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/proc_fs.h>
64 #include <linux/ioport.h>
65 #include <linux/uaccess.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>
88 #include <asm/pgtable.h>
89 #include <asm/processor.h>
92 static inline void task_name(struct seq_file *m, struct task_struct *p)
96 char tcomm[sizeof(p->comm)];
99 get_task_comm(tcomm, p);
101 seq_puts(m, "Name:\t");
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);
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
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 */
123 "Z (zombie)", /* 32 */
126 static inline const char *get_task_state(struct task_struct *tsk)
128 unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT;
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.
135 if (tsk->state == TASK_PARKED)
136 state = TASK_INTERRUPTIBLE;
138 BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1);
140 return task_state_array[fls(state)];
143 static inline int get_task_umask(struct task_struct *tsk)
145 struct fs_struct *fs;
156 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
157 struct pid *pid, struct task_struct *p)
159 struct user_namespace *user_ns = seq_user_ns(m);
160 struct group_info *group_info;
162 struct task_struct *tracer;
163 const struct cred *cred;
164 pid_t ppid, tpid = 0, tgid, ngid;
165 unsigned int max_fds = 0;
168 ppid = pid_alive(p) ?
169 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
171 tracer = ptrace_parent(p);
173 tpid = task_pid_nr_ns(tracer, ns);
175 tgid = task_tgid_nr_ns(p, ns);
176 ngid = task_numa_group_id(p);
177 cred = get_task_cred(p);
179 umask = get_task_umask(p);
181 seq_printf(m, "Umask:\t%#04o\n", umask);
185 max_fds = files_fdtable(p->files)->max_fds;
189 seq_printf(m, "State:\t%s", get_task_state(p));
191 seq_puts(m, "\nTgid:\t");
192 seq_put_decimal_ull(m, 0, tgid);
193 seq_puts(m, "\nNgid:\t");
194 seq_put_decimal_ull(m, 0, ngid);
195 seq_puts(m, "\nPid:\t");
196 seq_put_decimal_ull(m, 0, pid_nr_ns(pid, ns));
197 seq_puts(m, "\nPPid:\t");
198 seq_put_decimal_ull(m, 0, ppid);
199 seq_puts(m, "\nTracerPid:\t");
200 seq_put_decimal_ull(m, 0, tpid);
201 seq_puts(m, "\nUid:");
202 seq_put_decimal_ull(m, '\t', from_kuid_munged(user_ns, cred->uid));
203 seq_put_decimal_ull(m, '\t', from_kuid_munged(user_ns, cred->euid));
204 seq_put_decimal_ull(m, '\t', from_kuid_munged(user_ns, cred->suid));
205 seq_put_decimal_ull(m, '\t', from_kuid_munged(user_ns, cred->fsuid));
206 seq_puts(m, "\nGid:");
207 seq_put_decimal_ull(m, '\t', from_kgid_munged(user_ns, cred->gid));
208 seq_put_decimal_ull(m, '\t', from_kgid_munged(user_ns, cred->egid));
209 seq_put_decimal_ull(m, '\t', from_kgid_munged(user_ns, cred->sgid));
210 seq_put_decimal_ull(m, '\t', from_kgid_munged(user_ns, cred->fsgid));
211 seq_puts(m, "\nFDSize:\t");
212 seq_put_decimal_ull(m, 0, max_fds);
214 seq_puts(m, "\nGroups:\t");
215 group_info = cred->group_info;
216 for (g = 0; g < group_info->ngroups; g++)
217 seq_put_decimal_ull(m, g ? ' ' : 0, from_kgid_munged(user_ns, GROUP_AT(group_info, g)));
219 /* Trailing space shouldn't have been added in the first place. */
223 seq_puts(m, "\nNStgid:");
224 for (g = ns->level; g <= pid->level; g++)
225 seq_put_decimal_ull(m, '\t', task_tgid_nr_ns(p, pid->numbers[g].ns));
226 seq_puts(m, "\nNSpid:");
227 for (g = ns->level; g <= pid->level; g++)
228 seq_put_decimal_ull(m, '\t', task_pid_nr_ns(p, pid->numbers[g].ns));
229 seq_puts(m, "\nNSpgid:");
230 for (g = ns->level; g <= pid->level; g++)
231 seq_put_decimal_ull(m, '\t', task_pgrp_nr_ns(p, pid->numbers[g].ns));
232 seq_puts(m, "\nNSsid:");
233 for (g = ns->level; g <= pid->level; g++)
234 seq_put_decimal_ull(m, '\t', task_session_nr_ns(p, pid->numbers[g].ns));
239 void render_sigset_t(struct seq_file *m, const char *header,
251 if (sigismember(set, i+1)) x |= 1;
252 if (sigismember(set, i+2)) x |= 2;
253 if (sigismember(set, i+3)) x |= 4;
254 if (sigismember(set, i+4)) x |= 8;
255 seq_printf(m, "%x", x);
261 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
264 struct k_sigaction *k;
267 k = p->sighand->action;
268 for (i = 1; i <= _NSIG; ++i, ++k) {
269 if (k->sa.sa_handler == SIG_IGN)
271 else if (k->sa.sa_handler != SIG_DFL)
276 static inline void task_sig(struct seq_file *m, struct task_struct *p)
279 sigset_t pending, shpending, blocked, ignored, caught;
281 unsigned long qsize = 0;
282 unsigned long qlim = 0;
284 sigemptyset(&pending);
285 sigemptyset(&shpending);
286 sigemptyset(&blocked);
287 sigemptyset(&ignored);
288 sigemptyset(&caught);
290 if (lock_task_sighand(p, &flags)) {
291 pending = p->pending.signal;
292 shpending = p->signal->shared_pending.signal;
293 blocked = p->blocked;
294 collect_sigign_sigcatch(p, &ignored, &caught);
295 num_threads = get_nr_threads(p);
296 rcu_read_lock(); /* FIXME: is this correct? */
297 qsize = atomic_read(&__task_cred(p)->user->sigpending);
299 qlim = task_rlimit(p, RLIMIT_SIGPENDING);
300 unlock_task_sighand(p, &flags);
303 seq_puts(m, "Threads:\t");
304 seq_put_decimal_ull(m, 0, num_threads);
305 seq_puts(m, "\nSigQ:\t");
306 seq_put_decimal_ull(m, 0, qsize);
307 seq_put_decimal_ull(m, '/', qlim);
309 /* render them all */
310 render_sigset_t(m, "\nSigPnd:\t", &pending);
311 render_sigset_t(m, "ShdPnd:\t", &shpending);
312 render_sigset_t(m, "SigBlk:\t", &blocked);
313 render_sigset_t(m, "SigIgn:\t", &ignored);
314 render_sigset_t(m, "SigCgt:\t", &caught);
317 static void render_cap_t(struct seq_file *m, const char *header,
323 CAP_FOR_EACH_U32(__capi) {
324 seq_printf(m, "%08x",
325 a->cap[CAP_LAST_U32 - __capi]);
330 static inline void task_cap(struct seq_file *m, struct task_struct *p)
332 const struct cred *cred;
333 kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
334 cap_bset, cap_ambient;
337 cred = __task_cred(p);
338 cap_inheritable = cred->cap_inheritable;
339 cap_permitted = cred->cap_permitted;
340 cap_effective = cred->cap_effective;
341 cap_bset = cred->cap_bset;
342 cap_ambient = cred->cap_ambient;
345 render_cap_t(m, "CapInh:\t", &cap_inheritable);
346 render_cap_t(m, "CapPrm:\t", &cap_permitted);
347 render_cap_t(m, "CapEff:\t", &cap_effective);
348 render_cap_t(m, "CapBnd:\t", &cap_bset);
349 render_cap_t(m, "CapAmb:\t", &cap_ambient);
352 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
354 #ifdef CONFIG_SECCOMP
355 seq_puts(m, "Seccomp:\t");
356 seq_put_decimal_ull(m, 0, p->seccomp.mode);
361 static inline void task_context_switch_counts(struct seq_file *m,
362 struct task_struct *p)
364 seq_puts(m, "voluntary_ctxt_switches:\t");
365 seq_put_decimal_ull(m, 0, p->nvcsw);
366 seq_puts(m, "\nnonvoluntary_ctxt_switches:\t");
367 seq_put_decimal_ull(m, 0, p->nivcsw);
371 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
373 seq_printf(m, "Cpus_allowed:\t%*pb\n",
374 cpumask_pr_args(&task->cpus_allowed));
375 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
376 cpumask_pr_args(&task->cpus_allowed));
379 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
380 struct pid *pid, struct task_struct *task)
382 struct mm_struct *mm = get_task_mm(task);
385 task_state(m, ns, pid, task);
393 task_seccomp(m, task);
394 task_cpus_allowed(m, task);
395 cpuset_task_status_allowed(m, task);
396 task_context_switch_counts(m, task);
400 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
401 struct pid *pid, struct task_struct *task, int whole)
403 unsigned long vsize, eip, esp, wchan = 0;
405 int tty_pgrp = -1, tty_nr = 0;
406 sigset_t sigign, sigcatch;
408 pid_t ppid = 0, pgid = -1, sid = -1;
411 struct mm_struct *mm;
412 unsigned long long start_time;
413 unsigned long cmin_flt = 0, cmaj_flt = 0;
414 unsigned long min_flt = 0, maj_flt = 0;
415 cputime_t cutime, cstime, utime, stime;
416 cputime_t cgtime, gtime;
417 unsigned long rsslim = 0;
418 char tcomm[sizeof(task->comm)];
421 state = *get_task_state(task);
422 vsize = eip = esp = 0;
423 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
424 mm = get_task_mm(task);
426 vsize = task_vsize(mm);
428 eip = KSTK_EIP(task);
429 esp = KSTK_ESP(task);
433 get_task_comm(tcomm, task);
435 sigemptyset(&sigign);
436 sigemptyset(&sigcatch);
437 cutime = cstime = utime = stime = 0;
440 if (lock_task_sighand(task, &flags)) {
441 struct signal_struct *sig = task->signal;
444 struct pid *pgrp = tty_get_pgrp(sig->tty);
445 tty_pgrp = pid_nr_ns(pgrp, ns);
447 tty_nr = new_encode_dev(tty_devnum(sig->tty));
450 num_threads = get_nr_threads(task);
451 collect_sigign_sigcatch(task, &sigign, &sigcatch);
453 cmin_flt = sig->cmin_flt;
454 cmaj_flt = sig->cmaj_flt;
455 cutime = sig->cutime;
456 cstime = sig->cstime;
457 cgtime = sig->cgtime;
458 rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
460 /* add up live thread stats at the group level */
462 struct task_struct *t = task;
464 min_flt += t->min_flt;
465 maj_flt += t->maj_flt;
466 gtime += task_gtime(t);
467 } while_each_thread(task, t);
469 min_flt += sig->min_flt;
470 maj_flt += sig->maj_flt;
471 thread_group_cputime_adjusted(task, &utime, &stime);
475 sid = task_session_nr_ns(task, ns);
476 ppid = task_tgid_nr_ns(task->real_parent, ns);
477 pgid = task_pgrp_nr_ns(task, ns);
479 unlock_task_sighand(task, &flags);
482 if (permitted && (!whole || num_threads < 2))
483 wchan = get_wchan(task);
485 min_flt = task->min_flt;
486 maj_flt = task->maj_flt;
487 task_cputime_adjusted(task, &utime, &stime);
488 gtime = task_gtime(task);
491 /* scale priority and nice values from timeslices to -20..20 */
492 /* to make it look like a "normal" Unix priority/nice value */
493 priority = task_prio(task);
494 nice = task_nice(task);
496 /* convert nsec -> ticks */
497 start_time = nsec_to_clock_t(task->real_start_time);
499 seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state);
500 seq_put_decimal_ll(m, ' ', ppid);
501 seq_put_decimal_ll(m, ' ', pgid);
502 seq_put_decimal_ll(m, ' ', sid);
503 seq_put_decimal_ll(m, ' ', tty_nr);
504 seq_put_decimal_ll(m, ' ', tty_pgrp);
505 seq_put_decimal_ull(m, ' ', task->flags);
506 seq_put_decimal_ull(m, ' ', min_flt);
507 seq_put_decimal_ull(m, ' ', cmin_flt);
508 seq_put_decimal_ull(m, ' ', maj_flt);
509 seq_put_decimal_ull(m, ' ', cmaj_flt);
510 seq_put_decimal_ull(m, ' ', cputime_to_clock_t(utime));
511 seq_put_decimal_ull(m, ' ', cputime_to_clock_t(stime));
512 seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cutime));
513 seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cstime));
514 seq_put_decimal_ll(m, ' ', priority);
515 seq_put_decimal_ll(m, ' ', nice);
516 seq_put_decimal_ll(m, ' ', num_threads);
517 seq_put_decimal_ull(m, ' ', 0);
518 seq_put_decimal_ull(m, ' ', start_time);
519 seq_put_decimal_ull(m, ' ', vsize);
520 seq_put_decimal_ull(m, ' ', mm ? get_mm_rss(mm) : 0);
521 seq_put_decimal_ull(m, ' ', rsslim);
522 seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->start_code : 1) : 0);
523 seq_put_decimal_ull(m, ' ', mm ? (permitted ? mm->end_code : 1) : 0);
524 seq_put_decimal_ull(m, ' ', (permitted && mm) ? mm->start_stack : 0);
525 seq_put_decimal_ull(m, ' ', esp);
526 seq_put_decimal_ull(m, ' ', eip);
527 /* The signal information here is obsolete.
528 * It must be decimal for Linux 2.0 compatibility.
529 * Use /proc/#/status for real-time signals.
531 seq_put_decimal_ull(m, ' ', task->pending.signal.sig[0] & 0x7fffffffUL);
532 seq_put_decimal_ull(m, ' ', task->blocked.sig[0] & 0x7fffffffUL);
533 seq_put_decimal_ull(m, ' ', sigign.sig[0] & 0x7fffffffUL);
534 seq_put_decimal_ull(m, ' ', sigcatch.sig[0] & 0x7fffffffUL);
537 * We used to output the absolute kernel address, but that's an
538 * information leak - so instead we show a 0/1 flag here, to signal
539 * to user-space whether there's a wchan field in /proc/PID/wchan.
541 * This works with older implementations of procps as well.
548 seq_put_decimal_ull(m, ' ', 0);
549 seq_put_decimal_ull(m, ' ', 0);
550 seq_put_decimal_ll(m, ' ', task->exit_signal);
551 seq_put_decimal_ll(m, ' ', task_cpu(task));
552 seq_put_decimal_ull(m, ' ', task->rt_priority);
553 seq_put_decimal_ull(m, ' ', task->policy);
554 seq_put_decimal_ull(m, ' ', delayacct_blkio_ticks(task));
555 seq_put_decimal_ull(m, ' ', cputime_to_clock_t(gtime));
556 seq_put_decimal_ll(m, ' ', cputime_to_clock_t(cgtime));
558 if (mm && permitted) {
559 seq_put_decimal_ull(m, ' ', mm->start_data);
560 seq_put_decimal_ull(m, ' ', mm->end_data);
561 seq_put_decimal_ull(m, ' ', mm->start_brk);
562 seq_put_decimal_ull(m, ' ', mm->arg_start);
563 seq_put_decimal_ull(m, ' ', mm->arg_end);
564 seq_put_decimal_ull(m, ' ', mm->env_start);
565 seq_put_decimal_ull(m, ' ', mm->env_end);
567 seq_printf(m, " 0 0 0 0 0 0 0");
570 seq_put_decimal_ll(m, ' ', task->exit_code);
572 seq_put_decimal_ll(m, ' ', 0);
580 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
581 struct pid *pid, struct task_struct *task)
583 return do_task_stat(m, ns, pid, task, 0);
586 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
587 struct pid *pid, struct task_struct *task)
589 return do_task_stat(m, ns, pid, task, 1);
592 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
593 struct pid *pid, struct task_struct *task)
595 unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
596 struct mm_struct *mm = get_task_mm(task);
599 size = task_statm(mm, &shared, &text, &data, &resident);
603 * For quick read, open code by putting numbers directly
605 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
606 * size, resident, shared, text, data);
608 seq_put_decimal_ull(m, 0, size);
609 seq_put_decimal_ull(m, ' ', resident);
610 seq_put_decimal_ull(m, ' ', shared);
611 seq_put_decimal_ull(m, ' ', text);
612 seq_put_decimal_ull(m, ' ', 0);
613 seq_put_decimal_ull(m, ' ', data);
614 seq_put_decimal_ull(m, ' ', 0);
620 #ifdef CONFIG_PROC_CHILDREN
622 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
624 struct task_struct *start, *task;
625 struct pid *pid = NULL;
627 read_lock(&tasklist_lock);
629 start = pid_task(proc_pid(inode), PIDTYPE_PID);
634 * Lets try to continue searching first, this gives
635 * us significant speedup on children-rich processes.
638 task = pid_task(pid_prev, PIDTYPE_PID);
639 if (task && task->real_parent == start &&
640 !(list_empty(&task->sibling))) {
641 if (list_is_last(&task->sibling, &start->children))
643 task = list_first_entry(&task->sibling,
644 struct task_struct, sibling);
645 pid = get_pid(task_pid(task));
653 * We might miss some children here if children
654 * are exited while we were not holding the lock,
655 * but it was never promised to be accurate that
658 * "Just suppose that the parent sleeps, but N children
659 * exit after we printed their tids. Now the slow paths
660 * skips N extra children, we miss N tasks." (c)
662 * So one need to stop or freeze the leader and all
663 * its children to get a precise result.
665 list_for_each_entry(task, &start->children, sibling) {
667 pid = get_pid(task_pid(task));
673 read_unlock(&tasklist_lock);
677 static int children_seq_show(struct seq_file *seq, void *v)
679 struct inode *inode = seq->private;
682 pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
683 seq_printf(seq, "%d ", pid);
688 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
690 return get_children_pid(seq->private, NULL, *pos);
693 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
697 pid = get_children_pid(seq->private, v, *pos + 1);
704 static void children_seq_stop(struct seq_file *seq, void *v)
709 static const struct seq_operations children_seq_ops = {
710 .start = children_seq_start,
711 .next = children_seq_next,
712 .stop = children_seq_stop,
713 .show = children_seq_show,
716 static int children_seq_open(struct inode *inode, struct file *file)
721 ret = seq_open(file, &children_seq_ops);
725 m = file->private_data;
731 int children_seq_release(struct inode *inode, struct file *file)
733 seq_release(inode, file);
737 const struct file_operations proc_tid_children_operations = {
738 .open = children_seq_open,
741 .release = children_seq_release,
743 #endif /* CONFIG_PROC_CHILDREN */