context_switch_counts(struct seq_file *m,
356 						struct task_struct *p)
357 {
358 	seq_printf(m,	"voluntary_ctxt_switches:\t%lu\n"
359 			"nonvoluntary_ctxt_switches:\t%lu\n",
360 			p->nvcsw,
361 			p->nivcsw);
362 }
363 
364 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
365 {
366 	seq_printf(m, "Cpus_allowed:\t%*pb\n",
367 		   cpumask_pr_args(&task->cpus_allowed));
368 	seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
369 		   cpumask_pr_args(&task->cpus_allowed));
370 }
371 
372 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
373 			struct pid *pid, struct task_struct *task)
374 {
375 	struct mm_struct *mm = get_task_mm(task);
376 
377 	task_name(m, task);
378 	task_state(m, ns, pid, task);
379 
380 	if (mm) {
381 		task_mem(m, mm);
382 		mmput(mm);
383 	}
384 	task_sig(m, task);
385 	task_cap(m, task);
386 	task_seccomp(m, task);
387 	task_cpus_allowed(m, task);
388 	cpuset_task_status_allowed(m, task);
389 	task_context_switch_counts(m, task);
390 	return 0;
391 }
392 
393 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
394 			struct pid *pid, struct task_struct *task, int whole)
395 {
396 	unsigned long vsize, eip, esp, wchan = 0;
397 	int priority, nice;
398 	int tty_pgrp = -1, tty_nr = 0;
399 	sigset_t sigign, sigcatch;
400 	char state;
401 	pid_t ppid = 0, pgid = -1, sid = -1;
402 	int num_threads = 0;
403 	int permitted;
404 	struct mm_struct *mm;
405 	unsigned long long start_time;
406 	unsigned long cmin_flt = 0, cmaj_flt = 0;
407 	unsigned long  min_flt = 0,  maj_flt = 0;
408 	cputime_t cutime, cstime, utime, stime;
409 	cputime_t cgtime, gtime;
410 	unsigned long rsslim = 0;
411 	char tcomm[sizeof(task->comm)];
412 	unsigned long flags;
413 
414 	state = *get_task_state(task);
415 	vsize = eip = esp = 0;
416 	permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
417 	mm = get_task_mm(task);
418 	if (mm) {
419 		vsize = task_vsize(mm);
420 		if (permitted) {
421 			eip = KSTK_EIP(task);
422 			esp = KSTK_ESP(task);
423 		}
424 	}
425 
426 	get_task_comm(tcomm, task);
427 
428 	sigemptyset(&sigign);
429 	sigemptyset(&sigcatch);
430 	cutime = cstime = utime = stime = 0;
431 	cgtime = gtime = 0;
432 
433 	if (lock_task_sighand(task, &flags)) {
434 		struct signal_struct *sig = task->signal;
435 
436 		if (sig->tty) {
437 			struct pid *pgrp = tty_get_pgrp(sig->tty);
438 			tty_pgrp = pid_nr_ns(pgrp, ns);
439 			put_pid(pgrp);
440 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
441 		}
442 
443 		num_threads = get_nr_threads(task);
444 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
445 
446 		cmin_flt = sig->cmin_flt;
447 		cmaj_flt = sig->cmaj_flt;
448 		cutime = sig->cutime;
449 		cstime = sig->cstime;
450 		cgtime = sig->cgtime;
451 		rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
452 
453 		/* add up live thread stats at the group level */
454 		if (whole) {
455 			struct task_struct *t = task;
456 			do {
457 				min_flt += t->min_flt;
458 				maj_flt += t->maj_flt;
459 				gtime += task_gtime(t);
460 			} while_each_thread(task, t);
461 
462 			min_flt += sig->min_flt;
463 			maj_flt += sig->maj_flt;
464 			thread_group_cputime_adjusted(task, &utime, &stime);
465 			gtime += sig->gtime;
466 		}
467 
468 		sid = task_session_nr_ns(task, ns);
469 		ppid = task_tgid_nr_ns(task->real_parent, ns);
470 		pgid = task_pgrp_nr_ns(task, ns);
471 
472 		unlock_task_sighand(task, &flags);
473 	}
474 
475 	if (permitted && (!whole || num_threads < 2))
476 		wchan = get_wchan(task);
477 	if (!whole) {
478 		min_flt = task->min_flt;
479 		maj_flt = task->maj_flt;
480 		task_cputime_adjusted(task, &utime, &stime);
481 		gtime = task_gtime(task);
482 	}
483 
484 	/* scale priority and nice values from timeslices to -20..20 */
485 	/* to make it look like a "normal" Unix priority/nice value  */
486 	priority = task_prio(task);
487 	nice = task_nice(task);
488 
489 	/* convert nsec -> ticks */
490 	start_time = nsec_to_clock_t(task->real_start_time);
491 
492 	seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
493 	seq_put_decimal_ll(m, ' ', ppid);
494 	seq_put_decimal_ll(m, ' ', pgid);
495 	seq_put_decimal_ll(m, ' ', sid);
496 	seq_put_decimal_ll(m, ' ', tty_nr);
497 	seq_put_decimal_ll(m, ' ', tty_pgrp);
498 	seq_put_decimal_ull(m, ' ', task->flags);
499 	seq_put_decimal_ull(m, ' ', min_flt);
500 	seq_put_decimal_ull(m, ' ', cmin_flt);
501 	seq_put_decimal_ull(m, ' ', maj_flt);
502 	seq_put_decimal_ull(m, ' ', cmaj_flt);
503 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
504 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
505 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
506 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
507 	seq_put_decimal_ll(m, ' ', priority);
508 	seq_put_decimal_ll(m, ' ', nice);
509 	seq_put_decimal_ll(m, ' ', num_threads);
510 	seq_put_decimal_ull(m, ' ', 0);
511 	seq_put_decimal_ull(m, ' ', start_time);
512 	seq_put_decimal_ull(m, ' ', vsize);
513 	seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
514 	seq_put_decimal_ull(m, ' ', rsslim);
515 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
516 	seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
517 	seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
518 	seq_put_decimal_ull(m, ' ', esp);
519 	seq_put_decimal_ull(m, ' ', eip);
520 	/* The signal information here is obsolete.
521 	 * It must be decimal for Linux 2.0 compatibility.
522 	 * Use /proc/#/status for real-time signals.
523 	 */
524 	seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
525 	seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
526 	seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
527 	seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
528 
529 	/*
530 	 * We used to output the absolute kernel address, but that's an
531 	 * information leak - so instead we show a 0/1 flag here, to signal
532 	 * to user-space whether there's a wchan field in /proc/PID/wchan.
533 	 *
534 	 * This works with older implementations of procps as well.
535 	 */
536 	if (wchan)
537 		seq_puts(m, " 1");
538 	else
539 		seq_puts(m, " 0");
540 
541 	seq_put_decimal_ull(m, ' ', 0);
542 	seq_put_decimal_ull(m, ' ', 0);
543 	seq_put_decimal_ll(m, ' ', task->exit_signal);
544 	seq_put_decimal_ll(m, ' ', task_cpu(task));
545 	seq_put_decimal_ull(m, ' ', task->rt_priority);
546 	seq_put_decimal_ull(m, ' ', task->policy);
547 	seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
548 	seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
549 	seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
550 
551 	if (mm && permitted) {
552 		seq_put_decimal_ull(m, ' ', mm->start_data);
553 		seq_put_decimal_ull(m, ' ', mm->end_data);
554 		seq_put_decimal_ull(m, ' ', mm->start_brk);
555 		seq_put_decimal_ull(m, ' ', mm->arg_start);
556 		seq_put_decimal_ull(m, ' ', mm->arg_end);
557 		seq_put_decimal_ull(m, ' ', mm->env_start);
558 		seq_put_decimal_ull(m, ' ', mm->env_end);
559 	} else
560 		seq_printf(m, " 0 0 0 0 0 0 0");
561 
562 	if (permitted)
563 		seq_put_decimal_ll(m, ' ', task->exit_code);
564 	else
565 		seq_put_decimal_ll(m, ' ', 0);
566 
567 	seq_putc(m, '\n');
568 	if (mm)
569 		mmput(mm);
570 	return 0;
571 }
572 
573 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
574 			struct pid *pid, struct task_struct *task)
575 {
576 	return do_task_stat(m, ns, pid, task, 0);
577 }
578 
579 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
580 			struct pid *pid, struct task_struct *task)
581 {
582 	return do_task_stat(m, ns, pid, task, 1);
583 }
584 
585 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
586 			struct pid *pid, struct task_struct *task)
587 {
588 	unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
589 	struct mm_struct *mm = get_task_mm(task);
590 
591 	if (mm) {
592 		size = task_statm(mm, &shared, &text, &data, &resident);
593 		mmput(mm);
594 	}
595 	/*
596 	 * For quick read, open code by putting numbers directly
597 	 * expected format is
598 	 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
599 	 *               size, resident, shared, text, data);
600 	 */
601 	seq_put_decimal_ull(m, 0, size);
602 	seq_put_decimal_ull(m, ' ', resident);
603 	seq_put_decimal_ull(m, ' ', shared);
604 	seq_put_decimal_ull(m, ' ', text);
605 	seq_put_decimal_ull(m, ' ', 0);
606 	seq_put_decimal_ull(m, ' ', data);
607 	seq_put_decimal_ull(m, ' ', 0);
608 	seq_putc(m, '\n');
609 
610 	return 0;
611 }
612 
613 #ifdef CONFIG_PROC_CHILDREN
614 static struct pid *
615 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
616 {
617 	struct task_struct *start, *task;
618 	struct pid *pid = NULL;
619 
620 	read_lock(&tasklist_lock);
621 
622 	start = pid_task(proc_pid(inode), PIDTYPE_PID);
623 	if (!start)
624 		goto out;
625 
626 	/*
627 	 * Lets try to continue searching first, this gives
628 	 * us significant speedup on children-rich processes.
629 	 */
630 	if (pid_prev) {
631 		task = pid_task(pid_prev, PIDTYPE_PID);
632 		if (task && task->real_parent == start &&
633 		    !(list_empty(&task->sibling))) {
634 			if (list_is_last(&task->sibling, &start->children))
635 				goto out;
636 			task = list_first_entry(&task->sibling,
637 						struct task_struct, sibling);
638 			pid = get_pid(task_pid(task));
639 			goto out;
640 		}
641 	}
642 
643 	/*
644 	 * Slow search case.
645 	 *
646 	 * We might miss some children here if children
647 	 * are exited while we were not holding the lock,
648 	 * but it was never promised to be accurate that
649 	 * much.
650 	 *
651 	 * "Just suppose that the parent sleeps, but N children
652 	 *  exit after we printed their tids. Now the slow paths
653 	 *  skips N extra children, we miss N tasks." (c)
654 	 *
655 	 * So one need to stop or freeze the leader and all
656 	 * its children to get a precise result.
657 	 */
658 	list_for_each_entry(task, &start->children, sibling) {
659 		if (pos-- == 0) {
660 			pid = get_pid(task_pid(task));
661 			break;
662 		}
663 	}
664 
665 out:
666 	read_unlock(&tasklist_lock);
667 	return pid;
668 }
669 
670 static int children_seq_show(struct seq_file *seq, void *v)
671 {
672 	struct inode *inode = seq->private;
673 	pid_t pid;
674 
675 	pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
676 	seq_printf(seq, "%d ", pid);
677 
678 	return 0;
679 }
680 
681 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
682 {
683 	return get_children_pid(seq->private, NULL, *pos);
684 }
685 
686 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
687 {
688 	struct pid *pid;
689 
690 	pid = get_children_pid(seq->private, v, *pos + 1);
691 	put_pid(v);
692 
693 	++*pos;
694 	return pid;
695 }
696 
697 static void children_seq_stop(struct seq_file *seq, void *v)
698 {
699 	put_pid(v);
700 }
701 
702 static const struct seq_operations children_seq_ops = {
703 	.start	= children_seq_start,
704 	.next	= children_seq_next,
705 	.stop	= children_seq_stop,
706 	.show	= children_seq_show,
707 };
708 
709 static int children_seq_open(struct inode *inode, struct file *file)
710 {
711 	struct seq_file *m;
712 	int ret;
713 
714 	ret = seq_open(file, &children_seq_ops);
715 	if (ret)
716 		return ret;
717 
718 	m = file->private_data;
719 	m->private = inode;
720 
721 	return ret;
722 }
723 
724 int children_seq_release(struct inode *inode, struct file *file)
725 {
726 	seq_release(inode, file);
727 	return 0;
728 }
729 
730 const struct file_operations proc_tid_children_operations = {
731 	.open    = children_seq_open,
732 	.read    = seq_read,
733 	.llseek  = seq_lseek,
734 	.release = children_seq_release,
735 };
736 #endif /* CONFIG_PROC_CHILDREN */
737