2 * linux/fs/proc/array.c
4 * Copyright (C) 1992 by Linus Torvalds
5 * based on ideas by Darren Senn
8 * Michael. K. Johnson: stat,statm extensions.
9 * <johnsonm@stolaf.edu>
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>
17 * Danny ter Haar : added cpuinfo
20 * Alessandro Rubini : profile extension.
21 * <rubini@ipvvis.unipv.it>
23 * Jeff Tranter : added BogoMips field to cpuinfo
24 * <Jeff_Tranter@Mitel.COM>
26 * Bruno Haible : remove 4K limit for the maps file
27 * <haible@ma2s2.mathematik.uni-karlsruhe.de>
29 * Yves Arrouye : remove removal of trailing spaces in get_array.
30 * <Yves.Arrouye@marin.fdn.fr>
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>
39 * aeb@cwi.nl : /proc/partitions
42 * Alan Cox : security fixes.
43 * <alan@lxorguk.ukuu.org.uk>
45 * Al Viro : safe handling of mm_struct
47 * Gerhard Wichert : added BIGMEM support
48 * Siemens AG <Gerhard.Wichert@pdb.siemens.de>
50 * Al Viro & Jeff Garzik : moved most of the thing into base.c and
51 * : proc_misc.c. The rest may eventually go into
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/sched/mm.h>
64 #include <linux/sched/numa_balancing.h>
65 #include <linux/sched/task.h>
66 #include <linux/proc_fs.h>
67 #include <linux/ioport.h>
68 #include <linux/uaccess.h>
71 #include <linux/hugetlb.h>
72 #include <linux/pagemap.h>
73 #include <linux/swap.h>
74 #include <linux/smp.h>
75 #include <linux/signal.h>
76 #include <linux/highmem.h>
77 #include <linux/file.h>
78 #include <linux/fdtable.h>
79 #include <linux/times.h>
80 #include <linux/cpuset.h>
81 #include <linux/rcupdate.h>
82 #include <linux/delayacct.h>
83 #include <linux/seq_file.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/ptrace.h>
86 #include <linux/tracehook.h>
87 #include <linux/string_helpers.h>
88 #include <linux/user_namespace.h>
89 #include <linux/fs_struct.h>
91 #include <asm/pgtable.h>
92 #include <asm/processor.h>
95 static inline void task_name(struct seq_file *m, struct task_struct *p)
99 char tcomm[sizeof(p->comm)];
102 get_task_comm(tcomm, p);
104 seq_puts(m, "Name:\t");
106 size = seq_get_buf(m, &buf);
107 ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
108 seq_commit(m, ret < size ? ret : -1);
114 * The task state array is a strange "bitmap" of
115 * reasons to sleep. Thus "running" is zero, and
116 * you can test for combinations of others with
119 static const char * const task_state_array[] = {
120 "R (running)", /* 0 */
121 "S (sleeping)", /* 1 */
122 "D (disk sleep)", /* 2 */
123 "T (stopped)", /* 4 */
124 "t (tracing stop)", /* 8 */
126 "Z (zombie)", /* 32 */
129 static inline const char *get_task_state(struct task_struct *tsk)
131 unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
134 * Parked tasks do not run; they sit in __kthread_parkme().
135 * Without this check, we would report them as running, which is
136 * clearly wrong, so we report them as sleeping instead.
138 if (tsk->state == TASK_PARKED)
139 state = TASK_INTERRUPTIBLE;
141 BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
143 return task_state_array[fls(state)];
146 static inline int get_task_umask(struct task_struct *tsk)
148 struct fs_struct *fs;
159 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
160 struct pid *pid, struct task_struct *p)
162 struct user_namespace *user_ns = seq_user_ns(m);
163 struct group_info *group_info;
165 struct task_struct *tracer;
166 const struct cred *cred;
167 pid_t ppid, tpid = 0, tgid, ngid;
168 unsigned int max_fds = 0;
171 ppid = pid_alive(p) ?
172 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
174 tracer = ptrace_parent(p);
176 tpid = task_pid_nr_ns(tracer, ns);
178 tgid = task_tgid_nr_ns(p, ns);
179 ngid = task_numa_group_id(p);
180 cred = get_task_cred(p);
182 umask = get_task_umask(p);
184 seq_printf(m, "Umask:\t%#04o\n", umask);
188 max_fds = files_fdtable(p->files)->max_fds;
192 seq_printf(m, "State:\t%s", get_task_state(p));
194 seq_put_decimal_ull(m, "\nTgid:\t", tgid);
195 seq_put_decimal_ull(m, "\nNgid:\t", ngid);
196 seq_put_decimal_ull(m, "\nPid:\t", pid_nr_ns(pid, ns));
197 seq_put_decimal_ull(m, "\nPPid:\t", ppid);
198 seq_put_decimal_ull(m, "\nTracerPid:\t", tpid);
199 seq_put_decimal_ull(m, "\nUid:\t", from_kuid_munged(user_ns, cred->uid));
200 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->euid));
201 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->suid));
202 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->fsuid));
203 seq_put_decimal_ull(m, "\nGid:\t", from_kgid_munged(user_ns, cred->gid));
204 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->egid));
205 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->sgid));
206 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->fsgid));
207 seq_put_decimal_ull(m, "\nFDSize:\t", max_fds);
209 seq_puts(m, "\nGroups:\t");
210 group_info = cred->group_info;
211 for (g = 0; g < group_info->ngroups; g++)
212 seq_put_decimal_ull(m, g ? " " : "",
213 from_kgid_munged(user_ns, group_info->gid[g]));
215 /* Trailing space shouldn't have been added in the first place. */
219 seq_puts(m, "\nNStgid:");
220 for (g = ns->level; g <= pid->level; g++)
221 seq_put_decimal_ull(m, "\t", 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_put_decimal_ull(m, "\t", task_pid_nr_ns(p, pid->numbers[g].ns));
225 seq_puts(m, "\nNSpgid:");
226 for (g = ns->level; g <= pid->level; g++)
227 seq_put_decimal_ull(m, "\t", task_pgrp_nr_ns(p, pid->numbers[g].ns));
228 seq_puts(m, "\nNSsid:");
229 for (g = ns->level; g <= pid->level; g++)
230 seq_put_decimal_ull(m, "\t", task_session_nr_ns(p, pid->numbers[g].ns));
235 void render_sigset_t(struct seq_file *m, const char *header,
247 if (sigismember(set, i+1)) x |= 1;
248 if (sigismember(set, i+2)) x |= 2;
249 if (sigismember(set, i+3)) x |= 4;
250 if (sigismember(set, i+4)) x |= 8;
251 seq_putc(m, hex_asc[x]);
257 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
260 struct k_sigaction *k;
263 k = p->sighand->action;
264 for (i = 1; i <= _NSIG; ++i, ++k) {
265 if (k->sa.sa_handler == SIG_IGN)
267 else if (k->sa.sa_handler != SIG_DFL)
272 static inline void task_sig(struct seq_file *m, struct task_struct *p)
275 sigset_t pending, shpending, blocked, ignored, caught;
277 unsigned long qsize = 0;
278 unsigned long qlim = 0;
280 sigemptyset(&pending);
281 sigemptyset(&shpending);
282 sigemptyset(&blocked);
283 sigemptyset(&ignored);
284 sigemptyset(&caught);
286 if (lock_task_sighand(p, &flags)) {
287 pending = p->pending.signal;
288 shpending = p->signal->shared_pending.signal;
289 blocked = p->blocked;
290 collect_sigign_sigcatch(p, &ignored, &caught);
291 num_threads = get_nr_threads(p);
292 rcu_read_lock(); /* FIXME: is this correct? */
293 qsize = atomic_read(&__task_cred(p)->user->sigpending);
295 qlim = task_rlimit(p, RLIMIT_SIGPENDING);
296 unlock_task_sighand(p, &flags);
299 seq_put_decimal_ull(m, "Threads:\t", num_threads);
300 seq_put_decimal_ull(m, "\nSigQ:\t", qsize);
301 seq_put_decimal_ull(m, "/", qlim);
303 /* render them all */
304 render_sigset_t(m, "\nSigPnd:\t", &pending);
305 render_sigset_t(m, "ShdPnd:\t", &shpending);
306 render_sigset_t(m, "SigBlk:\t", &blocked);
307 render_sigset_t(m, "SigIgn:\t", &ignored);
308 render_sigset_t(m, "SigCgt:\t", &caught);
311 static void render_cap_t(struct seq_file *m, const char *header,
317 CAP_FOR_EACH_U32(__capi) {
318 seq_printf(m, "%08x",
319 a->cap[CAP_LAST_U32 - __capi]);
324 static inline void task_cap(struct seq_file *m, struct task_struct *p)
326 const struct cred *cred;
327 kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
328 cap_bset, cap_ambient;
331 cred = __task_cred(p);
332 cap_inheritable = cred->cap_inheritable;
333 cap_permitted = cred->cap_permitted;
334 cap_effective = cred->cap_effective;
335 cap_bset = cred->cap_bset;
336 cap_ambient = cred->cap_ambient;
339 render_cap_t(m, "CapInh:\t", &cap_inheritable);
340 render_cap_t(m, "CapPrm:\t", &cap_permitted);
341 render_cap_t(m, "CapEff:\t", &cap_effective);
342 render_cap_t(m, "CapBnd:\t", &cap_bset);
343 render_cap_t(m, "CapAmb:\t", &cap_ambient);
346 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
348 seq_put_decimal_ull(m, "NoNewPrivs:\t", task_no_new_privs(p));
349 #ifdef CONFIG_SECCOMP
350 seq_put_decimal_ull(m, "\nSeccomp:\t", p->seccomp.mode);
355 static inline void task_context_switch_counts(struct seq_file *m,
356 struct task_struct *p)
358 seq_put_decimal_ull(m, "voluntary_ctxt_switches:\t", p->nvcsw);
359 seq_put_decimal_ull(m, "\nnonvoluntary_ctxt_switches:\t", p->nivcsw);
363 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
365 seq_printf(m, "Cpus_allowed:\t%*pb\n",
366 cpumask_pr_args(&task->cpus_allowed));
367 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
368 cpumask_pr_args(&task->cpus_allowed));
371 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
372 struct pid *pid, struct task_struct *task)
374 struct mm_struct *mm = get_task_mm(task);
377 task_state(m, ns, pid, task);
385 task_seccomp(m, task);
386 task_cpus_allowed(m, task);
387 cpuset_task_status_allowed(m, task);
388 task_context_switch_counts(m, task);
392 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
393 struct pid *pid, struct task_struct *task, int whole)
395 unsigned long vsize, eip, esp, wchan = 0;
397 int tty_pgrp = -1, tty_nr = 0;
398 sigset_t sigign, sigcatch;
400 pid_t ppid = 0, pgid = -1, sid = -1;
403 struct mm_struct *mm;
404 unsigned long long start_time;
405 unsigned long cmin_flt = 0, cmaj_flt = 0;
406 unsigned long min_flt = 0, maj_flt = 0;
407 u64 cutime, cstime, utime, stime;
409 unsigned long rsslim = 0;
410 char tcomm[sizeof(task->comm)];
413 state = *get_task_state(task);
414 vsize = eip = esp = 0;
415 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
416 mm = get_task_mm(task);
418 vsize = task_vsize(mm);
420 * esp and eip are intentionally zeroed out. There is no
421 * non-racy way to read them without freezing the task.
422 * Programs that need reliable values can use ptrace(2).
426 get_task_comm(tcomm, task);
428 sigemptyset(&sigign);
429 sigemptyset(&sigcatch);
430 cutime = cstime = utime = stime = 0;
433 if (lock_task_sighand(task, &flags)) {
434 struct signal_struct *sig = task->signal;
437 struct pid *pgrp = tty_get_pgrp(sig->tty);
438 tty_pgrp = pid_nr_ns(pgrp, ns);
440 tty_nr = new_encode_dev(tty_devnum(sig->tty));
443 num_threads = get_nr_threads(task);
444 collect_sigign_sigcatch(task, &sigign, &sigcatch);
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);
453 /* add up live thread stats at the group level */
455 struct task_struct *t = task;
457 min_flt += t->min_flt;
458 maj_flt += t->maj_flt;
459 gtime += task_gtime(t);
460 } while_each_thread(task, t);
462 min_flt += sig->min_flt;
463 maj_flt += sig->maj_flt;
464 thread_group_cputime_adjusted(task, &utime, &stime);
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);
472 unlock_task_sighand(task, &flags);
475 if (permitted && (!whole || num_threads < 2))
476 wchan = get_wchan(task);
478 min_flt = task->min_flt;
479 maj_flt = task->maj_flt;
480 task_cputime_adjusted(task, &utime, &stime);
481 gtime = task_gtime(task);
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);
489 /* convert nsec -> ticks */
490 start_time = nsec_to_clock_t(task->real_start_time);
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, " ", nsec_to_clock_t(utime));
504 seq_put_decimal_ull(m, " ", nsec_to_clock_t(stime));
505 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cutime));
506 seq_put_decimal_ll(m, " ", nsec_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.
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);
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.
534 * This works with older implementations of procps as well.
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, " ", nsec_to_clock_t(gtime));
549 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cgtime));
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);
560 seq_puts(m, " 0 0 0 0 0 0 0");
563 seq_put_decimal_ll(m, " ", task->exit_code);
573 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
574 struct pid *pid, struct task_struct *task)
576 return do_task_stat(m, ns, pid, task, 0);
579 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
580 struct pid *pid, struct task_struct *task)
582 return do_task_stat(m, ns, pid, task, 1);
585 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
586 struct pid *pid, struct task_struct *task)
588 unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
589 struct mm_struct *mm = get_task_mm(task);
592 size = task_statm(mm, &shared, &text, &data, &resident);
596 * For quick read, open code by putting numbers directly
598 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
599 * size, resident, shared, text, data);
601 seq_put_decimal_ull(m, "", 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);
613 #ifdef CONFIG_PROC_CHILDREN
615 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
617 struct task_struct *start, *task;
618 struct pid *pid = NULL;
620 read_lock(&tasklist_lock);
622 start = pid_task(proc_pid(inode), PIDTYPE_PID);
627 * Lets try to continue searching first, this gives
628 * us significant speedup on children-rich processes.
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))
636 task = list_first_entry(&task->sibling,
637 struct task_struct, sibling);
638 pid = get_pid(task_pid(task));
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
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)
655 * So one need to stop or freeze the leader and all
656 * its children to get a precise result.
658 list_for_each_entry(task, &start->children, sibling) {
660 pid = get_pid(task_pid(task));
666 read_unlock(&tasklist_lock);
670 static int children_seq_show(struct seq_file *seq, void *v)
672 struct inode *inode = seq->private;
675 pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
676 seq_printf(seq, "%d ", pid);
681 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
683 return get_children_pid(seq->private, NULL, *pos);
686 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
690 pid = get_children_pid(seq->private, v, *pos + 1);
697 static void children_seq_stop(struct seq_file *seq, void *v)
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,
709 static int children_seq_open(struct inode *inode, struct file *file)
714 ret = seq_open(file, &children_seq_ops);
718 m = file->private_data;
724 int children_seq_release(struct inode *inode, struct file *file)
726 seq_release(inode, file);
730 const struct file_operations proc_tid_children_operations = {
731 .open = children_seq_open,
734 .release = children_seq_release,
736 #endif /* CONFIG_PROC_CHILDREN */