1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/proc/array.c
5 * Copyright (C) 1992 by Linus Torvalds
6 * based on ideas by Darren Senn
9 * Michael. K. Johnson: stat,statm extensions.
10 * <johnsonm@stolaf.edu>
12 * Pauline Middelink : Made cmdline,envline only break at '\0's, to
13 * make sure SET_PROCTITLE works. Also removed
14 * bad '!' which forced address recalculation for
15 * EVERY character on the current page.
16 * <middelin@polyware.iaf.nl>
18 * Danny ter Haar : added cpuinfo
21 * Alessandro Rubini : profile extension.
22 * <rubini@ipvvis.unipv.it>
24 * Jeff Tranter : added BogoMips field to cpuinfo
25 * <Jeff_Tranter@Mitel.COM>
27 * Bruno Haible : remove 4K limit for the maps file
28 * <haible@ma2s2.mathematik.uni-karlsruhe.de>
30 * Yves Arrouye : remove removal of trailing spaces in get_array.
31 * <Yves.Arrouye@marin.fdn.fr>
33 * Jerome Forissier : added per-CPU time information to /proc/stat
34 * and /proc/<pid>/cpu extension
35 * <forissier@isia.cma.fr>
36 * - Incorporation and non-SMP safe operation
37 * of forissier patch in 2.1.78 by
38 * Hans Marcus <crowbar@concepts.nl>
40 * aeb@cwi.nl : /proc/partitions
43 * Alan Cox : security fixes.
44 * <alan@lxorguk.ukuu.org.uk>
46 * Al Viro : safe handling of mm_struct
48 * Gerhard Wichert : added BIGMEM support
49 * Siemens AG <Gerhard.Wichert@pdb.siemens.de>
51 * Al Viro & Jeff Garzik : moved most of the thing into base.c and
52 * : proc_misc.c. The rest may eventually go into
56 #include <linux/types.h>
57 #include <linux/errno.h>
58 #include <linux/time.h>
59 #include <linux/time_namespace.h>
60 #include <linux/kernel.h>
61 #include <linux/kernel_stat.h>
62 #include <linux/tty.h>
63 #include <linux/string.h>
64 #include <linux/mman.h>
65 #include <linux/sched/mm.h>
66 #include <linux/sched/numa_balancing.h>
67 #include <linux/sched/task_stack.h>
68 #include <linux/sched/task.h>
69 #include <linux/sched/cputime.h>
70 #include <linux/proc_fs.h>
71 #include <linux/ioport.h>
72 #include <linux/uaccess.h>
75 #include <linux/hugetlb.h>
76 #include <linux/pagemap.h>
77 #include <linux/swap.h>
78 #include <linux/smp.h>
79 #include <linux/signal.h>
80 #include <linux/highmem.h>
81 #include <linux/file.h>
82 #include <linux/fdtable.h>
83 #include <linux/times.h>
84 #include <linux/cpuset.h>
85 #include <linux/rcupdate.h>
86 #include <linux/delayacct.h>
87 #include <linux/seq_file.h>
88 #include <linux/pid_namespace.h>
89 #include <linux/prctl.h>
90 #include <linux/ptrace.h>
91 #include <linux/tracehook.h>
92 #include <linux/string_helpers.h>
93 #include <linux/user_namespace.h>
94 #include <linux/fs_struct.h>
95 #include <linux/kthread.h>
97 #include <asm/processor.h>
100 void proc_task_name(struct seq_file *m, struct task_struct *p, bool escape)
104 if (p->flags & PF_WQ_WORKER)
105 wq_worker_comm(tcomm, sizeof(tcomm), p);
106 else if (p->flags & PF_KTHREAD)
107 get_kthread_comm(tcomm, sizeof(tcomm), p);
109 __get_task_comm(tcomm, sizeof(tcomm), p);
112 seq_escape_str(m, tcomm, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
114 seq_printf(m, "%.64s", tcomm);
118 * The task state array is a strange "bitmap" of
119 * reasons to sleep. Thus "running" is zero, and
120 * you can test for combinations of others with
123 static const char * const task_state_array[] = {
125 /* states in TASK_REPORT: */
126 "R (running)", /* 0x00 */
127 "S (sleeping)", /* 0x01 */
128 "D (disk sleep)", /* 0x02 */
129 "T (stopped)", /* 0x04 */
130 "t (tracing stop)", /* 0x08 */
131 "X (dead)", /* 0x10 */
132 "Z (zombie)", /* 0x20 */
133 "P (parked)", /* 0x40 */
135 /* states beyond TASK_REPORT: */
136 "I (idle)", /* 0x80 */
139 static inline const char *get_task_state(struct task_struct *tsk)
141 BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != ARRAY_SIZE(task_state_array));
142 return task_state_array[task_state_index(tsk)];
145 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
146 struct pid *pid, struct task_struct *p)
148 struct user_namespace *user_ns = seq_user_ns(m);
149 struct group_info *group_info;
151 struct task_struct *tracer;
152 const struct cred *cred;
153 pid_t ppid, tpid = 0, tgid, ngid;
154 unsigned int max_fds = 0;
157 ppid = pid_alive(p) ?
158 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
160 tracer = ptrace_parent(p);
162 tpid = task_pid_nr_ns(tracer, ns);
164 tgid = task_tgid_nr_ns(p, ns);
165 ngid = task_numa_group_id(p);
166 cred = get_task_cred(p);
170 umask = p->fs->umask;
172 max_fds = files_fdtable(p->files)->max_fds;
177 seq_printf(m, "Umask:\t%#04o\n", umask);
178 seq_puts(m, "State:\t");
179 seq_puts(m, get_task_state(p));
181 seq_put_decimal_ull(m, "\nTgid:\t", tgid);
182 seq_put_decimal_ull(m, "\nNgid:\t", ngid);
183 seq_put_decimal_ull(m, "\nPid:\t", pid_nr_ns(pid, ns));
184 seq_put_decimal_ull(m, "\nPPid:\t", ppid);
185 seq_put_decimal_ull(m, "\nTracerPid:\t", tpid);
186 seq_put_decimal_ull(m, "\nUid:\t", from_kuid_munged(user_ns, cred->uid));
187 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->euid));
188 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->suid));
189 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->fsuid));
190 seq_put_decimal_ull(m, "\nGid:\t", from_kgid_munged(user_ns, cred->gid));
191 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->egid));
192 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->sgid));
193 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->fsgid));
194 seq_put_decimal_ull(m, "\nFDSize:\t", max_fds);
196 seq_puts(m, "\nGroups:\t");
197 group_info = cred->group_info;
198 for (g = 0; g < group_info->ngroups; g++)
199 seq_put_decimal_ull(m, g ? " " : "",
200 from_kgid_munged(user_ns, group_info->gid[g]));
202 /* Trailing space shouldn't have been added in the first place. */
206 seq_puts(m, "\nNStgid:");
207 for (g = ns->level; g <= pid->level; g++)
208 seq_put_decimal_ull(m, "\t", task_tgid_nr_ns(p, pid->numbers[g].ns));
209 seq_puts(m, "\nNSpid:");
210 for (g = ns->level; g <= pid->level; g++)
211 seq_put_decimal_ull(m, "\t", task_pid_nr_ns(p, pid->numbers[g].ns));
212 seq_puts(m, "\nNSpgid:");
213 for (g = ns->level; g <= pid->level; g++)
214 seq_put_decimal_ull(m, "\t", task_pgrp_nr_ns(p, pid->numbers[g].ns));
215 seq_puts(m, "\nNSsid:");
216 for (g = ns->level; g <= pid->level; g++)
217 seq_put_decimal_ull(m, "\t", task_session_nr_ns(p, pid->numbers[g].ns));
222 void render_sigset_t(struct seq_file *m, const char *header,
234 if (sigismember(set, i+1)) x |= 1;
235 if (sigismember(set, i+2)) x |= 2;
236 if (sigismember(set, i+3)) x |= 4;
237 if (sigismember(set, i+4)) x |= 8;
238 seq_putc(m, hex_asc[x]);
244 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *sigign,
247 struct k_sigaction *k;
250 k = p->sighand->action;
251 for (i = 1; i <= _NSIG; ++i, ++k) {
252 if (k->sa.sa_handler == SIG_IGN)
253 sigaddset(sigign, i);
254 else if (k->sa.sa_handler != SIG_DFL)
255 sigaddset(sigcatch, i);
259 static inline void task_sig(struct seq_file *m, struct task_struct *p)
262 sigset_t pending, shpending, blocked, ignored, caught;
264 unsigned int qsize = 0;
265 unsigned long qlim = 0;
267 sigemptyset(&pending);
268 sigemptyset(&shpending);
269 sigemptyset(&blocked);
270 sigemptyset(&ignored);
271 sigemptyset(&caught);
273 if (lock_task_sighand(p, &flags)) {
274 pending = p->pending.signal;
275 shpending = p->signal->shared_pending.signal;
276 blocked = p->blocked;
277 collect_sigign_sigcatch(p, &ignored, &caught);
278 num_threads = get_nr_threads(p);
279 rcu_read_lock(); /* FIXME: is this correct? */
280 qsize = get_ucounts_value(task_ucounts(p), UCOUNT_RLIMIT_SIGPENDING);
282 qlim = task_rlimit(p, RLIMIT_SIGPENDING);
283 unlock_task_sighand(p, &flags);
286 seq_put_decimal_ull(m, "Threads:\t", num_threads);
287 seq_put_decimal_ull(m, "\nSigQ:\t", qsize);
288 seq_put_decimal_ull(m, "/", qlim);
290 /* render them all */
291 render_sigset_t(m, "\nSigPnd:\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);
298 static void render_cap_t(struct seq_file *m, const char *header,
304 CAP_FOR_EACH_U32(__capi) {
305 seq_put_hex_ll(m, NULL,
306 a->cap[CAP_LAST_U32 - __capi], 8);
311 static inline void task_cap(struct seq_file *m, struct task_struct *p)
313 const struct cred *cred;
314 kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
315 cap_bset, cap_ambient;
318 cred = __task_cred(p);
319 cap_inheritable = cred->cap_inheritable;
320 cap_permitted = cred->cap_permitted;
321 cap_effective = cred->cap_effective;
322 cap_bset = cred->cap_bset;
323 cap_ambient = cred->cap_ambient;
326 render_cap_t(m, "CapInh:\t", &cap_inheritable);
327 render_cap_t(m, "CapPrm:\t", &cap_permitted);
328 render_cap_t(m, "CapEff:\t", &cap_effective);
329 render_cap_t(m, "CapBnd:\t", &cap_bset);
330 render_cap_t(m, "CapAmb:\t", &cap_ambient);
333 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
335 seq_put_decimal_ull(m, "NoNewPrivs:\t", task_no_new_privs(p));
336 #ifdef CONFIG_SECCOMP
337 seq_put_decimal_ull(m, "\nSeccomp:\t", p->seccomp.mode);
338 #ifdef CONFIG_SECCOMP_FILTER
339 seq_put_decimal_ull(m, "\nSeccomp_filters:\t",
340 atomic_read(&p->seccomp.filter_count));
343 seq_puts(m, "\nSpeculation_Store_Bypass:\t");
344 switch (arch_prctl_spec_ctrl_get(p, PR_SPEC_STORE_BYPASS)) {
346 seq_puts(m, "unknown");
348 case PR_SPEC_NOT_AFFECTED:
349 seq_puts(m, "not vulnerable");
351 case PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE:
352 seq_puts(m, "thread force mitigated");
354 case PR_SPEC_PRCTL | PR_SPEC_DISABLE:
355 seq_puts(m, "thread mitigated");
357 case PR_SPEC_PRCTL | PR_SPEC_ENABLE:
358 seq_puts(m, "thread vulnerable");
360 case PR_SPEC_DISABLE:
361 seq_puts(m, "globally mitigated");
364 seq_puts(m, "vulnerable");
368 seq_puts(m, "\nSpeculationIndirectBranch:\t");
369 switch (arch_prctl_spec_ctrl_get(p, PR_SPEC_INDIRECT_BRANCH)) {
371 seq_puts(m, "unsupported");
373 case PR_SPEC_NOT_AFFECTED:
374 seq_puts(m, "not affected");
376 case PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE:
377 seq_puts(m, "conditional force disabled");
379 case PR_SPEC_PRCTL | PR_SPEC_DISABLE:
380 seq_puts(m, "conditional disabled");
382 case PR_SPEC_PRCTL | PR_SPEC_ENABLE:
383 seq_puts(m, "conditional enabled");
386 seq_puts(m, "always enabled");
388 case PR_SPEC_DISABLE:
389 seq_puts(m, "always disabled");
392 seq_puts(m, "unknown");
398 static inline void task_context_switch_counts(struct seq_file *m,
399 struct task_struct *p)
401 seq_put_decimal_ull(m, "voluntary_ctxt_switches:\t", p->nvcsw);
402 seq_put_decimal_ull(m, "\nnonvoluntary_ctxt_switches:\t", p->nivcsw);
406 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
408 seq_printf(m, "Cpus_allowed:\t%*pb\n",
409 cpumask_pr_args(&task->cpus_mask));
410 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
411 cpumask_pr_args(&task->cpus_mask));
414 static inline void task_core_dumping(struct seq_file *m, struct task_struct *task)
416 seq_put_decimal_ull(m, "CoreDumping:\t", !!task->signal->core_state);
420 static inline void task_thp_status(struct seq_file *m, struct mm_struct *mm)
422 bool thp_enabled = IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE);
425 thp_enabled = !test_bit(MMF_DISABLE_THP, &mm->flags);
426 seq_printf(m, "THP_enabled:\t%d\n", thp_enabled);
429 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
430 struct pid *pid, struct task_struct *task)
432 struct mm_struct *mm = get_task_mm(task);
434 seq_puts(m, "Name:\t");
435 proc_task_name(m, task, true);
438 task_state(m, ns, pid, task);
442 task_core_dumping(m, task);
443 task_thp_status(m, mm);
448 task_seccomp(m, task);
449 task_cpus_allowed(m, task);
450 cpuset_task_status_allowed(m, task);
451 task_context_switch_counts(m, task);
455 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
456 struct pid *pid, struct task_struct *task, int whole)
458 unsigned long vsize, eip, esp, wchan = 0;
460 int tty_pgrp = -1, tty_nr = 0;
461 sigset_t sigign, sigcatch;
463 pid_t ppid = 0, pgid = -1, sid = -1;
466 struct mm_struct *mm;
467 unsigned long long start_time;
468 unsigned long cmin_flt = 0, cmaj_flt = 0;
469 unsigned long min_flt = 0, maj_flt = 0;
470 u64 cutime, cstime, utime, stime;
472 unsigned long rsslim = 0;
475 state = *get_task_state(task);
476 vsize = eip = esp = 0;
477 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
478 mm = get_task_mm(task);
480 vsize = task_vsize(mm);
482 * esp and eip are intentionally zeroed out. There is no
483 * non-racy way to read them without freezing the task.
484 * Programs that need reliable values can use ptrace(2).
486 * The only exception is if the task is core dumping because
487 * a program is not able to use ptrace(2) in that case. It is
488 * safe because the task has stopped executing permanently.
490 if (permitted && (task->flags & (PF_EXITING|PF_DUMPCORE))) {
491 if (try_get_task_stack(task)) {
492 eip = KSTK_EIP(task);
493 esp = KSTK_ESP(task);
494 put_task_stack(task);
499 sigemptyset(&sigign);
500 sigemptyset(&sigcatch);
501 cutime = cstime = utime = stime = 0;
504 if (lock_task_sighand(task, &flags)) {
505 struct signal_struct *sig = task->signal;
508 struct pid *pgrp = tty_get_pgrp(sig->tty);
509 tty_pgrp = pid_nr_ns(pgrp, ns);
511 tty_nr = new_encode_dev(tty_devnum(sig->tty));
514 num_threads = get_nr_threads(task);
515 collect_sigign_sigcatch(task, &sigign, &sigcatch);
517 cmin_flt = sig->cmin_flt;
518 cmaj_flt = sig->cmaj_flt;
519 cutime = sig->cutime;
520 cstime = sig->cstime;
521 cgtime = sig->cgtime;
522 rsslim = READ_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
524 /* add up live thread stats at the group level */
526 struct task_struct *t = task;
528 min_flt += t->min_flt;
529 maj_flt += t->maj_flt;
530 gtime += task_gtime(t);
531 } while_each_thread(task, t);
533 min_flt += sig->min_flt;
534 maj_flt += sig->maj_flt;
535 thread_group_cputime_adjusted(task, &utime, &stime);
539 sid = task_session_nr_ns(task, ns);
540 ppid = task_tgid_nr_ns(task->real_parent, ns);
541 pgid = task_pgrp_nr_ns(task, ns);
543 unlock_task_sighand(task, &flags);
546 if (permitted && (!whole || num_threads < 2))
547 wchan = !task_is_running(task);
549 min_flt = task->min_flt;
550 maj_flt = task->maj_flt;
551 task_cputime_adjusted(task, &utime, &stime);
552 gtime = task_gtime(task);
555 /* scale priority and nice values from timeslices to -20..20 */
556 /* to make it look like a "normal" Unix priority/nice value */
557 priority = task_prio(task);
558 nice = task_nice(task);
560 /* apply timens offset for boottime and convert nsec -> ticks */
562 nsec_to_clock_t(timens_add_boottime_ns(task->start_boottime));
564 seq_put_decimal_ull(m, "", pid_nr_ns(pid, ns));
566 proc_task_name(m, task, false);
569 seq_put_decimal_ll(m, " ", ppid);
570 seq_put_decimal_ll(m, " ", pgid);
571 seq_put_decimal_ll(m, " ", sid);
572 seq_put_decimal_ll(m, " ", tty_nr);
573 seq_put_decimal_ll(m, " ", tty_pgrp);
574 seq_put_decimal_ull(m, " ", task->flags);
575 seq_put_decimal_ull(m, " ", min_flt);
576 seq_put_decimal_ull(m, " ", cmin_flt);
577 seq_put_decimal_ull(m, " ", maj_flt);
578 seq_put_decimal_ull(m, " ", cmaj_flt);
579 seq_put_decimal_ull(m, " ", nsec_to_clock_t(utime));
580 seq_put_decimal_ull(m, " ", nsec_to_clock_t(stime));
581 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cutime));
582 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cstime));
583 seq_put_decimal_ll(m, " ", priority);
584 seq_put_decimal_ll(m, " ", nice);
585 seq_put_decimal_ll(m, " ", num_threads);
586 seq_put_decimal_ull(m, " ", 0);
587 seq_put_decimal_ull(m, " ", start_time);
588 seq_put_decimal_ull(m, " ", vsize);
589 seq_put_decimal_ull(m, " ", mm ? get_mm_rss(mm) : 0);
590 seq_put_decimal_ull(m, " ", rsslim);
591 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->start_code : 1) : 0);
592 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->end_code : 1) : 0);
593 seq_put_decimal_ull(m, " ", (permitted && mm) ? mm->start_stack : 0);
594 seq_put_decimal_ull(m, " ", esp);
595 seq_put_decimal_ull(m, " ", eip);
596 /* The signal information here is obsolete.
597 * It must be decimal for Linux 2.0 compatibility.
598 * Use /proc/#/status for real-time signals.
600 seq_put_decimal_ull(m, " ", task->pending.signal.sig[0] & 0x7fffffffUL);
601 seq_put_decimal_ull(m, " ", task->blocked.sig[0] & 0x7fffffffUL);
602 seq_put_decimal_ull(m, " ", sigign.sig[0] & 0x7fffffffUL);
603 seq_put_decimal_ull(m, " ", sigcatch.sig[0] & 0x7fffffffUL);
606 * We used to output the absolute kernel address, but that's an
607 * information leak - so instead we show a 0/1 flag here, to signal
608 * to user-space whether there's a wchan field in /proc/PID/wchan.
610 * This works with older implementations of procps as well.
612 seq_put_decimal_ull(m, " ", wchan);
614 seq_put_decimal_ull(m, " ", 0);
615 seq_put_decimal_ull(m, " ", 0);
616 seq_put_decimal_ll(m, " ", task->exit_signal);
617 seq_put_decimal_ll(m, " ", task_cpu(task));
618 seq_put_decimal_ull(m, " ", task->rt_priority);
619 seq_put_decimal_ull(m, " ", task->policy);
620 seq_put_decimal_ull(m, " ", delayacct_blkio_ticks(task));
621 seq_put_decimal_ull(m, " ", nsec_to_clock_t(gtime));
622 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cgtime));
624 if (mm && permitted) {
625 seq_put_decimal_ull(m, " ", mm->start_data);
626 seq_put_decimal_ull(m, " ", mm->end_data);
627 seq_put_decimal_ull(m, " ", mm->start_brk);
628 seq_put_decimal_ull(m, " ", mm->arg_start);
629 seq_put_decimal_ull(m, " ", mm->arg_end);
630 seq_put_decimal_ull(m, " ", mm->env_start);
631 seq_put_decimal_ull(m, " ", mm->env_end);
633 seq_puts(m, " 0 0 0 0 0 0 0");
636 seq_put_decimal_ll(m, " ", task->exit_code);
646 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
647 struct pid *pid, struct task_struct *task)
649 return do_task_stat(m, ns, pid, task, 0);
652 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
653 struct pid *pid, struct task_struct *task)
655 return do_task_stat(m, ns, pid, task, 1);
658 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
659 struct pid *pid, struct task_struct *task)
661 struct mm_struct *mm = get_task_mm(task);
665 unsigned long resident = 0;
666 unsigned long shared = 0;
667 unsigned long text = 0;
668 unsigned long data = 0;
670 size = task_statm(mm, &shared, &text, &data, &resident);
674 * For quick read, open code by putting numbers directly
676 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
677 * size, resident, shared, text, data);
679 seq_put_decimal_ull(m, "", size);
680 seq_put_decimal_ull(m, " ", resident);
681 seq_put_decimal_ull(m, " ", shared);
682 seq_put_decimal_ull(m, " ", text);
683 seq_put_decimal_ull(m, " ", 0);
684 seq_put_decimal_ull(m, " ", data);
685 seq_put_decimal_ull(m, " ", 0);
688 seq_write(m, "0 0 0 0 0 0 0\n", 14);
693 #ifdef CONFIG_PROC_CHILDREN
695 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
697 struct task_struct *start, *task;
698 struct pid *pid = NULL;
700 read_lock(&tasklist_lock);
702 start = pid_task(proc_pid(inode), PIDTYPE_PID);
707 * Lets try to continue searching first, this gives
708 * us significant speedup on children-rich processes.
711 task = pid_task(pid_prev, PIDTYPE_PID);
712 if (task && task->real_parent == start &&
713 !(list_empty(&task->sibling))) {
714 if (list_is_last(&task->sibling, &start->children))
716 task = list_first_entry(&task->sibling,
717 struct task_struct, sibling);
718 pid = get_pid(task_pid(task));
726 * We might miss some children here if children
727 * are exited while we were not holding the lock,
728 * but it was never promised to be accurate that
731 * "Just suppose that the parent sleeps, but N children
732 * exit after we printed their tids. Now the slow paths
733 * skips N extra children, we miss N tasks." (c)
735 * So one need to stop or freeze the leader and all
736 * its children to get a precise result.
738 list_for_each_entry(task, &start->children, sibling) {
740 pid = get_pid(task_pid(task));
746 read_unlock(&tasklist_lock);
750 static int children_seq_show(struct seq_file *seq, void *v)
752 struct inode *inode = file_inode(seq->file);
754 seq_printf(seq, "%d ", pid_nr_ns(v, proc_pid_ns(inode->i_sb)));
758 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
760 return get_children_pid(file_inode(seq->file), NULL, *pos);
763 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
767 pid = get_children_pid(file_inode(seq->file), v, *pos + 1);
774 static void children_seq_stop(struct seq_file *seq, void *v)
779 static const struct seq_operations children_seq_ops = {
780 .start = children_seq_start,
781 .next = children_seq_next,
782 .stop = children_seq_stop,
783 .show = children_seq_show,
786 static int children_seq_open(struct inode *inode, struct file *file)
788 return seq_open(file, &children_seq_ops);
791 const struct file_operations proc_tid_children_operations = {
792 .open = children_seq_open,
795 .release = seq_release,
797 #endif /* CONFIG_PROC_CHILDREN */