1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/proc/array.c
5 * Copyright (C) 1992 by Linus Torvalds
6 * based on ideas by Darren Senn
9 * Michael. K. Johnson: stat,statm extensions.
12 * Pauline Middelink : Made cmdline,envline only break at '\0's, to
13 * make sure SET_PROCTITLE works. Also removed
14 * bad '!' which forced address recalculation for
15 * EVERY character on the current page.
18 * Danny ter Haar : added cpuinfo
21 * Alessandro Rubini : profile extension.
24 * Jeff Tranter : added BogoMips field to cpuinfo
27 * Bruno Haible : remove 4K limit for the maps file
30 * Yves Arrouye : remove removal of trailing spaces in get_array.
33 * Jerome Forissier : added per-CPU time information to /proc/stat
34 * and /proc/<pid>/cpu extension
36 * - Incorporation and non-SMP safe operation
37 * of forissier patch in 2.1.78 by
43 * Alan Cox : security fixes.
46 * Al Viro : safe handling of mm_struct
48 * Gerhard Wichert : added BIGMEM support
51 * Al Viro & Jeff Garzik : moved most of the thing into base.c and
52 * : proc_misc.c. The rest may eventually go into
56 #include <linux/types.h>
57 #include <linux/errno.h>
58 #include <linux/time.h>
59 #include <linux/kernel.h>
60 #include <linux/kernel_stat.h>
61 #include <linux/tty.h>
62 #include <linux/string.h>
63 #include <linux/mman.h>
64 #include <linux/sched/mm.h>
65 #include <linux/sched/numa_balancing.h>
66 #include <linux/sched/task_stack.h>
67 #include <linux/sched/task.h>
68 #include <linux/sched/cputime.h>
69 #include <linux/proc_fs.h>
70 #include <linux/ioport.h>
71 #include <linux/uaccess.h>
74 #include <linux/hugetlb.h>
75 #include <linux/pagemap.h>
76 #include <linux/swap.h>
77 #include <linux/smp.h>
78 #include <linux/signal.h>
79 #include <linux/highmem.h>
80 #include <linux/file.h>
81 #include <linux/fdtable.h>
82 #include <linux/times.h>
83 #include <linux/cpuset.h>
84 #include <linux/rcupdate.h>
85 #include <linux/delayacct.h>
86 #include <linux/seq_file.h>
87 #include <linux/pid_namespace.h>
88 #include <linux/ptrace.h>
89 #include <linux/tracehook.h>
90 #include <linux/string_helpers.h>
91 #include <linux/user_namespace.h>
92 #include <linux/fs_struct.h>
94 #include <asm/pgtable.h>
95 #include <asm/processor.h>
98 static inline void task_name(struct seq_file *m, struct task_struct *p)
102 char tcomm[sizeof(p->comm)];
105 get_task_comm(tcomm, p);
107 seq_puts(m, "Name:\t");
109 size = seq_get_buf(m, &buf);
110 ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\");
111 seq_commit(m, ret < size ? ret : -1);
117 * The task state array is a strange "bitmap" of
118 * reasons to sleep. Thus "running" is zero, and
119 * you can test for combinations of others with
122 static const char * const task_state_array[] = {
124 /* states in TASK_REPORT: */
125 "R (running)", /* 0x00 */
126 "S (sleeping)", /* 0x01 */
127 "D (disk sleep)", /* 0x02 */
128 "T (stopped)", /* 0x04 */
129 "t (tracing stop)", /* 0x08 */
130 "X (dead)", /* 0x10 */
131 "Z (zombie)", /* 0x20 */
132 "P (parked)", /* 0x40 */
134 /* states beyond TASK_REPORT: */
135 "I (idle)", /* 0x80 */
138 static inline const char *get_task_state(struct task_struct *tsk)
140 BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != ARRAY_SIZE(task_state_array));
141 return task_state_array[task_state_index(tsk)];
144 static inline void task_state(struct seq_file *m, struct pid_namespace *ns,
145 struct pid *pid, struct task_struct *p)
147 struct user_namespace *user_ns = seq_user_ns(m);
148 struct group_info *group_info;
150 struct task_struct *tracer;
151 const struct cred *cred;
152 pid_t ppid, tpid = 0, tgid, ngid;
153 unsigned int max_fds = 0;
156 ppid = pid_alive(p) ?
157 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0;
159 tracer = ptrace_parent(p);
161 tpid = task_pid_nr_ns(tracer, ns);
163 tgid = task_tgid_nr_ns(p, ns);
164 ngid = task_numa_group_id(p);
165 cred = get_task_cred(p);
169 umask = p->fs->umask;
171 max_fds = files_fdtable(p->files)->max_fds;
176 seq_printf(m, "Umask:\t%#04o\n", umask);
177 seq_puts(m, "State:\t");
178 seq_puts(m, get_task_state(p));
180 seq_put_decimal_ull(m, "\nTgid:\t", tgid);
181 seq_put_decimal_ull(m, "\nNgid:\t", ngid);
182 seq_put_decimal_ull(m, "\nPid:\t", pid_nr_ns(pid, ns));
183 seq_put_decimal_ull(m, "\nPPid:\t", ppid);
184 seq_put_decimal_ull(m, "\nTracerPid:\t", tpid);
185 seq_put_decimal_ull(m, "\nUid:\t", from_kuid_munged(user_ns, cred->uid));
186 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->euid));
187 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->suid));
188 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->fsuid));
189 seq_put_decimal_ull(m, "\nGid:\t", from_kgid_munged(user_ns, cred->gid));
190 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->egid));
191 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->sgid));
192 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->fsgid));
193 seq_put_decimal_ull(m, "\nFDSize:\t", max_fds);
195 seq_puts(m, "\nGroups:\t");
196 group_info = cred->group_info;
197 for (g = 0; g < group_info->ngroups; g++)
198 seq_put_decimal_ull(m, g ? " " : "",
199 from_kgid_munged(user_ns, group_info->gid[g]));
201 /* Trailing space shouldn't have been added in the first place. */
205 seq_puts(m, "\nNStgid:");
206 for (g = ns->level; g <= pid->level; g++)
207 seq_put_decimal_ull(m, "\t", task_tgid_nr_ns(p, pid->numbers[g].ns));
208 seq_puts(m, "\nNSpid:");
209 for (g = ns->level; g <= pid->level; g++)
210 seq_put_decimal_ull(m, "\t", task_pid_nr_ns(p, pid->numbers[g].ns));
211 seq_puts(m, "\nNSpgid:");
212 for (g = ns->level; g <= pid->level; g++)
213 seq_put_decimal_ull(m, "\t", task_pgrp_nr_ns(p, pid->numbers[g].ns));
214 seq_puts(m, "\nNSsid:");
215 for (g = ns->level; g <= pid->level; g++)
216 seq_put_decimal_ull(m, "\t", task_session_nr_ns(p, pid->numbers[g].ns));
221 void render_sigset_t(struct seq_file *m, const char *header,
233 if (sigismember(set, i+1)) x |= 1;
234 if (sigismember(set, i+2)) x |= 2;
235 if (sigismember(set, i+3)) x |= 4;
236 if (sigismember(set, i+4)) x |= 8;
237 seq_putc(m, hex_asc[x]);
243 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
246 struct k_sigaction *k;
249 k = p->sighand->action;
250 for (i = 1; i <= _NSIG; ++i, ++k) {
251 if (k->sa.sa_handler == SIG_IGN)
253 else if (k->sa.sa_handler != SIG_DFL)
258 static inline void task_sig(struct seq_file *m, struct task_struct *p)
261 sigset_t pending, shpending, blocked, ignored, caught;
263 unsigned long qsize = 0;
264 unsigned long qlim = 0;
266 sigemptyset(&pending);
267 sigemptyset(&shpending);
268 sigemptyset(&blocked);
269 sigemptyset(&ignored);
270 sigemptyset(&caught);
272 if (lock_task_sighand(p, &flags)) {
273 pending = p->pending.signal;
274 shpending = p->signal->shared_pending.signal;
275 blocked = p->blocked;
276 collect_sigign_sigcatch(p, &ignored, &caught);
277 num_threads = get_nr_threads(p);
278 rcu_read_lock(); /* FIXME: is this correct? */
279 qsize = atomic_read(&__task_cred(p)->user->sigpending);
281 qlim = task_rlimit(p, RLIMIT_SIGPENDING);
282 unlock_task_sighand(p, &flags);
285 seq_put_decimal_ull(m, "Threads:\t", num_threads);
286 seq_put_decimal_ull(m, "\nSigQ:\t", qsize);
287 seq_put_decimal_ull(m, "/", qlim);
289 /* render them all */
290 render_sigset_t(m, "\nSigPnd:\t", &pending);
291 render_sigset_t(m, "ShdPnd:\t", &shpending);
292 render_sigset_t(m, "SigBlk:\t", &blocked);
293 render_sigset_t(m, "SigIgn:\t", &ignored);
294 render_sigset_t(m, "SigCgt:\t", &caught);
297 static void render_cap_t(struct seq_file *m, const char *header,
303 CAP_FOR_EACH_U32(__capi) {
304 seq_put_hex_ll(m, NULL,
305 a->cap[CAP_LAST_U32 - __capi], 8);
310 static inline void task_cap(struct seq_file *m, struct task_struct *p)
312 const struct cred *cred;
313 kernel_cap_t cap_inheritable, cap_permitted, cap_effective,
314 cap_bset, cap_ambient;
317 cred = __task_cred(p);
318 cap_inheritable = cred->cap_inheritable;
319 cap_permitted = cred->cap_permitted;
320 cap_effective = cred->cap_effective;
321 cap_bset = cred->cap_bset;
322 cap_ambient = cred->cap_ambient;
325 render_cap_t(m, "CapInh:\t", &cap_inheritable);
326 render_cap_t(m, "CapPrm:\t", &cap_permitted);
327 render_cap_t(m, "CapEff:\t", &cap_effective);
328 render_cap_t(m, "CapBnd:\t", &cap_bset);
329 render_cap_t(m, "CapAmb:\t", &cap_ambient);
332 static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
334 seq_put_decimal_ull(m, "NoNewPrivs:\t", task_no_new_privs(p));
335 #ifdef CONFIG_SECCOMP
336 seq_put_decimal_ull(m, "\nSeccomp:\t", p->seccomp.mode);
341 static inline void task_context_switch_counts(struct seq_file *m,
342 struct task_struct *p)
344 seq_put_decimal_ull(m, "voluntary_ctxt_switches:\t", p->nvcsw);
345 seq_put_decimal_ull(m, "\nnonvoluntary_ctxt_switches:\t", p->nivcsw);
349 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
351 seq_printf(m, "Cpus_allowed:\t%*pb\n",
352 cpumask_pr_args(&task->cpus_allowed));
353 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n",
354 cpumask_pr_args(&task->cpus_allowed));
357 static inline void task_core_dumping(struct seq_file *m, struct mm_struct *mm)
359 seq_put_decimal_ull(m, "CoreDumping:\t", !!mm->core_state);
363 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
364 struct pid *pid, struct task_struct *task)
366 struct mm_struct *mm = get_task_mm(task);
369 task_state(m, ns, pid, task);
373 task_core_dumping(m, mm);
378 task_seccomp(m, task);
379 task_cpus_allowed(m, task);
380 cpuset_task_status_allowed(m, task);
381 task_context_switch_counts(m, task);
385 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
386 struct pid *pid, struct task_struct *task, int whole)
388 unsigned long vsize, eip, esp, wchan = 0;
390 int tty_pgrp = -1, tty_nr = 0;
391 sigset_t sigign, sigcatch;
393 pid_t ppid = 0, pgid = -1, sid = -1;
396 struct mm_struct *mm;
397 unsigned long long start_time;
398 unsigned long cmin_flt = 0, cmaj_flt = 0;
399 unsigned long min_flt = 0, maj_flt = 0;
400 u64 cutime, cstime, utime, stime;
402 unsigned long rsslim = 0;
403 char tcomm[sizeof(task->comm)];
406 state = *get_task_state(task);
407 vsize = eip = esp = 0;
408 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT);
409 mm = get_task_mm(task);
411 vsize = task_vsize(mm);
413 * esp and eip are intentionally zeroed out. There is no
414 * non-racy way to read them without freezing the task.
415 * Programs that need reliable values can use ptrace(2).
417 * The only exception is if the task is core dumping because
418 * a program is not able to use ptrace(2) in that case. It is
419 * safe because the task has stopped executing permanently.
421 if (permitted && (task->flags & PF_DUMPCORE)) {
422 if (try_get_task_stack(task)) {
423 eip = KSTK_EIP(task);
424 esp = KSTK_ESP(task);
425 put_task_stack(task);
430 get_task_comm(tcomm, task);
432 sigemptyset(&sigign);
433 sigemptyset(&sigcatch);
434 cutime = cstime = utime = stime = 0;
437 if (lock_task_sighand(task, &flags)) {
438 struct signal_struct *sig = task->signal;
441 struct pid *pgrp = tty_get_pgrp(sig->tty);
442 tty_pgrp = pid_nr_ns(pgrp, ns);
444 tty_nr = new_encode_dev(tty_devnum(sig->tty));
447 num_threads = get_nr_threads(task);
448 collect_sigign_sigcatch(task, &sigign, &sigcatch);
450 cmin_flt = sig->cmin_flt;
451 cmaj_flt = sig->cmaj_flt;
452 cutime = sig->cutime;
453 cstime = sig->cstime;
454 cgtime = sig->cgtime;
455 rsslim = READ_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur);
457 /* add up live thread stats at the group level */
459 struct task_struct *t = task;
461 min_flt += t->min_flt;
462 maj_flt += t->maj_flt;
463 gtime += task_gtime(t);
464 } while_each_thread(task, t);
466 min_flt += sig->min_flt;
467 maj_flt += sig->maj_flt;
468 thread_group_cputime_adjusted(task, &utime, &stime);
472 sid = task_session_nr_ns(task, ns);
473 ppid = task_tgid_nr_ns(task->real_parent, ns);
474 pgid = task_pgrp_nr_ns(task, ns);
476 unlock_task_sighand(task, &flags);
479 if (permitted && (!whole || num_threads < 2))
480 wchan = get_wchan(task);
482 min_flt = task->min_flt;
483 maj_flt = task->maj_flt;
484 task_cputime_adjusted(task, &utime, &stime);
485 gtime = task_gtime(task);
488 /* scale priority and nice values from timeslices to -20..20 */
489 /* to make it look like a "normal" Unix priority/nice value */
490 priority = task_prio(task);
491 nice = task_nice(task);
493 /* convert nsec -> ticks */
494 start_time = nsec_to_clock_t(task->real_start_time);
496 seq_put_decimal_ull(m, "", pid_nr_ns(pid, ns));
501 seq_put_decimal_ll(m, " ", ppid);
502 seq_put_decimal_ll(m, " ", pgid);
503 seq_put_decimal_ll(m, " ", sid);
504 seq_put_decimal_ll(m, " ", tty_nr);
505 seq_put_decimal_ll(m, " ", tty_pgrp);
506 seq_put_decimal_ull(m, " ", task->flags);
507 seq_put_decimal_ull(m, " ", min_flt);
508 seq_put_decimal_ull(m, " ", cmin_flt);
509 seq_put_decimal_ull(m, " ", maj_flt);
510 seq_put_decimal_ull(m, " ", cmaj_flt);
511 seq_put_decimal_ull(m, " ", nsec_to_clock_t(utime));
512 seq_put_decimal_ull(m, " ", nsec_to_clock_t(stime));
513 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cutime));
514 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cstime));
515 seq_put_decimal_ll(m, " ", priority);
516 seq_put_decimal_ll(m, " ", nice);
517 seq_put_decimal_ll(m, " ", num_threads);
518 seq_put_decimal_ull(m, " ", 0);
519 seq_put_decimal_ull(m, " ", start_time);
520 seq_put_decimal_ull(m, " ", vsize);
521 seq_put_decimal_ull(m, " ", mm ? get_mm_rss(mm) : 0);
522 seq_put_decimal_ull(m, " ", rsslim);
523 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->start_code : 1) : 0);
524 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->end_code : 1) : 0);
525 seq_put_decimal_ull(m, " ", (permitted && mm) ? mm->start_stack : 0);
526 seq_put_decimal_ull(m, " ", esp);
527 seq_put_decimal_ull(m, " ", eip);
528 /* The signal information here is obsolete.
529 * It must be decimal for Linux 2.0 compatibility.
530 * Use /proc/#/status for real-time signals.
532 seq_put_decimal_ull(m, " ", task->pending.signal.sig[0] & 0x7fffffffUL);
533 seq_put_decimal_ull(m, " ", task->blocked.sig[0] & 0x7fffffffUL);
534 seq_put_decimal_ull(m, " ", sigign.sig[0] & 0x7fffffffUL);
535 seq_put_decimal_ull(m, " ", sigcatch.sig[0] & 0x7fffffffUL);
538 * We used to output the absolute kernel address, but that's an
539 * information leak - so instead we show a 0/1 flag here, to signal
540 * to user-space whether there's a wchan field in /proc/PID/wchan.
542 * This works with older implementations of procps as well.
549 seq_put_decimal_ull(m, " ", 0);
550 seq_put_decimal_ull(m, " ", 0);
551 seq_put_decimal_ll(m, " ", task->exit_signal);
552 seq_put_decimal_ll(m, " ", task_cpu(task));
553 seq_put_decimal_ull(m, " ", task->rt_priority);
554 seq_put_decimal_ull(m, " ", task->policy);
555 seq_put_decimal_ull(m, " ", delayacct_blkio_ticks(task));
556 seq_put_decimal_ull(m, " ", nsec_to_clock_t(gtime));
557 seq_put_decimal_ll(m, " ", nsec_to_clock_t(cgtime));
559 if (mm && permitted) {
560 seq_put_decimal_ull(m, " ", mm->start_data);
561 seq_put_decimal_ull(m, " ", mm->end_data);
562 seq_put_decimal_ull(m, " ", mm->start_brk);
563 seq_put_decimal_ull(m, " ", mm->arg_start);
564 seq_put_decimal_ull(m, " ", mm->arg_end);
565 seq_put_decimal_ull(m, " ", mm->env_start);
566 seq_put_decimal_ull(m, " ", mm->env_end);
568 seq_puts(m, " 0 0 0 0 0 0 0");
571 seq_put_decimal_ll(m, " ", task->exit_code);
581 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns,
582 struct pid *pid, struct task_struct *task)
584 return do_task_stat(m, ns, pid, task, 0);
587 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns,
588 struct pid *pid, struct task_struct *task)
590 return do_task_stat(m, ns, pid, task, 1);
593 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns,
594 struct pid *pid, struct task_struct *task)
596 unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0;
597 struct mm_struct *mm = get_task_mm(task);
600 size = task_statm(mm, &shared, &text, &data, &resident);
604 * For quick read, open code by putting numbers directly
606 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
607 * size, resident, shared, text, data);
609 seq_put_decimal_ull(m, "", size);
610 seq_put_decimal_ull(m, " ", resident);
611 seq_put_decimal_ull(m, " ", shared);
612 seq_put_decimal_ull(m, " ", text);
613 seq_put_decimal_ull(m, " ", 0);
614 seq_put_decimal_ull(m, " ", data);
615 seq_put_decimal_ull(m, " ", 0);
621 #ifdef CONFIG_PROC_CHILDREN
623 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos)
625 struct task_struct *start, *task;
626 struct pid *pid = NULL;
628 read_lock(&tasklist_lock);
630 start = pid_task(proc_pid(inode), PIDTYPE_PID);
635 * Lets try to continue searching first, this gives
636 * us significant speedup on children-rich processes.
639 task = pid_task(pid_prev, PIDTYPE_PID);
640 if (task && task->real_parent == start &&
641 !(list_empty(&task->sibling))) {
642 if (list_is_last(&task->sibling, &start->children))
644 task = list_first_entry(&task->sibling,
645 struct task_struct, sibling);
646 pid = get_pid(task_pid(task));
654 * We might miss some children here if children
655 * are exited while we were not holding the lock,
656 * but it was never promised to be accurate that
659 * "Just suppose that the parent sleeps, but N children
660 * exit after we printed their tids. Now the slow paths
661 * skips N extra children, we miss N tasks." (c)
663 * So one need to stop or freeze the leader and all
664 * its children to get a precise result.
666 list_for_each_entry(task, &start->children, sibling) {
668 pid = get_pid(task_pid(task));
674 read_unlock(&tasklist_lock);
678 static int children_seq_show(struct seq_file *seq, void *v)
680 struct inode *inode = seq->private;
683 pid = pid_nr_ns(v, inode->i_sb->s_fs_info);
684 seq_printf(seq, "%d ", pid);
689 static void *children_seq_start(struct seq_file *seq, loff_t *pos)
691 return get_children_pid(seq->private, NULL, *pos);
694 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos)
698 pid = get_children_pid(seq->private, v, *pos + 1);
705 static void children_seq_stop(struct seq_file *seq, void *v)
710 static const struct seq_operations children_seq_ops = {
711 .start = children_seq_start,
712 .next = children_seq_next,
713 .stop = children_seq_stop,
714 .show = children_seq_show,
717 static int children_seq_open(struct inode *inode, struct file *file)
722 ret = seq_open(file, &children_seq_ops);
726 m = file->private_data;
732 const struct file_operations proc_tid_children_operations = {
733 .open = children_seq_open,
736 .release = seq_release,
738 #endif /* CONFIG_PROC_CHILDREN */