Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51#include <linux/uaccess.h>
52
53#include <linux/errno.h>
54#include <linux/time.h>
55#include <linux/proc_fs.h>
56#include <linux/stat.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/init.h>
59#include <linux/capability.h>
60#include <linux/file.h>
61#include <linux/fdtable.h>
62#include <linux/generic-radix-tree.h>
63#include <linux/string.h>
64#include <linux/seq_file.h>
65#include <linux/namei.h>
66#include <linux/mnt_namespace.h>
67#include <linux/mm.h>
68#include <linux/swap.h>
69#include <linux/rcupdate.h>
70#include <linux/stacktrace.h>
71#include <linux/resource.h>
72#include <linux/module.h>
73#include <linux/mount.h>
74#include <linux/security.h>
75#include <linux/ptrace.h>
76#include <linux/tracehook.h>
77#include <linux/printk.h>
78#include <linux/cache.h>
79#include <linux/cgroup.h>
80#include <linux/cpuset.h>
81#include <linux/audit.h>
82#include <linux/poll.h>
83#include <linux/nsproxy.h>
84#include <linux/oom.h>
85#include <linux/elf.h>
86#include <linux/pid_namespace.h>
87#include <linux/user_namespace.h>
88#include <linux/fs_struct.h>
89#include <linux/slab.h>
90#include <linux/sched/autogroup.h>
91#include <linux/sched/mm.h>
92#include <linux/sched/coredump.h>
93#include <linux/sched/debug.h>
94#include <linux/sched/stat.h>
95#include <linux/posix-timers.h>
96#include <linux/time_namespace.h>
97#include <linux/resctrl.h>
98#include <trace/events/oom.h>
99#include "internal.h"
100#include "fd.h"
101
102#include "../../lib/kstrtox.h"
103
104/* NOTE:
105 * Implementing inode permission operations in /proc is almost
106 * certainly an error. Permission checks need to happen during
107 * each system call not at open time. The reason is that most of
108 * what we wish to check for permissions in /proc varies at runtime.
109 *
110 * The classic example of a problem is opening file descriptors
111 * in /proc for a task before it execs a suid executable.
112 */
113
114static u8 nlink_tid __ro_after_init;
115static u8 nlink_tgid __ro_after_init;
116
117struct pid_entry {
118 const char *name;
119 unsigned int len;
120 umode_t mode;
121 const struct inode_operations *iop;
122 const struct file_operations *fop;
123 union proc_op op;
124};
125
126#define NOD(NAME, MODE, IOP, FOP, OP) { \
127 .name = (NAME), \
128 .len = sizeof(NAME) - 1, \
129 .mode = MODE, \
130 .iop = IOP, \
131 .fop = FOP, \
132 .op = OP, \
133}
134
135#define DIR(NAME, MODE, iops, fops) \
136 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
137#define LNK(NAME, get_link) \
138 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
139 &proc_pid_link_inode_operations, NULL, \
140 { .proc_get_link = get_link } )
141#define REG(NAME, MODE, fops) \
142 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
143#define ONE(NAME, MODE, show) \
144 NOD(NAME, (S_IFREG|(MODE)), \
145 NULL, &proc_single_file_operations, \
146 { .proc_show = show } )
147#define ATTR(LSM, NAME, MODE) \
148 NOD(NAME, (S_IFREG|(MODE)), \
149 NULL, &proc_pid_attr_operations, \
150 { .lsm = LSM })
151
152/*
153 * Count the number of hardlinks for the pid_entry table, excluding the .
154 * and .. links.
155 */
156static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
157 unsigned int n)
158{
159 unsigned int i;
160 unsigned int count;
161
162 count = 2;
163 for (i = 0; i < n; ++i) {
164 if (S_ISDIR(entries[i].mode))
165 ++count;
166 }
167
168 return count;
169}
170
171static int get_task_root(struct task_struct *task, struct path *root)
172{
173 int result = -ENOENT;
174
175 task_lock(task);
176 if (task->fs) {
177 get_fs_root(task->fs, root);
178 result = 0;
179 }
180 task_unlock(task);
181 return result;
182}
183
184static int proc_cwd_link(struct dentry *dentry, struct path *path)
185{
186 struct task_struct *task = get_proc_task(d_inode(dentry));
187 int result = -ENOENT;
188
189 if (task) {
190 task_lock(task);
191 if (task->fs) {
192 get_fs_pwd(task->fs, path);
193 result = 0;
194 }
195 task_unlock(task);
196 put_task_struct(task);
197 }
198 return result;
199}
200
201static int proc_root_link(struct dentry *dentry, struct path *path)
202{
203 struct task_struct *task = get_proc_task(d_inode(dentry));
204 int result = -ENOENT;
205
206 if (task) {
207 result = get_task_root(task, path);
208 put_task_struct(task);
209 }
210 return result;
211}
212
213/*
214 * If the user used setproctitle(), we just get the string from
215 * user space at arg_start, and limit it to a maximum of one page.
216 */
217static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
218 size_t count, unsigned long pos,
219 unsigned long arg_start)
220{
221 char *page;
222 int ret, got;
223
224 if (pos >= PAGE_SIZE)
225 return 0;
226
227 page = (char *)__get_free_page(GFP_KERNEL);
228 if (!page)
229 return -ENOMEM;
230
231 ret = 0;
232 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
233 if (got > 0) {
234 int len = strnlen(page, got);
235
236 /* Include the NUL character if it was found */
237 if (len < got)
238 len++;
239
240 if (len > pos) {
241 len -= pos;
242 if (len > count)
243 len = count;
244 len -= copy_to_user(buf, page+pos, len);
245 if (!len)
246 len = -EFAULT;
247 ret = len;
248 }
249 }
250 free_page((unsigned long)page);
251 return ret;
252}
253
254static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
255 size_t count, loff_t *ppos)
256{
257 unsigned long arg_start, arg_end, env_start, env_end;
258 unsigned long pos, len;
259 char *page, c;
260
261 /* Check if process spawned far enough to have cmdline. */
262 if (!mm->env_end)
263 return 0;
264
265 spin_lock(&mm->arg_lock);
266 arg_start = mm->arg_start;
267 arg_end = mm->arg_end;
268 env_start = mm->env_start;
269 env_end = mm->env_end;
270 spin_unlock(&mm->arg_lock);
271
272 if (arg_start >= arg_end)
273 return 0;
274
275 /*
276 * We allow setproctitle() to overwrite the argument
277 * strings, and overflow past the original end. But
278 * only when it overflows into the environment area.
279 */
280 if (env_start != arg_end || env_end < env_start)
281 env_start = env_end = arg_end;
282 len = env_end - arg_start;
283
284 /* We're not going to care if "*ppos" has high bits set */
285 pos = *ppos;
286 if (pos >= len)
287 return 0;
288 if (count > len - pos)
289 count = len - pos;
290 if (!count)
291 return 0;
292
293 /*
294 * Magical special case: if the argv[] end byte is not
295 * zero, the user has overwritten it with setproctitle(3).
296 *
297 * Possible future enhancement: do this only once when
298 * pos is 0, and set a flag in the 'struct file'.
299 */
300 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
301 return get_mm_proctitle(mm, buf, count, pos, arg_start);
302
303 /*
304 * For the non-setproctitle() case we limit things strictly
305 * to the [arg_start, arg_end[ range.
306 */
307 pos += arg_start;
308 if (pos < arg_start || pos >= arg_end)
309 return 0;
310 if (count > arg_end - pos)
311 count = arg_end - pos;
312
313 page = (char *)__get_free_page(GFP_KERNEL);
314 if (!page)
315 return -ENOMEM;
316
317 len = 0;
318 while (count) {
319 int got;
320 size_t size = min_t(size_t, PAGE_SIZE, count);
321
322 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
323 if (got <= 0)
324 break;
325 got -= copy_to_user(buf, page, got);
326 if (unlikely(!got)) {
327 if (!len)
328 len = -EFAULT;
329 break;
330 }
331 pos += got;
332 buf += got;
333 len += got;
334 count -= got;
335 }
336
337 free_page((unsigned long)page);
338 return len;
339}
340
341static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
342 size_t count, loff_t *pos)
343{
344 struct mm_struct *mm;
345 ssize_t ret;
346
347 mm = get_task_mm(tsk);
348 if (!mm)
349 return 0;
350
351 ret = get_mm_cmdline(mm, buf, count, pos);
352 mmput(mm);
353 return ret;
354}
355
356static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
357 size_t count, loff_t *pos)
358{
359 struct task_struct *tsk;
360 ssize_t ret;
361
362 BUG_ON(*pos < 0);
363
364 tsk = get_proc_task(file_inode(file));
365 if (!tsk)
366 return -ESRCH;
367 ret = get_task_cmdline(tsk, buf, count, pos);
368 put_task_struct(tsk);
369 if (ret > 0)
370 *pos += ret;
371 return ret;
372}
373
374static const struct file_operations proc_pid_cmdline_ops = {
375 .read = proc_pid_cmdline_read,
376 .llseek = generic_file_llseek,
377};
378
379#ifdef CONFIG_KALLSYMS
380/*
381 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
382 * Returns the resolved symbol. If that fails, simply return the address.
383 */
384static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
385 struct pid *pid, struct task_struct *task)
386{
387 unsigned long wchan;
388
389 if (ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
390 wchan = get_wchan(task);
391 else
392 wchan = 0;
393
394 if (wchan)
395 seq_printf(m, "%ps", (void *) wchan);
396 else
397 seq_putc(m, '0');
398
399 return 0;
400}
401#endif /* CONFIG_KALLSYMS */
402
403static int lock_trace(struct task_struct *task)
404{
405 int err = down_read_killable(&task->signal->exec_update_lock);
406 if (err)
407 return err;
408 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
409 up_read(&task->signal->exec_update_lock);
410 return -EPERM;
411 }
412 return 0;
413}
414
415static void unlock_trace(struct task_struct *task)
416{
417 up_read(&task->signal->exec_update_lock);
418}
419
420#ifdef CONFIG_STACKTRACE
421
422#define MAX_STACK_TRACE_DEPTH 64
423
424static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
425 struct pid *pid, struct task_struct *task)
426{
427 unsigned long *entries;
428 int err;
429
430 /*
431 * The ability to racily run the kernel stack unwinder on a running task
432 * and then observe the unwinder output is scary; while it is useful for
433 * debugging kernel issues, it can also allow an attacker to leak kernel
434 * stack contents.
435 * Doing this in a manner that is at least safe from races would require
436 * some work to ensure that the remote task can not be scheduled; and
437 * even then, this would still expose the unwinder as local attack
438 * surface.
439 * Therefore, this interface is restricted to root.
440 */
441 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
442 return -EACCES;
443
444 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
445 GFP_KERNEL);
446 if (!entries)
447 return -ENOMEM;
448
449 err = lock_trace(task);
450 if (!err) {
451 unsigned int i, nr_entries;
452
453 nr_entries = stack_trace_save_tsk(task, entries,
454 MAX_STACK_TRACE_DEPTH, 0);
455
456 for (i = 0; i < nr_entries; i++) {
457 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
458 }
459
460 unlock_trace(task);
461 }
462 kfree(entries);
463
464 return err;
465}
466#endif
467
468#ifdef CONFIG_SCHED_INFO
469/*
470 * Provides /proc/PID/schedstat
471 */
472static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
473 struct pid *pid, struct task_struct *task)
474{
475 if (unlikely(!sched_info_on()))
476 seq_puts(m, "0 0 0\n");
477 else
478 seq_printf(m, "%llu %llu %lu\n",
479 (unsigned long long)task->se.sum_exec_runtime,
480 (unsigned long long)task->sched_info.run_delay,
481 task->sched_info.pcount);
482
483 return 0;
484}
485#endif
486
487#ifdef CONFIG_LATENCYTOP
488static int lstats_show_proc(struct seq_file *m, void *v)
489{
490 int i;
491 struct inode *inode = m->private;
492 struct task_struct *task = get_proc_task(inode);
493
494 if (!task)
495 return -ESRCH;
496 seq_puts(m, "Latency Top version : v0.1\n");
497 for (i = 0; i < LT_SAVECOUNT; i++) {
498 struct latency_record *lr = &task->latency_record[i];
499 if (lr->backtrace[0]) {
500 int q;
501 seq_printf(m, "%i %li %li",
502 lr->count, lr->time, lr->max);
503 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
504 unsigned long bt = lr->backtrace[q];
505
506 if (!bt)
507 break;
508 seq_printf(m, " %ps", (void *)bt);
509 }
510 seq_putc(m, '\n');
511 }
512
513 }
514 put_task_struct(task);
515 return 0;
516}
517
518static int lstats_open(struct inode *inode, struct file *file)
519{
520 return single_open(file, lstats_show_proc, inode);
521}
522
523static ssize_t lstats_write(struct file *file, const char __user *buf,
524 size_t count, loff_t *offs)
525{
526 struct task_struct *task = get_proc_task(file_inode(file));
527
528 if (!task)
529 return -ESRCH;
530 clear_tsk_latency_tracing(task);
531 put_task_struct(task);
532
533 return count;
534}
535
536static const struct file_operations proc_lstats_operations = {
537 .open = lstats_open,
538 .read = seq_read,
539 .write = lstats_write,
540 .llseek = seq_lseek,
541 .release = single_release,
542};
543
544#endif
545
546static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
547 struct pid *pid, struct task_struct *task)
548{
549 unsigned long totalpages = totalram_pages() + total_swap_pages;
550 unsigned long points = 0;
551 long badness;
552
553 badness = oom_badness(task, totalpages);
554 /*
555 * Special case OOM_SCORE_ADJ_MIN for all others scale the
556 * badness value into [0, 2000] range which we have been
557 * exporting for a long time so userspace might depend on it.
558 */
559 if (badness != LONG_MIN)
560 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
561
562 seq_printf(m, "%lu\n", points);
563
564 return 0;
565}
566
567struct limit_names {
568 const char *name;
569 const char *unit;
570};
571
572static const struct limit_names lnames[RLIM_NLIMITS] = {
573 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
574 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
575 [RLIMIT_DATA] = {"Max data size", "bytes"},
576 [RLIMIT_STACK] = {"Max stack size", "bytes"},
577 [RLIMIT_CORE] = {"Max core file size", "bytes"},
578 [RLIMIT_RSS] = {"Max resident set", "bytes"},
579 [RLIMIT_NPROC] = {"Max processes", "processes"},
580 [RLIMIT_NOFILE] = {"Max open files", "files"},
581 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
582 [RLIMIT_AS] = {"Max address space", "bytes"},
583 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
584 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
585 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
586 [RLIMIT_NICE] = {"Max nice priority", NULL},
587 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
588 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
589};
590
591/* Display limits for a process */
592static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
593 struct pid *pid, struct task_struct *task)
594{
595 unsigned int i;
596 unsigned long flags;
597
598 struct rlimit rlim[RLIM_NLIMITS];
599
600 if (!lock_task_sighand(task, &flags))
601 return 0;
602 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
603 unlock_task_sighand(task, &flags);
604
605 /*
606 * print the file header
607 */
608 seq_puts(m, "Limit "
609 "Soft Limit "
610 "Hard Limit "
611 "Units \n");
612
613 for (i = 0; i < RLIM_NLIMITS; i++) {
614 if (rlim[i].rlim_cur == RLIM_INFINITY)
615 seq_printf(m, "%-25s %-20s ",
616 lnames[i].name, "unlimited");
617 else
618 seq_printf(m, "%-25s %-20lu ",
619 lnames[i].name, rlim[i].rlim_cur);
620
621 if (rlim[i].rlim_max == RLIM_INFINITY)
622 seq_printf(m, "%-20s ", "unlimited");
623 else
624 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
625
626 if (lnames[i].unit)
627 seq_printf(m, "%-10s\n", lnames[i].unit);
628 else
629 seq_putc(m, '\n');
630 }
631
632 return 0;
633}
634
635#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
636static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
637 struct pid *pid, struct task_struct *task)
638{
639 struct syscall_info info;
640 u64 *args = &info.data.args[0];
641 int res;
642
643 res = lock_trace(task);
644 if (res)
645 return res;
646
647 if (task_current_syscall(task, &info))
648 seq_puts(m, "running\n");
649 else if (info.data.nr < 0)
650 seq_printf(m, "%d 0x%llx 0x%llx\n",
651 info.data.nr, info.sp, info.data.instruction_pointer);
652 else
653 seq_printf(m,
654 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
655 info.data.nr,
656 args[0], args[1], args[2], args[3], args[4], args[5],
657 info.sp, info.data.instruction_pointer);
658 unlock_trace(task);
659
660 return 0;
661}
662#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
663
664/************************************************************************/
665/* Here the fs part begins */
666/************************************************************************/
667
668/* permission checks */
669static int proc_fd_access_allowed(struct inode *inode)
670{
671 struct task_struct *task;
672 int allowed = 0;
673 /* Allow access to a task's file descriptors if it is us or we
674 * may use ptrace attach to the process and find out that
675 * information.
676 */
677 task = get_proc_task(inode);
678 if (task) {
679 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
680 put_task_struct(task);
681 }
682 return allowed;
683}
684
685int proc_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
686 struct iattr *attr)
687{
688 int error;
689 struct inode *inode = d_inode(dentry);
690
691 if (attr->ia_valid & ATTR_MODE)
692 return -EPERM;
693
694 error = setattr_prepare(&init_user_ns, dentry, attr);
695 if (error)
696 return error;
697
698 setattr_copy(&init_user_ns, inode, attr);
699 mark_inode_dirty(inode);
700 return 0;
701}
702
703/*
704 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
705 * or euid/egid (for hide_pid_min=2)?
706 */
707static bool has_pid_permissions(struct proc_fs_info *fs_info,
708 struct task_struct *task,
709 enum proc_hidepid hide_pid_min)
710{
711 /*
712 * If 'hidpid' mount option is set force a ptrace check,
713 * we indicate that we are using a filesystem syscall
714 * by passing PTRACE_MODE_READ_FSCREDS
715 */
716 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
717 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
718
719 if (fs_info->hide_pid < hide_pid_min)
720 return true;
721 if (in_group_p(fs_info->pid_gid))
722 return true;
723 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
724}
725
726
727static int proc_pid_permission(struct user_namespace *mnt_userns,
728 struct inode *inode, int mask)
729{
730 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
731 struct task_struct *task;
732 bool has_perms;
733
734 task = get_proc_task(inode);
735 if (!task)
736 return -ESRCH;
737 has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
738 put_task_struct(task);
739
740 if (!has_perms) {
741 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
742 /*
743 * Let's make getdents(), stat(), and open()
744 * consistent with each other. If a process
745 * may not stat() a file, it shouldn't be seen
746 * in procfs at all.
747 */
748 return -ENOENT;
749 }
750
751 return -EPERM;
752 }
753 return generic_permission(&init_user_ns, inode, mask);
754}
755
756
757
758static const struct inode_operations proc_def_inode_operations = {
759 .setattr = proc_setattr,
760};
761
762static int proc_single_show(struct seq_file *m, void *v)
763{
764 struct inode *inode = m->private;
765 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
766 struct pid *pid = proc_pid(inode);
767 struct task_struct *task;
768 int ret;
769
770 task = get_pid_task(pid, PIDTYPE_PID);
771 if (!task)
772 return -ESRCH;
773
774 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
775
776 put_task_struct(task);
777 return ret;
778}
779
780static int proc_single_open(struct inode *inode, struct file *filp)
781{
782 return single_open(filp, proc_single_show, inode);
783}
784
785static const struct file_operations proc_single_file_operations = {
786 .open = proc_single_open,
787 .read = seq_read,
788 .llseek = seq_lseek,
789 .release = single_release,
790};
791
792
793struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
794{
795 struct task_struct *task = get_proc_task(inode);
796 struct mm_struct *mm = ERR_PTR(-ESRCH);
797
798 if (task) {
799 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
800 put_task_struct(task);
801
802 if (!IS_ERR_OR_NULL(mm)) {
803 /* ensure this mm_struct can't be freed */
804 mmgrab(mm);
805 /* but do not pin its memory */
806 mmput(mm);
807 }
808 }
809
810 return mm;
811}
812
813static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
814{
815 struct mm_struct *mm = proc_mem_open(inode, mode);
816
817 if (IS_ERR(mm))
818 return PTR_ERR(mm);
819
820 file->private_data = mm;
821 return 0;
822}
823
824static int mem_open(struct inode *inode, struct file *file)
825{
826 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
827
828 /* OK to pass negative loff_t, we can catch out-of-range */
829 file->f_mode |= FMODE_UNSIGNED_OFFSET;
830
831 return ret;
832}
833
834static ssize_t mem_rw(struct file *file, char __user *buf,
835 size_t count, loff_t *ppos, int write)
836{
837 struct mm_struct *mm = file->private_data;
838 unsigned long addr = *ppos;
839 ssize_t copied;
840 char *page;
841 unsigned int flags;
842
843 if (!mm)
844 return 0;
845
846 page = (char *)__get_free_page(GFP_KERNEL);
847 if (!page)
848 return -ENOMEM;
849
850 copied = 0;
851 if (!mmget_not_zero(mm))
852 goto free;
853
854 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
855
856 while (count > 0) {
857 size_t this_len = min_t(size_t, count, PAGE_SIZE);
858
859 if (write && copy_from_user(page, buf, this_len)) {
860 copied = -EFAULT;
861 break;
862 }
863
864 this_len = access_remote_vm(mm, addr, page, this_len, flags);
865 if (!this_len) {
866 if (!copied)
867 copied = -EIO;
868 break;
869 }
870
871 if (!write && copy_to_user(buf, page, this_len)) {
872 copied = -EFAULT;
873 break;
874 }
875
876 buf += this_len;
877 addr += this_len;
878 copied += this_len;
879 count -= this_len;
880 }
881 *ppos = addr;
882
883 mmput(mm);
884free:
885 free_page((unsigned long) page);
886 return copied;
887}
888
889static ssize_t mem_read(struct file *file, char __user *buf,
890 size_t count, loff_t *ppos)
891{
892 return mem_rw(file, buf, count, ppos, 0);
893}
894
895static ssize_t mem_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
897{
898 return mem_rw(file, (char __user*)buf, count, ppos, 1);
899}
900
901loff_t mem_lseek(struct file *file, loff_t offset, int orig)
902{
903 switch (orig) {
904 case 0:
905 file->f_pos = offset;
906 break;
907 case 1:
908 file->f_pos += offset;
909 break;
910 default:
911 return -EINVAL;
912 }
913 force_successful_syscall_return();
914 return file->f_pos;
915}
916
917static int mem_release(struct inode *inode, struct file *file)
918{
919 struct mm_struct *mm = file->private_data;
920 if (mm)
921 mmdrop(mm);
922 return 0;
923}
924
925static const struct file_operations proc_mem_operations = {
926 .llseek = mem_lseek,
927 .read = mem_read,
928 .write = mem_write,
929 .open = mem_open,
930 .release = mem_release,
931};
932
933static int environ_open(struct inode *inode, struct file *file)
934{
935 return __mem_open(inode, file, PTRACE_MODE_READ);
936}
937
938static ssize_t environ_read(struct file *file, char __user *buf,
939 size_t count, loff_t *ppos)
940{
941 char *page;
942 unsigned long src = *ppos;
943 int ret = 0;
944 struct mm_struct *mm = file->private_data;
945 unsigned long env_start, env_end;
946
947 /* Ensure the process spawned far enough to have an environment. */
948 if (!mm || !mm->env_end)
949 return 0;
950
951 page = (char *)__get_free_page(GFP_KERNEL);
952 if (!page)
953 return -ENOMEM;
954
955 ret = 0;
956 if (!mmget_not_zero(mm))
957 goto free;
958
959 spin_lock(&mm->arg_lock);
960 env_start = mm->env_start;
961 env_end = mm->env_end;
962 spin_unlock(&mm->arg_lock);
963
964 while (count > 0) {
965 size_t this_len, max_len;
966 int retval;
967
968 if (src >= (env_end - env_start))
969 break;
970
971 this_len = env_end - (env_start + src);
972
973 max_len = min_t(size_t, PAGE_SIZE, count);
974 this_len = min(max_len, this_len);
975
976 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
977
978 if (retval <= 0) {
979 ret = retval;
980 break;
981 }
982
983 if (copy_to_user(buf, page, retval)) {
984 ret = -EFAULT;
985 break;
986 }
987
988 ret += retval;
989 src += retval;
990 buf += retval;
991 count -= retval;
992 }
993 *ppos = src;
994 mmput(mm);
995
996free:
997 free_page((unsigned long) page);
998 return ret;
999}
1000
1001static const struct file_operations proc_environ_operations = {
1002 .open = environ_open,
1003 .read = environ_read,
1004 .llseek = generic_file_llseek,
1005 .release = mem_release,
1006};
1007
1008static int auxv_open(struct inode *inode, struct file *file)
1009{
1010 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1011}
1012
1013static ssize_t auxv_read(struct file *file, char __user *buf,
1014 size_t count, loff_t *ppos)
1015{
1016 struct mm_struct *mm = file->private_data;
1017 unsigned int nwords = 0;
1018
1019 if (!mm)
1020 return 0;
1021 do {
1022 nwords += 2;
1023 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1024 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1025 nwords * sizeof(mm->saved_auxv[0]));
1026}
1027
1028static const struct file_operations proc_auxv_operations = {
1029 .open = auxv_open,
1030 .read = auxv_read,
1031 .llseek = generic_file_llseek,
1032 .release = mem_release,
1033};
1034
1035static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1036 loff_t *ppos)
1037{
1038 struct task_struct *task = get_proc_task(file_inode(file));
1039 char buffer[PROC_NUMBUF];
1040 int oom_adj = OOM_ADJUST_MIN;
1041 size_t len;
1042
1043 if (!task)
1044 return -ESRCH;
1045 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1046 oom_adj = OOM_ADJUST_MAX;
1047 else
1048 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1049 OOM_SCORE_ADJ_MAX;
1050 put_task_struct(task);
1051 if (oom_adj > OOM_ADJUST_MAX)
1052 oom_adj = OOM_ADJUST_MAX;
1053 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1054 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1055}
1056
1057static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1058{
1059 struct mm_struct *mm = NULL;
1060 struct task_struct *task;
1061 int err = 0;
1062
1063 task = get_proc_task(file_inode(file));
1064 if (!task)
1065 return -ESRCH;
1066
1067 mutex_lock(&oom_adj_mutex);
1068 if (legacy) {
1069 if (oom_adj < task->signal->oom_score_adj &&
1070 !capable(CAP_SYS_RESOURCE)) {
1071 err = -EACCES;
1072 goto err_unlock;
1073 }
1074 /*
1075 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1076 * /proc/pid/oom_score_adj instead.
1077 */
1078 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1079 current->comm, task_pid_nr(current), task_pid_nr(task),
1080 task_pid_nr(task));
1081 } else {
1082 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1083 !capable(CAP_SYS_RESOURCE)) {
1084 err = -EACCES;
1085 goto err_unlock;
1086 }
1087 }
1088
1089 /*
1090 * Make sure we will check other processes sharing the mm if this is
1091 * not vfrok which wants its own oom_score_adj.
1092 * pin the mm so it doesn't go away and get reused after task_unlock
1093 */
1094 if (!task->vfork_done) {
1095 struct task_struct *p = find_lock_task_mm(task);
1096
1097 if (p) {
1098 if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1099 mm = p->mm;
1100 mmgrab(mm);
1101 }
1102 task_unlock(p);
1103 }
1104 }
1105
1106 task->signal->oom_score_adj = oom_adj;
1107 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1108 task->signal->oom_score_adj_min = (short)oom_adj;
1109 trace_oom_score_adj_update(task);
1110
1111 if (mm) {
1112 struct task_struct *p;
1113
1114 rcu_read_lock();
1115 for_each_process(p) {
1116 if (same_thread_group(task, p))
1117 continue;
1118
1119 /* do not touch kernel threads or the global init */
1120 if (p->flags & PF_KTHREAD || is_global_init(p))
1121 continue;
1122
1123 task_lock(p);
1124 if (!p->vfork_done && process_shares_mm(p, mm)) {
1125 p->signal->oom_score_adj = oom_adj;
1126 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1127 p->signal->oom_score_adj_min = (short)oom_adj;
1128 }
1129 task_unlock(p);
1130 }
1131 rcu_read_unlock();
1132 mmdrop(mm);
1133 }
1134err_unlock:
1135 mutex_unlock(&oom_adj_mutex);
1136 put_task_struct(task);
1137 return err;
1138}
1139
1140/*
1141 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1142 * kernels. The effective policy is defined by oom_score_adj, which has a
1143 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1144 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1145 * Processes that become oom disabled via oom_adj will still be oom disabled
1146 * with this implementation.
1147 *
1148 * oom_adj cannot be removed since existing userspace binaries use it.
1149 */
1150static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1151 size_t count, loff_t *ppos)
1152{
1153 char buffer[PROC_NUMBUF];
1154 int oom_adj;
1155 int err;
1156
1157 memset(buffer, 0, sizeof(buffer));
1158 if (count > sizeof(buffer) - 1)
1159 count = sizeof(buffer) - 1;
1160 if (copy_from_user(buffer, buf, count)) {
1161 err = -EFAULT;
1162 goto out;
1163 }
1164
1165 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1166 if (err)
1167 goto out;
1168 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1169 oom_adj != OOM_DISABLE) {
1170 err = -EINVAL;
1171 goto out;
1172 }
1173
1174 /*
1175 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1176 * value is always attainable.
1177 */
1178 if (oom_adj == OOM_ADJUST_MAX)
1179 oom_adj = OOM_SCORE_ADJ_MAX;
1180 else
1181 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1182
1183 err = __set_oom_adj(file, oom_adj, true);
1184out:
1185 return err < 0 ? err : count;
1186}
1187
1188static const struct file_operations proc_oom_adj_operations = {
1189 .read = oom_adj_read,
1190 .write = oom_adj_write,
1191 .llseek = generic_file_llseek,
1192};
1193
1194static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1195 size_t count, loff_t *ppos)
1196{
1197 struct task_struct *task = get_proc_task(file_inode(file));
1198 char buffer[PROC_NUMBUF];
1199 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1200 size_t len;
1201
1202 if (!task)
1203 return -ESRCH;
1204 oom_score_adj = task->signal->oom_score_adj;
1205 put_task_struct(task);
1206 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1207 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1208}
1209
1210static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1211 size_t count, loff_t *ppos)
1212{
1213 char buffer[PROC_NUMBUF];
1214 int oom_score_adj;
1215 int err;
1216
1217 memset(buffer, 0, sizeof(buffer));
1218 if (count > sizeof(buffer) - 1)
1219 count = sizeof(buffer) - 1;
1220 if (copy_from_user(buffer, buf, count)) {
1221 err = -EFAULT;
1222 goto out;
1223 }
1224
1225 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1226 if (err)
1227 goto out;
1228 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1229 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1230 err = -EINVAL;
1231 goto out;
1232 }
1233
1234 err = __set_oom_adj(file, oom_score_adj, false);
1235out:
1236 return err < 0 ? err : count;
1237}
1238
1239static const struct file_operations proc_oom_score_adj_operations = {
1240 .read = oom_score_adj_read,
1241 .write = oom_score_adj_write,
1242 .llseek = default_llseek,
1243};
1244
1245#ifdef CONFIG_AUDIT
1246#define TMPBUFLEN 11
1247static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1248 size_t count, loff_t *ppos)
1249{
1250 struct inode * inode = file_inode(file);
1251 struct task_struct *task = get_proc_task(inode);
1252 ssize_t length;
1253 char tmpbuf[TMPBUFLEN];
1254
1255 if (!task)
1256 return -ESRCH;
1257 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1258 from_kuid(file->f_cred->user_ns,
1259 audit_get_loginuid(task)));
1260 put_task_struct(task);
1261 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1262}
1263
1264static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1265 size_t count, loff_t *ppos)
1266{
1267 struct inode * inode = file_inode(file);
1268 uid_t loginuid;
1269 kuid_t kloginuid;
1270 int rv;
1271
1272 /* Don't let kthreads write their own loginuid */
1273 if (current->flags & PF_KTHREAD)
1274 return -EPERM;
1275
1276 rcu_read_lock();
1277 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1278 rcu_read_unlock();
1279 return -EPERM;
1280 }
1281 rcu_read_unlock();
1282
1283 if (*ppos != 0) {
1284 /* No partial writes. */
1285 return -EINVAL;
1286 }
1287
1288 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1289 if (rv < 0)
1290 return rv;
1291
1292 /* is userspace tring to explicitly UNSET the loginuid? */
1293 if (loginuid == AUDIT_UID_UNSET) {
1294 kloginuid = INVALID_UID;
1295 } else {
1296 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1297 if (!uid_valid(kloginuid))
1298 return -EINVAL;
1299 }
1300
1301 rv = audit_set_loginuid(kloginuid);
1302 if (rv < 0)
1303 return rv;
1304 return count;
1305}
1306
1307static const struct file_operations proc_loginuid_operations = {
1308 .read = proc_loginuid_read,
1309 .write = proc_loginuid_write,
1310 .llseek = generic_file_llseek,
1311};
1312
1313static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1314 size_t count, loff_t *ppos)
1315{
1316 struct inode * inode = file_inode(file);
1317 struct task_struct *task = get_proc_task(inode);
1318 ssize_t length;
1319 char tmpbuf[TMPBUFLEN];
1320
1321 if (!task)
1322 return -ESRCH;
1323 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1324 audit_get_sessionid(task));
1325 put_task_struct(task);
1326 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1327}
1328
1329static const struct file_operations proc_sessionid_operations = {
1330 .read = proc_sessionid_read,
1331 .llseek = generic_file_llseek,
1332};
1333#endif
1334
1335#ifdef CONFIG_FAULT_INJECTION
1336static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1337 size_t count, loff_t *ppos)
1338{
1339 struct task_struct *task = get_proc_task(file_inode(file));
1340 char buffer[PROC_NUMBUF];
1341 size_t len;
1342 int make_it_fail;
1343
1344 if (!task)
1345 return -ESRCH;
1346 make_it_fail = task->make_it_fail;
1347 put_task_struct(task);
1348
1349 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1350
1351 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1352}
1353
1354static ssize_t proc_fault_inject_write(struct file * file,
1355 const char __user * buf, size_t count, loff_t *ppos)
1356{
1357 struct task_struct *task;
1358 char buffer[PROC_NUMBUF];
1359 int make_it_fail;
1360 int rv;
1361
1362 if (!capable(CAP_SYS_RESOURCE))
1363 return -EPERM;
1364 memset(buffer, 0, sizeof(buffer));
1365 if (count > sizeof(buffer) - 1)
1366 count = sizeof(buffer) - 1;
1367 if (copy_from_user(buffer, buf, count))
1368 return -EFAULT;
1369 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1370 if (rv < 0)
1371 return rv;
1372 if (make_it_fail < 0 || make_it_fail > 1)
1373 return -EINVAL;
1374
1375 task = get_proc_task(file_inode(file));
1376 if (!task)
1377 return -ESRCH;
1378 task->make_it_fail = make_it_fail;
1379 put_task_struct(task);
1380
1381 return count;
1382}
1383
1384static const struct file_operations proc_fault_inject_operations = {
1385 .read = proc_fault_inject_read,
1386 .write = proc_fault_inject_write,
1387 .llseek = generic_file_llseek,
1388};
1389
1390static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1391 size_t count, loff_t *ppos)
1392{
1393 struct task_struct *task;
1394 int err;
1395 unsigned int n;
1396
1397 err = kstrtouint_from_user(buf, count, 0, &n);
1398 if (err)
1399 return err;
1400
1401 task = get_proc_task(file_inode(file));
1402 if (!task)
1403 return -ESRCH;
1404 task->fail_nth = n;
1405 put_task_struct(task);
1406
1407 return count;
1408}
1409
1410static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1411 size_t count, loff_t *ppos)
1412{
1413 struct task_struct *task;
1414 char numbuf[PROC_NUMBUF];
1415 ssize_t len;
1416
1417 task = get_proc_task(file_inode(file));
1418 if (!task)
1419 return -ESRCH;
1420 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1421 put_task_struct(task);
1422 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1423}
1424
1425static const struct file_operations proc_fail_nth_operations = {
1426 .read = proc_fail_nth_read,
1427 .write = proc_fail_nth_write,
1428};
1429#endif
1430
1431
1432#ifdef CONFIG_SCHED_DEBUG
1433/*
1434 * Print out various scheduling related per-task fields:
1435 */
1436static int sched_show(struct seq_file *m, void *v)
1437{
1438 struct inode *inode = m->private;
1439 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1440 struct task_struct *p;
1441
1442 p = get_proc_task(inode);
1443 if (!p)
1444 return -ESRCH;
1445 proc_sched_show_task(p, ns, m);
1446
1447 put_task_struct(p);
1448
1449 return 0;
1450}
1451
1452static ssize_t
1453sched_write(struct file *file, const char __user *buf,
1454 size_t count, loff_t *offset)
1455{
1456 struct inode *inode = file_inode(file);
1457 struct task_struct *p;
1458
1459 p = get_proc_task(inode);
1460 if (!p)
1461 return -ESRCH;
1462 proc_sched_set_task(p);
1463
1464 put_task_struct(p);
1465
1466 return count;
1467}
1468
1469static int sched_open(struct inode *inode, struct file *filp)
1470{
1471 return single_open(filp, sched_show, inode);
1472}
1473
1474static const struct file_operations proc_pid_sched_operations = {
1475 .open = sched_open,
1476 .read = seq_read,
1477 .write = sched_write,
1478 .llseek = seq_lseek,
1479 .release = single_release,
1480};
1481
1482#endif
1483
1484#ifdef CONFIG_SCHED_AUTOGROUP
1485/*
1486 * Print out autogroup related information:
1487 */
1488static int sched_autogroup_show(struct seq_file *m, void *v)
1489{
1490 struct inode *inode = m->private;
1491 struct task_struct *p;
1492
1493 p = get_proc_task(inode);
1494 if (!p)
1495 return -ESRCH;
1496 proc_sched_autogroup_show_task(p, m);
1497
1498 put_task_struct(p);
1499
1500 return 0;
1501}
1502
1503static ssize_t
1504sched_autogroup_write(struct file *file, const char __user *buf,
1505 size_t count, loff_t *offset)
1506{
1507 struct inode *inode = file_inode(file);
1508 struct task_struct *p;
1509 char buffer[PROC_NUMBUF];
1510 int nice;
1511 int err;
1512
1513 memset(buffer, 0, sizeof(buffer));
1514 if (count > sizeof(buffer) - 1)
1515 count = sizeof(buffer) - 1;
1516 if (copy_from_user(buffer, buf, count))
1517 return -EFAULT;
1518
1519 err = kstrtoint(strstrip(buffer), 0, &nice);
1520 if (err < 0)
1521 return err;
1522
1523 p = get_proc_task(inode);
1524 if (!p)
1525 return -ESRCH;
1526
1527 err = proc_sched_autogroup_set_nice(p, nice);
1528 if (err)
1529 count = err;
1530
1531 put_task_struct(p);
1532
1533 return count;
1534}
1535
1536static int sched_autogroup_open(struct inode *inode, struct file *filp)
1537{
1538 int ret;
1539
1540 ret = single_open(filp, sched_autogroup_show, NULL);
1541 if (!ret) {
1542 struct seq_file *m = filp->private_data;
1543
1544 m->private = inode;
1545 }
1546 return ret;
1547}
1548
1549static const struct file_operations proc_pid_sched_autogroup_operations = {
1550 .open = sched_autogroup_open,
1551 .read = seq_read,
1552 .write = sched_autogroup_write,
1553 .llseek = seq_lseek,
1554 .release = single_release,
1555};
1556
1557#endif /* CONFIG_SCHED_AUTOGROUP */
1558
1559#ifdef CONFIG_TIME_NS
1560static int timens_offsets_show(struct seq_file *m, void *v)
1561{
1562 struct task_struct *p;
1563
1564 p = get_proc_task(file_inode(m->file));
1565 if (!p)
1566 return -ESRCH;
1567 proc_timens_show_offsets(p, m);
1568
1569 put_task_struct(p);
1570
1571 return 0;
1572}
1573
1574static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1575 size_t count, loff_t *ppos)
1576{
1577 struct inode *inode = file_inode(file);
1578 struct proc_timens_offset offsets[2];
1579 char *kbuf = NULL, *pos, *next_line;
1580 struct task_struct *p;
1581 int ret, noffsets;
1582
1583 /* Only allow < page size writes at the beginning of the file */
1584 if ((*ppos != 0) || (count >= PAGE_SIZE))
1585 return -EINVAL;
1586
1587 /* Slurp in the user data */
1588 kbuf = memdup_user_nul(buf, count);
1589 if (IS_ERR(kbuf))
1590 return PTR_ERR(kbuf);
1591
1592 /* Parse the user data */
1593 ret = -EINVAL;
1594 noffsets = 0;
1595 for (pos = kbuf; pos; pos = next_line) {
1596 struct proc_timens_offset *off = &offsets[noffsets];
1597 char clock[10];
1598 int err;
1599
1600 /* Find the end of line and ensure we don't look past it */
1601 next_line = strchr(pos, '\n');
1602 if (next_line) {
1603 *next_line = '\0';
1604 next_line++;
1605 if (*next_line == '\0')
1606 next_line = NULL;
1607 }
1608
1609 err = sscanf(pos, "%9s %lld %lu", clock,
1610 &off->val.tv_sec, &off->val.tv_nsec);
1611 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1612 goto out;
1613
1614 clock[sizeof(clock) - 1] = 0;
1615 if (strcmp(clock, "monotonic") == 0 ||
1616 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1617 off->clockid = CLOCK_MONOTONIC;
1618 else if (strcmp(clock, "boottime") == 0 ||
1619 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1620 off->clockid = CLOCK_BOOTTIME;
1621 else
1622 goto out;
1623
1624 noffsets++;
1625 if (noffsets == ARRAY_SIZE(offsets)) {
1626 if (next_line)
1627 count = next_line - kbuf;
1628 break;
1629 }
1630 }
1631
1632 ret = -ESRCH;
1633 p = get_proc_task(inode);
1634 if (!p)
1635 goto out;
1636 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1637 put_task_struct(p);
1638 if (ret)
1639 goto out;
1640
1641 ret = count;
1642out:
1643 kfree(kbuf);
1644 return ret;
1645}
1646
1647static int timens_offsets_open(struct inode *inode, struct file *filp)
1648{
1649 return single_open(filp, timens_offsets_show, inode);
1650}
1651
1652static const struct file_operations proc_timens_offsets_operations = {
1653 .open = timens_offsets_open,
1654 .read = seq_read,
1655 .write = timens_offsets_write,
1656 .llseek = seq_lseek,
1657 .release = single_release,
1658};
1659#endif /* CONFIG_TIME_NS */
1660
1661static ssize_t comm_write(struct file *file, const char __user *buf,
1662 size_t count, loff_t *offset)
1663{
1664 struct inode *inode = file_inode(file);
1665 struct task_struct *p;
1666 char buffer[TASK_COMM_LEN];
1667 const size_t maxlen = sizeof(buffer) - 1;
1668
1669 memset(buffer, 0, sizeof(buffer));
1670 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1671 return -EFAULT;
1672
1673 p = get_proc_task(inode);
1674 if (!p)
1675 return -ESRCH;
1676
1677 if (same_thread_group(current, p))
1678 set_task_comm(p, buffer);
1679 else
1680 count = -EINVAL;
1681
1682 put_task_struct(p);
1683
1684 return count;
1685}
1686
1687static int comm_show(struct seq_file *m, void *v)
1688{
1689 struct inode *inode = m->private;
1690 struct task_struct *p;
1691
1692 p = get_proc_task(inode);
1693 if (!p)
1694 return -ESRCH;
1695
1696 proc_task_name(m, p, false);
1697 seq_putc(m, '\n');
1698
1699 put_task_struct(p);
1700
1701 return 0;
1702}
1703
1704static int comm_open(struct inode *inode, struct file *filp)
1705{
1706 return single_open(filp, comm_show, inode);
1707}
1708
1709static const struct file_operations proc_pid_set_comm_operations = {
1710 .open = comm_open,
1711 .read = seq_read,
1712 .write = comm_write,
1713 .llseek = seq_lseek,
1714 .release = single_release,
1715};
1716
1717static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1718{
1719 struct task_struct *task;
1720 struct file *exe_file;
1721
1722 task = get_proc_task(d_inode(dentry));
1723 if (!task)
1724 return -ENOENT;
1725 exe_file = get_task_exe_file(task);
1726 put_task_struct(task);
1727 if (exe_file) {
1728 *exe_path = exe_file->f_path;
1729 path_get(&exe_file->f_path);
1730 fput(exe_file);
1731 return 0;
1732 } else
1733 return -ENOENT;
1734}
1735
1736static const char *proc_pid_get_link(struct dentry *dentry,
1737 struct inode *inode,
1738 struct delayed_call *done)
1739{
1740 struct path path;
1741 int error = -EACCES;
1742
1743 if (!dentry)
1744 return ERR_PTR(-ECHILD);
1745
1746 /* Are we allowed to snoop on the tasks file descriptors? */
1747 if (!proc_fd_access_allowed(inode))
1748 goto out;
1749
1750 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1751 if (error)
1752 goto out;
1753
1754 error = nd_jump_link(&path);
1755out:
1756 return ERR_PTR(error);
1757}
1758
1759static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1760{
1761 char *tmp = (char *)__get_free_page(GFP_KERNEL);
1762 char *pathname;
1763 int len;
1764
1765 if (!tmp)
1766 return -ENOMEM;
1767
1768 pathname = d_path(path, tmp, PAGE_SIZE);
1769 len = PTR_ERR(pathname);
1770 if (IS_ERR(pathname))
1771 goto out;
1772 len = tmp + PAGE_SIZE - 1 - pathname;
1773
1774 if (len > buflen)
1775 len = buflen;
1776 if (copy_to_user(buffer, pathname, len))
1777 len = -EFAULT;
1778 out:
1779 free_page((unsigned long)tmp);
1780 return len;
1781}
1782
1783static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1784{
1785 int error = -EACCES;
1786 struct inode *inode = d_inode(dentry);
1787 struct path path;
1788
1789 /* Are we allowed to snoop on the tasks file descriptors? */
1790 if (!proc_fd_access_allowed(inode))
1791 goto out;
1792
1793 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1794 if (error)
1795 goto out;
1796
1797 error = do_proc_readlink(&path, buffer, buflen);
1798 path_put(&path);
1799out:
1800 return error;
1801}
1802
1803const struct inode_operations proc_pid_link_inode_operations = {
1804 .readlink = proc_pid_readlink,
1805 .get_link = proc_pid_get_link,
1806 .setattr = proc_setattr,
1807};
1808
1809
1810/* building an inode */
1811
1812void task_dump_owner(struct task_struct *task, umode_t mode,
1813 kuid_t *ruid, kgid_t *rgid)
1814{
1815 /* Depending on the state of dumpable compute who should own a
1816 * proc file for a task.
1817 */
1818 const struct cred *cred;
1819 kuid_t uid;
1820 kgid_t gid;
1821
1822 if (unlikely(task->flags & PF_KTHREAD)) {
1823 *ruid = GLOBAL_ROOT_UID;
1824 *rgid = GLOBAL_ROOT_GID;
1825 return;
1826 }
1827
1828 /* Default to the tasks effective ownership */
1829 rcu_read_lock();
1830 cred = __task_cred(task);
1831 uid = cred->euid;
1832 gid = cred->egid;
1833 rcu_read_unlock();
1834
1835 /*
1836 * Before the /proc/pid/status file was created the only way to read
1837 * the effective uid of a /process was to stat /proc/pid. Reading
1838 * /proc/pid/status is slow enough that procps and other packages
1839 * kept stating /proc/pid. To keep the rules in /proc simple I have
1840 * made this apply to all per process world readable and executable
1841 * directories.
1842 */
1843 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1844 struct mm_struct *mm;
1845 task_lock(task);
1846 mm = task->mm;
1847 /* Make non-dumpable tasks owned by some root */
1848 if (mm) {
1849 if (get_dumpable(mm) != SUID_DUMP_USER) {
1850 struct user_namespace *user_ns = mm->user_ns;
1851
1852 uid = make_kuid(user_ns, 0);
1853 if (!uid_valid(uid))
1854 uid = GLOBAL_ROOT_UID;
1855
1856 gid = make_kgid(user_ns, 0);
1857 if (!gid_valid(gid))
1858 gid = GLOBAL_ROOT_GID;
1859 }
1860 } else {
1861 uid = GLOBAL_ROOT_UID;
1862 gid = GLOBAL_ROOT_GID;
1863 }
1864 task_unlock(task);
1865 }
1866 *ruid = uid;
1867 *rgid = gid;
1868}
1869
1870void proc_pid_evict_inode(struct proc_inode *ei)
1871{
1872 struct pid *pid = ei->pid;
1873
1874 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1875 spin_lock(&pid->lock);
1876 hlist_del_init_rcu(&ei->sibling_inodes);
1877 spin_unlock(&pid->lock);
1878 }
1879
1880 put_pid(pid);
1881}
1882
1883struct inode *proc_pid_make_inode(struct super_block * sb,
1884 struct task_struct *task, umode_t mode)
1885{
1886 struct inode * inode;
1887 struct proc_inode *ei;
1888 struct pid *pid;
1889
1890 /* We need a new inode */
1891
1892 inode = new_inode(sb);
1893 if (!inode)
1894 goto out;
1895
1896 /* Common stuff */
1897 ei = PROC_I(inode);
1898 inode->i_mode = mode;
1899 inode->i_ino = get_next_ino();
1900 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1901 inode->i_op = &proc_def_inode_operations;
1902
1903 /*
1904 * grab the reference to task.
1905 */
1906 pid = get_task_pid(task, PIDTYPE_PID);
1907 if (!pid)
1908 goto out_unlock;
1909
1910 /* Let the pid remember us for quick removal */
1911 ei->pid = pid;
1912 if (S_ISDIR(mode)) {
1913 spin_lock(&pid->lock);
1914 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1915 spin_unlock(&pid->lock);
1916 }
1917
1918 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1919 security_task_to_inode(task, inode);
1920
1921out:
1922 return inode;
1923
1924out_unlock:
1925 iput(inode);
1926 return NULL;
1927}
1928
1929int pid_getattr(struct user_namespace *mnt_userns, const struct path *path,
1930 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1931{
1932 struct inode *inode = d_inode(path->dentry);
1933 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
1934 struct task_struct *task;
1935
1936 generic_fillattr(&init_user_ns, inode, stat);
1937
1938 stat->uid = GLOBAL_ROOT_UID;
1939 stat->gid = GLOBAL_ROOT_GID;
1940 rcu_read_lock();
1941 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1942 if (task) {
1943 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
1944 rcu_read_unlock();
1945 /*
1946 * This doesn't prevent learning whether PID exists,
1947 * it only makes getattr() consistent with readdir().
1948 */
1949 return -ENOENT;
1950 }
1951 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1952 }
1953 rcu_read_unlock();
1954 return 0;
1955}
1956
1957/* dentry stuff */
1958
1959/*
1960 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1961 */
1962void pid_update_inode(struct task_struct *task, struct inode *inode)
1963{
1964 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1965
1966 inode->i_mode &= ~(S_ISUID | S_ISGID);
1967 security_task_to_inode(task, inode);
1968}
1969
1970/*
1971 * Rewrite the inode's ownerships here because the owning task may have
1972 * performed a setuid(), etc.
1973 *
1974 */
1975static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1976{
1977 struct inode *inode;
1978 struct task_struct *task;
1979
1980 if (flags & LOOKUP_RCU)
1981 return -ECHILD;
1982
1983 inode = d_inode(dentry);
1984 task = get_proc_task(inode);
1985
1986 if (task) {
1987 pid_update_inode(task, inode);
1988 put_task_struct(task);
1989 return 1;
1990 }
1991 return 0;
1992}
1993
1994static inline bool proc_inode_is_dead(struct inode *inode)
1995{
1996 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1997}
1998
1999int pid_delete_dentry(const struct dentry *dentry)
2000{
2001 /* Is the task we represent dead?
2002 * If so, then don't put the dentry on the lru list,
2003 * kill it immediately.
2004 */
2005 return proc_inode_is_dead(d_inode(dentry));
2006}
2007
2008const struct dentry_operations pid_dentry_operations =
2009{
2010 .d_revalidate = pid_revalidate,
2011 .d_delete = pid_delete_dentry,
2012};
2013
2014/* Lookups */
2015
2016/*
2017 * Fill a directory entry.
2018 *
2019 * If possible create the dcache entry and derive our inode number and
2020 * file type from dcache entry.
2021 *
2022 * Since all of the proc inode numbers are dynamically generated, the inode
2023 * numbers do not exist until the inode is cache. This means creating
2024 * the dcache entry in readdir is necessary to keep the inode numbers
2025 * reported by readdir in sync with the inode numbers reported
2026 * by stat.
2027 */
2028bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2029 const char *name, unsigned int len,
2030 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2031{
2032 struct dentry *child, *dir = file->f_path.dentry;
2033 struct qstr qname = QSTR_INIT(name, len);
2034 struct inode *inode;
2035 unsigned type = DT_UNKNOWN;
2036 ino_t ino = 1;
2037
2038 child = d_hash_and_lookup(dir, &qname);
2039 if (!child) {
2040 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2041 child = d_alloc_parallel(dir, &qname, &wq);
2042 if (IS_ERR(child))
2043 goto end_instantiate;
2044 if (d_in_lookup(child)) {
2045 struct dentry *res;
2046 res = instantiate(child, task, ptr);
2047 d_lookup_done(child);
2048 if (unlikely(res)) {
2049 dput(child);
2050 child = res;
2051 if (IS_ERR(child))
2052 goto end_instantiate;
2053 }
2054 }
2055 }
2056 inode = d_inode(child);
2057 ino = inode->i_ino;
2058 type = inode->i_mode >> 12;
2059 dput(child);
2060end_instantiate:
2061 return dir_emit(ctx, name, len, ino, type);
2062}
2063
2064/*
2065 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2066 * which represent vma start and end addresses.
2067 */
2068static int dname_to_vma_addr(struct dentry *dentry,
2069 unsigned long *start, unsigned long *end)
2070{
2071 const char *str = dentry->d_name.name;
2072 unsigned long long sval, eval;
2073 unsigned int len;
2074
2075 if (str[0] == '0' && str[1] != '-')
2076 return -EINVAL;
2077 len = _parse_integer(str, 16, &sval);
2078 if (len & KSTRTOX_OVERFLOW)
2079 return -EINVAL;
2080 if (sval != (unsigned long)sval)
2081 return -EINVAL;
2082 str += len;
2083
2084 if (*str != '-')
2085 return -EINVAL;
2086 str++;
2087
2088 if (str[0] == '0' && str[1])
2089 return -EINVAL;
2090 len = _parse_integer(str, 16, &eval);
2091 if (len & KSTRTOX_OVERFLOW)
2092 return -EINVAL;
2093 if (eval != (unsigned long)eval)
2094 return -EINVAL;
2095 str += len;
2096
2097 if (*str != '\0')
2098 return -EINVAL;
2099
2100 *start = sval;
2101 *end = eval;
2102
2103 return 0;
2104}
2105
2106static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2107{
2108 unsigned long vm_start, vm_end;
2109 bool exact_vma_exists = false;
2110 struct mm_struct *mm = NULL;
2111 struct task_struct *task;
2112 struct inode *inode;
2113 int status = 0;
2114
2115 if (flags & LOOKUP_RCU)
2116 return -ECHILD;
2117
2118 inode = d_inode(dentry);
2119 task = get_proc_task(inode);
2120 if (!task)
2121 goto out_notask;
2122
2123 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2124 if (IS_ERR_OR_NULL(mm))
2125 goto out;
2126
2127 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2128 status = mmap_read_lock_killable(mm);
2129 if (!status) {
2130 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2131 vm_end);
2132 mmap_read_unlock(mm);
2133 }
2134 }
2135
2136 mmput(mm);
2137
2138 if (exact_vma_exists) {
2139 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2140
2141 security_task_to_inode(task, inode);
2142 status = 1;
2143 }
2144
2145out:
2146 put_task_struct(task);
2147
2148out_notask:
2149 return status;
2150}
2151
2152static const struct dentry_operations tid_map_files_dentry_operations = {
2153 .d_revalidate = map_files_d_revalidate,
2154 .d_delete = pid_delete_dentry,
2155};
2156
2157static int map_files_get_link(struct dentry *dentry, struct path *path)
2158{
2159 unsigned long vm_start, vm_end;
2160 struct vm_area_struct *vma;
2161 struct task_struct *task;
2162 struct mm_struct *mm;
2163 int rc;
2164
2165 rc = -ENOENT;
2166 task = get_proc_task(d_inode(dentry));
2167 if (!task)
2168 goto out;
2169
2170 mm = get_task_mm(task);
2171 put_task_struct(task);
2172 if (!mm)
2173 goto out;
2174
2175 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2176 if (rc)
2177 goto out_mmput;
2178
2179 rc = mmap_read_lock_killable(mm);
2180 if (rc)
2181 goto out_mmput;
2182
2183 rc = -ENOENT;
2184 vma = find_exact_vma(mm, vm_start, vm_end);
2185 if (vma && vma->vm_file) {
2186 *path = vma->vm_file->f_path;
2187 path_get(path);
2188 rc = 0;
2189 }
2190 mmap_read_unlock(mm);
2191
2192out_mmput:
2193 mmput(mm);
2194out:
2195 return rc;
2196}
2197
2198struct map_files_info {
2199 unsigned long start;
2200 unsigned long end;
2201 fmode_t mode;
2202};
2203
2204/*
2205 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2206 * to concerns about how the symlinks may be used to bypass permissions on
2207 * ancestor directories in the path to the file in question.
2208 */
2209static const char *
2210proc_map_files_get_link(struct dentry *dentry,
2211 struct inode *inode,
2212 struct delayed_call *done)
2213{
2214 if (!checkpoint_restore_ns_capable(&init_user_ns))
2215 return ERR_PTR(-EPERM);
2216
2217 return proc_pid_get_link(dentry, inode, done);
2218}
2219
2220/*
2221 * Identical to proc_pid_link_inode_operations except for get_link()
2222 */
2223static const struct inode_operations proc_map_files_link_inode_operations = {
2224 .readlink = proc_pid_readlink,
2225 .get_link = proc_map_files_get_link,
2226 .setattr = proc_setattr,
2227};
2228
2229static struct dentry *
2230proc_map_files_instantiate(struct dentry *dentry,
2231 struct task_struct *task, const void *ptr)
2232{
2233 fmode_t mode = (fmode_t)(unsigned long)ptr;
2234 struct proc_inode *ei;
2235 struct inode *inode;
2236
2237 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2238 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2239 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2240 if (!inode)
2241 return ERR_PTR(-ENOENT);
2242
2243 ei = PROC_I(inode);
2244 ei->op.proc_get_link = map_files_get_link;
2245
2246 inode->i_op = &proc_map_files_link_inode_operations;
2247 inode->i_size = 64;
2248
2249 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2250 return d_splice_alias(inode, dentry);
2251}
2252
2253static struct dentry *proc_map_files_lookup(struct inode *dir,
2254 struct dentry *dentry, unsigned int flags)
2255{
2256 unsigned long vm_start, vm_end;
2257 struct vm_area_struct *vma;
2258 struct task_struct *task;
2259 struct dentry *result;
2260 struct mm_struct *mm;
2261
2262 result = ERR_PTR(-ENOENT);
2263 task = get_proc_task(dir);
2264 if (!task)
2265 goto out;
2266
2267 result = ERR_PTR(-EACCES);
2268 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2269 goto out_put_task;
2270
2271 result = ERR_PTR(-ENOENT);
2272 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2273 goto out_put_task;
2274
2275 mm = get_task_mm(task);
2276 if (!mm)
2277 goto out_put_task;
2278
2279 result = ERR_PTR(-EINTR);
2280 if (mmap_read_lock_killable(mm))
2281 goto out_put_mm;
2282
2283 result = ERR_PTR(-ENOENT);
2284 vma = find_exact_vma(mm, vm_start, vm_end);
2285 if (!vma)
2286 goto out_no_vma;
2287
2288 if (vma->vm_file)
2289 result = proc_map_files_instantiate(dentry, task,
2290 (void *)(unsigned long)vma->vm_file->f_mode);
2291
2292out_no_vma:
2293 mmap_read_unlock(mm);
2294out_put_mm:
2295 mmput(mm);
2296out_put_task:
2297 put_task_struct(task);
2298out:
2299 return result;
2300}
2301
2302static const struct inode_operations proc_map_files_inode_operations = {
2303 .lookup = proc_map_files_lookup,
2304 .permission = proc_fd_permission,
2305 .setattr = proc_setattr,
2306};
2307
2308static int
2309proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2310{
2311 struct vm_area_struct *vma;
2312 struct task_struct *task;
2313 struct mm_struct *mm;
2314 unsigned long nr_files, pos, i;
2315 GENRADIX(struct map_files_info) fa;
2316 struct map_files_info *p;
2317 int ret;
2318
2319 genradix_init(&fa);
2320
2321 ret = -ENOENT;
2322 task = get_proc_task(file_inode(file));
2323 if (!task)
2324 goto out;
2325
2326 ret = -EACCES;
2327 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2328 goto out_put_task;
2329
2330 ret = 0;
2331 if (!dir_emit_dots(file, ctx))
2332 goto out_put_task;
2333
2334 mm = get_task_mm(task);
2335 if (!mm)
2336 goto out_put_task;
2337
2338 ret = mmap_read_lock_killable(mm);
2339 if (ret) {
2340 mmput(mm);
2341 goto out_put_task;
2342 }
2343
2344 nr_files = 0;
2345
2346 /*
2347 * We need two passes here:
2348 *
2349 * 1) Collect vmas of mapped files with mmap_lock taken
2350 * 2) Release mmap_lock and instantiate entries
2351 *
2352 * otherwise we get lockdep complained, since filldir()
2353 * routine might require mmap_lock taken in might_fault().
2354 */
2355
2356 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2357 if (!vma->vm_file)
2358 continue;
2359 if (++pos <= ctx->pos)
2360 continue;
2361
2362 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2363 if (!p) {
2364 ret = -ENOMEM;
2365 mmap_read_unlock(mm);
2366 mmput(mm);
2367 goto out_put_task;
2368 }
2369
2370 p->start = vma->vm_start;
2371 p->end = vma->vm_end;
2372 p->mode = vma->vm_file->f_mode;
2373 }
2374 mmap_read_unlock(mm);
2375 mmput(mm);
2376
2377 for (i = 0; i < nr_files; i++) {
2378 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2379 unsigned int len;
2380
2381 p = genradix_ptr(&fa, i);
2382 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2383 if (!proc_fill_cache(file, ctx,
2384 buf, len,
2385 proc_map_files_instantiate,
2386 task,
2387 (void *)(unsigned long)p->mode))
2388 break;
2389 ctx->pos++;
2390 }
2391
2392out_put_task:
2393 put_task_struct(task);
2394out:
2395 genradix_free(&fa);
2396 return ret;
2397}
2398
2399static const struct file_operations proc_map_files_operations = {
2400 .read = generic_read_dir,
2401 .iterate_shared = proc_map_files_readdir,
2402 .llseek = generic_file_llseek,
2403};
2404
2405#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2406struct timers_private {
2407 struct pid *pid;
2408 struct task_struct *task;
2409 struct sighand_struct *sighand;
2410 struct pid_namespace *ns;
2411 unsigned long flags;
2412};
2413
2414static void *timers_start(struct seq_file *m, loff_t *pos)
2415{
2416 struct timers_private *tp = m->private;
2417
2418 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2419 if (!tp->task)
2420 return ERR_PTR(-ESRCH);
2421
2422 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2423 if (!tp->sighand)
2424 return ERR_PTR(-ESRCH);
2425
2426 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2427}
2428
2429static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2430{
2431 struct timers_private *tp = m->private;
2432 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2433}
2434
2435static void timers_stop(struct seq_file *m, void *v)
2436{
2437 struct timers_private *tp = m->private;
2438
2439 if (tp->sighand) {
2440 unlock_task_sighand(tp->task, &tp->flags);
2441 tp->sighand = NULL;
2442 }
2443
2444 if (tp->task) {
2445 put_task_struct(tp->task);
2446 tp->task = NULL;
2447 }
2448}
2449
2450static int show_timer(struct seq_file *m, void *v)
2451{
2452 struct k_itimer *timer;
2453 struct timers_private *tp = m->private;
2454 int notify;
2455 static const char * const nstr[] = {
2456 [SIGEV_SIGNAL] = "signal",
2457 [SIGEV_NONE] = "none",
2458 [SIGEV_THREAD] = "thread",
2459 };
2460
2461 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2462 notify = timer->it_sigev_notify;
2463
2464 seq_printf(m, "ID: %d\n", timer->it_id);
2465 seq_printf(m, "signal: %d/%px\n",
2466 timer->sigq->info.si_signo,
2467 timer->sigq->info.si_value.sival_ptr);
2468 seq_printf(m, "notify: %s/%s.%d\n",
2469 nstr[notify & ~SIGEV_THREAD_ID],
2470 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2471 pid_nr_ns(timer->it_pid, tp->ns));
2472 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2473
2474 return 0;
2475}
2476
2477static const struct seq_operations proc_timers_seq_ops = {
2478 .start = timers_start,
2479 .next = timers_next,
2480 .stop = timers_stop,
2481 .show = show_timer,
2482};
2483
2484static int proc_timers_open(struct inode *inode, struct file *file)
2485{
2486 struct timers_private *tp;
2487
2488 tp = __seq_open_private(file, &proc_timers_seq_ops,
2489 sizeof(struct timers_private));
2490 if (!tp)
2491 return -ENOMEM;
2492
2493 tp->pid = proc_pid(inode);
2494 tp->ns = proc_pid_ns(inode->i_sb);
2495 return 0;
2496}
2497
2498static const struct file_operations proc_timers_operations = {
2499 .open = proc_timers_open,
2500 .read = seq_read,
2501 .llseek = seq_lseek,
2502 .release = seq_release_private,
2503};
2504#endif
2505
2506static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2507 size_t count, loff_t *offset)
2508{
2509 struct inode *inode = file_inode(file);
2510 struct task_struct *p;
2511 u64 slack_ns;
2512 int err;
2513
2514 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2515 if (err < 0)
2516 return err;
2517
2518 p = get_proc_task(inode);
2519 if (!p)
2520 return -ESRCH;
2521
2522 if (p != current) {
2523 rcu_read_lock();
2524 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2525 rcu_read_unlock();
2526 count = -EPERM;
2527 goto out;
2528 }
2529 rcu_read_unlock();
2530
2531 err = security_task_setscheduler(p);
2532 if (err) {
2533 count = err;
2534 goto out;
2535 }
2536 }
2537
2538 task_lock(p);
2539 if (slack_ns == 0)
2540 p->timer_slack_ns = p->default_timer_slack_ns;
2541 else
2542 p->timer_slack_ns = slack_ns;
2543 task_unlock(p);
2544
2545out:
2546 put_task_struct(p);
2547
2548 return count;
2549}
2550
2551static int timerslack_ns_show(struct seq_file *m, void *v)
2552{
2553 struct inode *inode = m->private;
2554 struct task_struct *p;
2555 int err = 0;
2556
2557 p = get_proc_task(inode);
2558 if (!p)
2559 return -ESRCH;
2560
2561 if (p != current) {
2562 rcu_read_lock();
2563 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2564 rcu_read_unlock();
2565 err = -EPERM;
2566 goto out;
2567 }
2568 rcu_read_unlock();
2569
2570 err = security_task_getscheduler(p);
2571 if (err)
2572 goto out;
2573 }
2574
2575 task_lock(p);
2576 seq_printf(m, "%llu\n", p->timer_slack_ns);
2577 task_unlock(p);
2578
2579out:
2580 put_task_struct(p);
2581
2582 return err;
2583}
2584
2585static int timerslack_ns_open(struct inode *inode, struct file *filp)
2586{
2587 return single_open(filp, timerslack_ns_show, inode);
2588}
2589
2590static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2591 .open = timerslack_ns_open,
2592 .read = seq_read,
2593 .write = timerslack_ns_write,
2594 .llseek = seq_lseek,
2595 .release = single_release,
2596};
2597
2598static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2599 struct task_struct *task, const void *ptr)
2600{
2601 const struct pid_entry *p = ptr;
2602 struct inode *inode;
2603 struct proc_inode *ei;
2604
2605 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2606 if (!inode)
2607 return ERR_PTR(-ENOENT);
2608
2609 ei = PROC_I(inode);
2610 if (S_ISDIR(inode->i_mode))
2611 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2612 if (p->iop)
2613 inode->i_op = p->iop;
2614 if (p->fop)
2615 inode->i_fop = p->fop;
2616 ei->op = p->op;
2617 pid_update_inode(task, inode);
2618 d_set_d_op(dentry, &pid_dentry_operations);
2619 return d_splice_alias(inode, dentry);
2620}
2621
2622static struct dentry *proc_pident_lookup(struct inode *dir,
2623 struct dentry *dentry,
2624 const struct pid_entry *p,
2625 const struct pid_entry *end)
2626{
2627 struct task_struct *task = get_proc_task(dir);
2628 struct dentry *res = ERR_PTR(-ENOENT);
2629
2630 if (!task)
2631 goto out_no_task;
2632
2633 /*
2634 * Yes, it does not scale. And it should not. Don't add
2635 * new entries into /proc/<tgid>/ without very good reasons.
2636 */
2637 for (; p < end; p++) {
2638 if (p->len != dentry->d_name.len)
2639 continue;
2640 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2641 res = proc_pident_instantiate(dentry, task, p);
2642 break;
2643 }
2644 }
2645 put_task_struct(task);
2646out_no_task:
2647 return res;
2648}
2649
2650static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2651 const struct pid_entry *ents, unsigned int nents)
2652{
2653 struct task_struct *task = get_proc_task(file_inode(file));
2654 const struct pid_entry *p;
2655
2656 if (!task)
2657 return -ENOENT;
2658
2659 if (!dir_emit_dots(file, ctx))
2660 goto out;
2661
2662 if (ctx->pos >= nents + 2)
2663 goto out;
2664
2665 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2666 if (!proc_fill_cache(file, ctx, p->name, p->len,
2667 proc_pident_instantiate, task, p))
2668 break;
2669 ctx->pos++;
2670 }
2671out:
2672 put_task_struct(task);
2673 return 0;
2674}
2675
2676#ifdef CONFIG_SECURITY
2677static int proc_pid_attr_open(struct inode *inode, struct file *file)
2678{
2679 file->private_data = NULL;
2680 __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2681 return 0;
2682}
2683
2684static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2685 size_t count, loff_t *ppos)
2686{
2687 struct inode * inode = file_inode(file);
2688 char *p = NULL;
2689 ssize_t length;
2690 struct task_struct *task = get_proc_task(inode);
2691
2692 if (!task)
2693 return -ESRCH;
2694
2695 length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2696 (char*)file->f_path.dentry->d_name.name,
2697 &p);
2698 put_task_struct(task);
2699 if (length > 0)
2700 length = simple_read_from_buffer(buf, count, ppos, p, length);
2701 kfree(p);
2702 return length;
2703}
2704
2705static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2706 size_t count, loff_t *ppos)
2707{
2708 struct inode * inode = file_inode(file);
2709 struct task_struct *task;
2710 void *page;
2711 int rv;
2712
2713 /* A task may only write when it was the opener. */
2714 if (file->private_data != current->mm)
2715 return -EPERM;
2716
2717 rcu_read_lock();
2718 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2719 if (!task) {
2720 rcu_read_unlock();
2721 return -ESRCH;
2722 }
2723 /* A task may only write its own attributes. */
2724 if (current != task) {
2725 rcu_read_unlock();
2726 return -EACCES;
2727 }
2728 /* Prevent changes to overridden credentials. */
2729 if (current_cred() != current_real_cred()) {
2730 rcu_read_unlock();
2731 return -EBUSY;
2732 }
2733 rcu_read_unlock();
2734
2735 if (count > PAGE_SIZE)
2736 count = PAGE_SIZE;
2737
2738 /* No partial writes. */
2739 if (*ppos != 0)
2740 return -EINVAL;
2741
2742 page = memdup_user(buf, count);
2743 if (IS_ERR(page)) {
2744 rv = PTR_ERR(page);
2745 goto out;
2746 }
2747
2748 /* Guard against adverse ptrace interaction */
2749 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2750 if (rv < 0)
2751 goto out_free;
2752
2753 rv = security_setprocattr(PROC_I(inode)->op.lsm,
2754 file->f_path.dentry->d_name.name, page,
2755 count);
2756 mutex_unlock(¤t->signal->cred_guard_mutex);
2757out_free:
2758 kfree(page);
2759out:
2760 return rv;
2761}
2762
2763static const struct file_operations proc_pid_attr_operations = {
2764 .open = proc_pid_attr_open,
2765 .read = proc_pid_attr_read,
2766 .write = proc_pid_attr_write,
2767 .llseek = generic_file_llseek,
2768 .release = mem_release,
2769};
2770
2771#define LSM_DIR_OPS(LSM) \
2772static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2773 struct dir_context *ctx) \
2774{ \
2775 return proc_pident_readdir(filp, ctx, \
2776 LSM##_attr_dir_stuff, \
2777 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2778} \
2779\
2780static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2781 .read = generic_read_dir, \
2782 .iterate = proc_##LSM##_attr_dir_iterate, \
2783 .llseek = default_llseek, \
2784}; \
2785\
2786static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2787 struct dentry *dentry, unsigned int flags) \
2788{ \
2789 return proc_pident_lookup(dir, dentry, \
2790 LSM##_attr_dir_stuff, \
2791 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2792} \
2793\
2794static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2795 .lookup = proc_##LSM##_attr_dir_lookup, \
2796 .getattr = pid_getattr, \
2797 .setattr = proc_setattr, \
2798}
2799
2800#ifdef CONFIG_SECURITY_SMACK
2801static const struct pid_entry smack_attr_dir_stuff[] = {
2802 ATTR("smack", "current", 0666),
2803};
2804LSM_DIR_OPS(smack);
2805#endif
2806
2807#ifdef CONFIG_SECURITY_APPARMOR
2808static const struct pid_entry apparmor_attr_dir_stuff[] = {
2809 ATTR("apparmor", "current", 0666),
2810 ATTR("apparmor", "prev", 0444),
2811 ATTR("apparmor", "exec", 0666),
2812};
2813LSM_DIR_OPS(apparmor);
2814#endif
2815
2816static const struct pid_entry attr_dir_stuff[] = {
2817 ATTR(NULL, "current", 0666),
2818 ATTR(NULL, "prev", 0444),
2819 ATTR(NULL, "exec", 0666),
2820 ATTR(NULL, "fscreate", 0666),
2821 ATTR(NULL, "keycreate", 0666),
2822 ATTR(NULL, "sockcreate", 0666),
2823#ifdef CONFIG_SECURITY_SMACK
2824 DIR("smack", 0555,
2825 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2826#endif
2827#ifdef CONFIG_SECURITY_APPARMOR
2828 DIR("apparmor", 0555,
2829 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2830#endif
2831};
2832
2833static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2834{
2835 return proc_pident_readdir(file, ctx,
2836 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2837}
2838
2839static const struct file_operations proc_attr_dir_operations = {
2840 .read = generic_read_dir,
2841 .iterate_shared = proc_attr_dir_readdir,
2842 .llseek = generic_file_llseek,
2843};
2844
2845static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2846 struct dentry *dentry, unsigned int flags)
2847{
2848 return proc_pident_lookup(dir, dentry,
2849 attr_dir_stuff,
2850 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2851}
2852
2853static const struct inode_operations proc_attr_dir_inode_operations = {
2854 .lookup = proc_attr_dir_lookup,
2855 .getattr = pid_getattr,
2856 .setattr = proc_setattr,
2857};
2858
2859#endif
2860
2861#ifdef CONFIG_ELF_CORE
2862static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2863 size_t count, loff_t *ppos)
2864{
2865 struct task_struct *task = get_proc_task(file_inode(file));
2866 struct mm_struct *mm;
2867 char buffer[PROC_NUMBUF];
2868 size_t len;
2869 int ret;
2870
2871 if (!task)
2872 return -ESRCH;
2873
2874 ret = 0;
2875 mm = get_task_mm(task);
2876 if (mm) {
2877 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2878 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2879 MMF_DUMP_FILTER_SHIFT));
2880 mmput(mm);
2881 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2882 }
2883
2884 put_task_struct(task);
2885
2886 return ret;
2887}
2888
2889static ssize_t proc_coredump_filter_write(struct file *file,
2890 const char __user *buf,
2891 size_t count,
2892 loff_t *ppos)
2893{
2894 struct task_struct *task;
2895 struct mm_struct *mm;
2896 unsigned int val;
2897 int ret;
2898 int i;
2899 unsigned long mask;
2900
2901 ret = kstrtouint_from_user(buf, count, 0, &val);
2902 if (ret < 0)
2903 return ret;
2904
2905 ret = -ESRCH;
2906 task = get_proc_task(file_inode(file));
2907 if (!task)
2908 goto out_no_task;
2909
2910 mm = get_task_mm(task);
2911 if (!mm)
2912 goto out_no_mm;
2913 ret = 0;
2914
2915 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2916 if (val & mask)
2917 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2918 else
2919 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2920 }
2921
2922 mmput(mm);
2923 out_no_mm:
2924 put_task_struct(task);
2925 out_no_task:
2926 if (ret < 0)
2927 return ret;
2928 return count;
2929}
2930
2931static const struct file_operations proc_coredump_filter_operations = {
2932 .read = proc_coredump_filter_read,
2933 .write = proc_coredump_filter_write,
2934 .llseek = generic_file_llseek,
2935};
2936#endif
2937
2938#ifdef CONFIG_TASK_IO_ACCOUNTING
2939static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2940{
2941 struct task_io_accounting acct = task->ioac;
2942 unsigned long flags;
2943 int result;
2944
2945 result = down_read_killable(&task->signal->exec_update_lock);
2946 if (result)
2947 return result;
2948
2949 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2950 result = -EACCES;
2951 goto out_unlock;
2952 }
2953
2954 if (whole && lock_task_sighand(task, &flags)) {
2955 struct task_struct *t = task;
2956
2957 task_io_accounting_add(&acct, &task->signal->ioac);
2958 while_each_thread(task, t)
2959 task_io_accounting_add(&acct, &t->ioac);
2960
2961 unlock_task_sighand(task, &flags);
2962 }
2963 seq_printf(m,
2964 "rchar: %llu\n"
2965 "wchar: %llu\n"
2966 "syscr: %llu\n"
2967 "syscw: %llu\n"
2968 "read_bytes: %llu\n"
2969 "write_bytes: %llu\n"
2970 "cancelled_write_bytes: %llu\n",
2971 (unsigned long long)acct.rchar,
2972 (unsigned long long)acct.wchar,
2973 (unsigned long long)acct.syscr,
2974 (unsigned long long)acct.syscw,
2975 (unsigned long long)acct.read_bytes,
2976 (unsigned long long)acct.write_bytes,
2977 (unsigned long long)acct.cancelled_write_bytes);
2978 result = 0;
2979
2980out_unlock:
2981 up_read(&task->signal->exec_update_lock);
2982 return result;
2983}
2984
2985static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2986 struct pid *pid, struct task_struct *task)
2987{
2988 return do_io_accounting(task, m, 0);
2989}
2990
2991static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2992 struct pid *pid, struct task_struct *task)
2993{
2994 return do_io_accounting(task, m, 1);
2995}
2996#endif /* CONFIG_TASK_IO_ACCOUNTING */
2997
2998#ifdef CONFIG_USER_NS
2999static int proc_id_map_open(struct inode *inode, struct file *file,
3000 const struct seq_operations *seq_ops)
3001{
3002 struct user_namespace *ns = NULL;
3003 struct task_struct *task;
3004 struct seq_file *seq;
3005 int ret = -EINVAL;
3006
3007 task = get_proc_task(inode);
3008 if (task) {
3009 rcu_read_lock();
3010 ns = get_user_ns(task_cred_xxx(task, user_ns));
3011 rcu_read_unlock();
3012 put_task_struct(task);
3013 }
3014 if (!ns)
3015 goto err;
3016
3017 ret = seq_open(file, seq_ops);
3018 if (ret)
3019 goto err_put_ns;
3020
3021 seq = file->private_data;
3022 seq->private = ns;
3023
3024 return 0;
3025err_put_ns:
3026 put_user_ns(ns);
3027err:
3028 return ret;
3029}
3030
3031static int proc_id_map_release(struct inode *inode, struct file *file)
3032{
3033 struct seq_file *seq = file->private_data;
3034 struct user_namespace *ns = seq->private;
3035 put_user_ns(ns);
3036 return seq_release(inode, file);
3037}
3038
3039static int proc_uid_map_open(struct inode *inode, struct file *file)
3040{
3041 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3042}
3043
3044static int proc_gid_map_open(struct inode *inode, struct file *file)
3045{
3046 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3047}
3048
3049static int proc_projid_map_open(struct inode *inode, struct file *file)
3050{
3051 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3052}
3053
3054static const struct file_operations proc_uid_map_operations = {
3055 .open = proc_uid_map_open,
3056 .write = proc_uid_map_write,
3057 .read = seq_read,
3058 .llseek = seq_lseek,
3059 .release = proc_id_map_release,
3060};
3061
3062static const struct file_operations proc_gid_map_operations = {
3063 .open = proc_gid_map_open,
3064 .write = proc_gid_map_write,
3065 .read = seq_read,
3066 .llseek = seq_lseek,
3067 .release = proc_id_map_release,
3068};
3069
3070static const struct file_operations proc_projid_map_operations = {
3071 .open = proc_projid_map_open,
3072 .write = proc_projid_map_write,
3073 .read = seq_read,
3074 .llseek = seq_lseek,
3075 .release = proc_id_map_release,
3076};
3077
3078static int proc_setgroups_open(struct inode *inode, struct file *file)
3079{
3080 struct user_namespace *ns = NULL;
3081 struct task_struct *task;
3082 int ret;
3083
3084 ret = -ESRCH;
3085 task = get_proc_task(inode);
3086 if (task) {
3087 rcu_read_lock();
3088 ns = get_user_ns(task_cred_xxx(task, user_ns));
3089 rcu_read_unlock();
3090 put_task_struct(task);
3091 }
3092 if (!ns)
3093 goto err;
3094
3095 if (file->f_mode & FMODE_WRITE) {
3096 ret = -EACCES;
3097 if (!ns_capable(ns, CAP_SYS_ADMIN))
3098 goto err_put_ns;
3099 }
3100
3101 ret = single_open(file, &proc_setgroups_show, ns);
3102 if (ret)
3103 goto err_put_ns;
3104
3105 return 0;
3106err_put_ns:
3107 put_user_ns(ns);
3108err:
3109 return ret;
3110}
3111
3112static int proc_setgroups_release(struct inode *inode, struct file *file)
3113{
3114 struct seq_file *seq = file->private_data;
3115 struct user_namespace *ns = seq->private;
3116 int ret = single_release(inode, file);
3117 put_user_ns(ns);
3118 return ret;
3119}
3120
3121static const struct file_operations proc_setgroups_operations = {
3122 .open = proc_setgroups_open,
3123 .write = proc_setgroups_write,
3124 .read = seq_read,
3125 .llseek = seq_lseek,
3126 .release = proc_setgroups_release,
3127};
3128#endif /* CONFIG_USER_NS */
3129
3130static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3131 struct pid *pid, struct task_struct *task)
3132{
3133 int err = lock_trace(task);
3134 if (!err) {
3135 seq_printf(m, "%08x\n", task->personality);
3136 unlock_trace(task);
3137 }
3138 return err;
3139}
3140
3141#ifdef CONFIG_LIVEPATCH
3142static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3143 struct pid *pid, struct task_struct *task)
3144{
3145 seq_printf(m, "%d\n", task->patch_state);
3146 return 0;
3147}
3148#endif /* CONFIG_LIVEPATCH */
3149
3150#ifdef CONFIG_STACKLEAK_METRICS
3151static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3152 struct pid *pid, struct task_struct *task)
3153{
3154 unsigned long prev_depth = THREAD_SIZE -
3155 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3156 unsigned long depth = THREAD_SIZE -
3157 (task->lowest_stack & (THREAD_SIZE - 1));
3158
3159 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3160 prev_depth, depth);
3161 return 0;
3162}
3163#endif /* CONFIG_STACKLEAK_METRICS */
3164
3165/*
3166 * Thread groups
3167 */
3168static const struct file_operations proc_task_operations;
3169static const struct inode_operations proc_task_inode_operations;
3170
3171static const struct pid_entry tgid_base_stuff[] = {
3172 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3173 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3174 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3175 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3176 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3177#ifdef CONFIG_NET
3178 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3179#endif
3180 REG("environ", S_IRUSR, proc_environ_operations),
3181 REG("auxv", S_IRUSR, proc_auxv_operations),
3182 ONE("status", S_IRUGO, proc_pid_status),
3183 ONE("personality", S_IRUSR, proc_pid_personality),
3184 ONE("limits", S_IRUGO, proc_pid_limits),
3185#ifdef CONFIG_SCHED_DEBUG
3186 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3187#endif
3188#ifdef CONFIG_SCHED_AUTOGROUP
3189 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3190#endif
3191#ifdef CONFIG_TIME_NS
3192 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3193#endif
3194 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3195#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3196 ONE("syscall", S_IRUSR, proc_pid_syscall),
3197#endif
3198 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3199 ONE("stat", S_IRUGO, proc_tgid_stat),
3200 ONE("statm", S_IRUGO, proc_pid_statm),
3201 REG("maps", S_IRUGO, proc_pid_maps_operations),
3202#ifdef CONFIG_NUMA
3203 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3204#endif
3205 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3206 LNK("cwd", proc_cwd_link),
3207 LNK("root", proc_root_link),
3208 LNK("exe", proc_exe_link),
3209 REG("mounts", S_IRUGO, proc_mounts_operations),
3210 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3211 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3212#ifdef CONFIG_PROC_PAGE_MONITOR
3213 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3214 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3215 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3216 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3217#endif
3218#ifdef CONFIG_SECURITY
3219 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3220#endif
3221#ifdef CONFIG_KALLSYMS
3222 ONE("wchan", S_IRUGO, proc_pid_wchan),
3223#endif
3224#ifdef CONFIG_STACKTRACE
3225 ONE("stack", S_IRUSR, proc_pid_stack),
3226#endif
3227#ifdef CONFIG_SCHED_INFO
3228 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3229#endif
3230#ifdef CONFIG_LATENCYTOP
3231 REG("latency", S_IRUGO, proc_lstats_operations),
3232#endif
3233#ifdef CONFIG_PROC_PID_CPUSET
3234 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3235#endif
3236#ifdef CONFIG_CGROUPS
3237 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3238#endif
3239#ifdef CONFIG_PROC_CPU_RESCTRL
3240 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3241#endif
3242 ONE("oom_score", S_IRUGO, proc_oom_score),
3243 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3244 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3245#ifdef CONFIG_AUDIT
3246 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3247 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3248#endif
3249#ifdef CONFIG_FAULT_INJECTION
3250 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3251 REG("fail-nth", 0644, proc_fail_nth_operations),
3252#endif
3253#ifdef CONFIG_ELF_CORE
3254 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3255#endif
3256#ifdef CONFIG_TASK_IO_ACCOUNTING
3257 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3258#endif
3259#ifdef CONFIG_USER_NS
3260 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3261 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3262 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3263 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3264#endif
3265#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3266 REG("timers", S_IRUGO, proc_timers_operations),
3267#endif
3268 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3269#ifdef CONFIG_LIVEPATCH
3270 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3271#endif
3272#ifdef CONFIG_STACKLEAK_METRICS
3273 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3274#endif
3275#ifdef CONFIG_PROC_PID_ARCH_STATUS
3276 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3277#endif
3278#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3279 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3280#endif
3281};
3282
3283static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3284{
3285 return proc_pident_readdir(file, ctx,
3286 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3287}
3288
3289static const struct file_operations proc_tgid_base_operations = {
3290 .read = generic_read_dir,
3291 .iterate_shared = proc_tgid_base_readdir,
3292 .llseek = generic_file_llseek,
3293};
3294
3295struct pid *tgid_pidfd_to_pid(const struct file *file)
3296{
3297 if (file->f_op != &proc_tgid_base_operations)
3298 return ERR_PTR(-EBADF);
3299
3300 return proc_pid(file_inode(file));
3301}
3302
3303static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3304{
3305 return proc_pident_lookup(dir, dentry,
3306 tgid_base_stuff,
3307 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3308}
3309
3310static const struct inode_operations proc_tgid_base_inode_operations = {
3311 .lookup = proc_tgid_base_lookup,
3312 .getattr = pid_getattr,
3313 .setattr = proc_setattr,
3314 .permission = proc_pid_permission,
3315};
3316
3317/**
3318 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3319 * @pid: pid that should be flushed.
3320 *
3321 * This function walks a list of inodes (that belong to any proc
3322 * filesystem) that are attached to the pid and flushes them from
3323 * the dentry cache.
3324 *
3325 * It is safe and reasonable to cache /proc entries for a task until
3326 * that task exits. After that they just clog up the dcache with
3327 * useless entries, possibly causing useful dcache entries to be
3328 * flushed instead. This routine is provided to flush those useless
3329 * dcache entries when a process is reaped.
3330 *
3331 * NOTE: This routine is just an optimization so it does not guarantee
3332 * that no dcache entries will exist after a process is reaped
3333 * it just makes it very unlikely that any will persist.
3334 */
3335
3336void proc_flush_pid(struct pid *pid)
3337{
3338 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3339}
3340
3341static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3342 struct task_struct *task, const void *ptr)
3343{
3344 struct inode *inode;
3345
3346 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3347 if (!inode)
3348 return ERR_PTR(-ENOENT);
3349
3350 inode->i_op = &proc_tgid_base_inode_operations;
3351 inode->i_fop = &proc_tgid_base_operations;
3352 inode->i_flags|=S_IMMUTABLE;
3353
3354 set_nlink(inode, nlink_tgid);
3355 pid_update_inode(task, inode);
3356
3357 d_set_d_op(dentry, &pid_dentry_operations);
3358 return d_splice_alias(inode, dentry);
3359}
3360
3361struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3362{
3363 struct task_struct *task;
3364 unsigned tgid;
3365 struct proc_fs_info *fs_info;
3366 struct pid_namespace *ns;
3367 struct dentry *result = ERR_PTR(-ENOENT);
3368
3369 tgid = name_to_int(&dentry->d_name);
3370 if (tgid == ~0U)
3371 goto out;
3372
3373 fs_info = proc_sb_info(dentry->d_sb);
3374 ns = fs_info->pid_ns;
3375 rcu_read_lock();
3376 task = find_task_by_pid_ns(tgid, ns);
3377 if (task)
3378 get_task_struct(task);
3379 rcu_read_unlock();
3380 if (!task)
3381 goto out;
3382
3383 /* Limit procfs to only ptraceable tasks */
3384 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3385 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3386 goto out_put_task;
3387 }
3388
3389 result = proc_pid_instantiate(dentry, task, NULL);
3390out_put_task:
3391 put_task_struct(task);
3392out:
3393 return result;
3394}
3395
3396/*
3397 * Find the first task with tgid >= tgid
3398 *
3399 */
3400struct tgid_iter {
3401 unsigned int tgid;
3402 struct task_struct *task;
3403};
3404static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3405{
3406 struct pid *pid;
3407
3408 if (iter.task)
3409 put_task_struct(iter.task);
3410 rcu_read_lock();
3411retry:
3412 iter.task = NULL;
3413 pid = find_ge_pid(iter.tgid, ns);
3414 if (pid) {
3415 iter.tgid = pid_nr_ns(pid, ns);
3416 iter.task = pid_task(pid, PIDTYPE_TGID);
3417 if (!iter.task) {
3418 iter.tgid += 1;
3419 goto retry;
3420 }
3421 get_task_struct(iter.task);
3422 }
3423 rcu_read_unlock();
3424 return iter;
3425}
3426
3427#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3428
3429/* for the /proc/ directory itself, after non-process stuff has been done */
3430int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3431{
3432 struct tgid_iter iter;
3433 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3434 struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3435 loff_t pos = ctx->pos;
3436
3437 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3438 return 0;
3439
3440 if (pos == TGID_OFFSET - 2) {
3441 struct inode *inode = d_inode(fs_info->proc_self);
3442 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3443 return 0;
3444 ctx->pos = pos = pos + 1;
3445 }
3446 if (pos == TGID_OFFSET - 1) {
3447 struct inode *inode = d_inode(fs_info->proc_thread_self);
3448 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3449 return 0;
3450 ctx->pos = pos = pos + 1;
3451 }
3452 iter.tgid = pos - TGID_OFFSET;
3453 iter.task = NULL;
3454 for (iter = next_tgid(ns, iter);
3455 iter.task;
3456 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3457 char name[10 + 1];
3458 unsigned int len;
3459
3460 cond_resched();
3461 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3462 continue;
3463
3464 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3465 ctx->pos = iter.tgid + TGID_OFFSET;
3466 if (!proc_fill_cache(file, ctx, name, len,
3467 proc_pid_instantiate, iter.task, NULL)) {
3468 put_task_struct(iter.task);
3469 return 0;
3470 }
3471 }
3472 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3473 return 0;
3474}
3475
3476/*
3477 * proc_tid_comm_permission is a special permission function exclusively
3478 * used for the node /proc/<pid>/task/<tid>/comm.
3479 * It bypasses generic permission checks in the case where a task of the same
3480 * task group attempts to access the node.
3481 * The rationale behind this is that glibc and bionic access this node for
3482 * cross thread naming (pthread_set/getname_np(!self)). However, if
3483 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3484 * which locks out the cross thread naming implementation.
3485 * This function makes sure that the node is always accessible for members of
3486 * same thread group.
3487 */
3488static int proc_tid_comm_permission(struct user_namespace *mnt_userns,
3489 struct inode *inode, int mask)
3490{
3491 bool is_same_tgroup;
3492 struct task_struct *task;
3493
3494 task = get_proc_task(inode);
3495 if (!task)
3496 return -ESRCH;
3497 is_same_tgroup = same_thread_group(current, task);
3498 put_task_struct(task);
3499
3500 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3501 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3502 * read or written by the members of the corresponding
3503 * thread group.
3504 */
3505 return 0;
3506 }
3507
3508 return generic_permission(&init_user_ns, inode, mask);
3509}
3510
3511static const struct inode_operations proc_tid_comm_inode_operations = {
3512 .permission = proc_tid_comm_permission,
3513};
3514
3515/*
3516 * Tasks
3517 */
3518static const struct pid_entry tid_base_stuff[] = {
3519 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3520 DIR("fdinfo", S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3521 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3522#ifdef CONFIG_NET
3523 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3524#endif
3525 REG("environ", S_IRUSR, proc_environ_operations),
3526 REG("auxv", S_IRUSR, proc_auxv_operations),
3527 ONE("status", S_IRUGO, proc_pid_status),
3528 ONE("personality", S_IRUSR, proc_pid_personality),
3529 ONE("limits", S_IRUGO, proc_pid_limits),
3530#ifdef CONFIG_SCHED_DEBUG
3531 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3532#endif
3533 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3534 &proc_tid_comm_inode_operations,
3535 &proc_pid_set_comm_operations, {}),
3536#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3537 ONE("syscall", S_IRUSR, proc_pid_syscall),
3538#endif
3539 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3540 ONE("stat", S_IRUGO, proc_tid_stat),
3541 ONE("statm", S_IRUGO, proc_pid_statm),
3542 REG("maps", S_IRUGO, proc_pid_maps_operations),
3543#ifdef CONFIG_PROC_CHILDREN
3544 REG("children", S_IRUGO, proc_tid_children_operations),
3545#endif
3546#ifdef CONFIG_NUMA
3547 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3548#endif
3549 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3550 LNK("cwd", proc_cwd_link),
3551 LNK("root", proc_root_link),
3552 LNK("exe", proc_exe_link),
3553 REG("mounts", S_IRUGO, proc_mounts_operations),
3554 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3555#ifdef CONFIG_PROC_PAGE_MONITOR
3556 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3557 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3558 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3559 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3560#endif
3561#ifdef CONFIG_SECURITY
3562 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3563#endif
3564#ifdef CONFIG_KALLSYMS
3565 ONE("wchan", S_IRUGO, proc_pid_wchan),
3566#endif
3567#ifdef CONFIG_STACKTRACE
3568 ONE("stack", S_IRUSR, proc_pid_stack),
3569#endif
3570#ifdef CONFIG_SCHED_INFO
3571 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3572#endif
3573#ifdef CONFIG_LATENCYTOP
3574 REG("latency", S_IRUGO, proc_lstats_operations),
3575#endif
3576#ifdef CONFIG_PROC_PID_CPUSET
3577 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3578#endif
3579#ifdef CONFIG_CGROUPS
3580 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3581#endif
3582#ifdef CONFIG_PROC_CPU_RESCTRL
3583 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3584#endif
3585 ONE("oom_score", S_IRUGO, proc_oom_score),
3586 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3587 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3588#ifdef CONFIG_AUDIT
3589 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3590 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3591#endif
3592#ifdef CONFIG_FAULT_INJECTION
3593 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3594 REG("fail-nth", 0644, proc_fail_nth_operations),
3595#endif
3596#ifdef CONFIG_TASK_IO_ACCOUNTING
3597 ONE("io", S_IRUSR, proc_tid_io_accounting),
3598#endif
3599#ifdef CONFIG_USER_NS
3600 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3601 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3602 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3603 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3604#endif
3605#ifdef CONFIG_LIVEPATCH
3606 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3607#endif
3608#ifdef CONFIG_PROC_PID_ARCH_STATUS
3609 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3610#endif
3611#ifdef CONFIG_SECCOMP_CACHE_DEBUG
3612 ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3613#endif
3614};
3615
3616static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3617{
3618 return proc_pident_readdir(file, ctx,
3619 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3620}
3621
3622static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3623{
3624 return proc_pident_lookup(dir, dentry,
3625 tid_base_stuff,
3626 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3627}
3628
3629static const struct file_operations proc_tid_base_operations = {
3630 .read = generic_read_dir,
3631 .iterate_shared = proc_tid_base_readdir,
3632 .llseek = generic_file_llseek,
3633};
3634
3635static const struct inode_operations proc_tid_base_inode_operations = {
3636 .lookup = proc_tid_base_lookup,
3637 .getattr = pid_getattr,
3638 .setattr = proc_setattr,
3639};
3640
3641static struct dentry *proc_task_instantiate(struct dentry *dentry,
3642 struct task_struct *task, const void *ptr)
3643{
3644 struct inode *inode;
3645 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3646 if (!inode)
3647 return ERR_PTR(-ENOENT);
3648
3649 inode->i_op = &proc_tid_base_inode_operations;
3650 inode->i_fop = &proc_tid_base_operations;
3651 inode->i_flags |= S_IMMUTABLE;
3652
3653 set_nlink(inode, nlink_tid);
3654 pid_update_inode(task, inode);
3655
3656 d_set_d_op(dentry, &pid_dentry_operations);
3657 return d_splice_alias(inode, dentry);
3658}
3659
3660static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3661{
3662 struct task_struct *task;
3663 struct task_struct *leader = get_proc_task(dir);
3664 unsigned tid;
3665 struct proc_fs_info *fs_info;
3666 struct pid_namespace *ns;
3667 struct dentry *result = ERR_PTR(-ENOENT);
3668
3669 if (!leader)
3670 goto out_no_task;
3671
3672 tid = name_to_int(&dentry->d_name);
3673 if (tid == ~0U)
3674 goto out;
3675
3676 fs_info = proc_sb_info(dentry->d_sb);
3677 ns = fs_info->pid_ns;
3678 rcu_read_lock();
3679 task = find_task_by_pid_ns(tid, ns);
3680 if (task)
3681 get_task_struct(task);
3682 rcu_read_unlock();
3683 if (!task)
3684 goto out;
3685 if (!same_thread_group(leader, task))
3686 goto out_drop_task;
3687
3688 result = proc_task_instantiate(dentry, task, NULL);
3689out_drop_task:
3690 put_task_struct(task);
3691out:
3692 put_task_struct(leader);
3693out_no_task:
3694 return result;
3695}
3696
3697/*
3698 * Find the first tid of a thread group to return to user space.
3699 *
3700 * Usually this is just the thread group leader, but if the users
3701 * buffer was too small or there was a seek into the middle of the
3702 * directory we have more work todo.
3703 *
3704 * In the case of a short read we start with find_task_by_pid.
3705 *
3706 * In the case of a seek we start with the leader and walk nr
3707 * threads past it.
3708 */
3709static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3710 struct pid_namespace *ns)
3711{
3712 struct task_struct *pos, *task;
3713 unsigned long nr = f_pos;
3714
3715 if (nr != f_pos) /* 32bit overflow? */
3716 return NULL;
3717
3718 rcu_read_lock();
3719 task = pid_task(pid, PIDTYPE_PID);
3720 if (!task)
3721 goto fail;
3722
3723 /* Attempt to start with the tid of a thread */
3724 if (tid && nr) {
3725 pos = find_task_by_pid_ns(tid, ns);
3726 if (pos && same_thread_group(pos, task))
3727 goto found;
3728 }
3729
3730 /* If nr exceeds the number of threads there is nothing todo */
3731 if (nr >= get_nr_threads(task))
3732 goto fail;
3733
3734 /* If we haven't found our starting place yet start
3735 * with the leader and walk nr threads forward.
3736 */
3737 pos = task = task->group_leader;
3738 do {
3739 if (!nr--)
3740 goto found;
3741 } while_each_thread(task, pos);
3742fail:
3743 pos = NULL;
3744 goto out;
3745found:
3746 get_task_struct(pos);
3747out:
3748 rcu_read_unlock();
3749 return pos;
3750}
3751
3752/*
3753 * Find the next thread in the thread list.
3754 * Return NULL if there is an error or no next thread.
3755 *
3756 * The reference to the input task_struct is released.
3757 */
3758static struct task_struct *next_tid(struct task_struct *start)
3759{
3760 struct task_struct *pos = NULL;
3761 rcu_read_lock();
3762 if (pid_alive(start)) {
3763 pos = next_thread(start);
3764 if (thread_group_leader(pos))
3765 pos = NULL;
3766 else
3767 get_task_struct(pos);
3768 }
3769 rcu_read_unlock();
3770 put_task_struct(start);
3771 return pos;
3772}
3773
3774/* for the /proc/TGID/task/ directories */
3775static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3776{
3777 struct inode *inode = file_inode(file);
3778 struct task_struct *task;
3779 struct pid_namespace *ns;
3780 int tid;
3781
3782 if (proc_inode_is_dead(inode))
3783 return -ENOENT;
3784
3785 if (!dir_emit_dots(file, ctx))
3786 return 0;
3787
3788 /* f_version caches the tgid value that the last readdir call couldn't
3789 * return. lseek aka telldir automagically resets f_version to 0.
3790 */
3791 ns = proc_pid_ns(inode->i_sb);
3792 tid = (int)file->f_version;
3793 file->f_version = 0;
3794 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3795 task;
3796 task = next_tid(task), ctx->pos++) {
3797 char name[10 + 1];
3798 unsigned int len;
3799 tid = task_pid_nr_ns(task, ns);
3800 len = snprintf(name, sizeof(name), "%u", tid);
3801 if (!proc_fill_cache(file, ctx, name, len,
3802 proc_task_instantiate, task, NULL)) {
3803 /* returning this tgid failed, save it as the first
3804 * pid for the next readir call */
3805 file->f_version = (u64)tid;
3806 put_task_struct(task);
3807 break;
3808 }
3809 }
3810
3811 return 0;
3812}
3813
3814static int proc_task_getattr(struct user_namespace *mnt_userns,
3815 const struct path *path, struct kstat *stat,
3816 u32 request_mask, unsigned int query_flags)
3817{
3818 struct inode *inode = d_inode(path->dentry);
3819 struct task_struct *p = get_proc_task(inode);
3820 generic_fillattr(&init_user_ns, inode, stat);
3821
3822 if (p) {
3823 stat->nlink += get_nr_threads(p);
3824 put_task_struct(p);
3825 }
3826
3827 return 0;
3828}
3829
3830static const struct inode_operations proc_task_inode_operations = {
3831 .lookup = proc_task_lookup,
3832 .getattr = proc_task_getattr,
3833 .setattr = proc_setattr,
3834 .permission = proc_pid_permission,
3835};
3836
3837static const struct file_operations proc_task_operations = {
3838 .read = generic_read_dir,
3839 .iterate_shared = proc_task_readdir,
3840 .llseek = generic_file_llseek,
3841};
3842
3843void __init set_proc_pid_nlink(void)
3844{
3845 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3846 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3847}
1/*
2 * linux/fs/proc/base.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * proc base directory handling functions
7 *
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
14 *
15 *
16 * Changelog:
17 * 17-Jan-2005
18 * Allan Bezerra
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
32 *
33 * Changelog:
34 * 21-Feb-2005
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
37 *
38 * ChangeLog:
39 * 10-Mar-2005
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
45 *
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
48 */
49
50#include <linux/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/printk.h>
77#include <linux/cgroup.h>
78#include <linux/cpuset.h>
79#include <linux/audit.h>
80#include <linux/poll.h>
81#include <linux/nsproxy.h>
82#include <linux/oom.h>
83#include <linux/elf.h>
84#include <linux/pid_namespace.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87#include <linux/slab.h>
88#include <linux/flex_array.h>
89#include <linux/posix-timers.h>
90#ifdef CONFIG_HARDWALL
91#include <asm/hardwall.h>
92#endif
93#include <trace/events/oom.h>
94#include "internal.h"
95#include "fd.h"
96
97/* NOTE:
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
102 *
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
105 */
106
107static u8 nlink_tid;
108static u8 nlink_tgid;
109
110struct pid_entry {
111 const char *name;
112 unsigned int len;
113 umode_t mode;
114 const struct inode_operations *iop;
115 const struct file_operations *fop;
116 union proc_op op;
117};
118
119#define NOD(NAME, MODE, IOP, FOP, OP) { \
120 .name = (NAME), \
121 .len = sizeof(NAME) - 1, \
122 .mode = MODE, \
123 .iop = IOP, \
124 .fop = FOP, \
125 .op = OP, \
126}
127
128#define DIR(NAME, MODE, iops, fops) \
129 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
130#define LNK(NAME, get_link) \
131 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
132 &proc_pid_link_inode_operations, NULL, \
133 { .proc_get_link = get_link } )
134#define REG(NAME, MODE, fops) \
135 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
136#define ONE(NAME, MODE, show) \
137 NOD(NAME, (S_IFREG|(MODE)), \
138 NULL, &proc_single_file_operations, \
139 { .proc_show = show } )
140
141/*
142 * Count the number of hardlinks for the pid_entry table, excluding the .
143 * and .. links.
144 */
145static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
146 unsigned int n)
147{
148 unsigned int i;
149 unsigned int count;
150
151 count = 2;
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
154 ++count;
155 }
156
157 return count;
158}
159
160static int get_task_root(struct task_struct *task, struct path *root)
161{
162 int result = -ENOENT;
163
164 task_lock(task);
165 if (task->fs) {
166 get_fs_root(task->fs, root);
167 result = 0;
168 }
169 task_unlock(task);
170 return result;
171}
172
173static int proc_cwd_link(struct dentry *dentry, struct path *path)
174{
175 struct task_struct *task = get_proc_task(d_inode(dentry));
176 int result = -ENOENT;
177
178 if (task) {
179 task_lock(task);
180 if (task->fs) {
181 get_fs_pwd(task->fs, path);
182 result = 0;
183 }
184 task_unlock(task);
185 put_task_struct(task);
186 }
187 return result;
188}
189
190static int proc_root_link(struct dentry *dentry, struct path *path)
191{
192 struct task_struct *task = get_proc_task(d_inode(dentry));
193 int result = -ENOENT;
194
195 if (task) {
196 result = get_task_root(task, path);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
203 size_t _count, loff_t *pos)
204{
205 struct task_struct *tsk;
206 struct mm_struct *mm;
207 char *page;
208 unsigned long count = _count;
209 unsigned long arg_start, arg_end, env_start, env_end;
210 unsigned long len1, len2, len;
211 unsigned long p;
212 char c;
213 ssize_t rv;
214
215 BUG_ON(*pos < 0);
216
217 tsk = get_proc_task(file_inode(file));
218 if (!tsk)
219 return -ESRCH;
220 mm = get_task_mm(tsk);
221 put_task_struct(tsk);
222 if (!mm)
223 return 0;
224 /* Check if process spawned far enough to have cmdline. */
225 if (!mm->env_end) {
226 rv = 0;
227 goto out_mmput;
228 }
229
230 page = (char *)__get_free_page(GFP_TEMPORARY);
231 if (!page) {
232 rv = -ENOMEM;
233 goto out_mmput;
234 }
235
236 down_read(&mm->mmap_sem);
237 arg_start = mm->arg_start;
238 arg_end = mm->arg_end;
239 env_start = mm->env_start;
240 env_end = mm->env_end;
241 up_read(&mm->mmap_sem);
242
243 BUG_ON(arg_start > arg_end);
244 BUG_ON(env_start > env_end);
245
246 len1 = arg_end - arg_start;
247 len2 = env_end - env_start;
248
249 /* Empty ARGV. */
250 if (len1 == 0) {
251 rv = 0;
252 goto out_free_page;
253 }
254 /*
255 * Inherently racy -- command line shares address space
256 * with code and data.
257 */
258 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
259 if (rv <= 0)
260 goto out_free_page;
261
262 rv = 0;
263
264 if (c == '\0') {
265 /* Command line (set of strings) occupies whole ARGV. */
266 if (len1 <= *pos)
267 goto out_free_page;
268
269 p = arg_start + *pos;
270 len = len1 - *pos;
271 while (count > 0 && len > 0) {
272 unsigned int _count;
273 int nr_read;
274
275 _count = min3(count, len, PAGE_SIZE);
276 nr_read = access_remote_vm(mm, p, page, _count, 0);
277 if (nr_read < 0)
278 rv = nr_read;
279 if (nr_read <= 0)
280 goto out_free_page;
281
282 if (copy_to_user(buf, page, nr_read)) {
283 rv = -EFAULT;
284 goto out_free_page;
285 }
286
287 p += nr_read;
288 len -= nr_read;
289 buf += nr_read;
290 count -= nr_read;
291 rv += nr_read;
292 }
293 } else {
294 /*
295 * Command line (1 string) occupies ARGV and maybe
296 * extends into ENVP.
297 */
298 if (len1 + len2 <= *pos)
299 goto skip_argv_envp;
300 if (len1 <= *pos)
301 goto skip_argv;
302
303 p = arg_start + *pos;
304 len = len1 - *pos;
305 while (count > 0 && len > 0) {
306 unsigned int _count, l;
307 int nr_read;
308 bool final;
309
310 _count = min3(count, len, PAGE_SIZE);
311 nr_read = access_remote_vm(mm, p, page, _count, 0);
312 if (nr_read < 0)
313 rv = nr_read;
314 if (nr_read <= 0)
315 goto out_free_page;
316
317 /*
318 * Command line can be shorter than whole ARGV
319 * even if last "marker" byte says it is not.
320 */
321 final = false;
322 l = strnlen(page, nr_read);
323 if (l < nr_read) {
324 nr_read = l;
325 final = true;
326 }
327
328 if (copy_to_user(buf, page, nr_read)) {
329 rv = -EFAULT;
330 goto out_free_page;
331 }
332
333 p += nr_read;
334 len -= nr_read;
335 buf += nr_read;
336 count -= nr_read;
337 rv += nr_read;
338
339 if (final)
340 goto out_free_page;
341 }
342skip_argv:
343 /*
344 * Command line (1 string) occupies ARGV and
345 * extends into ENVP.
346 */
347 if (len1 <= *pos) {
348 p = env_start + *pos - len1;
349 len = len1 + len2 - *pos;
350 } else {
351 p = env_start;
352 len = len2;
353 }
354 while (count > 0 && len > 0) {
355 unsigned int _count, l;
356 int nr_read;
357 bool final;
358
359 _count = min3(count, len, PAGE_SIZE);
360 nr_read = access_remote_vm(mm, p, page, _count, 0);
361 if (nr_read < 0)
362 rv = nr_read;
363 if (nr_read <= 0)
364 goto out_free_page;
365
366 /* Find EOS. */
367 final = false;
368 l = strnlen(page, nr_read);
369 if (l < nr_read) {
370 nr_read = l;
371 final = true;
372 }
373
374 if (copy_to_user(buf, page, nr_read)) {
375 rv = -EFAULT;
376 goto out_free_page;
377 }
378
379 p += nr_read;
380 len -= nr_read;
381 buf += nr_read;
382 count -= nr_read;
383 rv += nr_read;
384
385 if (final)
386 goto out_free_page;
387 }
388skip_argv_envp:
389 ;
390 }
391
392out_free_page:
393 free_page((unsigned long)page);
394out_mmput:
395 mmput(mm);
396 if (rv > 0)
397 *pos += rv;
398 return rv;
399}
400
401static const struct file_operations proc_pid_cmdline_ops = {
402 .read = proc_pid_cmdline_read,
403 .llseek = generic_file_llseek,
404};
405
406#ifdef CONFIG_KALLSYMS
407/*
408 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
409 * Returns the resolved symbol. If that fails, simply return the address.
410 */
411static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
412 struct pid *pid, struct task_struct *task)
413{
414 unsigned long wchan;
415 char symname[KSYM_NAME_LEN];
416
417 wchan = get_wchan(task);
418
419 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
420 && !lookup_symbol_name(wchan, symname))
421 seq_printf(m, "%s", symname);
422 else
423 seq_putc(m, '0');
424
425 return 0;
426}
427#endif /* CONFIG_KALLSYMS */
428
429static int lock_trace(struct task_struct *task)
430{
431 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
432 if (err)
433 return err;
434 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
435 mutex_unlock(&task->signal->cred_guard_mutex);
436 return -EPERM;
437 }
438 return 0;
439}
440
441static void unlock_trace(struct task_struct *task)
442{
443 mutex_unlock(&task->signal->cred_guard_mutex);
444}
445
446#ifdef CONFIG_STACKTRACE
447
448#define MAX_STACK_TRACE_DEPTH 64
449
450static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
451 struct pid *pid, struct task_struct *task)
452{
453 struct stack_trace trace;
454 unsigned long *entries;
455 int err;
456 int i;
457
458 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
459 if (!entries)
460 return -ENOMEM;
461
462 trace.nr_entries = 0;
463 trace.max_entries = MAX_STACK_TRACE_DEPTH;
464 trace.entries = entries;
465 trace.skip = 0;
466
467 err = lock_trace(task);
468 if (!err) {
469 save_stack_trace_tsk(task, &trace);
470
471 for (i = 0; i < trace.nr_entries; i++) {
472 seq_printf(m, "[<%pK>] %pB\n",
473 (void *)entries[i], (void *)entries[i]);
474 }
475 unlock_trace(task);
476 }
477 kfree(entries);
478
479 return err;
480}
481#endif
482
483#ifdef CONFIG_SCHED_INFO
484/*
485 * Provides /proc/PID/schedstat
486 */
487static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
488 struct pid *pid, struct task_struct *task)
489{
490 if (unlikely(!sched_info_on()))
491 seq_printf(m, "0 0 0\n");
492 else
493 seq_printf(m, "%llu %llu %lu\n",
494 (unsigned long long)task->se.sum_exec_runtime,
495 (unsigned long long)task->sched_info.run_delay,
496 task->sched_info.pcount);
497
498 return 0;
499}
500#endif
501
502#ifdef CONFIG_LATENCYTOP
503static int lstats_show_proc(struct seq_file *m, void *v)
504{
505 int i;
506 struct inode *inode = m->private;
507 struct task_struct *task = get_proc_task(inode);
508
509 if (!task)
510 return -ESRCH;
511 seq_puts(m, "Latency Top version : v0.1\n");
512 for (i = 0; i < 32; i++) {
513 struct latency_record *lr = &task->latency_record[i];
514 if (lr->backtrace[0]) {
515 int q;
516 seq_printf(m, "%i %li %li",
517 lr->count, lr->time, lr->max);
518 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
519 unsigned long bt = lr->backtrace[q];
520 if (!bt)
521 break;
522 if (bt == ULONG_MAX)
523 break;
524 seq_printf(m, " %ps", (void *)bt);
525 }
526 seq_putc(m, '\n');
527 }
528
529 }
530 put_task_struct(task);
531 return 0;
532}
533
534static int lstats_open(struct inode *inode, struct file *file)
535{
536 return single_open(file, lstats_show_proc, inode);
537}
538
539static ssize_t lstats_write(struct file *file, const char __user *buf,
540 size_t count, loff_t *offs)
541{
542 struct task_struct *task = get_proc_task(file_inode(file));
543
544 if (!task)
545 return -ESRCH;
546 clear_all_latency_tracing(task);
547 put_task_struct(task);
548
549 return count;
550}
551
552static const struct file_operations proc_lstats_operations = {
553 .open = lstats_open,
554 .read = seq_read,
555 .write = lstats_write,
556 .llseek = seq_lseek,
557 .release = single_release,
558};
559
560#endif
561
562static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
563 struct pid *pid, struct task_struct *task)
564{
565 unsigned long totalpages = totalram_pages + total_swap_pages;
566 unsigned long points = 0;
567
568 points = oom_badness(task, NULL, NULL, totalpages) *
569 1000 / totalpages;
570 seq_printf(m, "%lu\n", points);
571
572 return 0;
573}
574
575struct limit_names {
576 const char *name;
577 const char *unit;
578};
579
580static const struct limit_names lnames[RLIM_NLIMITS] = {
581 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
582 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
583 [RLIMIT_DATA] = {"Max data size", "bytes"},
584 [RLIMIT_STACK] = {"Max stack size", "bytes"},
585 [RLIMIT_CORE] = {"Max core file size", "bytes"},
586 [RLIMIT_RSS] = {"Max resident set", "bytes"},
587 [RLIMIT_NPROC] = {"Max processes", "processes"},
588 [RLIMIT_NOFILE] = {"Max open files", "files"},
589 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
590 [RLIMIT_AS] = {"Max address space", "bytes"},
591 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
592 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
593 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
594 [RLIMIT_NICE] = {"Max nice priority", NULL},
595 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
596 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
597};
598
599/* Display limits for a process */
600static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
601 struct pid *pid, struct task_struct *task)
602{
603 unsigned int i;
604 unsigned long flags;
605
606 struct rlimit rlim[RLIM_NLIMITS];
607
608 if (!lock_task_sighand(task, &flags))
609 return 0;
610 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
611 unlock_task_sighand(task, &flags);
612
613 /*
614 * print the file header
615 */
616 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
617 "Limit", "Soft Limit", "Hard Limit", "Units");
618
619 for (i = 0; i < RLIM_NLIMITS; i++) {
620 if (rlim[i].rlim_cur == RLIM_INFINITY)
621 seq_printf(m, "%-25s %-20s ",
622 lnames[i].name, "unlimited");
623 else
624 seq_printf(m, "%-25s %-20lu ",
625 lnames[i].name, rlim[i].rlim_cur);
626
627 if (rlim[i].rlim_max == RLIM_INFINITY)
628 seq_printf(m, "%-20s ", "unlimited");
629 else
630 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
631
632 if (lnames[i].unit)
633 seq_printf(m, "%-10s\n", lnames[i].unit);
634 else
635 seq_putc(m, '\n');
636 }
637
638 return 0;
639}
640
641#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
642static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
643 struct pid *pid, struct task_struct *task)
644{
645 long nr;
646 unsigned long args[6], sp, pc;
647 int res;
648
649 res = lock_trace(task);
650 if (res)
651 return res;
652
653 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
654 seq_puts(m, "running\n");
655 else if (nr < 0)
656 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
657 else
658 seq_printf(m,
659 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
660 nr,
661 args[0], args[1], args[2], args[3], args[4], args[5],
662 sp, pc);
663 unlock_trace(task);
664
665 return 0;
666}
667#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
668
669/************************************************************************/
670/* Here the fs part begins */
671/************************************************************************/
672
673/* permission checks */
674static int proc_fd_access_allowed(struct inode *inode)
675{
676 struct task_struct *task;
677 int allowed = 0;
678 /* Allow access to a task's file descriptors if it is us or we
679 * may use ptrace attach to the process and find out that
680 * information.
681 */
682 task = get_proc_task(inode);
683 if (task) {
684 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
685 put_task_struct(task);
686 }
687 return allowed;
688}
689
690int proc_setattr(struct dentry *dentry, struct iattr *attr)
691{
692 int error;
693 struct inode *inode = d_inode(dentry);
694
695 if (attr->ia_valid & ATTR_MODE)
696 return -EPERM;
697
698 error = setattr_prepare(dentry, attr);
699 if (error)
700 return error;
701
702 setattr_copy(inode, attr);
703 mark_inode_dirty(inode);
704 return 0;
705}
706
707/*
708 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
709 * or euid/egid (for hide_pid_min=2)?
710 */
711static bool has_pid_permissions(struct pid_namespace *pid,
712 struct task_struct *task,
713 int hide_pid_min)
714{
715 if (pid->hide_pid < hide_pid_min)
716 return true;
717 if (in_group_p(pid->pid_gid))
718 return true;
719 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
720}
721
722
723static int proc_pid_permission(struct inode *inode, int mask)
724{
725 struct pid_namespace *pid = inode->i_sb->s_fs_info;
726 struct task_struct *task;
727 bool has_perms;
728
729 task = get_proc_task(inode);
730 if (!task)
731 return -ESRCH;
732 has_perms = has_pid_permissions(pid, task, 1);
733 put_task_struct(task);
734
735 if (!has_perms) {
736 if (pid->hide_pid == 2) {
737 /*
738 * Let's make getdents(), stat(), and open()
739 * consistent with each other. If a process
740 * may not stat() a file, it shouldn't be seen
741 * in procfs at all.
742 */
743 return -ENOENT;
744 }
745
746 return -EPERM;
747 }
748 return generic_permission(inode, mask);
749}
750
751
752
753static const struct inode_operations proc_def_inode_operations = {
754 .setattr = proc_setattr,
755};
756
757static int proc_single_show(struct seq_file *m, void *v)
758{
759 struct inode *inode = m->private;
760 struct pid_namespace *ns;
761 struct pid *pid;
762 struct task_struct *task;
763 int ret;
764
765 ns = inode->i_sb->s_fs_info;
766 pid = proc_pid(inode);
767 task = get_pid_task(pid, PIDTYPE_PID);
768 if (!task)
769 return -ESRCH;
770
771 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
772
773 put_task_struct(task);
774 return ret;
775}
776
777static int proc_single_open(struct inode *inode, struct file *filp)
778{
779 return single_open(filp, proc_single_show, inode);
780}
781
782static const struct file_operations proc_single_file_operations = {
783 .open = proc_single_open,
784 .read = seq_read,
785 .llseek = seq_lseek,
786 .release = single_release,
787};
788
789
790struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
791{
792 struct task_struct *task = get_proc_task(inode);
793 struct mm_struct *mm = ERR_PTR(-ESRCH);
794
795 if (task) {
796 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
797 put_task_struct(task);
798
799 if (!IS_ERR_OR_NULL(mm)) {
800 /* ensure this mm_struct can't be freed */
801 atomic_inc(&mm->mm_count);
802 /* but do not pin its memory */
803 mmput(mm);
804 }
805 }
806
807 return mm;
808}
809
810static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
811{
812 struct mm_struct *mm = proc_mem_open(inode, mode);
813
814 if (IS_ERR(mm))
815 return PTR_ERR(mm);
816
817 file->private_data = mm;
818 return 0;
819}
820
821static int mem_open(struct inode *inode, struct file *file)
822{
823 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
824
825 /* OK to pass negative loff_t, we can catch out-of-range */
826 file->f_mode |= FMODE_UNSIGNED_OFFSET;
827
828 return ret;
829}
830
831static ssize_t mem_rw(struct file *file, char __user *buf,
832 size_t count, loff_t *ppos, int write)
833{
834 struct mm_struct *mm = file->private_data;
835 unsigned long addr = *ppos;
836 ssize_t copied;
837 char *page;
838 unsigned int flags;
839
840 if (!mm)
841 return 0;
842
843 page = (char *)__get_free_page(GFP_TEMPORARY);
844 if (!page)
845 return -ENOMEM;
846
847 copied = 0;
848 if (!atomic_inc_not_zero(&mm->mm_users))
849 goto free;
850
851 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
852 flags = FOLL_FORCE;
853 if (write)
854 flags |= FOLL_WRITE;
855
856 while (count > 0) {
857 int this_len = min_t(int, count, PAGE_SIZE);
858
859 if (write && copy_from_user(page, buf, this_len)) {
860 copied = -EFAULT;
861 break;
862 }
863
864 this_len = access_remote_vm(mm, addr, page, this_len, flags);
865 if (!this_len) {
866 if (!copied)
867 copied = -EIO;
868 break;
869 }
870
871 if (!write && copy_to_user(buf, page, this_len)) {
872 copied = -EFAULT;
873 break;
874 }
875
876 buf += this_len;
877 addr += this_len;
878 copied += this_len;
879 count -= this_len;
880 }
881 *ppos = addr;
882
883 mmput(mm);
884free:
885 free_page((unsigned long) page);
886 return copied;
887}
888
889static ssize_t mem_read(struct file *file, char __user *buf,
890 size_t count, loff_t *ppos)
891{
892 return mem_rw(file, buf, count, ppos, 0);
893}
894
895static ssize_t mem_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
897{
898 return mem_rw(file, (char __user*)buf, count, ppos, 1);
899}
900
901loff_t mem_lseek(struct file *file, loff_t offset, int orig)
902{
903 switch (orig) {
904 case 0:
905 file->f_pos = offset;
906 break;
907 case 1:
908 file->f_pos += offset;
909 break;
910 default:
911 return -EINVAL;
912 }
913 force_successful_syscall_return();
914 return file->f_pos;
915}
916
917static int mem_release(struct inode *inode, struct file *file)
918{
919 struct mm_struct *mm = file->private_data;
920 if (mm)
921 mmdrop(mm);
922 return 0;
923}
924
925static const struct file_operations proc_mem_operations = {
926 .llseek = mem_lseek,
927 .read = mem_read,
928 .write = mem_write,
929 .open = mem_open,
930 .release = mem_release,
931};
932
933static int environ_open(struct inode *inode, struct file *file)
934{
935 return __mem_open(inode, file, PTRACE_MODE_READ);
936}
937
938static ssize_t environ_read(struct file *file, char __user *buf,
939 size_t count, loff_t *ppos)
940{
941 char *page;
942 unsigned long src = *ppos;
943 int ret = 0;
944 struct mm_struct *mm = file->private_data;
945 unsigned long env_start, env_end;
946
947 /* Ensure the process spawned far enough to have an environment. */
948 if (!mm || !mm->env_end)
949 return 0;
950
951 page = (char *)__get_free_page(GFP_TEMPORARY);
952 if (!page)
953 return -ENOMEM;
954
955 ret = 0;
956 if (!atomic_inc_not_zero(&mm->mm_users))
957 goto free;
958
959 down_read(&mm->mmap_sem);
960 env_start = mm->env_start;
961 env_end = mm->env_end;
962 up_read(&mm->mmap_sem);
963
964 while (count > 0) {
965 size_t this_len, max_len;
966 int retval;
967
968 if (src >= (env_end - env_start))
969 break;
970
971 this_len = env_end - (env_start + src);
972
973 max_len = min_t(size_t, PAGE_SIZE, count);
974 this_len = min(max_len, this_len);
975
976 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
977
978 if (retval <= 0) {
979 ret = retval;
980 break;
981 }
982
983 if (copy_to_user(buf, page, retval)) {
984 ret = -EFAULT;
985 break;
986 }
987
988 ret += retval;
989 src += retval;
990 buf += retval;
991 count -= retval;
992 }
993 *ppos = src;
994 mmput(mm);
995
996free:
997 free_page((unsigned long) page);
998 return ret;
999}
1000
1001static const struct file_operations proc_environ_operations = {
1002 .open = environ_open,
1003 .read = environ_read,
1004 .llseek = generic_file_llseek,
1005 .release = mem_release,
1006};
1007
1008static int auxv_open(struct inode *inode, struct file *file)
1009{
1010 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1011}
1012
1013static ssize_t auxv_read(struct file *file, char __user *buf,
1014 size_t count, loff_t *ppos)
1015{
1016 struct mm_struct *mm = file->private_data;
1017 unsigned int nwords = 0;
1018
1019 if (!mm)
1020 return 0;
1021 do {
1022 nwords += 2;
1023 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1024 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1025 nwords * sizeof(mm->saved_auxv[0]));
1026}
1027
1028static const struct file_operations proc_auxv_operations = {
1029 .open = auxv_open,
1030 .read = auxv_read,
1031 .llseek = generic_file_llseek,
1032 .release = mem_release,
1033};
1034
1035static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1036 loff_t *ppos)
1037{
1038 struct task_struct *task = get_proc_task(file_inode(file));
1039 char buffer[PROC_NUMBUF];
1040 int oom_adj = OOM_ADJUST_MIN;
1041 size_t len;
1042
1043 if (!task)
1044 return -ESRCH;
1045 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1046 oom_adj = OOM_ADJUST_MAX;
1047 else
1048 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1049 OOM_SCORE_ADJ_MAX;
1050 put_task_struct(task);
1051 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1052 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1053}
1054
1055static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1056{
1057 static DEFINE_MUTEX(oom_adj_mutex);
1058 struct mm_struct *mm = NULL;
1059 struct task_struct *task;
1060 int err = 0;
1061
1062 task = get_proc_task(file_inode(file));
1063 if (!task)
1064 return -ESRCH;
1065
1066 mutex_lock(&oom_adj_mutex);
1067 if (legacy) {
1068 if (oom_adj < task->signal->oom_score_adj &&
1069 !capable(CAP_SYS_RESOURCE)) {
1070 err = -EACCES;
1071 goto err_unlock;
1072 }
1073 /*
1074 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1075 * /proc/pid/oom_score_adj instead.
1076 */
1077 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1078 current->comm, task_pid_nr(current), task_pid_nr(task),
1079 task_pid_nr(task));
1080 } else {
1081 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1082 !capable(CAP_SYS_RESOURCE)) {
1083 err = -EACCES;
1084 goto err_unlock;
1085 }
1086 }
1087
1088 /*
1089 * Make sure we will check other processes sharing the mm if this is
1090 * not vfrok which wants its own oom_score_adj.
1091 * pin the mm so it doesn't go away and get reused after task_unlock
1092 */
1093 if (!task->vfork_done) {
1094 struct task_struct *p = find_lock_task_mm(task);
1095
1096 if (p) {
1097 if (atomic_read(&p->mm->mm_users) > 1) {
1098 mm = p->mm;
1099 atomic_inc(&mm->mm_count);
1100 }
1101 task_unlock(p);
1102 }
1103 }
1104
1105 task->signal->oom_score_adj = oom_adj;
1106 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1107 task->signal->oom_score_adj_min = (short)oom_adj;
1108 trace_oom_score_adj_update(task);
1109
1110 if (mm) {
1111 struct task_struct *p;
1112
1113 rcu_read_lock();
1114 for_each_process(p) {
1115 if (same_thread_group(task, p))
1116 continue;
1117
1118 /* do not touch kernel threads or the global init */
1119 if (p->flags & PF_KTHREAD || is_global_init(p))
1120 continue;
1121
1122 task_lock(p);
1123 if (!p->vfork_done && process_shares_mm(p, mm)) {
1124 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1125 task_pid_nr(p), p->comm,
1126 p->signal->oom_score_adj, oom_adj,
1127 task_pid_nr(task), task->comm);
1128 p->signal->oom_score_adj = oom_adj;
1129 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1130 p->signal->oom_score_adj_min = (short)oom_adj;
1131 }
1132 task_unlock(p);
1133 }
1134 rcu_read_unlock();
1135 mmdrop(mm);
1136 }
1137err_unlock:
1138 mutex_unlock(&oom_adj_mutex);
1139 put_task_struct(task);
1140 return err;
1141}
1142
1143/*
1144 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1145 * kernels. The effective policy is defined by oom_score_adj, which has a
1146 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1147 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1148 * Processes that become oom disabled via oom_adj will still be oom disabled
1149 * with this implementation.
1150 *
1151 * oom_adj cannot be removed since existing userspace binaries use it.
1152 */
1153static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1154 size_t count, loff_t *ppos)
1155{
1156 char buffer[PROC_NUMBUF];
1157 int oom_adj;
1158 int err;
1159
1160 memset(buffer, 0, sizeof(buffer));
1161 if (count > sizeof(buffer) - 1)
1162 count = sizeof(buffer) - 1;
1163 if (copy_from_user(buffer, buf, count)) {
1164 err = -EFAULT;
1165 goto out;
1166 }
1167
1168 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1169 if (err)
1170 goto out;
1171 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1172 oom_adj != OOM_DISABLE) {
1173 err = -EINVAL;
1174 goto out;
1175 }
1176
1177 /*
1178 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1179 * value is always attainable.
1180 */
1181 if (oom_adj == OOM_ADJUST_MAX)
1182 oom_adj = OOM_SCORE_ADJ_MAX;
1183 else
1184 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1185
1186 err = __set_oom_adj(file, oom_adj, true);
1187out:
1188 return err < 0 ? err : count;
1189}
1190
1191static const struct file_operations proc_oom_adj_operations = {
1192 .read = oom_adj_read,
1193 .write = oom_adj_write,
1194 .llseek = generic_file_llseek,
1195};
1196
1197static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1198 size_t count, loff_t *ppos)
1199{
1200 struct task_struct *task = get_proc_task(file_inode(file));
1201 char buffer[PROC_NUMBUF];
1202 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1203 size_t len;
1204
1205 if (!task)
1206 return -ESRCH;
1207 oom_score_adj = task->signal->oom_score_adj;
1208 put_task_struct(task);
1209 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1210 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1211}
1212
1213static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1214 size_t count, loff_t *ppos)
1215{
1216 char buffer[PROC_NUMBUF];
1217 int oom_score_adj;
1218 int err;
1219
1220 memset(buffer, 0, sizeof(buffer));
1221 if (count > sizeof(buffer) - 1)
1222 count = sizeof(buffer) - 1;
1223 if (copy_from_user(buffer, buf, count)) {
1224 err = -EFAULT;
1225 goto out;
1226 }
1227
1228 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1229 if (err)
1230 goto out;
1231 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1232 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1233 err = -EINVAL;
1234 goto out;
1235 }
1236
1237 err = __set_oom_adj(file, oom_score_adj, false);
1238out:
1239 return err < 0 ? err : count;
1240}
1241
1242static const struct file_operations proc_oom_score_adj_operations = {
1243 .read = oom_score_adj_read,
1244 .write = oom_score_adj_write,
1245 .llseek = default_llseek,
1246};
1247
1248#ifdef CONFIG_AUDITSYSCALL
1249#define TMPBUFLEN 11
1250static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1251 size_t count, loff_t *ppos)
1252{
1253 struct inode * inode = file_inode(file);
1254 struct task_struct *task = get_proc_task(inode);
1255 ssize_t length;
1256 char tmpbuf[TMPBUFLEN];
1257
1258 if (!task)
1259 return -ESRCH;
1260 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1261 from_kuid(file->f_cred->user_ns,
1262 audit_get_loginuid(task)));
1263 put_task_struct(task);
1264 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1265}
1266
1267static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1268 size_t count, loff_t *ppos)
1269{
1270 struct inode * inode = file_inode(file);
1271 uid_t loginuid;
1272 kuid_t kloginuid;
1273 int rv;
1274
1275 rcu_read_lock();
1276 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1277 rcu_read_unlock();
1278 return -EPERM;
1279 }
1280 rcu_read_unlock();
1281
1282 if (*ppos != 0) {
1283 /* No partial writes. */
1284 return -EINVAL;
1285 }
1286
1287 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1288 if (rv < 0)
1289 return rv;
1290
1291 /* is userspace tring to explicitly UNSET the loginuid? */
1292 if (loginuid == AUDIT_UID_UNSET) {
1293 kloginuid = INVALID_UID;
1294 } else {
1295 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1296 if (!uid_valid(kloginuid))
1297 return -EINVAL;
1298 }
1299
1300 rv = audit_set_loginuid(kloginuid);
1301 if (rv < 0)
1302 return rv;
1303 return count;
1304}
1305
1306static const struct file_operations proc_loginuid_operations = {
1307 .read = proc_loginuid_read,
1308 .write = proc_loginuid_write,
1309 .llseek = generic_file_llseek,
1310};
1311
1312static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1313 size_t count, loff_t *ppos)
1314{
1315 struct inode * inode = file_inode(file);
1316 struct task_struct *task = get_proc_task(inode);
1317 ssize_t length;
1318 char tmpbuf[TMPBUFLEN];
1319
1320 if (!task)
1321 return -ESRCH;
1322 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1323 audit_get_sessionid(task));
1324 put_task_struct(task);
1325 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1326}
1327
1328static const struct file_operations proc_sessionid_operations = {
1329 .read = proc_sessionid_read,
1330 .llseek = generic_file_llseek,
1331};
1332#endif
1333
1334#ifdef CONFIG_FAULT_INJECTION
1335static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1336 size_t count, loff_t *ppos)
1337{
1338 struct task_struct *task = get_proc_task(file_inode(file));
1339 char buffer[PROC_NUMBUF];
1340 size_t len;
1341 int make_it_fail;
1342
1343 if (!task)
1344 return -ESRCH;
1345 make_it_fail = task->make_it_fail;
1346 put_task_struct(task);
1347
1348 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1349
1350 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1351}
1352
1353static ssize_t proc_fault_inject_write(struct file * file,
1354 const char __user * buf, size_t count, loff_t *ppos)
1355{
1356 struct task_struct *task;
1357 char buffer[PROC_NUMBUF];
1358 int make_it_fail;
1359 int rv;
1360
1361 if (!capable(CAP_SYS_RESOURCE))
1362 return -EPERM;
1363 memset(buffer, 0, sizeof(buffer));
1364 if (count > sizeof(buffer) - 1)
1365 count = sizeof(buffer) - 1;
1366 if (copy_from_user(buffer, buf, count))
1367 return -EFAULT;
1368 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1369 if (rv < 0)
1370 return rv;
1371 if (make_it_fail < 0 || make_it_fail > 1)
1372 return -EINVAL;
1373
1374 task = get_proc_task(file_inode(file));
1375 if (!task)
1376 return -ESRCH;
1377 task->make_it_fail = make_it_fail;
1378 put_task_struct(task);
1379
1380 return count;
1381}
1382
1383static const struct file_operations proc_fault_inject_operations = {
1384 .read = proc_fault_inject_read,
1385 .write = proc_fault_inject_write,
1386 .llseek = generic_file_llseek,
1387};
1388#endif
1389
1390
1391#ifdef CONFIG_SCHED_DEBUG
1392/*
1393 * Print out various scheduling related per-task fields:
1394 */
1395static int sched_show(struct seq_file *m, void *v)
1396{
1397 struct inode *inode = m->private;
1398 struct task_struct *p;
1399
1400 p = get_proc_task(inode);
1401 if (!p)
1402 return -ESRCH;
1403 proc_sched_show_task(p, m);
1404
1405 put_task_struct(p);
1406
1407 return 0;
1408}
1409
1410static ssize_t
1411sched_write(struct file *file, const char __user *buf,
1412 size_t count, loff_t *offset)
1413{
1414 struct inode *inode = file_inode(file);
1415 struct task_struct *p;
1416
1417 p = get_proc_task(inode);
1418 if (!p)
1419 return -ESRCH;
1420 proc_sched_set_task(p);
1421
1422 put_task_struct(p);
1423
1424 return count;
1425}
1426
1427static int sched_open(struct inode *inode, struct file *filp)
1428{
1429 return single_open(filp, sched_show, inode);
1430}
1431
1432static const struct file_operations proc_pid_sched_operations = {
1433 .open = sched_open,
1434 .read = seq_read,
1435 .write = sched_write,
1436 .llseek = seq_lseek,
1437 .release = single_release,
1438};
1439
1440#endif
1441
1442#ifdef CONFIG_SCHED_AUTOGROUP
1443/*
1444 * Print out autogroup related information:
1445 */
1446static int sched_autogroup_show(struct seq_file *m, void *v)
1447{
1448 struct inode *inode = m->private;
1449 struct task_struct *p;
1450
1451 p = get_proc_task(inode);
1452 if (!p)
1453 return -ESRCH;
1454 proc_sched_autogroup_show_task(p, m);
1455
1456 put_task_struct(p);
1457
1458 return 0;
1459}
1460
1461static ssize_t
1462sched_autogroup_write(struct file *file, const char __user *buf,
1463 size_t count, loff_t *offset)
1464{
1465 struct inode *inode = file_inode(file);
1466 struct task_struct *p;
1467 char buffer[PROC_NUMBUF];
1468 int nice;
1469 int err;
1470
1471 memset(buffer, 0, sizeof(buffer));
1472 if (count > sizeof(buffer) - 1)
1473 count = sizeof(buffer) - 1;
1474 if (copy_from_user(buffer, buf, count))
1475 return -EFAULT;
1476
1477 err = kstrtoint(strstrip(buffer), 0, &nice);
1478 if (err < 0)
1479 return err;
1480
1481 p = get_proc_task(inode);
1482 if (!p)
1483 return -ESRCH;
1484
1485 err = proc_sched_autogroup_set_nice(p, nice);
1486 if (err)
1487 count = err;
1488
1489 put_task_struct(p);
1490
1491 return count;
1492}
1493
1494static int sched_autogroup_open(struct inode *inode, struct file *filp)
1495{
1496 int ret;
1497
1498 ret = single_open(filp, sched_autogroup_show, NULL);
1499 if (!ret) {
1500 struct seq_file *m = filp->private_data;
1501
1502 m->private = inode;
1503 }
1504 return ret;
1505}
1506
1507static const struct file_operations proc_pid_sched_autogroup_operations = {
1508 .open = sched_autogroup_open,
1509 .read = seq_read,
1510 .write = sched_autogroup_write,
1511 .llseek = seq_lseek,
1512 .release = single_release,
1513};
1514
1515#endif /* CONFIG_SCHED_AUTOGROUP */
1516
1517static ssize_t comm_write(struct file *file, const char __user *buf,
1518 size_t count, loff_t *offset)
1519{
1520 struct inode *inode = file_inode(file);
1521 struct task_struct *p;
1522 char buffer[TASK_COMM_LEN];
1523 const size_t maxlen = sizeof(buffer) - 1;
1524
1525 memset(buffer, 0, sizeof(buffer));
1526 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1527 return -EFAULT;
1528
1529 p = get_proc_task(inode);
1530 if (!p)
1531 return -ESRCH;
1532
1533 if (same_thread_group(current, p))
1534 set_task_comm(p, buffer);
1535 else
1536 count = -EINVAL;
1537
1538 put_task_struct(p);
1539
1540 return count;
1541}
1542
1543static int comm_show(struct seq_file *m, void *v)
1544{
1545 struct inode *inode = m->private;
1546 struct task_struct *p;
1547
1548 p = get_proc_task(inode);
1549 if (!p)
1550 return -ESRCH;
1551
1552 task_lock(p);
1553 seq_printf(m, "%s\n", p->comm);
1554 task_unlock(p);
1555
1556 put_task_struct(p);
1557
1558 return 0;
1559}
1560
1561static int comm_open(struct inode *inode, struct file *filp)
1562{
1563 return single_open(filp, comm_show, inode);
1564}
1565
1566static const struct file_operations proc_pid_set_comm_operations = {
1567 .open = comm_open,
1568 .read = seq_read,
1569 .write = comm_write,
1570 .llseek = seq_lseek,
1571 .release = single_release,
1572};
1573
1574static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1575{
1576 struct task_struct *task;
1577 struct file *exe_file;
1578
1579 task = get_proc_task(d_inode(dentry));
1580 if (!task)
1581 return -ENOENT;
1582 exe_file = get_task_exe_file(task);
1583 put_task_struct(task);
1584 if (exe_file) {
1585 *exe_path = exe_file->f_path;
1586 path_get(&exe_file->f_path);
1587 fput(exe_file);
1588 return 0;
1589 } else
1590 return -ENOENT;
1591}
1592
1593static const char *proc_pid_get_link(struct dentry *dentry,
1594 struct inode *inode,
1595 struct delayed_call *done)
1596{
1597 struct path path;
1598 int error = -EACCES;
1599
1600 if (!dentry)
1601 return ERR_PTR(-ECHILD);
1602
1603 /* Are we allowed to snoop on the tasks file descriptors? */
1604 if (!proc_fd_access_allowed(inode))
1605 goto out;
1606
1607 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1608 if (error)
1609 goto out;
1610
1611 nd_jump_link(&path);
1612 return NULL;
1613out:
1614 return ERR_PTR(error);
1615}
1616
1617static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1618{
1619 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1620 char *pathname;
1621 int len;
1622
1623 if (!tmp)
1624 return -ENOMEM;
1625
1626 pathname = d_path(path, tmp, PAGE_SIZE);
1627 len = PTR_ERR(pathname);
1628 if (IS_ERR(pathname))
1629 goto out;
1630 len = tmp + PAGE_SIZE - 1 - pathname;
1631
1632 if (len > buflen)
1633 len = buflen;
1634 if (copy_to_user(buffer, pathname, len))
1635 len = -EFAULT;
1636 out:
1637 free_page((unsigned long)tmp);
1638 return len;
1639}
1640
1641static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1642{
1643 int error = -EACCES;
1644 struct inode *inode = d_inode(dentry);
1645 struct path path;
1646
1647 /* Are we allowed to snoop on the tasks file descriptors? */
1648 if (!proc_fd_access_allowed(inode))
1649 goto out;
1650
1651 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1652 if (error)
1653 goto out;
1654
1655 error = do_proc_readlink(&path, buffer, buflen);
1656 path_put(&path);
1657out:
1658 return error;
1659}
1660
1661const struct inode_operations proc_pid_link_inode_operations = {
1662 .readlink = proc_pid_readlink,
1663 .get_link = proc_pid_get_link,
1664 .setattr = proc_setattr,
1665};
1666
1667
1668/* building an inode */
1669
1670struct inode *proc_pid_make_inode(struct super_block * sb,
1671 struct task_struct *task, umode_t mode)
1672{
1673 struct inode * inode;
1674 struct proc_inode *ei;
1675 const struct cred *cred;
1676
1677 /* We need a new inode */
1678
1679 inode = new_inode(sb);
1680 if (!inode)
1681 goto out;
1682
1683 /* Common stuff */
1684 ei = PROC_I(inode);
1685 inode->i_mode = mode;
1686 inode->i_ino = get_next_ino();
1687 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1688 inode->i_op = &proc_def_inode_operations;
1689
1690 /*
1691 * grab the reference to task.
1692 */
1693 ei->pid = get_task_pid(task, PIDTYPE_PID);
1694 if (!ei->pid)
1695 goto out_unlock;
1696
1697 if (task_dumpable(task)) {
1698 rcu_read_lock();
1699 cred = __task_cred(task);
1700 inode->i_uid = cred->euid;
1701 inode->i_gid = cred->egid;
1702 rcu_read_unlock();
1703 }
1704 security_task_to_inode(task, inode);
1705
1706out:
1707 return inode;
1708
1709out_unlock:
1710 iput(inode);
1711 return NULL;
1712}
1713
1714int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1715{
1716 struct inode *inode = d_inode(dentry);
1717 struct task_struct *task;
1718 const struct cred *cred;
1719 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1720
1721 generic_fillattr(inode, stat);
1722
1723 rcu_read_lock();
1724 stat->uid = GLOBAL_ROOT_UID;
1725 stat->gid = GLOBAL_ROOT_GID;
1726 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1727 if (task) {
1728 if (!has_pid_permissions(pid, task, 2)) {
1729 rcu_read_unlock();
1730 /*
1731 * This doesn't prevent learning whether PID exists,
1732 * it only makes getattr() consistent with readdir().
1733 */
1734 return -ENOENT;
1735 }
1736 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1737 task_dumpable(task)) {
1738 cred = __task_cred(task);
1739 stat->uid = cred->euid;
1740 stat->gid = cred->egid;
1741 }
1742 }
1743 rcu_read_unlock();
1744 return 0;
1745}
1746
1747/* dentry stuff */
1748
1749/*
1750 * Exceptional case: normally we are not allowed to unhash a busy
1751 * directory. In this case, however, we can do it - no aliasing problems
1752 * due to the way we treat inodes.
1753 *
1754 * Rewrite the inode's ownerships here because the owning task may have
1755 * performed a setuid(), etc.
1756 *
1757 * Before the /proc/pid/status file was created the only way to read
1758 * the effective uid of a /process was to stat /proc/pid. Reading
1759 * /proc/pid/status is slow enough that procps and other packages
1760 * kept stating /proc/pid. To keep the rules in /proc simple I have
1761 * made this apply to all per process world readable and executable
1762 * directories.
1763 */
1764int pid_revalidate(struct dentry *dentry, unsigned int flags)
1765{
1766 struct inode *inode;
1767 struct task_struct *task;
1768 const struct cred *cred;
1769
1770 if (flags & LOOKUP_RCU)
1771 return -ECHILD;
1772
1773 inode = d_inode(dentry);
1774 task = get_proc_task(inode);
1775
1776 if (task) {
1777 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1778 task_dumpable(task)) {
1779 rcu_read_lock();
1780 cred = __task_cred(task);
1781 inode->i_uid = cred->euid;
1782 inode->i_gid = cred->egid;
1783 rcu_read_unlock();
1784 } else {
1785 inode->i_uid = GLOBAL_ROOT_UID;
1786 inode->i_gid = GLOBAL_ROOT_GID;
1787 }
1788 inode->i_mode &= ~(S_ISUID | S_ISGID);
1789 security_task_to_inode(task, inode);
1790 put_task_struct(task);
1791 return 1;
1792 }
1793 return 0;
1794}
1795
1796static inline bool proc_inode_is_dead(struct inode *inode)
1797{
1798 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1799}
1800
1801int pid_delete_dentry(const struct dentry *dentry)
1802{
1803 /* Is the task we represent dead?
1804 * If so, then don't put the dentry on the lru list,
1805 * kill it immediately.
1806 */
1807 return proc_inode_is_dead(d_inode(dentry));
1808}
1809
1810const struct dentry_operations pid_dentry_operations =
1811{
1812 .d_revalidate = pid_revalidate,
1813 .d_delete = pid_delete_dentry,
1814};
1815
1816/* Lookups */
1817
1818/*
1819 * Fill a directory entry.
1820 *
1821 * If possible create the dcache entry and derive our inode number and
1822 * file type from dcache entry.
1823 *
1824 * Since all of the proc inode numbers are dynamically generated, the inode
1825 * numbers do not exist until the inode is cache. This means creating the
1826 * the dcache entry in readdir is necessary to keep the inode numbers
1827 * reported by readdir in sync with the inode numbers reported
1828 * by stat.
1829 */
1830bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1831 const char *name, int len,
1832 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1833{
1834 struct dentry *child, *dir = file->f_path.dentry;
1835 struct qstr qname = QSTR_INIT(name, len);
1836 struct inode *inode;
1837 unsigned type;
1838 ino_t ino;
1839
1840 child = d_hash_and_lookup(dir, &qname);
1841 if (!child) {
1842 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1843 child = d_alloc_parallel(dir, &qname, &wq);
1844 if (IS_ERR(child))
1845 goto end_instantiate;
1846 if (d_in_lookup(child)) {
1847 int err = instantiate(d_inode(dir), child, task, ptr);
1848 d_lookup_done(child);
1849 if (err < 0) {
1850 dput(child);
1851 goto end_instantiate;
1852 }
1853 }
1854 }
1855 inode = d_inode(child);
1856 ino = inode->i_ino;
1857 type = inode->i_mode >> 12;
1858 dput(child);
1859 return dir_emit(ctx, name, len, ino, type);
1860
1861end_instantiate:
1862 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1863}
1864
1865/*
1866 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1867 * which represent vma start and end addresses.
1868 */
1869static int dname_to_vma_addr(struct dentry *dentry,
1870 unsigned long *start, unsigned long *end)
1871{
1872 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1873 return -EINVAL;
1874
1875 return 0;
1876}
1877
1878static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1879{
1880 unsigned long vm_start, vm_end;
1881 bool exact_vma_exists = false;
1882 struct mm_struct *mm = NULL;
1883 struct task_struct *task;
1884 const struct cred *cred;
1885 struct inode *inode;
1886 int status = 0;
1887
1888 if (flags & LOOKUP_RCU)
1889 return -ECHILD;
1890
1891 inode = d_inode(dentry);
1892 task = get_proc_task(inode);
1893 if (!task)
1894 goto out_notask;
1895
1896 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1897 if (IS_ERR_OR_NULL(mm))
1898 goto out;
1899
1900 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1901 down_read(&mm->mmap_sem);
1902 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1903 up_read(&mm->mmap_sem);
1904 }
1905
1906 mmput(mm);
1907
1908 if (exact_vma_exists) {
1909 if (task_dumpable(task)) {
1910 rcu_read_lock();
1911 cred = __task_cred(task);
1912 inode->i_uid = cred->euid;
1913 inode->i_gid = cred->egid;
1914 rcu_read_unlock();
1915 } else {
1916 inode->i_uid = GLOBAL_ROOT_UID;
1917 inode->i_gid = GLOBAL_ROOT_GID;
1918 }
1919 security_task_to_inode(task, inode);
1920 status = 1;
1921 }
1922
1923out:
1924 put_task_struct(task);
1925
1926out_notask:
1927 return status;
1928}
1929
1930static const struct dentry_operations tid_map_files_dentry_operations = {
1931 .d_revalidate = map_files_d_revalidate,
1932 .d_delete = pid_delete_dentry,
1933};
1934
1935static int map_files_get_link(struct dentry *dentry, struct path *path)
1936{
1937 unsigned long vm_start, vm_end;
1938 struct vm_area_struct *vma;
1939 struct task_struct *task;
1940 struct mm_struct *mm;
1941 int rc;
1942
1943 rc = -ENOENT;
1944 task = get_proc_task(d_inode(dentry));
1945 if (!task)
1946 goto out;
1947
1948 mm = get_task_mm(task);
1949 put_task_struct(task);
1950 if (!mm)
1951 goto out;
1952
1953 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1954 if (rc)
1955 goto out_mmput;
1956
1957 rc = -ENOENT;
1958 down_read(&mm->mmap_sem);
1959 vma = find_exact_vma(mm, vm_start, vm_end);
1960 if (vma && vma->vm_file) {
1961 *path = vma->vm_file->f_path;
1962 path_get(path);
1963 rc = 0;
1964 }
1965 up_read(&mm->mmap_sem);
1966
1967out_mmput:
1968 mmput(mm);
1969out:
1970 return rc;
1971}
1972
1973struct map_files_info {
1974 fmode_t mode;
1975 unsigned int len;
1976 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1977};
1978
1979/*
1980 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1981 * symlinks may be used to bypass permissions on ancestor directories in the
1982 * path to the file in question.
1983 */
1984static const char *
1985proc_map_files_get_link(struct dentry *dentry,
1986 struct inode *inode,
1987 struct delayed_call *done)
1988{
1989 if (!capable(CAP_SYS_ADMIN))
1990 return ERR_PTR(-EPERM);
1991
1992 return proc_pid_get_link(dentry, inode, done);
1993}
1994
1995/*
1996 * Identical to proc_pid_link_inode_operations except for get_link()
1997 */
1998static const struct inode_operations proc_map_files_link_inode_operations = {
1999 .readlink = proc_pid_readlink,
2000 .get_link = proc_map_files_get_link,
2001 .setattr = proc_setattr,
2002};
2003
2004static int
2005proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2006 struct task_struct *task, const void *ptr)
2007{
2008 fmode_t mode = (fmode_t)(unsigned long)ptr;
2009 struct proc_inode *ei;
2010 struct inode *inode;
2011
2012 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2013 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2014 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2015 if (!inode)
2016 return -ENOENT;
2017
2018 ei = PROC_I(inode);
2019 ei->op.proc_get_link = map_files_get_link;
2020
2021 inode->i_op = &proc_map_files_link_inode_operations;
2022 inode->i_size = 64;
2023
2024 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2025 d_add(dentry, inode);
2026
2027 return 0;
2028}
2029
2030static struct dentry *proc_map_files_lookup(struct inode *dir,
2031 struct dentry *dentry, unsigned int flags)
2032{
2033 unsigned long vm_start, vm_end;
2034 struct vm_area_struct *vma;
2035 struct task_struct *task;
2036 int result;
2037 struct mm_struct *mm;
2038
2039 result = -ENOENT;
2040 task = get_proc_task(dir);
2041 if (!task)
2042 goto out;
2043
2044 result = -EACCES;
2045 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2046 goto out_put_task;
2047
2048 result = -ENOENT;
2049 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2050 goto out_put_task;
2051
2052 mm = get_task_mm(task);
2053 if (!mm)
2054 goto out_put_task;
2055
2056 down_read(&mm->mmap_sem);
2057 vma = find_exact_vma(mm, vm_start, vm_end);
2058 if (!vma)
2059 goto out_no_vma;
2060
2061 if (vma->vm_file)
2062 result = proc_map_files_instantiate(dir, dentry, task,
2063 (void *)(unsigned long)vma->vm_file->f_mode);
2064
2065out_no_vma:
2066 up_read(&mm->mmap_sem);
2067 mmput(mm);
2068out_put_task:
2069 put_task_struct(task);
2070out:
2071 return ERR_PTR(result);
2072}
2073
2074static const struct inode_operations proc_map_files_inode_operations = {
2075 .lookup = proc_map_files_lookup,
2076 .permission = proc_fd_permission,
2077 .setattr = proc_setattr,
2078};
2079
2080static int
2081proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2082{
2083 struct vm_area_struct *vma;
2084 struct task_struct *task;
2085 struct mm_struct *mm;
2086 unsigned long nr_files, pos, i;
2087 struct flex_array *fa = NULL;
2088 struct map_files_info info;
2089 struct map_files_info *p;
2090 int ret;
2091
2092 ret = -ENOENT;
2093 task = get_proc_task(file_inode(file));
2094 if (!task)
2095 goto out;
2096
2097 ret = -EACCES;
2098 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2099 goto out_put_task;
2100
2101 ret = 0;
2102 if (!dir_emit_dots(file, ctx))
2103 goto out_put_task;
2104
2105 mm = get_task_mm(task);
2106 if (!mm)
2107 goto out_put_task;
2108 down_read(&mm->mmap_sem);
2109
2110 nr_files = 0;
2111
2112 /*
2113 * We need two passes here:
2114 *
2115 * 1) Collect vmas of mapped files with mmap_sem taken
2116 * 2) Release mmap_sem and instantiate entries
2117 *
2118 * otherwise we get lockdep complained, since filldir()
2119 * routine might require mmap_sem taken in might_fault().
2120 */
2121
2122 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2123 if (vma->vm_file && ++pos > ctx->pos)
2124 nr_files++;
2125 }
2126
2127 if (nr_files) {
2128 fa = flex_array_alloc(sizeof(info), nr_files,
2129 GFP_KERNEL);
2130 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2131 GFP_KERNEL)) {
2132 ret = -ENOMEM;
2133 if (fa)
2134 flex_array_free(fa);
2135 up_read(&mm->mmap_sem);
2136 mmput(mm);
2137 goto out_put_task;
2138 }
2139 for (i = 0, vma = mm->mmap, pos = 2; vma;
2140 vma = vma->vm_next) {
2141 if (!vma->vm_file)
2142 continue;
2143 if (++pos <= ctx->pos)
2144 continue;
2145
2146 info.mode = vma->vm_file->f_mode;
2147 info.len = snprintf(info.name,
2148 sizeof(info.name), "%lx-%lx",
2149 vma->vm_start, vma->vm_end);
2150 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2151 BUG();
2152 }
2153 }
2154 up_read(&mm->mmap_sem);
2155
2156 for (i = 0; i < nr_files; i++) {
2157 p = flex_array_get(fa, i);
2158 if (!proc_fill_cache(file, ctx,
2159 p->name, p->len,
2160 proc_map_files_instantiate,
2161 task,
2162 (void *)(unsigned long)p->mode))
2163 break;
2164 ctx->pos++;
2165 }
2166 if (fa)
2167 flex_array_free(fa);
2168 mmput(mm);
2169
2170out_put_task:
2171 put_task_struct(task);
2172out:
2173 return ret;
2174}
2175
2176static const struct file_operations proc_map_files_operations = {
2177 .read = generic_read_dir,
2178 .iterate_shared = proc_map_files_readdir,
2179 .llseek = generic_file_llseek,
2180};
2181
2182#ifdef CONFIG_CHECKPOINT_RESTORE
2183struct timers_private {
2184 struct pid *pid;
2185 struct task_struct *task;
2186 struct sighand_struct *sighand;
2187 struct pid_namespace *ns;
2188 unsigned long flags;
2189};
2190
2191static void *timers_start(struct seq_file *m, loff_t *pos)
2192{
2193 struct timers_private *tp = m->private;
2194
2195 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2196 if (!tp->task)
2197 return ERR_PTR(-ESRCH);
2198
2199 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2200 if (!tp->sighand)
2201 return ERR_PTR(-ESRCH);
2202
2203 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2204}
2205
2206static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2207{
2208 struct timers_private *tp = m->private;
2209 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2210}
2211
2212static void timers_stop(struct seq_file *m, void *v)
2213{
2214 struct timers_private *tp = m->private;
2215
2216 if (tp->sighand) {
2217 unlock_task_sighand(tp->task, &tp->flags);
2218 tp->sighand = NULL;
2219 }
2220
2221 if (tp->task) {
2222 put_task_struct(tp->task);
2223 tp->task = NULL;
2224 }
2225}
2226
2227static int show_timer(struct seq_file *m, void *v)
2228{
2229 struct k_itimer *timer;
2230 struct timers_private *tp = m->private;
2231 int notify;
2232 static const char * const nstr[] = {
2233 [SIGEV_SIGNAL] = "signal",
2234 [SIGEV_NONE] = "none",
2235 [SIGEV_THREAD] = "thread",
2236 };
2237
2238 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2239 notify = timer->it_sigev_notify;
2240
2241 seq_printf(m, "ID: %d\n", timer->it_id);
2242 seq_printf(m, "signal: %d/%p\n",
2243 timer->sigq->info.si_signo,
2244 timer->sigq->info.si_value.sival_ptr);
2245 seq_printf(m, "notify: %s/%s.%d\n",
2246 nstr[notify & ~SIGEV_THREAD_ID],
2247 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2248 pid_nr_ns(timer->it_pid, tp->ns));
2249 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2250
2251 return 0;
2252}
2253
2254static const struct seq_operations proc_timers_seq_ops = {
2255 .start = timers_start,
2256 .next = timers_next,
2257 .stop = timers_stop,
2258 .show = show_timer,
2259};
2260
2261static int proc_timers_open(struct inode *inode, struct file *file)
2262{
2263 struct timers_private *tp;
2264
2265 tp = __seq_open_private(file, &proc_timers_seq_ops,
2266 sizeof(struct timers_private));
2267 if (!tp)
2268 return -ENOMEM;
2269
2270 tp->pid = proc_pid(inode);
2271 tp->ns = inode->i_sb->s_fs_info;
2272 return 0;
2273}
2274
2275static const struct file_operations proc_timers_operations = {
2276 .open = proc_timers_open,
2277 .read = seq_read,
2278 .llseek = seq_lseek,
2279 .release = seq_release_private,
2280};
2281#endif
2282
2283static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2284 size_t count, loff_t *offset)
2285{
2286 struct inode *inode = file_inode(file);
2287 struct task_struct *p;
2288 u64 slack_ns;
2289 int err;
2290
2291 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2292 if (err < 0)
2293 return err;
2294
2295 p = get_proc_task(inode);
2296 if (!p)
2297 return -ESRCH;
2298
2299 if (p != current) {
2300 if (!capable(CAP_SYS_NICE)) {
2301 count = -EPERM;
2302 goto out;
2303 }
2304
2305 err = security_task_setscheduler(p);
2306 if (err) {
2307 count = err;
2308 goto out;
2309 }
2310 }
2311
2312 task_lock(p);
2313 if (slack_ns == 0)
2314 p->timer_slack_ns = p->default_timer_slack_ns;
2315 else
2316 p->timer_slack_ns = slack_ns;
2317 task_unlock(p);
2318
2319out:
2320 put_task_struct(p);
2321
2322 return count;
2323}
2324
2325static int timerslack_ns_show(struct seq_file *m, void *v)
2326{
2327 struct inode *inode = m->private;
2328 struct task_struct *p;
2329 int err = 0;
2330
2331 p = get_proc_task(inode);
2332 if (!p)
2333 return -ESRCH;
2334
2335 if (p != current) {
2336
2337 if (!capable(CAP_SYS_NICE)) {
2338 err = -EPERM;
2339 goto out;
2340 }
2341 err = security_task_getscheduler(p);
2342 if (err)
2343 goto out;
2344 }
2345
2346 task_lock(p);
2347 seq_printf(m, "%llu\n", p->timer_slack_ns);
2348 task_unlock(p);
2349
2350out:
2351 put_task_struct(p);
2352
2353 return err;
2354}
2355
2356static int timerslack_ns_open(struct inode *inode, struct file *filp)
2357{
2358 return single_open(filp, timerslack_ns_show, inode);
2359}
2360
2361static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2362 .open = timerslack_ns_open,
2363 .read = seq_read,
2364 .write = timerslack_ns_write,
2365 .llseek = seq_lseek,
2366 .release = single_release,
2367};
2368
2369static int proc_pident_instantiate(struct inode *dir,
2370 struct dentry *dentry, struct task_struct *task, const void *ptr)
2371{
2372 const struct pid_entry *p = ptr;
2373 struct inode *inode;
2374 struct proc_inode *ei;
2375
2376 inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2377 if (!inode)
2378 goto out;
2379
2380 ei = PROC_I(inode);
2381 if (S_ISDIR(inode->i_mode))
2382 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2383 if (p->iop)
2384 inode->i_op = p->iop;
2385 if (p->fop)
2386 inode->i_fop = p->fop;
2387 ei->op = p->op;
2388 d_set_d_op(dentry, &pid_dentry_operations);
2389 d_add(dentry, inode);
2390 /* Close the race of the process dying before we return the dentry */
2391 if (pid_revalidate(dentry, 0))
2392 return 0;
2393out:
2394 return -ENOENT;
2395}
2396
2397static struct dentry *proc_pident_lookup(struct inode *dir,
2398 struct dentry *dentry,
2399 const struct pid_entry *ents,
2400 unsigned int nents)
2401{
2402 int error;
2403 struct task_struct *task = get_proc_task(dir);
2404 const struct pid_entry *p, *last;
2405
2406 error = -ENOENT;
2407
2408 if (!task)
2409 goto out_no_task;
2410
2411 /*
2412 * Yes, it does not scale. And it should not. Don't add
2413 * new entries into /proc/<tgid>/ without very good reasons.
2414 */
2415 last = &ents[nents];
2416 for (p = ents; p < last; p++) {
2417 if (p->len != dentry->d_name.len)
2418 continue;
2419 if (!memcmp(dentry->d_name.name, p->name, p->len))
2420 break;
2421 }
2422 if (p >= last)
2423 goto out;
2424
2425 error = proc_pident_instantiate(dir, dentry, task, p);
2426out:
2427 put_task_struct(task);
2428out_no_task:
2429 return ERR_PTR(error);
2430}
2431
2432static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2433 const struct pid_entry *ents, unsigned int nents)
2434{
2435 struct task_struct *task = get_proc_task(file_inode(file));
2436 const struct pid_entry *p;
2437
2438 if (!task)
2439 return -ENOENT;
2440
2441 if (!dir_emit_dots(file, ctx))
2442 goto out;
2443
2444 if (ctx->pos >= nents + 2)
2445 goto out;
2446
2447 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2448 if (!proc_fill_cache(file, ctx, p->name, p->len,
2449 proc_pident_instantiate, task, p))
2450 break;
2451 ctx->pos++;
2452 }
2453out:
2454 put_task_struct(task);
2455 return 0;
2456}
2457
2458#ifdef CONFIG_SECURITY
2459static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2460 size_t count, loff_t *ppos)
2461{
2462 struct inode * inode = file_inode(file);
2463 char *p = NULL;
2464 ssize_t length;
2465 struct task_struct *task = get_proc_task(inode);
2466
2467 if (!task)
2468 return -ESRCH;
2469
2470 length = security_getprocattr(task,
2471 (char*)file->f_path.dentry->d_name.name,
2472 &p);
2473 put_task_struct(task);
2474 if (length > 0)
2475 length = simple_read_from_buffer(buf, count, ppos, p, length);
2476 kfree(p);
2477 return length;
2478}
2479
2480static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2481 size_t count, loff_t *ppos)
2482{
2483 struct inode * inode = file_inode(file);
2484 void *page;
2485 ssize_t length;
2486 struct task_struct *task = get_proc_task(inode);
2487
2488 length = -ESRCH;
2489 if (!task)
2490 goto out_no_task;
2491 if (count > PAGE_SIZE)
2492 count = PAGE_SIZE;
2493
2494 /* No partial writes. */
2495 length = -EINVAL;
2496 if (*ppos != 0)
2497 goto out;
2498
2499 page = memdup_user(buf, count);
2500 if (IS_ERR(page)) {
2501 length = PTR_ERR(page);
2502 goto out;
2503 }
2504
2505 /* Guard against adverse ptrace interaction */
2506 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2507 if (length < 0)
2508 goto out_free;
2509
2510 length = security_setprocattr(task,
2511 (char*)file->f_path.dentry->d_name.name,
2512 page, count);
2513 mutex_unlock(&task->signal->cred_guard_mutex);
2514out_free:
2515 kfree(page);
2516out:
2517 put_task_struct(task);
2518out_no_task:
2519 return length;
2520}
2521
2522static const struct file_operations proc_pid_attr_operations = {
2523 .read = proc_pid_attr_read,
2524 .write = proc_pid_attr_write,
2525 .llseek = generic_file_llseek,
2526};
2527
2528static const struct pid_entry attr_dir_stuff[] = {
2529 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2530 REG("prev", S_IRUGO, proc_pid_attr_operations),
2531 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2532 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2534 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2535};
2536
2537static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2538{
2539 return proc_pident_readdir(file, ctx,
2540 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2541}
2542
2543static const struct file_operations proc_attr_dir_operations = {
2544 .read = generic_read_dir,
2545 .iterate_shared = proc_attr_dir_readdir,
2546 .llseek = generic_file_llseek,
2547};
2548
2549static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2550 struct dentry *dentry, unsigned int flags)
2551{
2552 return proc_pident_lookup(dir, dentry,
2553 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2554}
2555
2556static const struct inode_operations proc_attr_dir_inode_operations = {
2557 .lookup = proc_attr_dir_lookup,
2558 .getattr = pid_getattr,
2559 .setattr = proc_setattr,
2560};
2561
2562#endif
2563
2564#ifdef CONFIG_ELF_CORE
2565static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2566 size_t count, loff_t *ppos)
2567{
2568 struct task_struct *task = get_proc_task(file_inode(file));
2569 struct mm_struct *mm;
2570 char buffer[PROC_NUMBUF];
2571 size_t len;
2572 int ret;
2573
2574 if (!task)
2575 return -ESRCH;
2576
2577 ret = 0;
2578 mm = get_task_mm(task);
2579 if (mm) {
2580 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2581 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2582 MMF_DUMP_FILTER_SHIFT));
2583 mmput(mm);
2584 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2585 }
2586
2587 put_task_struct(task);
2588
2589 return ret;
2590}
2591
2592static ssize_t proc_coredump_filter_write(struct file *file,
2593 const char __user *buf,
2594 size_t count,
2595 loff_t *ppos)
2596{
2597 struct task_struct *task;
2598 struct mm_struct *mm;
2599 unsigned int val;
2600 int ret;
2601 int i;
2602 unsigned long mask;
2603
2604 ret = kstrtouint_from_user(buf, count, 0, &val);
2605 if (ret < 0)
2606 return ret;
2607
2608 ret = -ESRCH;
2609 task = get_proc_task(file_inode(file));
2610 if (!task)
2611 goto out_no_task;
2612
2613 mm = get_task_mm(task);
2614 if (!mm)
2615 goto out_no_mm;
2616 ret = 0;
2617
2618 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2619 if (val & mask)
2620 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2621 else
2622 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2623 }
2624
2625 mmput(mm);
2626 out_no_mm:
2627 put_task_struct(task);
2628 out_no_task:
2629 if (ret < 0)
2630 return ret;
2631 return count;
2632}
2633
2634static const struct file_operations proc_coredump_filter_operations = {
2635 .read = proc_coredump_filter_read,
2636 .write = proc_coredump_filter_write,
2637 .llseek = generic_file_llseek,
2638};
2639#endif
2640
2641#ifdef CONFIG_TASK_IO_ACCOUNTING
2642static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2643{
2644 struct task_io_accounting acct = task->ioac;
2645 unsigned long flags;
2646 int result;
2647
2648 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2649 if (result)
2650 return result;
2651
2652 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2653 result = -EACCES;
2654 goto out_unlock;
2655 }
2656
2657 if (whole && lock_task_sighand(task, &flags)) {
2658 struct task_struct *t = task;
2659
2660 task_io_accounting_add(&acct, &task->signal->ioac);
2661 while_each_thread(task, t)
2662 task_io_accounting_add(&acct, &t->ioac);
2663
2664 unlock_task_sighand(task, &flags);
2665 }
2666 seq_printf(m,
2667 "rchar: %llu\n"
2668 "wchar: %llu\n"
2669 "syscr: %llu\n"
2670 "syscw: %llu\n"
2671 "read_bytes: %llu\n"
2672 "write_bytes: %llu\n"
2673 "cancelled_write_bytes: %llu\n",
2674 (unsigned long long)acct.rchar,
2675 (unsigned long long)acct.wchar,
2676 (unsigned long long)acct.syscr,
2677 (unsigned long long)acct.syscw,
2678 (unsigned long long)acct.read_bytes,
2679 (unsigned long long)acct.write_bytes,
2680 (unsigned long long)acct.cancelled_write_bytes);
2681 result = 0;
2682
2683out_unlock:
2684 mutex_unlock(&task->signal->cred_guard_mutex);
2685 return result;
2686}
2687
2688static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2689 struct pid *pid, struct task_struct *task)
2690{
2691 return do_io_accounting(task, m, 0);
2692}
2693
2694static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2695 struct pid *pid, struct task_struct *task)
2696{
2697 return do_io_accounting(task, m, 1);
2698}
2699#endif /* CONFIG_TASK_IO_ACCOUNTING */
2700
2701#ifdef CONFIG_USER_NS
2702static int proc_id_map_open(struct inode *inode, struct file *file,
2703 const struct seq_operations *seq_ops)
2704{
2705 struct user_namespace *ns = NULL;
2706 struct task_struct *task;
2707 struct seq_file *seq;
2708 int ret = -EINVAL;
2709
2710 task = get_proc_task(inode);
2711 if (task) {
2712 rcu_read_lock();
2713 ns = get_user_ns(task_cred_xxx(task, user_ns));
2714 rcu_read_unlock();
2715 put_task_struct(task);
2716 }
2717 if (!ns)
2718 goto err;
2719
2720 ret = seq_open(file, seq_ops);
2721 if (ret)
2722 goto err_put_ns;
2723
2724 seq = file->private_data;
2725 seq->private = ns;
2726
2727 return 0;
2728err_put_ns:
2729 put_user_ns(ns);
2730err:
2731 return ret;
2732}
2733
2734static int proc_id_map_release(struct inode *inode, struct file *file)
2735{
2736 struct seq_file *seq = file->private_data;
2737 struct user_namespace *ns = seq->private;
2738 put_user_ns(ns);
2739 return seq_release(inode, file);
2740}
2741
2742static int proc_uid_map_open(struct inode *inode, struct file *file)
2743{
2744 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2745}
2746
2747static int proc_gid_map_open(struct inode *inode, struct file *file)
2748{
2749 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2750}
2751
2752static int proc_projid_map_open(struct inode *inode, struct file *file)
2753{
2754 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2755}
2756
2757static const struct file_operations proc_uid_map_operations = {
2758 .open = proc_uid_map_open,
2759 .write = proc_uid_map_write,
2760 .read = seq_read,
2761 .llseek = seq_lseek,
2762 .release = proc_id_map_release,
2763};
2764
2765static const struct file_operations proc_gid_map_operations = {
2766 .open = proc_gid_map_open,
2767 .write = proc_gid_map_write,
2768 .read = seq_read,
2769 .llseek = seq_lseek,
2770 .release = proc_id_map_release,
2771};
2772
2773static const struct file_operations proc_projid_map_operations = {
2774 .open = proc_projid_map_open,
2775 .write = proc_projid_map_write,
2776 .read = seq_read,
2777 .llseek = seq_lseek,
2778 .release = proc_id_map_release,
2779};
2780
2781static int proc_setgroups_open(struct inode *inode, struct file *file)
2782{
2783 struct user_namespace *ns = NULL;
2784 struct task_struct *task;
2785 int ret;
2786
2787 ret = -ESRCH;
2788 task = get_proc_task(inode);
2789 if (task) {
2790 rcu_read_lock();
2791 ns = get_user_ns(task_cred_xxx(task, user_ns));
2792 rcu_read_unlock();
2793 put_task_struct(task);
2794 }
2795 if (!ns)
2796 goto err;
2797
2798 if (file->f_mode & FMODE_WRITE) {
2799 ret = -EACCES;
2800 if (!ns_capable(ns, CAP_SYS_ADMIN))
2801 goto err_put_ns;
2802 }
2803
2804 ret = single_open(file, &proc_setgroups_show, ns);
2805 if (ret)
2806 goto err_put_ns;
2807
2808 return 0;
2809err_put_ns:
2810 put_user_ns(ns);
2811err:
2812 return ret;
2813}
2814
2815static int proc_setgroups_release(struct inode *inode, struct file *file)
2816{
2817 struct seq_file *seq = file->private_data;
2818 struct user_namespace *ns = seq->private;
2819 int ret = single_release(inode, file);
2820 put_user_ns(ns);
2821 return ret;
2822}
2823
2824static const struct file_operations proc_setgroups_operations = {
2825 .open = proc_setgroups_open,
2826 .write = proc_setgroups_write,
2827 .read = seq_read,
2828 .llseek = seq_lseek,
2829 .release = proc_setgroups_release,
2830};
2831#endif /* CONFIG_USER_NS */
2832
2833static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2834 struct pid *pid, struct task_struct *task)
2835{
2836 int err = lock_trace(task);
2837 if (!err) {
2838 seq_printf(m, "%08x\n", task->personality);
2839 unlock_trace(task);
2840 }
2841 return err;
2842}
2843
2844/*
2845 * Thread groups
2846 */
2847static const struct file_operations proc_task_operations;
2848static const struct inode_operations proc_task_inode_operations;
2849
2850static const struct pid_entry tgid_base_stuff[] = {
2851 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2852 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2853 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2854 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2855 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2856#ifdef CONFIG_NET
2857 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2858#endif
2859 REG("environ", S_IRUSR, proc_environ_operations),
2860 REG("auxv", S_IRUSR, proc_auxv_operations),
2861 ONE("status", S_IRUGO, proc_pid_status),
2862 ONE("personality", S_IRUSR, proc_pid_personality),
2863 ONE("limits", S_IRUGO, proc_pid_limits),
2864#ifdef CONFIG_SCHED_DEBUG
2865 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2866#endif
2867#ifdef CONFIG_SCHED_AUTOGROUP
2868 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2869#endif
2870 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2871#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2872 ONE("syscall", S_IRUSR, proc_pid_syscall),
2873#endif
2874 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2875 ONE("stat", S_IRUGO, proc_tgid_stat),
2876 ONE("statm", S_IRUGO, proc_pid_statm),
2877 REG("maps", S_IRUGO, proc_pid_maps_operations),
2878#ifdef CONFIG_NUMA
2879 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2880#endif
2881 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2882 LNK("cwd", proc_cwd_link),
2883 LNK("root", proc_root_link),
2884 LNK("exe", proc_exe_link),
2885 REG("mounts", S_IRUGO, proc_mounts_operations),
2886 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2887 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2888#ifdef CONFIG_PROC_PAGE_MONITOR
2889 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2890 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2891 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2892#endif
2893#ifdef CONFIG_SECURITY
2894 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2895#endif
2896#ifdef CONFIG_KALLSYMS
2897 ONE("wchan", S_IRUGO, proc_pid_wchan),
2898#endif
2899#ifdef CONFIG_STACKTRACE
2900 ONE("stack", S_IRUSR, proc_pid_stack),
2901#endif
2902#ifdef CONFIG_SCHED_INFO
2903 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2904#endif
2905#ifdef CONFIG_LATENCYTOP
2906 REG("latency", S_IRUGO, proc_lstats_operations),
2907#endif
2908#ifdef CONFIG_PROC_PID_CPUSET
2909 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2910#endif
2911#ifdef CONFIG_CGROUPS
2912 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2913#endif
2914 ONE("oom_score", S_IRUGO, proc_oom_score),
2915 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2916 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2917#ifdef CONFIG_AUDITSYSCALL
2918 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2919 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2920#endif
2921#ifdef CONFIG_FAULT_INJECTION
2922 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2923#endif
2924#ifdef CONFIG_ELF_CORE
2925 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2926#endif
2927#ifdef CONFIG_TASK_IO_ACCOUNTING
2928 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2929#endif
2930#ifdef CONFIG_HARDWALL
2931 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2932#endif
2933#ifdef CONFIG_USER_NS
2934 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2935 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2936 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2937 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2938#endif
2939#ifdef CONFIG_CHECKPOINT_RESTORE
2940 REG("timers", S_IRUGO, proc_timers_operations),
2941#endif
2942 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2943};
2944
2945static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2946{
2947 return proc_pident_readdir(file, ctx,
2948 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2949}
2950
2951static const struct file_operations proc_tgid_base_operations = {
2952 .read = generic_read_dir,
2953 .iterate_shared = proc_tgid_base_readdir,
2954 .llseek = generic_file_llseek,
2955};
2956
2957static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2958{
2959 return proc_pident_lookup(dir, dentry,
2960 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2961}
2962
2963static const struct inode_operations proc_tgid_base_inode_operations = {
2964 .lookup = proc_tgid_base_lookup,
2965 .getattr = pid_getattr,
2966 .setattr = proc_setattr,
2967 .permission = proc_pid_permission,
2968};
2969
2970static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2971{
2972 struct dentry *dentry, *leader, *dir;
2973 char buf[PROC_NUMBUF];
2974 struct qstr name;
2975
2976 name.name = buf;
2977 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2978 /* no ->d_hash() rejects on procfs */
2979 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2980 if (dentry) {
2981 d_invalidate(dentry);
2982 dput(dentry);
2983 }
2984
2985 if (pid == tgid)
2986 return;
2987
2988 name.name = buf;
2989 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2990 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2991 if (!leader)
2992 goto out;
2993
2994 name.name = "task";
2995 name.len = strlen(name.name);
2996 dir = d_hash_and_lookup(leader, &name);
2997 if (!dir)
2998 goto out_put_leader;
2999
3000 name.name = buf;
3001 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3002 dentry = d_hash_and_lookup(dir, &name);
3003 if (dentry) {
3004 d_invalidate(dentry);
3005 dput(dentry);
3006 }
3007
3008 dput(dir);
3009out_put_leader:
3010 dput(leader);
3011out:
3012 return;
3013}
3014
3015/**
3016 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3017 * @task: task that should be flushed.
3018 *
3019 * When flushing dentries from proc, one needs to flush them from global
3020 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3021 * in. This call is supposed to do all of this job.
3022 *
3023 * Looks in the dcache for
3024 * /proc/@pid
3025 * /proc/@tgid/task/@pid
3026 * if either directory is present flushes it and all of it'ts children
3027 * from the dcache.
3028 *
3029 * It is safe and reasonable to cache /proc entries for a task until
3030 * that task exits. After that they just clog up the dcache with
3031 * useless entries, possibly causing useful dcache entries to be
3032 * flushed instead. This routine is proved to flush those useless
3033 * dcache entries at process exit time.
3034 *
3035 * NOTE: This routine is just an optimization so it does not guarantee
3036 * that no dcache entries will exist at process exit time it
3037 * just makes it very unlikely that any will persist.
3038 */
3039
3040void proc_flush_task(struct task_struct *task)
3041{
3042 int i;
3043 struct pid *pid, *tgid;
3044 struct upid *upid;
3045
3046 pid = task_pid(task);
3047 tgid = task_tgid(task);
3048
3049 for (i = 0; i <= pid->level; i++) {
3050 upid = &pid->numbers[i];
3051 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3052 tgid->numbers[i].nr);
3053 }
3054}
3055
3056static int proc_pid_instantiate(struct inode *dir,
3057 struct dentry * dentry,
3058 struct task_struct *task, const void *ptr)
3059{
3060 struct inode *inode;
3061
3062 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3063 if (!inode)
3064 goto out;
3065
3066 inode->i_op = &proc_tgid_base_inode_operations;
3067 inode->i_fop = &proc_tgid_base_operations;
3068 inode->i_flags|=S_IMMUTABLE;
3069
3070 set_nlink(inode, nlink_tgid);
3071
3072 d_set_d_op(dentry, &pid_dentry_operations);
3073
3074 d_add(dentry, inode);
3075 /* Close the race of the process dying before we return the dentry */
3076 if (pid_revalidate(dentry, 0))
3077 return 0;
3078out:
3079 return -ENOENT;
3080}
3081
3082struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3083{
3084 int result = -ENOENT;
3085 struct task_struct *task;
3086 unsigned tgid;
3087 struct pid_namespace *ns;
3088
3089 tgid = name_to_int(&dentry->d_name);
3090 if (tgid == ~0U)
3091 goto out;
3092
3093 ns = dentry->d_sb->s_fs_info;
3094 rcu_read_lock();
3095 task = find_task_by_pid_ns(tgid, ns);
3096 if (task)
3097 get_task_struct(task);
3098 rcu_read_unlock();
3099 if (!task)
3100 goto out;
3101
3102 result = proc_pid_instantiate(dir, dentry, task, NULL);
3103 put_task_struct(task);
3104out:
3105 return ERR_PTR(result);
3106}
3107
3108/*
3109 * Find the first task with tgid >= tgid
3110 *
3111 */
3112struct tgid_iter {
3113 unsigned int tgid;
3114 struct task_struct *task;
3115};
3116static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3117{
3118 struct pid *pid;
3119
3120 if (iter.task)
3121 put_task_struct(iter.task);
3122 rcu_read_lock();
3123retry:
3124 iter.task = NULL;
3125 pid = find_ge_pid(iter.tgid, ns);
3126 if (pid) {
3127 iter.tgid = pid_nr_ns(pid, ns);
3128 iter.task = pid_task(pid, PIDTYPE_PID);
3129 /* What we to know is if the pid we have find is the
3130 * pid of a thread_group_leader. Testing for task
3131 * being a thread_group_leader is the obvious thing
3132 * todo but there is a window when it fails, due to
3133 * the pid transfer logic in de_thread.
3134 *
3135 * So we perform the straight forward test of seeing
3136 * if the pid we have found is the pid of a thread
3137 * group leader, and don't worry if the task we have
3138 * found doesn't happen to be a thread group leader.
3139 * As we don't care in the case of readdir.
3140 */
3141 if (!iter.task || !has_group_leader_pid(iter.task)) {
3142 iter.tgid += 1;
3143 goto retry;
3144 }
3145 get_task_struct(iter.task);
3146 }
3147 rcu_read_unlock();
3148 return iter;
3149}
3150
3151#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3152
3153/* for the /proc/ directory itself, after non-process stuff has been done */
3154int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3155{
3156 struct tgid_iter iter;
3157 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3158 loff_t pos = ctx->pos;
3159
3160 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3161 return 0;
3162
3163 if (pos == TGID_OFFSET - 2) {
3164 struct inode *inode = d_inode(ns->proc_self);
3165 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3166 return 0;
3167 ctx->pos = pos = pos + 1;
3168 }
3169 if (pos == TGID_OFFSET - 1) {
3170 struct inode *inode = d_inode(ns->proc_thread_self);
3171 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3172 return 0;
3173 ctx->pos = pos = pos + 1;
3174 }
3175 iter.tgid = pos - TGID_OFFSET;
3176 iter.task = NULL;
3177 for (iter = next_tgid(ns, iter);
3178 iter.task;
3179 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3180 char name[PROC_NUMBUF];
3181 int len;
3182
3183 cond_resched();
3184 if (!has_pid_permissions(ns, iter.task, 2))
3185 continue;
3186
3187 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3188 ctx->pos = iter.tgid + TGID_OFFSET;
3189 if (!proc_fill_cache(file, ctx, name, len,
3190 proc_pid_instantiate, iter.task, NULL)) {
3191 put_task_struct(iter.task);
3192 return 0;
3193 }
3194 }
3195 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3196 return 0;
3197}
3198
3199/*
3200 * proc_tid_comm_permission is a special permission function exclusively
3201 * used for the node /proc/<pid>/task/<tid>/comm.
3202 * It bypasses generic permission checks in the case where a task of the same
3203 * task group attempts to access the node.
3204 * The rationale behind this is that glibc and bionic access this node for
3205 * cross thread naming (pthread_set/getname_np(!self)). However, if
3206 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3207 * which locks out the cross thread naming implementation.
3208 * This function makes sure that the node is always accessible for members of
3209 * same thread group.
3210 */
3211static int proc_tid_comm_permission(struct inode *inode, int mask)
3212{
3213 bool is_same_tgroup;
3214 struct task_struct *task;
3215
3216 task = get_proc_task(inode);
3217 if (!task)
3218 return -ESRCH;
3219 is_same_tgroup = same_thread_group(current, task);
3220 put_task_struct(task);
3221
3222 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3223 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3224 * read or written by the members of the corresponding
3225 * thread group.
3226 */
3227 return 0;
3228 }
3229
3230 return generic_permission(inode, mask);
3231}
3232
3233static const struct inode_operations proc_tid_comm_inode_operations = {
3234 .permission = proc_tid_comm_permission,
3235};
3236
3237/*
3238 * Tasks
3239 */
3240static const struct pid_entry tid_base_stuff[] = {
3241 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3242 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3243 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3244#ifdef CONFIG_NET
3245 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3246#endif
3247 REG("environ", S_IRUSR, proc_environ_operations),
3248 REG("auxv", S_IRUSR, proc_auxv_operations),
3249 ONE("status", S_IRUGO, proc_pid_status),
3250 ONE("personality", S_IRUSR, proc_pid_personality),
3251 ONE("limits", S_IRUGO, proc_pid_limits),
3252#ifdef CONFIG_SCHED_DEBUG
3253 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3254#endif
3255 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3256 &proc_tid_comm_inode_operations,
3257 &proc_pid_set_comm_operations, {}),
3258#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3259 ONE("syscall", S_IRUSR, proc_pid_syscall),
3260#endif
3261 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3262 ONE("stat", S_IRUGO, proc_tid_stat),
3263 ONE("statm", S_IRUGO, proc_pid_statm),
3264 REG("maps", S_IRUGO, proc_tid_maps_operations),
3265#ifdef CONFIG_PROC_CHILDREN
3266 REG("children", S_IRUGO, proc_tid_children_operations),
3267#endif
3268#ifdef CONFIG_NUMA
3269 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3270#endif
3271 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3272 LNK("cwd", proc_cwd_link),
3273 LNK("root", proc_root_link),
3274 LNK("exe", proc_exe_link),
3275 REG("mounts", S_IRUGO, proc_mounts_operations),
3276 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3277#ifdef CONFIG_PROC_PAGE_MONITOR
3278 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3279 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3280 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3281#endif
3282#ifdef CONFIG_SECURITY
3283 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3284#endif
3285#ifdef CONFIG_KALLSYMS
3286 ONE("wchan", S_IRUGO, proc_pid_wchan),
3287#endif
3288#ifdef CONFIG_STACKTRACE
3289 ONE("stack", S_IRUSR, proc_pid_stack),
3290#endif
3291#ifdef CONFIG_SCHED_INFO
3292 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3293#endif
3294#ifdef CONFIG_LATENCYTOP
3295 REG("latency", S_IRUGO, proc_lstats_operations),
3296#endif
3297#ifdef CONFIG_PROC_PID_CPUSET
3298 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3299#endif
3300#ifdef CONFIG_CGROUPS
3301 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3302#endif
3303 ONE("oom_score", S_IRUGO, proc_oom_score),
3304 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3305 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3306#ifdef CONFIG_AUDITSYSCALL
3307 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3308 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3309#endif
3310#ifdef CONFIG_FAULT_INJECTION
3311 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3312#endif
3313#ifdef CONFIG_TASK_IO_ACCOUNTING
3314 ONE("io", S_IRUSR, proc_tid_io_accounting),
3315#endif
3316#ifdef CONFIG_HARDWALL
3317 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3318#endif
3319#ifdef CONFIG_USER_NS
3320 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3321 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3322 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3323 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3324#endif
3325};
3326
3327static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3328{
3329 return proc_pident_readdir(file, ctx,
3330 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3331}
3332
3333static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3334{
3335 return proc_pident_lookup(dir, dentry,
3336 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3337}
3338
3339static const struct file_operations proc_tid_base_operations = {
3340 .read = generic_read_dir,
3341 .iterate_shared = proc_tid_base_readdir,
3342 .llseek = generic_file_llseek,
3343};
3344
3345static const struct inode_operations proc_tid_base_inode_operations = {
3346 .lookup = proc_tid_base_lookup,
3347 .getattr = pid_getattr,
3348 .setattr = proc_setattr,
3349};
3350
3351static int proc_task_instantiate(struct inode *dir,
3352 struct dentry *dentry, struct task_struct *task, const void *ptr)
3353{
3354 struct inode *inode;
3355 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3356
3357 if (!inode)
3358 goto out;
3359 inode->i_op = &proc_tid_base_inode_operations;
3360 inode->i_fop = &proc_tid_base_operations;
3361 inode->i_flags|=S_IMMUTABLE;
3362
3363 set_nlink(inode, nlink_tid);
3364
3365 d_set_d_op(dentry, &pid_dentry_operations);
3366
3367 d_add(dentry, inode);
3368 /* Close the race of the process dying before we return the dentry */
3369 if (pid_revalidate(dentry, 0))
3370 return 0;
3371out:
3372 return -ENOENT;
3373}
3374
3375static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3376{
3377 int result = -ENOENT;
3378 struct task_struct *task;
3379 struct task_struct *leader = get_proc_task(dir);
3380 unsigned tid;
3381 struct pid_namespace *ns;
3382
3383 if (!leader)
3384 goto out_no_task;
3385
3386 tid = name_to_int(&dentry->d_name);
3387 if (tid == ~0U)
3388 goto out;
3389
3390 ns = dentry->d_sb->s_fs_info;
3391 rcu_read_lock();
3392 task = find_task_by_pid_ns(tid, ns);
3393 if (task)
3394 get_task_struct(task);
3395 rcu_read_unlock();
3396 if (!task)
3397 goto out;
3398 if (!same_thread_group(leader, task))
3399 goto out_drop_task;
3400
3401 result = proc_task_instantiate(dir, dentry, task, NULL);
3402out_drop_task:
3403 put_task_struct(task);
3404out:
3405 put_task_struct(leader);
3406out_no_task:
3407 return ERR_PTR(result);
3408}
3409
3410/*
3411 * Find the first tid of a thread group to return to user space.
3412 *
3413 * Usually this is just the thread group leader, but if the users
3414 * buffer was too small or there was a seek into the middle of the
3415 * directory we have more work todo.
3416 *
3417 * In the case of a short read we start with find_task_by_pid.
3418 *
3419 * In the case of a seek we start with the leader and walk nr
3420 * threads past it.
3421 */
3422static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3423 struct pid_namespace *ns)
3424{
3425 struct task_struct *pos, *task;
3426 unsigned long nr = f_pos;
3427
3428 if (nr != f_pos) /* 32bit overflow? */
3429 return NULL;
3430
3431 rcu_read_lock();
3432 task = pid_task(pid, PIDTYPE_PID);
3433 if (!task)
3434 goto fail;
3435
3436 /* Attempt to start with the tid of a thread */
3437 if (tid && nr) {
3438 pos = find_task_by_pid_ns(tid, ns);
3439 if (pos && same_thread_group(pos, task))
3440 goto found;
3441 }
3442
3443 /* If nr exceeds the number of threads there is nothing todo */
3444 if (nr >= get_nr_threads(task))
3445 goto fail;
3446
3447 /* If we haven't found our starting place yet start
3448 * with the leader and walk nr threads forward.
3449 */
3450 pos = task = task->group_leader;
3451 do {
3452 if (!nr--)
3453 goto found;
3454 } while_each_thread(task, pos);
3455fail:
3456 pos = NULL;
3457 goto out;
3458found:
3459 get_task_struct(pos);
3460out:
3461 rcu_read_unlock();
3462 return pos;
3463}
3464
3465/*
3466 * Find the next thread in the thread list.
3467 * Return NULL if there is an error or no next thread.
3468 *
3469 * The reference to the input task_struct is released.
3470 */
3471static struct task_struct *next_tid(struct task_struct *start)
3472{
3473 struct task_struct *pos = NULL;
3474 rcu_read_lock();
3475 if (pid_alive(start)) {
3476 pos = next_thread(start);
3477 if (thread_group_leader(pos))
3478 pos = NULL;
3479 else
3480 get_task_struct(pos);
3481 }
3482 rcu_read_unlock();
3483 put_task_struct(start);
3484 return pos;
3485}
3486
3487/* for the /proc/TGID/task/ directories */
3488static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3489{
3490 struct inode *inode = file_inode(file);
3491 struct task_struct *task;
3492 struct pid_namespace *ns;
3493 int tid;
3494
3495 if (proc_inode_is_dead(inode))
3496 return -ENOENT;
3497
3498 if (!dir_emit_dots(file, ctx))
3499 return 0;
3500
3501 /* f_version caches the tgid value that the last readdir call couldn't
3502 * return. lseek aka telldir automagically resets f_version to 0.
3503 */
3504 ns = inode->i_sb->s_fs_info;
3505 tid = (int)file->f_version;
3506 file->f_version = 0;
3507 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3508 task;
3509 task = next_tid(task), ctx->pos++) {
3510 char name[PROC_NUMBUF];
3511 int len;
3512 tid = task_pid_nr_ns(task, ns);
3513 len = snprintf(name, sizeof(name), "%d", tid);
3514 if (!proc_fill_cache(file, ctx, name, len,
3515 proc_task_instantiate, task, NULL)) {
3516 /* returning this tgid failed, save it as the first
3517 * pid for the next readir call */
3518 file->f_version = (u64)tid;
3519 put_task_struct(task);
3520 break;
3521 }
3522 }
3523
3524 return 0;
3525}
3526
3527static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3528{
3529 struct inode *inode = d_inode(dentry);
3530 struct task_struct *p = get_proc_task(inode);
3531 generic_fillattr(inode, stat);
3532
3533 if (p) {
3534 stat->nlink += get_nr_threads(p);
3535 put_task_struct(p);
3536 }
3537
3538 return 0;
3539}
3540
3541static const struct inode_operations proc_task_inode_operations = {
3542 .lookup = proc_task_lookup,
3543 .getattr = proc_task_getattr,
3544 .setattr = proc_setattr,
3545 .permission = proc_pid_permission,
3546};
3547
3548static const struct file_operations proc_task_operations = {
3549 .read = generic_read_dir,
3550 .iterate_shared = proc_task_readdir,
3551 .llseek = generic_file_llseek,
3552};
3553
3554void __init set_proc_pid_nlink(void)
3555{
3556 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3557 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3558}