Merge git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched
[linux-2.6.git] / fs / proc / array.c
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.Cox@linux.org>
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/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/times.h>
77 #include <linux/cpuset.h>
78 #include <linux/rcupdate.h>
79 #include <linux/delayacct.h>
80 #include <linux/pid_namespace.h>
81
82 #include <asm/pgtable.h>
83 #include <asm/processor.h>
84 #include "internal.h"
85
86 /* Gcc optimizes away "strlen(x)" for constant x */
87 #define ADDBUF(buffer, string) \
88 do { memcpy(buffer, string, strlen(string)); \
89      buffer += strlen(string); } while (0)
90
91 static inline char *task_name(struct task_struct *p, char *buf)
92 {
93         int i;
94         char *name;
95         char tcomm[sizeof(p->comm)];
96
97         get_task_comm(tcomm, p);
98
99         ADDBUF(buf, "Name:\t");
100         name = tcomm;
101         i = sizeof(tcomm);
102         do {
103                 unsigned char c = *name;
104                 name++;
105                 i--;
106                 *buf = c;
107                 if (!c)
108                         break;
109                 if (c == '\\') {
110                         buf[1] = c;
111                         buf += 2;
112                         continue;
113                 }
114                 if (c == '\n') {
115                         buf[0] = '\\';
116                         buf[1] = 'n';
117                         buf += 2;
118                         continue;
119                 }
120                 buf++;
121         } while (i);
122         *buf = '\n';
123         return buf+1;
124 }
125
126 /*
127  * The task state array is a strange "bitmap" of
128  * reasons to sleep. Thus "running" is zero, and
129  * you can test for combinations of others with
130  * simple bit tests.
131  */
132 static const char *task_state_array[] = {
133         "R (running)",          /*  0 */
134         "S (sleeping)",         /*  1 */
135         "D (disk sleep)",       /*  2 */
136         "T (stopped)",          /*  4 */
137         "T (tracing stop)",     /*  8 */
138         "Z (zombie)",           /* 16 */
139         "X (dead)"              /* 32 */
140 };
141
142 static inline const char *get_task_state(struct task_struct *tsk)
143 {
144         unsigned int state = (tsk->state & (TASK_RUNNING |
145                                             TASK_INTERRUPTIBLE |
146                                             TASK_UNINTERRUPTIBLE |
147                                             TASK_STOPPED |
148                                             TASK_TRACED)) |
149                                            tsk->exit_state;
150         const char **p = &task_state_array[0];
151
152         while (state) {
153                 p++;
154                 state >>= 1;
155         }
156         return *p;
157 }
158
159 static inline char *task_state(struct task_struct *p, char *buffer)
160 {
161         struct group_info *group_info;
162         int g;
163         struct fdtable *fdt = NULL;
164         struct pid_namespace *ns;
165         pid_t ppid, tpid;
166
167         ns = current->nsproxy->pid_ns;
168         rcu_read_lock();
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_ppid_nr_ns(rcu_dereference(p->parent), ns) : 0;
173         buffer += sprintf(buffer,
174                 "State:\t%s\n"
175                 "Tgid:\t%d\n"
176                 "Pid:\t%d\n"
177                 "PPid:\t%d\n"
178                 "TracerPid:\t%d\n"
179                 "Uid:\t%d\t%d\t%d\t%d\n"
180                 "Gid:\t%d\t%d\t%d\t%d\n",
181                 get_task_state(p),
182                 task_tgid_nr_ns(p, ns),
183                 task_pid_nr_ns(p, ns),
184                 ppid, tpid,
185                 p->uid, p->euid, p->suid, p->fsuid,
186                 p->gid, p->egid, p->sgid, p->fsgid);
187
188         task_lock(p);
189         if (p->files)
190                 fdt = files_fdtable(p->files);
191         buffer += sprintf(buffer,
192                 "FDSize:\t%d\n"
193                 "Groups:\t",
194                 fdt ? fdt->max_fds : 0);
195         rcu_read_unlock();
196
197         group_info = p->group_info;
198         get_group_info(group_info);
199         task_unlock(p);
200
201         for (g = 0; g < min(group_info->ngroups, NGROUPS_SMALL); g++)
202                 buffer += sprintf(buffer, "%d ", GROUP_AT(group_info, g));
203         put_group_info(group_info);
204
205         buffer += sprintf(buffer, "\n");
206         return buffer;
207 }
208
209 static char *render_sigset_t(const char *header, sigset_t *set, char *buffer)
210 {
211         int i, len;
212
213         len = strlen(header);
214         memcpy(buffer, header, len);
215         buffer += len;
216
217         i = _NSIG;
218         do {
219                 int x = 0;
220
221                 i -= 4;
222                 if (sigismember(set, i+1)) x |= 1;
223                 if (sigismember(set, i+2)) x |= 2;
224                 if (sigismember(set, i+3)) x |= 4;
225                 if (sigismember(set, i+4)) x |= 8;
226                 *buffer++ = (x < 10 ? '0' : 'a' - 10) + x;
227         } while (i >= 4);
228
229         *buffer++ = '\n';
230         *buffer = 0;
231         return buffer;
232 }
233
234 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
235                                     sigset_t *catch)
236 {
237         struct k_sigaction *k;
238         int i;
239
240         k = p->sighand->action;
241         for (i = 1; i <= _NSIG; ++i, ++k) {
242                 if (k->sa.sa_handler == SIG_IGN)
243                         sigaddset(ign, i);
244                 else if (k->sa.sa_handler != SIG_DFL)
245                         sigaddset(catch, i);
246         }
247 }
248
249 static inline char *task_sig(struct task_struct *p, char *buffer)
250 {
251         unsigned long flags;
252         sigset_t pending, shpending, blocked, ignored, caught;
253         int num_threads = 0;
254         unsigned long qsize = 0;
255         unsigned long qlim = 0;
256
257         sigemptyset(&pending);
258         sigemptyset(&shpending);
259         sigemptyset(&blocked);
260         sigemptyset(&ignored);
261         sigemptyset(&caught);
262
263         rcu_read_lock();
264         if (lock_task_sighand(p, &flags)) {
265                 pending = p->pending.signal;
266                 shpending = p->signal->shared_pending.signal;
267                 blocked = p->blocked;
268                 collect_sigign_sigcatch(p, &ignored, &caught);
269                 num_threads = atomic_read(&p->signal->count);
270                 qsize = atomic_read(&p->user->sigpending);
271                 qlim = p->signal->rlim[RLIMIT_SIGPENDING].rlim_cur;
272                 unlock_task_sighand(p, &flags);
273         }
274         rcu_read_unlock();
275
276         buffer += sprintf(buffer, "Threads:\t%d\n", num_threads);
277         buffer += sprintf(buffer, "SigQ:\t%lu/%lu\n", qsize, qlim);
278
279         /* render them all */
280         buffer = render_sigset_t("SigPnd:\t", &pending, buffer);
281         buffer = render_sigset_t("ShdPnd:\t", &shpending, buffer);
282         buffer = render_sigset_t("SigBlk:\t", &blocked, buffer);
283         buffer = render_sigset_t("SigIgn:\t", &ignored, buffer);
284         buffer = render_sigset_t("SigCgt:\t", &caught, buffer);
285
286         return buffer;
287 }
288
289 static inline char *task_cap(struct task_struct *p, char *buffer)
290 {
291     return buffer + sprintf(buffer, "CapInh:\t%016x\n"
292                             "CapPrm:\t%016x\n"
293                             "CapEff:\t%016x\n",
294                             cap_t(p->cap_inheritable),
295                             cap_t(p->cap_permitted),
296                             cap_t(p->cap_effective));
297 }
298
299 static inline char *task_context_switch_counts(struct task_struct *p,
300                                                 char *buffer)
301 {
302         return buffer + sprintf(buffer, "voluntary_ctxt_switches:\t%lu\n"
303                             "nonvoluntary_ctxt_switches:\t%lu\n",
304                             p->nvcsw,
305                             p->nivcsw);
306 }
307
308 int proc_pid_status(struct task_struct *task, char *buffer)
309 {
310         char *orig = buffer;
311         struct mm_struct *mm = get_task_mm(task);
312
313         buffer = task_name(task, buffer);
314         buffer = task_state(task, buffer);
315
316         if (mm) {
317                 buffer = task_mem(mm, buffer);
318                 mmput(mm);
319         }
320         buffer = task_sig(task, buffer);
321         buffer = task_cap(task, buffer);
322         buffer = cpuset_task_status_allowed(task, buffer);
323 #if defined(CONFIG_S390)
324         buffer = task_show_regs(task, buffer);
325 #endif
326         buffer = task_context_switch_counts(task, buffer);
327         return buffer - orig;
328 }
329
330 /*
331  * Use precise platform statistics if available:
332  */
333 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
334 static cputime_t task_utime(struct task_struct *p)
335 {
336         return p->utime;
337 }
338
339 static cputime_t task_stime(struct task_struct *p)
340 {
341         return p->stime;
342 }
343 #else
344 static cputime_t task_utime(struct task_struct *p)
345 {
346         clock_t utime = cputime_to_clock_t(p->utime),
347                 total = utime + cputime_to_clock_t(p->stime);
348         u64 temp;
349
350         /*
351          * Use CFS's precise accounting:
352          */
353         temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);
354
355         if (total) {
356                 temp *= utime;
357                 do_div(temp, total);
358         }
359         utime = (clock_t)temp;
360
361         return clock_t_to_cputime(utime);
362 }
363
364 static cputime_t task_stime(struct task_struct *p)
365 {
366         clock_t stime;
367
368         /*
369          * Use CFS's precise accounting. (we subtract utime from
370          * the total, to make sure the total observed by userspace
371          * grows monotonically - apps rely on that):
372          */
373         stime = nsec_to_clock_t(p->se.sum_exec_runtime) -
374                         cputime_to_clock_t(task_utime(p));
375
376         return clock_t_to_cputime(stime);
377 }
378 #endif
379
380 static cputime_t task_gtime(struct task_struct *p)
381 {
382         return p->gtime;
383 }
384
385 static int do_task_stat(struct task_struct *task, char *buffer, int whole)
386 {
387         unsigned long vsize, eip, esp, wchan = ~0UL;
388         long priority, nice;
389         int tty_pgrp = -1, tty_nr = 0;
390         sigset_t sigign, sigcatch;
391         char state;
392         int res;
393         pid_t ppid = 0, pgid = -1, sid = -1;
394         int num_threads = 0;
395         struct mm_struct *mm;
396         unsigned long long start_time;
397         unsigned long cmin_flt = 0, cmaj_flt = 0;
398         unsigned long  min_flt = 0,  maj_flt = 0;
399         cputime_t cutime, cstime, utime, stime;
400         cputime_t cgtime, gtime;
401         unsigned long rsslim = 0;
402         char tcomm[sizeof(task->comm)];
403         unsigned long flags;
404         struct pid_namespace *ns;
405
406         ns = current->nsproxy->pid_ns;
407
408         state = *get_task_state(task);
409         vsize = eip = esp = 0;
410         mm = get_task_mm(task);
411         if (mm) {
412                 vsize = task_vsize(mm);
413                 eip = KSTK_EIP(task);
414                 esp = KSTK_ESP(task);
415         }
416
417         get_task_comm(tcomm, task);
418
419         sigemptyset(&sigign);
420         sigemptyset(&sigcatch);
421         cutime = cstime = utime = stime = cputime_zero;
422         cgtime = gtime = cputime_zero;
423
424         rcu_read_lock();
425         if (lock_task_sighand(task, &flags)) {
426                 struct signal_struct *sig = task->signal;
427
428                 if (sig->tty) {
429                         tty_pgrp = pid_nr_ns(sig->tty->pgrp, ns);
430                         tty_nr = new_encode_dev(tty_devnum(sig->tty));
431                 }
432
433                 num_threads = atomic_read(&sig->count);
434                 collect_sigign_sigcatch(task, &sigign, &sigcatch);
435
436                 cmin_flt = sig->cmin_flt;
437                 cmaj_flt = sig->cmaj_flt;
438                 cutime = sig->cutime;
439                 cstime = sig->cstime;
440                 cgtime = sig->cgtime;
441                 rsslim = sig->rlim[RLIMIT_RSS].rlim_cur;
442
443                 /* add up live thread stats at the group level */
444                 if (whole) {
445                         struct task_struct *t = task;
446                         do {
447                                 min_flt += t->min_flt;
448                                 maj_flt += t->maj_flt;
449                                 utime = cputime_add(utime, task_utime(t));
450                                 stime = cputime_add(stime, task_stime(t));
451                                 gtime = cputime_add(gtime, task_gtime(t));
452                                 t = next_thread(t);
453                         } while (t != task);
454
455                         min_flt += sig->min_flt;
456                         maj_flt += sig->maj_flt;
457                         utime = cputime_add(utime, sig->utime);
458                         stime = cputime_add(stime, sig->stime);
459                         gtime = cputime_add(gtime, sig->gtime);
460                 }
461
462                 sid = task_session_nr_ns(task, ns);
463                 pgid = task_pgrp_nr_ns(task, ns);
464                 ppid = task_ppid_nr_ns(task, ns);
465
466                 unlock_task_sighand(task, &flags);
467         }
468         rcu_read_unlock();
469
470         if (!whole || num_threads < 2)
471                 wchan = get_wchan(task);
472         if (!whole) {
473                 min_flt = task->min_flt;
474                 maj_flt = task->maj_flt;
475                 utime = task_utime(task);
476                 stime = task_stime(task);
477                 gtime = task_gtime(task);
478         }
479
480         /* scale priority and nice values from timeslices to -20..20 */
481         /* to make it look like a "normal" Unix priority/nice value  */
482         priority = task_prio(task);
483         nice = task_nice(task);
484
485         /* Temporary variable needed for gcc-2.96 */
486         /* convert timespec -> nsec*/
487         start_time =
488                 (unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
489                                 + task->real_start_time.tv_nsec;
490         /* convert nsec -> ticks */
491         start_time = nsec_to_clock_t(start_time);
492
493         res = sprintf(buffer, "%d (%s) %c %d %d %d %d %d %u %lu \
494 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
495 %lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu %lu %ld\n",
496                 task_pid_nr_ns(task, ns),
497                 tcomm,
498                 state,
499                 ppid,
500                 pgid,
501                 sid,
502                 tty_nr,
503                 tty_pgrp,
504                 task->flags,
505                 min_flt,
506                 cmin_flt,
507                 maj_flt,
508                 cmaj_flt,
509                 cputime_to_clock_t(utime),
510                 cputime_to_clock_t(stime),
511                 cputime_to_clock_t(cutime),
512                 cputime_to_clock_t(cstime),
513                 priority,
514                 nice,
515                 num_threads,
516                 start_time,
517                 vsize,
518                 mm ? get_mm_rss(mm) : 0,
519                 rsslim,
520                 mm ? mm->start_code : 0,
521                 mm ? mm->end_code : 0,
522                 mm ? mm->start_stack : 0,
523                 esp,
524                 eip,
525                 /* The signal information here is obsolete.
526                  * It must be decimal for Linux 2.0 compatibility.
527                  * Use /proc/#/status for real-time signals.
528                  */
529                 task->pending.signal.sig[0] & 0x7fffffffUL,
530                 task->blocked.sig[0] & 0x7fffffffUL,
531                 sigign      .sig[0] & 0x7fffffffUL,
532                 sigcatch    .sig[0] & 0x7fffffffUL,
533                 wchan,
534                 0UL,
535                 0UL,
536                 task->exit_signal,
537                 task_cpu(task),
538                 task->rt_priority,
539                 task->policy,
540                 (unsigned long long)delayacct_blkio_ticks(task),
541                 cputime_to_clock_t(gtime),
542                 cputime_to_clock_t(cgtime));
543         if (mm)
544                 mmput(mm);
545         return res;
546 }
547
548 int proc_tid_stat(struct task_struct *task, char *buffer)
549 {
550         return do_task_stat(task, buffer, 0);
551 }
552
553 int proc_tgid_stat(struct task_struct *task, char *buffer)
554 {
555         return do_task_stat(task, buffer, 1);
556 }
557
558 int proc_pid_statm(struct task_struct *task, char *buffer)
559 {
560         int size = 0, resident = 0, shared = 0, text = 0, lib = 0, data = 0;
561         struct mm_struct *mm = get_task_mm(task);
562
563         if (mm) {
564                 size = task_statm(mm, &shared, &text, &data, &resident);
565                 mmput(mm);
566         }
567
568         return sprintf(buffer, "%d %d %d %d %d %d %d\n",
569                        size, resident, shared, text, lib, data, 0);
570 }