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.Cox@linux.org>
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/slab.h>
72 #include <linux/smp.h>
73 #include <linux/signal.h>
74 #include <linux/highmem.h>
75 #include <linux/file.h>
76 #include <linux/fdtable.h>
77 #include <linux/times.h>
78 #include <linux/cpuset.h>
79 #include <linux/rcupdate.h>
80 #include <linux/delayacct.h>
81 #include <linux/seq_file.h>
82 #include <linux/pid_namespace.h>
84 #include <asm/pgtable.h>
85 #include <asm/processor.h>
88 /* Gcc optimizes away "strlen(x)" for constant x */
89 #define ADDBUF(buffer, string) \
90 do { memcpy(buffer, string, strlen(string)); \
91 buffer += strlen(string); } while (0)
93 static inline void task_name(struct seq_file *m, struct task_struct *p)
98 char tcomm[sizeof(p->comm)];
100 get_task_comm(tcomm, p);
102 seq_printf(m, "Name:\t");
103 end = m->buf + m->size;
104 buf = m->buf + m->count;
107 while (i && (buf < end)) {
108 unsigned char c = *name;
128 m->count = buf - m->buf;
133 * The task state array is a strange "bitmap" of
134 * reasons to sleep. Thus "running" is zero, and
135 * you can test for combinations of others with
138 static const char *task_state_array[] = {
139 "R (running)", /* 0 */
140 "S (sleeping)", /* 1 */
141 "D (disk sleep)", /* 2 */
142 "T (stopped)", /* 4 */
143 "T (tracing stop)", /* 8 */
144 "Z (zombie)", /* 16 */
148 static inline const char *get_task_state(struct task_struct *tsk)
150 unsigned int state = (tsk->state & TASK_REPORT) | tsk->exit_state;
151 const char **p = &task_state_array[0];
160 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
161 struct pid *pid, struct task_struct *p)
163 struct group_info *group_info;
165 struct fdtable *fdt = NULL;
169 ppid = pid_alive(p) ?
170 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
171 tpid = pid_alive(p) && p->ptrace ?
172 task_pid_nr_ns(rcu_dereference(p->parent), ns) : 0;
179 "Uid:\t%d\t%d\t%d\t%d\n"
180 "Gid:\t%d\t%d\t%d\t%d\n",
182 task_tgid_nr_ns(p, ns),
185 p->uid, p->euid, p->suid, p->fsuid,
186 p->gid, p->egid, p->sgid, p->fsgid);
190 fdt = files_fdtable(p->files);
194 fdt ? fdt->max_fds : 0);
197 group_info = p->group_info;
198 get_group_info(group_info);
201 for (g = 0; g < min(group_info->ngroups, NGROUPS_SMALL); g++)
202 seq_printf(m, "%d ", GROUP_AT(group_info, g));
203 put_group_info(group_info);
208 static void render_sigset_t(struct seq_file *m, const char *header,
213 seq_printf(m, "%s", header);
220 if (sigismember(set, i+1)) x |= 1;
221 if (sigismember(set, i+2)) x |= 2;
222 if (sigismember(set, i+3)) x |= 4;
223 if (sigismember(set, i+4)) x |= 8;
224 seq_printf(m, "%x", x);
230 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
233 struct k_sigaction *k;
236 k = p->sighand->action;
237 for (i = 1; i <= _NSIG; ++i, ++k) {
238 if (k->sa.sa_handler == SIG_IGN)
240 else if (k->sa.sa_handler != SIG_DFL)
245 static inline void task_sig(struct seq_file *m, struct task_struct *p)
248 sigset_t pending, shpending, blocked, ignored, caught;
250 unsigned long qsize = 0;
251 unsigned long qlim = 0;
253 sigemptyset(&pending);
254 sigemptyset(&shpending);
255 sigemptyset(&blocked);
256 sigemptyset(&ignored);
257 sigemptyset(&caught);
260 if (lock_task_sighand(p, &flags)) {
261 pending = p->pending.signal;
262 shpending = p->signal->shared_pending.signal;
263 blocked = p->blocked;
264 collect_sigign_sigcatch(p, &ignored, &caught);
265 num_threads = atomic_read(&p->signal->count);
266 qsize = atomic_read(&p->user->sigpending);
267 qlim = p->signal->rlim[RLIMIT_SIGPENDING].rlim_cur;
268 unlock_task_sighand(p, &flags);
272 seq_printf(m, "Threads:\t%d\n", num_threads);
273 seq_printf(m, "SigQ:\t%lu/%lu\n", qsize, qlim);
275 /* render them all */
276 render_sigset_t(m, "SigPnd:\t", &pending);
277 render_sigset_t(m, "ShdPnd:\t", &shpending);
278 render_sigset_t(m, "SigBlk:\t", &blocked);
279 render_sigset_t(m, "SigIgn:\t", &ignored);
280 render_sigset_t(m, "SigCgt:\t", &caught);
283 static void render_cap_t(struct seq_file *m, const char *header,
288 seq_printf(m, "%s", header);
289 CAP_FOR_EACH_U32(__capi) {
290 seq_printf(m, "%08x",
291 a->cap[(_LINUX_CAPABILITY_U32S-1) - __capi]);
296 static inline void task_cap(struct seq_file *m, struct task_struct *p)
298 render_cap_t(m, "CapInh:\t", &p->cap_inheritable);
299 render_cap_t(m, "CapPrm:\t", &p->cap_permitted);
300 render_cap_t(m, "CapEff:\t", &p->cap_effective);
301 render_cap_t(m, "CapBnd:\t", &p->cap_bset);
304 static inline void task_context_switch_counts(struct seq_file *m,
305 struct task_struct *p)
307 seq_printf(m, "voluntary_ctxt_switches:\t%lu\n"
308 "nonvoluntary_ctxt_switches:\t%lu\n",
313 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
314 struct pid *pid, struct task_struct *task)
316 struct mm_struct *mm = get_task_mm(task);
319 task_state(m, ns, pid, task);
327 cpuset_task_status_allowed(m, task);
328 #if defined(CONFIG_S390)
329 task_show_regs(m, task);
331 task_context_switch_counts(m, task);
336 * Use precise platform statistics if available:
338 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
339 static cputime_t task_utime(struct task_struct *p)
344 static cputime_t task_stime(struct task_struct *p)
349 static cputime_t task_utime(struct task_struct *p)
351 clock_t utime = cputime_to_clock_t(p->utime),
352 total = utime + cputime_to_clock_t(p->stime);
356 * Use CFS's precise accounting:
358 temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);
364 utime = (clock_t)temp;
366 p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime));
367 return p->prev_utime;
370 static cputime_t task_stime(struct task_struct *p)
375 * Use CFS's precise accounting. (we subtract utime from
376 * the total, to make sure the total observed by userspace
377 * grows monotonically - apps rely on that):
379 stime = nsec_to_clock_t(p->se.sum_exec_runtime) -
380 cputime_to_clock_t(task_utime(p));
383 p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime));
385 return p->prev_stime;
389 static cputime_t task_gtime(struct task_struct *p)
394 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
395 struct pid *pid, struct task_struct *task, int whole)
397 unsigned long vsize, eip, esp, wchan = ~0UL;
399 int tty_pgrp = -1, tty_nr = 0;
400 sigset_t sigign, sigcatch;
402 pid_t ppid = 0, pgid = -1, sid = -1;
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)];
414 state = *get_task_state(task);
415 vsize = eip = esp = 0;
416 mm = get_task_mm(task);
418 vsize = task_vsize(mm);
419 eip = KSTK_EIP(task);
420 esp = KSTK_ESP(task);
423 get_task_comm(tcomm, task);
425 sigemptyset(&sigign);
426 sigemptyset(&sigcatch);
427 cutime = cstime = utime = stime = cputime_zero;
428 cgtime = gtime = cputime_zero;
430 if (lock_task_sighand(task, &flags)) {
431 struct signal_struct *sig = task->signal;
434 struct pid *pgrp = tty_get_pgrp(sig->tty);
435 tty_pgrp = pid_nr_ns(pgrp, ns);
437 tty_nr = new_encode_dev(tty_devnum(sig->tty));
440 num_threads = atomic_read(&sig->count);
441 collect_sigign_sigcatch(task, &sigign, &sigcatch);
443 cmin_flt = sig->cmin_flt;
444 cmaj_flt = sig->cmaj_flt;
445 cutime = sig->cutime;
446 cstime = sig->cstime;
447 cgtime = sig->cgtime;
448 rsslim = sig->rlim[RLIMIT_RSS].rlim_cur;
450 /* add up live thread stats at the group level */
452 struct task_struct *t = task;
454 min_flt += t->min_flt;
455 maj_flt += t->maj_flt;
456 utime = cputime_add(utime, task_utime(t));
457 stime = cputime_add(stime, task_stime(t));
458 gtime = cputime_add(gtime, task_gtime(t));
462 min_flt += sig->min_flt;
463 maj_flt += sig->maj_flt;
464 utime = cputime_add(utime, sig->utime);
465 stime = cputime_add(stime, sig->stime);
466 gtime = cputime_add(gtime, sig->gtime);
469 sid = task_session_nr_ns(task, ns);
470 ppid = task_tgid_nr_ns(task->real_parent, ns);
471 pgid = task_pgrp_nr_ns(task, ns);
473 unlock_task_sighand(task, &flags);
476 if (!whole || num_threads < 2)
477 wchan = get_wchan(task);
479 min_flt = task->min_flt;
480 maj_flt = task->maj_flt;
481 utime = task_utime(task);
482 stime = task_stime(task);
483 gtime = task_gtime(task);
486 /* scale priority and nice values from timeslices to -20..20 */
487 /* to make it look like a "normal" Unix priority/nice value */
488 priority = task_prio(task);
489 nice = task_nice(task);
491 /* Temporary variable needed for gcc-2.96 */
492 /* convert timespec -> nsec*/
494 (unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
495 + task->real_start_time.tv_nsec;
496 /* convert nsec -> ticks */
497 start_time = nsec_to_clock_t(start_time);
499 seq_printf(m, "%d (%s) %c %d %d %d %d %d %u %lu \
500 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
501 %lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu %lu %ld\n",
515 cputime_to_clock_t(utime),
516 cputime_to_clock_t(stime),
517 cputime_to_clock_t(cutime),
518 cputime_to_clock_t(cstime),
524 mm ? get_mm_rss(mm) : 0,
526 mm ? mm->start_code : 0,
527 mm ? mm->end_code : 0,
528 mm ? mm->start_stack : 0,
531 /* The signal information here is obsolete.
532 * It must be decimal for Linux 2.0 compatibility.
533 * Use /proc/#/status for real-time signals.
535 task->pending.signal.sig[0] & 0x7fffffffUL,
536 task->blocked.sig[0] & 0x7fffffffUL,
537 sigign .sig[0] & 0x7fffffffUL,
538 sigcatch .sig[0] & 0x7fffffffUL,
546 (unsigned long long)delayacct_blkio_ticks(task),
547 cputime_to_clock_t(gtime),
548 cputime_to_clock_t(cgtime));
554 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
555 struct pid *pid, struct task_struct *task)
557 return do_task_stat(m, ns, pid, task, 0);
560 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
561 struct pid *pid, struct task_struct *task)
563 return do_task_stat(m, ns, pid, task, 1);
566 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
567 struct pid *pid, struct task_struct *task)
569 int size = 0, resident = 0, shared = 0, text = 0, lib = 0, data = 0;
570 struct mm_struct *mm = get_task_mm(task);
573 size = task_statm(mm, &shared, &text, &data, &resident);
576 seq_printf(m, "%d %d %d %d %d %d %d\n",
577 size, resident, shared, text, lib, data, 0);