xref: /linux/fs/proc/array.c (revision 42fda66387daa53538ae13a2c858396aaf037158) 
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 
81 #include <asm/pgtable.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 	rcu_read_lock();
166 	buffer += sprintf(buffer,
167 		"State:\t%s\n"
168 		"Tgid:\t%d\n"
169 		"Pid:\t%d\n"
170 		"PPid:\t%d\n"
171 		"TracerPid:\t%d\n"
172 		"Uid:\t%d\t%d\t%d\t%d\n"
173 		"Gid:\t%d\t%d\t%d\t%d\n",
174 		get_task_state(p),
175 		p->tgid, p->pid,
176 		pid_alive(p) ? rcu_dereference(p->real_parent)->tgid : 0,
177 		pid_alive(p) && p->ptrace ? rcu_dereference(p->parent)->pid : 0,
178 		p->uid, p->euid, p->suid, p->fsuid,
179 		p->gid, p->egid, p->sgid, p->fsgid);
180 
181 	task_lock(p);
182 	if (p->files)
183 		fdt = files_fdtable(p->files);
184 	buffer += sprintf(buffer,
185 		"FDSize:\t%d\n"
186 		"Groups:\t",
187 		fdt ? fdt->max_fds : 0);
188 	rcu_read_unlock();
189 
190 	group_info = p->group_info;
191 	get_group_info(group_info);
192 	task_unlock(p);
193 
194 	for (g = 0; g < min(group_info->ngroups, NGROUPS_SMALL); g++)
195 		buffer += sprintf(buffer, "%d ", GROUP_AT(group_info, g));
196 	put_group_info(group_info);
197 
198 	buffer += sprintf(buffer, "\n");
199 	return buffer;
200 }
201 
202 static char *render_sigset_t(const char *header, sigset_t *set, char *buffer)
203 {
204 	int i, len;
205 
206 	len = strlen(header);
207 	memcpy(buffer, header, len);
208 	buffer += len;
209 
210 	i = _NSIG;
211 	do {
212 		int x = 0;
213 
214 		i -= 4;
215 		if (sigismember(set, i+1)) x |= 1;
216 		if (sigismember(set, i+2)) x |= 2;
217 		if (sigismember(set, i+3)) x |= 4;
218 		if (sigismember(set, i+4)) x |= 8;
219 		*buffer++ = (x < 10 ? '0' : 'a' - 10) + x;
220 	} while (i >= 4);
221 
222 	*buffer++ = '\n';
223 	*buffer = 0;
224 	return buffer;
225 }
226 
227 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
228 				    sigset_t *catch)
229 {
230 	struct k_sigaction *k;
231 	int i;
232 
233 	k = p->sighand->action;
234 	for (i = 1; i <= _NSIG; ++i, ++k) {
235 		if (k->sa.sa_handler == SIG_IGN)
236 			sigaddset(ign, i);
237 		else if (k->sa.sa_handler != SIG_DFL)
238 			sigaddset(catch, i);
239 	}
240 }
241 
242 static inline char *task_sig(struct task_struct *p, char *buffer)
243 {
244 	unsigned long flags;
245 	sigset_t pending, shpending, blocked, ignored, caught;
246 	int num_threads = 0;
247 	unsigned long qsize = 0;
248 	unsigned long qlim = 0;
249 
250 	sigemptyset(&pending);
251 	sigemptyset(&shpending);
252 	sigemptyset(&blocked);
253 	sigemptyset(&ignored);
254 	sigemptyset(&caught);
255 
256 	rcu_read_lock();
257 	if (lock_task_sighand(p, &flags)) {
258 		pending = p->pending.signal;
259 		shpending = p->signal->shared_pending.signal;
260 		blocked = p->blocked;
261 		collect_sigign_sigcatch(p, &ignored, &caught);
262 		num_threads = atomic_read(&p->signal->count);
263 		qsize = atomic_read(&p->user->sigpending);
264 		qlim = p->signal->rlim[RLIMIT_SIGPENDING].rlim_cur;
265 		unlock_task_sighand(p, &flags);
266 	}
267 	rcu_read_unlock();
268 
269 	buffer += sprintf(buffer, "Threads:\t%d\n", num_threads);
270 	buffer += sprintf(buffer, "SigQ:\t%lu/%lu\n", qsize, qlim);
271 
272 	/* render them all */
273 	buffer = render_sigset_t("SigPnd:\t", &pending, buffer);
274 	buffer = render_sigset_t("ShdPnd:\t", &shpending, buffer);
275 	buffer = render_sigset_t("SigBlk:\t", &blocked, buffer);
276 	buffer = render_sigset_t("SigIgn:\t", &ignored, buffer);
277 	buffer = render_sigset_t("SigCgt:\t", &caught, buffer);
278 
279 	return buffer;
280 }
281 
282 static inline char *task_cap(struct task_struct *p, char *buffer)
283 {
284     return buffer + sprintf(buffer, "CapInh:\t%016x\n"
285 			    "CapPrm:\t%016x\n"
286 			    "CapEff:\t%016x\n",
287 			    cap_t(p->cap_inheritable),
288 			    cap_t(p->cap_permitted),
289 			    cap_t(p->cap_effective));
290 }
291 
292 static inline char *task_context_switch_counts(struct task_struct *p,
293 						char *buffer)
294 {
295 	return buffer + sprintf(buffer, "voluntary_ctxt_switches:\t%lu\n"
296 			    "nonvoluntary_ctxt_switches:\t%lu\n",
297 			    p->nvcsw,
298 			    p->nivcsw);
299 }
300 
301 int proc_pid_status(struct task_struct *task, char *buffer)
302 {
303 	char *orig = buffer;
304 	struct mm_struct *mm = get_task_mm(task);
305 
306 	buffer = task_name(task, buffer);
307 	buffer = task_state(task, buffer);
308 
309 	if (mm) {
310 		buffer = task_mem(mm, buffer);
311 		mmput(mm);
312 	}
313 	buffer = task_sig(task, buffer);
314 	buffer = task_cap(task, buffer);
315 	buffer = cpuset_task_status_allowed(task, buffer);
316 #if defined(CONFIG_S390)
317 	buffer = task_show_regs(task, buffer);
318 #endif
319 	buffer = task_context_switch_counts(task, buffer);
320 	return buffer - orig;
321 }
322 
323 /*
324  * Use precise platform statistics if available:
325  */
326 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
327 static cputime_t task_utime(struct task_struct *p)
328 {
329 	return p->utime;
330 }
331 
332 static cputime_t task_stime(struct task_struct *p)
333 {
334 	return p->stime;
335 }
336 #else
337 static cputime_t task_utime(struct task_struct *p)
338 {
339 	clock_t utime = cputime_to_clock_t(p->utime),
340 		total = utime + cputime_to_clock_t(p->stime);
341 	u64 temp;
342 
343 	/*
344 	 * Use CFS's precise accounting:
345 	 */
346 	temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);
347 
348 	if (total) {
349 		temp *= utime;
350 		do_div(temp, total);
351 	}
352 	utime = (clock_t)temp;
353 
354 	return clock_t_to_cputime(utime);
355 }
356 
357 static cputime_t task_stime(struct task_struct *p)
358 {
359 	clock_t stime;
360 
361 	/*
362 	 * Use CFS's precise accounting. (we subtract utime from
363 	 * the total, to make sure the total observed by userspace
364 	 * grows monotonically - apps rely on that):
365 	 */
366 	stime = nsec_to_clock_t(p->se.sum_exec_runtime) -
367 			cputime_to_clock_t(task_utime(p));
368 
369 	return clock_t_to_cputime(stime);
370 }
371 #endif
372 
373 static cputime_t task_gtime(struct task_struct *p)
374 {
375 	return p->gtime;
376 }
377 
378 static int do_task_stat(struct task_struct *task, char *buffer, int whole)
379 {
380 	unsigned long vsize, eip, esp, wchan = ~0UL;
381 	long priority, nice;
382 	int tty_pgrp = -1, tty_nr = 0;
383 	sigset_t sigign, sigcatch;
384 	char state;
385 	int res;
386 	pid_t ppid = 0, pgid = -1, sid = -1;
387 	int num_threads = 0;
388 	struct mm_struct *mm;
389 	unsigned long long start_time;
390 	unsigned long cmin_flt = 0, cmaj_flt = 0;
391 	unsigned long  min_flt = 0,  maj_flt = 0;
392 	cputime_t cutime, cstime, utime, stime;
393 	cputime_t cgtime, gtime;
394 	unsigned long rsslim = 0;
395 	char tcomm[sizeof(task->comm)];
396 	unsigned long flags;
397 
398 	state = *get_task_state(task);
399 	vsize = eip = esp = 0;
400 	mm = get_task_mm(task);
401 	if (mm) {
402 		vsize = task_vsize(mm);
403 		eip = KSTK_EIP(task);
404 		esp = KSTK_ESP(task);
405 	}
406 
407 	get_task_comm(tcomm, task);
408 
409 	sigemptyset(&sigign);
410 	sigemptyset(&sigcatch);
411 	cutime = cstime = utime = stime = cputime_zero;
412 	cgtime = gtime = cputime_zero;
413 
414 	rcu_read_lock();
415 	if (lock_task_sighand(task, &flags)) {
416 		struct signal_struct *sig = task->signal;
417 
418 		if (sig->tty) {
419 			tty_pgrp = pid_nr(sig->tty->pgrp);
420 			tty_nr = new_encode_dev(tty_devnum(sig->tty));
421 		}
422 
423 		num_threads = atomic_read(&sig->count);
424 		collect_sigign_sigcatch(task, &sigign, &sigcatch);
425 
426 		cmin_flt = sig->cmin_flt;
427 		cmaj_flt = sig->cmaj_flt;
428 		cutime = sig->cutime;
429 		cstime = sig->cstime;
430 		cgtime = sig->cgtime;
431 		rsslim = sig->rlim[RLIMIT_RSS].rlim_cur;
432 
433 		/* add up live thread stats at the group level */
434 		if (whole) {
435 			struct task_struct *t = task;
436 			do {
437 				min_flt += t->min_flt;
438 				maj_flt += t->maj_flt;
439 				utime = cputime_add(utime, task_utime(t));
440 				stime = cputime_add(stime, task_stime(t));
441 				gtime = cputime_add(gtime, task_gtime(t));
442 				t = next_thread(t);
443 			} while (t != task);
444 
445 			min_flt += sig->min_flt;
446 			maj_flt += sig->maj_flt;
447 			utime = cputime_add(utime, sig->utime);
448 			stime = cputime_add(stime, sig->stime);
449 			gtime += cputime_add(gtime, sig->gtime);
450 		}
451 
452 		sid = signal_session(sig);
453 		pgid = process_group(task);
454 		ppid = rcu_dereference(task->real_parent)->tgid;
455 
456 		unlock_task_sighand(task, &flags);
457 	}
458 	rcu_read_unlock();
459 
460 	if (!whole || num_threads < 2)
461 		wchan = get_wchan(task);
462 	if (!whole) {
463 		min_flt = task->min_flt;
464 		maj_flt = task->maj_flt;
465 		utime = task_utime(task);
466 		stime = task_stime(task);
467 		gtime = task_gtime(task);
468 	}
469 
470 	/* scale priority and nice values from timeslices to -20..20 */
471 	/* to make it look like a "normal" Unix priority/nice value  */
472 	priority = task_prio(task);
473 	nice = task_nice(task);
474 
475 	/* Temporary variable needed for gcc-2.96 */
476 	/* convert timespec -> nsec*/
477 	start_time =
478 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
479 				+ task->real_start_time.tv_nsec;
480 	/* convert nsec -> ticks */
481 	start_time = nsec_to_clock_t(start_time);
482 
483 	res = sprintf(buffer, "%d (%s) %c %d %d %d %d %d %u %lu \
484 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
485 %lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu %lu %ld\n",
486 		task->pid,
487 		tcomm,
488 		state,
489 		ppid,
490 		pgid,
491 		sid,
492 		tty_nr,
493 		tty_pgrp,
494 		task->flags,
495 		min_flt,
496 		cmin_flt,
497 		maj_flt,
498 		cmaj_flt,
499 		cputime_to_clock_t(utime),
500 		cputime_to_clock_t(stime),
501 		cputime_to_clock_t(cutime),
502 		cputime_to_clock_t(cstime),
503 		priority,
504 		nice,
505 		num_threads,
506 		start_time,
507 		vsize,
508 		mm ? get_mm_rss(mm) : 0,
509 		rsslim,
510 		mm ? mm->start_code : 0,
511 		mm ? mm->end_code : 0,
512 		mm ? mm->start_stack : 0,
513 		esp,
514 		eip,
515 		/* The signal information here is obsolete.
516 		 * It must be decimal for Linux 2.0 compatibility.
517 		 * Use /proc/#/status for real-time signals.
518 		 */
519 		task->pending.signal.sig[0] & 0x7fffffffUL,
520 		task->blocked.sig[0] & 0x7fffffffUL,
521 		sigign      .sig[0] & 0x7fffffffUL,
522 		sigcatch    .sig[0] & 0x7fffffffUL,
523 		wchan,
524 		0UL,
525 		0UL,
526 		task->exit_signal,
527 		task_cpu(task),
528 		task->rt_priority,
529 		task->policy,
530 		(unsigned long long)delayacct_blkio_ticks(task),
531 		cputime_to_clock_t(gtime),
532 		cputime_to_clock_t(cgtime));
533 	if (mm)
534 		mmput(mm);
535 	return res;
536 }
537 
538 int proc_tid_stat(struct task_struct *task, char *buffer)
539 {
540 	return do_task_stat(task, buffer, 0);
541 }
542 
543 int proc_tgid_stat(struct task_struct *task, char *buffer)
544 {
545 	return do_task_stat(task, buffer, 1);
546 }
547 
548 int proc_pid_statm(struct task_struct *task, char *buffer)
549 {
550 	int size = 0, resident = 0, shared = 0, text = 0, lib = 0, data = 0;
551 	struct mm_struct *mm = get_task_mm(task);
552 
553 	if (mm) {
554 		size = task_statm(mm, &shared, &text, &data, &resident);
555 		mmput(mm);
556 	}
557 
558 	return sprintf(buffer, "%d %d %d %d %d %d %d\n",
559 		       size, resident, shared, text, lib, data, 0);
560 }
561