[PATCH] per-task-delay-accounting: /proc export of aggregated block I/O delays
[safe/jmp/linux-2.6] / 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/mm.h>
66 #include <linux/hugetlb.h>
67 #include <linux/pagemap.h>
68 #include <linux/swap.h>
69 #include <linux/slab.h>
70 #include <linux/smp.h>
71 #include <linux/signal.h>
72 #include <linux/highmem.h>
73 #include <linux/file.h>
74 #include <linux/times.h>
75 #include <linux/cpuset.h>
76 #include <linux/rcupdate.h>
77 #include <linux/delayacct.h>
78
79 #include <asm/uaccess.h>
80 #include <asm/pgtable.h>
81 #include <asm/io.h>
82 #include <asm/processor.h>
83 #include "internal.h"
84
85 /* Gcc optimizes away "strlen(x)" for constant x */
86 #define ADDBUF(buffer, string) \
87 do { memcpy(buffer, string, strlen(string)); \
88      buffer += strlen(string); } while (0)
89
90 static inline char * task_name(struct task_struct *p, char * buf)
91 {
92         int i;
93         char * name;
94         char tcomm[sizeof(p->comm)];
95
96         get_task_comm(tcomm, p);
97
98         ADDBUF(buf, "Name:\t");
99         name = tcomm;
100         i = sizeof(tcomm);
101         do {
102                 unsigned char c = *name;
103                 name++;
104                 i--;
105                 *buf = c;
106                 if (!c)
107                         break;
108                 if (c == '\\') {
109                         buf[1] = c;
110                         buf += 2;
111                         continue;
112                 }
113                 if (c == '\n') {
114                         buf[0] = '\\';
115                         buf[1] = 'n';
116                         buf += 2;
117                         continue;
118                 }
119                 buf++;
120         } while (i);
121         *buf = '\n';
122         return buf+1;
123 }
124
125 /*
126  * The task state array is a strange "bitmap" of
127  * reasons to sleep. Thus "running" is zero, and
128  * you can test for combinations of others with
129  * simple bit tests.
130  */
131 static const char *task_state_array[] = {
132         "R (running)",          /*  0 */
133         "S (sleeping)",         /*  1 */
134         "D (disk sleep)",       /*  2 */
135         "T (stopped)",          /*  4 */
136         "T (tracing stop)",     /*  8 */
137         "Z (zombie)",           /* 16 */
138         "X (dead)"              /* 32 */
139 };
140
141 static inline const char * get_task_state(struct task_struct *tsk)
142 {
143         unsigned int state = (tsk->state & (TASK_RUNNING |
144                                             TASK_INTERRUPTIBLE |
145                                             TASK_UNINTERRUPTIBLE |
146                                             TASK_STOPPED |
147                                             TASK_TRACED)) |
148                         (tsk->exit_state & (EXIT_ZOMBIE |
149                                             EXIT_DEAD));
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
165         read_lock(&tasklist_lock);
166         buffer += sprintf(buffer,
167                 "State:\t%s\n"
168                 "SleepAVG:\t%lu%%\n"
169                 "Tgid:\t%d\n"
170                 "Pid:\t%d\n"
171                 "PPid:\t%d\n"
172                 "TracerPid:\t%d\n"
173                 "Uid:\t%d\t%d\t%d\t%d\n"
174                 "Gid:\t%d\t%d\t%d\t%d\n",
175                 get_task_state(p),
176                 (p->sleep_avg/1024)*100/(1020000000/1024),
177                 p->tgid,
178                 p->pid, pid_alive(p) ? p->group_leader->real_parent->tgid : 0,
179                 pid_alive(p) && p->ptrace ? p->parent->pid : 0,
180                 p->uid, p->euid, p->suid, p->fsuid,
181                 p->gid, p->egid, p->sgid, p->fsgid);
182         read_unlock(&tasklist_lock);
183         task_lock(p);
184         rcu_read_lock();
185         if (p->files)
186                 fdt = files_fdtable(p->files);
187         buffer += sprintf(buffer,
188                 "FDSize:\t%d\n"
189                 "Groups:\t",
190                 fdt ? fdt->max_fds : 0);
191         rcu_read_unlock();
192
193         group_info = p->group_info;
194         get_group_info(group_info);
195         task_unlock(p);
196
197         for (g = 0; g < min(group_info->ngroups,NGROUPS_SMALL); g++)
198                 buffer += sprintf(buffer, "%d ", GROUP_AT(group_info,g));
199         put_group_info(group_info);
200
201         buffer += sprintf(buffer, "\n");
202         return buffer;
203 }
204
205 static char * render_sigset_t(const char *header, sigset_t *set, char *buffer)
206 {
207         int i, len;
208
209         len = strlen(header);
210         memcpy(buffer, header, len);
211         buffer += len;
212
213         i = _NSIG;
214         do {
215                 int x = 0;
216
217                 i -= 4;
218                 if (sigismember(set, i+1)) x |= 1;
219                 if (sigismember(set, i+2)) x |= 2;
220                 if (sigismember(set, i+3)) x |= 4;
221                 if (sigismember(set, i+4)) x |= 8;
222                 *buffer++ = (x < 10 ? '0' : 'a' - 10) + x;
223         } while (i >= 4);
224
225         *buffer++ = '\n';
226         *buffer = 0;
227         return buffer;
228 }
229
230 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
231                                     sigset_t *catch)
232 {
233         struct k_sigaction *k;
234         int i;
235
236         k = p->sighand->action;
237         for (i = 1; i <= _NSIG; ++i, ++k) {
238                 if (k->sa.sa_handler == SIG_IGN)
239                         sigaddset(ign, i);
240                 else if (k->sa.sa_handler != SIG_DFL)
241                         sigaddset(catch, i);
242         }
243 }
244
245 static inline char * task_sig(struct task_struct *p, char *buffer)
246 {
247         sigset_t pending, shpending, blocked, ignored, caught;
248         int num_threads = 0;
249         unsigned long qsize = 0;
250         unsigned long qlim = 0;
251
252         sigemptyset(&pending);
253         sigemptyset(&shpending);
254         sigemptyset(&blocked);
255         sigemptyset(&ignored);
256         sigemptyset(&caught);
257
258         /* Gather all the data with the appropriate locks held */
259         read_lock(&tasklist_lock);
260         if (p->sighand) {
261                 spin_lock_irq(&p->sighand->siglock);
262                 pending = p->pending.signal;
263                 shpending = p->signal->shared_pending.signal;
264                 blocked = p->blocked;
265                 collect_sigign_sigcatch(p, &ignored, &caught);
266                 num_threads = atomic_read(&p->signal->count);
267                 qsize = atomic_read(&p->user->sigpending);
268                 qlim = p->signal->rlim[RLIMIT_SIGPENDING].rlim_cur;
269                 spin_unlock_irq(&p->sighand->siglock);
270         }
271         read_unlock(&tasklist_lock);
272
273         buffer += sprintf(buffer, "Threads:\t%d\n", num_threads);
274         buffer += sprintf(buffer, "SigQ:\t%lu/%lu\n", qsize, qlim);
275
276         /* render them all */
277         buffer = render_sigset_t("SigPnd:\t", &pending, buffer);
278         buffer = render_sigset_t("ShdPnd:\t", &shpending, buffer);
279         buffer = render_sigset_t("SigBlk:\t", &blocked, buffer);
280         buffer = render_sigset_t("SigIgn:\t", &ignored, buffer);
281         buffer = render_sigset_t("SigCgt:\t", &caught, buffer);
282
283         return buffer;
284 }
285
286 static inline char *task_cap(struct task_struct *p, char *buffer)
287 {
288     return buffer + sprintf(buffer, "CapInh:\t%016x\n"
289                             "CapPrm:\t%016x\n"
290                             "CapEff:\t%016x\n",
291                             cap_t(p->cap_inheritable),
292                             cap_t(p->cap_permitted),
293                             cap_t(p->cap_effective));
294 }
295
296 int proc_pid_status(struct task_struct *task, char * buffer)
297 {
298         char * orig = buffer;
299         struct mm_struct *mm = get_task_mm(task);
300
301         buffer = task_name(task, buffer);
302         buffer = task_state(task, buffer);
303  
304         if (mm) {
305                 buffer = task_mem(mm, buffer);
306                 mmput(mm);
307         }
308         buffer = task_sig(task, buffer);
309         buffer = task_cap(task, buffer);
310         buffer = cpuset_task_status_allowed(task, buffer);
311 #if defined(CONFIG_S390)
312         buffer = task_show_regs(task, buffer);
313 #endif
314         return buffer - orig;
315 }
316
317 static int do_task_stat(struct task_struct *task, char * buffer, int whole)
318 {
319         unsigned long vsize, eip, esp, wchan = ~0UL;
320         long priority, nice;
321         int tty_pgrp = -1, tty_nr = 0;
322         sigset_t sigign, sigcatch;
323         char state;
324         int res;
325         pid_t ppid, pgid = -1, sid = -1;
326         int num_threads = 0;
327         struct mm_struct *mm;
328         unsigned long long start_time;
329         unsigned long cmin_flt = 0, cmaj_flt = 0;
330         unsigned long  min_flt = 0,  maj_flt = 0;
331         cputime_t cutime, cstime, utime, stime;
332         unsigned long rsslim = 0;
333         struct task_struct *t;
334         char tcomm[sizeof(task->comm)];
335
336         state = *get_task_state(task);
337         vsize = eip = esp = 0;
338         mm = get_task_mm(task);
339         if (mm) {
340                 vsize = task_vsize(mm);
341                 eip = KSTK_EIP(task);
342                 esp = KSTK_ESP(task);
343         }
344
345         get_task_comm(tcomm, task);
346
347         sigemptyset(&sigign);
348         sigemptyset(&sigcatch);
349         cutime = cstime = utime = stime = cputime_zero;
350         read_lock(&tasklist_lock);
351         if (task->sighand) {
352                 spin_lock_irq(&task->sighand->siglock);
353                 num_threads = atomic_read(&task->signal->count);
354                 collect_sigign_sigcatch(task, &sigign, &sigcatch);
355
356                 /* add up live thread stats at the group level */
357                 if (whole) {
358                         t = task;
359                         do {
360                                 min_flt += t->min_flt;
361                                 maj_flt += t->maj_flt;
362                                 utime = cputime_add(utime, t->utime);
363                                 stime = cputime_add(stime, t->stime);
364                                 t = next_thread(t);
365                         } while (t != task);
366                 }
367
368                 spin_unlock_irq(&task->sighand->siglock);
369         }
370         if (task->signal) {
371                 if (task->signal->tty) {
372                         tty_pgrp = task->signal->tty->pgrp;
373                         tty_nr = new_encode_dev(tty_devnum(task->signal->tty));
374                 }
375                 pgid = process_group(task);
376                 sid = task->signal->session;
377                 cmin_flt = task->signal->cmin_flt;
378                 cmaj_flt = task->signal->cmaj_flt;
379                 cutime = task->signal->cutime;
380                 cstime = task->signal->cstime;
381                 rsslim = task->signal->rlim[RLIMIT_RSS].rlim_cur;
382                 if (whole) {
383                         min_flt += task->signal->min_flt;
384                         maj_flt += task->signal->maj_flt;
385                         utime = cputime_add(utime, task->signal->utime);
386                         stime = cputime_add(stime, task->signal->stime);
387                 }
388         }
389         ppid = pid_alive(task) ? task->group_leader->real_parent->tgid : 0;
390         read_unlock(&tasklist_lock);
391
392         if (!whole || num_threads<2)
393                 wchan = get_wchan(task);
394         if (!whole) {
395                 min_flt = task->min_flt;
396                 maj_flt = task->maj_flt;
397                 utime = task->utime;
398                 stime = task->stime;
399         }
400
401         /* scale priority and nice values from timeslices to -20..20 */
402         /* to make it look like a "normal" Unix priority/nice value  */
403         priority = task_prio(task);
404         nice = task_nice(task);
405
406         /* Temporary variable needed for gcc-2.96 */
407         /* convert timespec -> nsec*/
408         start_time = (unsigned long long)task->start_time.tv_sec * NSEC_PER_SEC
409                                 + task->start_time.tv_nsec;
410         /* convert nsec -> ticks */
411         start_time = nsec_to_clock_t(start_time);
412
413         res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %lu %lu \
414 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
415 %lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu %llu\n",
416                 task->pid,
417                 tcomm,
418                 state,
419                 ppid,
420                 pgid,
421                 sid,
422                 tty_nr,
423                 tty_pgrp,
424                 task->flags,
425                 min_flt,
426                 cmin_flt,
427                 maj_flt,
428                 cmaj_flt,
429                 cputime_to_clock_t(utime),
430                 cputime_to_clock_t(stime),
431                 cputime_to_clock_t(cutime),
432                 cputime_to_clock_t(cstime),
433                 priority,
434                 nice,
435                 num_threads,
436                 start_time,
437                 vsize,
438                 mm ? get_mm_rss(mm) : 0,
439                 rsslim,
440                 mm ? mm->start_code : 0,
441                 mm ? mm->end_code : 0,
442                 mm ? mm->start_stack : 0,
443                 esp,
444                 eip,
445                 /* The signal information here is obsolete.
446                  * It must be decimal for Linux 2.0 compatibility.
447                  * Use /proc/#/status for real-time signals.
448                  */
449                 task->pending.signal.sig[0] & 0x7fffffffUL,
450                 task->blocked.sig[0] & 0x7fffffffUL,
451                 sigign      .sig[0] & 0x7fffffffUL,
452                 sigcatch    .sig[0] & 0x7fffffffUL,
453                 wchan,
454                 0UL,
455                 0UL,
456                 task->exit_signal,
457                 task_cpu(task),
458                 task->rt_priority,
459                 task->policy,
460                 (unsigned long long)delayacct_blkio_ticks(task));
461         if(mm)
462                 mmput(mm);
463         return res;
464 }
465
466 int proc_tid_stat(struct task_struct *task, char * buffer)
467 {
468         return do_task_stat(task, buffer, 0);
469 }
470
471 int proc_tgid_stat(struct task_struct *task, char * buffer)
472 {
473         return do_task_stat(task, buffer, 1);
474 }
475
476 int proc_pid_statm(struct task_struct *task, char *buffer)
477 {
478         int size = 0, resident = 0, shared = 0, text = 0, lib = 0, data = 0;
479         struct mm_struct *mm = get_task_mm(task);
480         
481         if (mm) {
482                 size = task_statm(mm, &shared, &text, &data, &resident);
483                 mmput(mm);
484         }
485
486         return sprintf(buffer,"%d %d %d %d %d %d %d\n",
487                        size, resident, shared, text, lib, data, 0);
488 }